# bok-ai-lab-20250509-research

full research [here](https://docs.google.com/document/d/1xYHACsxN9_8Bdrs_me7IzIty0ve7hvWUFh1aDcXyh1g/edit?tab=t.0)

# Identity, Replication, and Authenticity: A Historical and Cultural Exploration

## **Introduction**

Anxieties about identity and authenticity in the face of copies or “doubles” are as old as human culture. Across time and civilizations, people have grappled with the implications of seeing themselves or their world replicated: from ancient myths of spirit doubles to medieval debates over holy icons, and from the invention of photography to today’s AI-generated deepfakes. This report traces the evolving **cultural, philosophical, theological, folkloric, psychological, and technological** perspectives on **identity, replication, and authenticity**. We will survey how different eras and societies – Western and non-Western alike – have intuitively feared or theorized about **copies of people or reality**, and how each new medium of reproduction (idols, images, photography, film, and now AI clones) transforms those concerns. By comparing global traditions (e.g. Plato’s philosophy of mimesis, East Asian spirit-doubles, Indigenous shapeshifter lore, etc.) and key modern thinkers (Benjamin, Sontag, Lacan, etc.), we will see both **parallels and contrasts** in how authenticity and selfhood are understood. Finally, we discuss how these historical frameworks might shift as **AI cloning technologies** challenge the boundaries between real and replica in unprecedented ways.

## **Ancient Perspectives: Myths, Philosophy, and Spirit Doubles**

**Mythology and early cultures** often featured “doubles” that raised questions about the soul and reality. In **ancient Egypt**, for example, the concept of the *ka* was essentially a tangible spirit twin of a person – a duplicate of their soul. An Egyptian myth even tells of using a ka of Helen of Troy to mislead Paris, indicating a belief that a replica with the same memories and feelings as the original could influence real events. Similarly, Zoroastrian cosmology envisioned twin spirits (the good Ahura Mazda and evil Angra Mainyu) as co-eternal doubles embodying cosmic duality.

In the **Western philosophical canon**, Plato articulated one of the earliest skeptical views of artistic “copies.” In *The Republic*, he argues that all art is imitation (*mimesis*) and inherently a step removed from truth. The true reality is the realm of Forms (ideal essences); physical objects are mere imperfect copies of those Forms, and artists then copy those copies by making images. Thus, **imitative art** (like painting or poetry) only captures appearances, “far below truth,” and risks misleading viewers. Plato even likens a painter’s work to an **“imitation of appearance”**, noting that a painting of a couch, for instance, just reproduces how a couch looks from one angle, missing the full truth of the Form of “couch”. This inferiority of copies led Plato to **distrust art**: he warned that imitations can corrupt the soul and should be banned from the ideal city. In essence, for Plato, a copy (an image or mimicked persona) lacks the authentic essence of the original – a philosophical anxiety about replication that echoes through the ages.

Ancient thinkers also recognized a **psychological uncanniness** in encountering one’s double. The Greek myth of Narcissus – a youth entranced by his own reflection – might be seen as an early commentary on the seductive but illusory nature of one’s mirrored image. And multiple cultures held that seeing one’s exact double (sometimes called a *doppelgänger* in modern terms) is an omen of death or bad luck. The idea that a person’s **“second self”** could portend danger suggests an ingrained fear: if your identity can appear independently, does it signify the loss or end of the “real” you? In short, ancient worldviews already contained the seeds of a dilemma: **if a person or sacred essence is copied, is something vital lost or threatened?**

## **Religious Iconography and Medieval Debates on Images**

With the rise of organized religion, the question of **representation versus reality** became a theological battleground. Many faith traditions wrestled with whether creating an image of a person or deity was an act of reverence, or a dangerous illusion and idolatry.

In the **Abrahamic religions**, the **Second Commandment’s** prohibition of “graven images” set the tone for millennia of suspicion toward idols. The fear was that worshippers might mistake an **image or statue** for the divine itself, thus worshipping a lifeless copy instead of the true God. This tension climaxed in events like the **Byzantine Iconoclasm** (8th–9th centuries CE), when imperial authorities in Constantinople banned and destroyed religious icons. The iconoclasts’ arguments reveal a profound concern with **authenticity of holy images**. They declared that any painted or carved image of Christ or a saint is by nature **“lifeless…empty of spirit and life”**, and thus **incapable of truly representing the divine prototype**. Only something sharing the *actual substance* of Christ – such as the Eucharist – could be considered a true image of Him. In their view, painting Christ’s human form but not His divine nature either split His identity (a theological error) or confused it; any **material depiction** risked either heresy or misleading idolatry. For iconoclast thinkers, **no replica in wood or paint could capture the authentic agency or holiness of its subject**, making such images inherently deceptive. This was not merely an abstract argument – it was fueled by a visceral fear that believers might **transfer their devotion to an inanimate copy**, thus “worshipping the creature instead of the Creator”.

On the other side, the defenders of icons (the iconodules) developed sophisticated theories for why images could be **venerated but not worshipped** – effectively attempting to delineate how an image might **participate in the identity** of its subject without equating to it. St. John of Damascus famously argued that because God had become visible in the flesh (through the Incarnation of Christ), it was permissible to paint Christ’s human form; the honor given to an image, he said, passes to its prototype. This argument acknowledged the representational nature of images while insisting the **“authenticity” resides in the prototype**, not the wood and paint. Ultimately, the **iconophile view prevailed** in Eastern Orthodoxy (the Second Council of Nicaea, 787 CE, approved holy images), but not without leaving a legacy of caution: images must be treated carefully to avoid confusion between **symbol and reality**.

A similar wave of **image anxiety** swept parts of Europe during the **Protestant Reformation** (16th century). Protestants smashed statues and whitewashed frescoes in churches, fearing that Catholics had imbued these images with supernatural powers or undue reverence. Again, the core worry was that the **replica (the statue, the painting) could usurp the devotion** due only to the original (God or saints). This iconoclastic impulse underscores a recurring theme: **when a new form of reproduction or representation emerges (be it idol, icon, or printed image), society questions whether the image is “faithful” or a dangerous fake.** Who or what animates the copy? Does it have *agency* or spirit, or is it a soulless shell?

Outside Europe, many cultures had long-standing beliefs linking images to the soul’s well-being. In parts of Africa, Oceania, and the Americas, indigenous peoples were often reluctant to be sketched or photographed by early colonial visitors, due to a belief that **capturing one’s image could steal a part of one’s soul**. This belief – that a representation carries spiritual essence – actually mirrors the logic of both the iconophiles (who thought an icon shares in the saint’s holiness) and their opposite (indigenous fear that an unauthorized image-snatch robs personal power). In both cases, an **authentic connection between image and original** is assumed, and that raises the stakes: creating or possessing someone’s image becomes an action with spiritual or moral consequences.

In summary, medieval and early-modern debates on images were not just about art – they were about **identity and presence**. If an image of a holy figure is treated as *more than mere paint*, then either it is worryingly **powerful** (capable of “housing” the presence of its subject) or worryingly **deceptive** (a lifeless impostor mistaken for the real). Both possibilities fueled cultural anxiety around replication: an imitation might either **usurp the original’s role** or **mislead people into false relations** with an object. These theological and cultural intuitions set the stage for later secular worries about photographs, films, and avatars: in all cases, we ask, *what happens to authenticity and trust when we deal with a copy*?

## **Folklore of Doubles and Shapeshifters: Fear of the Mimic**

Beyond formal philosophy or religion, **folklores worldwide** have expressed visceral fears of **doppelgängers, changelings, and shapeshifters**. Such tales personify the anxiety that something which looks or acts just like a human might hide an inhuman or malicious truth – or that one’s own identity could be stolen or imitated by something else.

In European folklore, the **doppelgänger** (a German term meaning “double-goer”) is a spectral double of a living person, often seen as an omen of misfortune or death. These doubles are typically described as **wraith-like replicas** that **cast no shadow**, a detail emphasizing their unreality despite their perfect mimicry. Seeing one’s own doppelgänger was believed to foreshadow one’s imminent death. If a friend or relative saw your double, it could mean you were in peril. Folklore also warned that these doubles might **speak to you or advise you**, but their counsel would be deceptive or harmful, luring the victim into disaster. People were cautioned to **avoid communicating with their own double at all costs**, treating it as fundamentally *Other* despite its familiar appearance. Such stories reveal a deep psychological unease: an **encounter with oneself** (in image or person) becomes fatal. It is as if nature abhors two instances of the same identity coexisting – a theme we will see echoed in modern sci-fi tropes about meeting one’s clone.

Northern Europe had the **changeling** myth, in which fairies or trolls steal a human baby and leave behind their own offspring in its place, disguised to look human. In Orkney folklore, for example, small fairy creatures called *trows* would kidnap healthy human infants and replace them with sickly **replicas known as changelings**, who perfectly resembled the original baby. The panic here was very real for medieval parents – a child who suddenly behaved strangely or fell ill might be suspected not to be their “real” child at all, but an impostor. This speaks to a primal fear of **mimicry as deception**: the idea that an **imposter could seamlessly take the place of a loved one**. The changeling legend offered a supernatural explanation for developmental disorders or illness, but in doing so, it kept alive the notion that **a copy could pass for the original**, fooling even one’s closest family. To this day, the changeling has its sci-fi/horror descendants in stories of alien body-snatchers or evil twins.

Many **Indigenous cultures** likewise speak of **shape-shifters** that blur the line between human and other beings. The Navajo **Skinwalker** legend is a prominent example. A Skinwalker (*yee naaldlooshii* in Navajo) is typically a malevolent witch who has gained the power to transform into animals – or even **assume the form of other people**. This **ability to don any shape or voice** makes the Skinwalker especially feared; it can infiltrate communities in disguise or lure victims by appearing as someone they trust. Similar shape-shifter figures appear in other Native American traditions (Pueblo, Apache, Hopi, etc.) and often share the detail that these beings were once human (usually a corrupted medicine person) who **chose an evil path and thus gained dark transformative powers**. The idea that a human could **lose their true identity and roam about in others’ skins** is inherently unsettling – it combines fear of **identity loss** (the witch forsakes their humanity) with fear of **duplicitous appearances** (you cannot be sure that the creature wearing your neighbor’s face is actually them). It’s a caution that things (or people) are not always what they appear. Interestingly, even in these indigenous stories, the language of *stolen skin or form* hints that the authentic self (the rightful owner of the skin) is displaced by a malicious copy. This resonates with the modern dread of having one’s **digital identity “hijacked” by a deepfake** – an ancient archetype repurposed for new technology.

Non-Western folklore has many other notions of doubles: for instance, Japanese legend speaks of the *ikiryō*, a living ghost that can leave a person’s body and haunt others (essentially one’s double acting independently), and Chinese folklore includes the idea of **hun and po** (multiple souls where one might wander – creating the effect of a double appearing elsewhere). Norse mythology had the *vardøger*, a spirit that precedes a person, arriving moments or hours before the real individual and performing their actions in advance. Unlike the doppelgänger, the *vardøger* was often considered less sinister – almost a time-shifted double that is more playful than deadly. Nonetheless, it underscores a widespread intuition: if **a person’s likeness or essence can operate separately from the person**, it challenges our sense of a unified, singular identity.

These folk motifs of **doubles, imposters, and replicas** show that **fear of imitation** is not merely a product of modern technology – it is deeply human. We seem predisposed to imagine and **mythologize the possibility of being copied or mimicked**, and to view it with suspicion. The copy might be malevolent (the double with evil intent, the fairy changeling), or it might herald existential doom (an omen of death, or one’s soul leaving one’s body). Either way, the consistent message is that **one’s identity is unique and sacrosanct, and any duplication of it is a portentous event**. Culturally, this laid fertile ground for later reactions when actual technologies *did* emerge that could create doubles of our faces, voices, or actions.

## **Mirrors and the Self: Psychological and Psychoanalytic Insights**

While folklore externalized the double as a separate being, psychology and psychoanalysis have explored how the idea of a “double” is integral to one’s **own sense of self**. In early development, imitation and reflection play crucial roles – sometimes comforting, sometimes disorienting.

Consider a simple **childhood scenario**: one child starts copying another’s every word or gesture in a playful *mimicry game*. At first, this imitation might be funny, but very quickly children (and adults) become annoyed or unnerved by a perfect copycat. There’s a reason “Stop copying me\!” is a common playground complaint. Psychologically, being imitated can feel like an **encroachment on one’s identity** – as if the mimic is stealing one’s role or mocking one’s individuality. Developmental studies note that children learn by imitation from their earliest months, and indeed **mirroring behavior** by parents is crucial for infants. Yet, as self-awareness grows, so does a child’s desire to assert “I am me, not you.” A too-perfect mirror held up to our actions is uncomfortable because it blurs the boundary between self and other. This everyday phenomenon hints that humans have an innate tension with replication: we **depend on imitation** to learn and connect, but we also resent or fear it when it threatens our uniqueness or control.

The French psychoanalyst **Jacques Lacan** turned the mirror metaphor into a foundational theory of identity. In his account of the **Mirror Stage**, an infant (around 6–18 months old) first recognizes their reflection as themselves. This moment is joyous but also profoundly formative: the baby identifies with the **image** in the mirror – a seemingly coherent, whole person – even though the baby’s own internal experience is one of fragmentation and lack of coordination (an infant’s motor skills are minimal). The mirror image represents an **ideal “I” (the Ego)**: unified and in control, in contrast to the child’s felt reality of disjointed sensations. According to Lacan, the **child’s ego is essentially a construct based on this external image**, a **mental duplicate** it has taken to be itself. Crucially, this means our sense of a unified self is, from the start, bound up with an **illusion or outside perspective** – literally an **external copy** of ourselves (even if a truthful one in the mirror). Lacan notes this has an “alienating” effect: *“the image actually comes to take the place of the self… the sense of a unified self is acquired at the price of this self being an Other”*. In other words, we become whole by identifying with a **double of ourselves**. This paradox – *to be oneself, one must become like one’s mirror-double* – suggests that a kind of *internalized replication* underpins identity. Little wonder that later encounters with doubles (photographs, recordings, avatars) feel uncanny; they **resonate with the very origin of our self-image**.

The notion of the “fragmented self” that is patched together by images continued into modern psychology and literature. Psychoanalytic thinkers like **Sigmund Freud** examined why **doubles evoke eerie feelings**. In his 1919 essay *“The Uncanny”* (*Das Unheimliche*), Freud observed that seeing one’s own image or likeness unexpectedly (in a mirror at a wrong time, or in another person) can produce a profound unease. He linked this to repressed experiences and primitive beliefs. One striking idea Freud references is that an uncanny effect often arises when there is **“doubt whether something is alive or not, or whether an object might be animate”**. For example, a lifelike wax figure, or a doll that seems to move, blurs the line between living and inanimate in a way that spooks us – we momentarily can’t tell if it has a soul. He cites psychologist Jentsch’s example of automata and how a moving doll can be unsettling until one determines it’s just a mechanism. Freud then goes further to discuss the *double* specifically: he theorizes that primitive man took the double (reflections, shadows, twins) as an **“insurance of immortality”**, a safeguard against death (the thinking being: if there’s another of me, I live on). But as the modern ego developed, that notion was abandoned, and the double became a **harbinger of death** instead, an object of fear and loathing. In short, what was once comforting (an external self) turned **uncanny** once we no longer needed or believed in it. The double now represents the return of repressed self-love and the reality of one’s mortality – meeting your double hints that perhaps **you are already a ghost** or that your uniqueness is lost.

This psychoanalytic perspective helps explain why the **notion of being copied** – whether by a twin, a portrait, or a clone – can be so deeply unsettling. It touches on ancient existential questions: *Do I have a soul that is uniquely mine? Can it be split or duplicated?* If an inanimate thing looks or acts like me, is it just a clever fake, or does it somehow **borrow my essence** (and if so, do I lose some)? Freud’s contemporary, the writer E.T.A. Hoffmann, explored these questions in fiction – e.g. in *The Sandman*, a man falls in love with an automaton that looks human, and when he discovers the truth, it drives him mad. The tale dramatizes the collision of desire with doubt about authenticity: if your beloved isn’t “real,” what does that say about your own identity or sanity?

Modern psychology also suggests an **evolutionary basis** for our wariness of almost-human fakes. The concept of the **“uncanny valley”**, coined by roboticist Masahiro Mori, holds that as a robot or CGI figure approaches human realism, people’s comfort level rises – but just before perfect realism, there’s a sharp drop into revulsion or eeriness. One hypothesis is that our brains have finely tuned **face and behavior recognition systems** (since we are social animals), and when those systems get conflicting information – *most* cues say “human” but a few say “not quite” – it triggers a warning signal. We become hyper-alert to the possibility that something is **“wrong” with this almost-human entity**. This could be a byproduct of our general pattern-recognition, or it might have specific adaptive value – for example, one theory suggests it helped early humans avoid **diseased individuals or corpses** (which might look human but with subtly “off” features indicating sickness or death). Another theory posits it was useful to detect **dangerous impostors** – perhaps even individuals from an enemy tribe or another human species – where recognizing “this is not one of us” quickly was life-saving. While speculative, these ideas converge on a point: our minds are **acutely sensitive to authenticity in appearance and behavior**, and we experience visceral discomfort at **flawed copies**. We likely evolved to trust what’s real and to be cautious or investigative when something seems **“too perfect” or “just an imitation.”** In the deep past that might have meant differentiating a human from a predator in disguise, or a healthy person from a plague victim. Today, it might mean scrutinizing a video that *looks* exactly like your friend but somehow the voice cadence is off – a possible deepfake.

In sum, psychology teaches us that **our very self-concept is born from engaging with our own image (a kind of self-replication)**, and that we carry within us some innate alarms regarding imitations of humans. We love seeing ourselves reflected – it’s how we grow – but we are also wired to notice the cracks in a reflection. This duality in the psyche makes the issue of identity copies especially charged: on one hand, we are *drawn* to them (think of our fascination with identical twins, or with avatars that represent us), and on the other hand, we are *disturbed* by them (twins have often been seen as eerie in lore, and many find hyper-realistic androids creepy). As we proceed to the era of mechanical and digital reproduction, these psychological underpinnings will help explain society’s mixed reactions to new “mirrors” of ourselves.

## **The Age of Mechanical Reproduction: Photography and Film**

The 19th and 20th centuries introduced technologies that could replicate reality with astonishing fidelity: first **photography**, then audio recording and **film**. For the first time in history, humans could **create nearly exact visual (and later auditory) copies of scenes and people**. These inventions forced a reevaluation of authenticity and aura, provoking both enthusiasm and anxiety.

When **photography** emerged in the mid-1800s, people were thunderstruck by its realism. A camera could **freeze a moment in time**, seemingly **preserving reality itself** in a way no painting could. As essayist Susan Sontag famously put it, *“Photographs really are experience captured”*. A photograph feels like a **piece of the world**, not just a representation. Early photographers and observers were quick to note this **magical quality** – that a photo was not an interpretation (like a drawing), but an actual imprint or **index of reality** (light from the subject literally causes a chemical change on the plate). Sontag also remarked, *“To photograph is to appropriate the thing photographed”*, an act of claiming or miniaturizing a part of reality. With a camera, one could *acquire* the images of loved ones, famous sites, even fleeting news events, effectively **collecting pieces of the world**. This power to replicate brought immense excitement – photography was embraced for documentation, art, science – but it also brought new forms of **cultural anxiety**.

One common fear in the 19th century was almost a **modern retelling of the soul-stealing superstition**. Across continents, some individuals (especially from communities new to the technology) believed that being photographed might **steal one’s soul or vital essence**. This belief wasn’t universal, but it was “present to some extent in various cultures” as a broader **“cultural belief in the power of images… to preserve a person’s essence”**. For example, certain Native American and African traditions held that a photo could capture part of your spirit, weakening you. Why would this idea gain traction? Perhaps because a **photo looked so uncannily alive**, more so than any painting, that it seemed the person was actually **imprinted into the paper**. A static image might not walk or talk, but those staring eyes and lifelike features suggested something of the person was truly in there – or at least, that the photographer now *possessed* something of them. In essence, photography revived questions about **agency and consent in representation**: is it **violating someone’s personhood** to take their likeness without permission? Does the *subject* lose some control or “aura” once their image can be circulated freely? These questions persist even now (think of the ethics of taking someone’s photo in public, or the unease of having your image online without consent) and have only intensified with deepfakes – but the 19th century already grappled with them in embryo.

Philosopher **Walter Benjamin**, in his 1936 essay *“The Work of Art in the Age of Mechanical Reproduction”*, offered a seminal analysis of what happens when art (and by extension, reality) becomes infinitely reproducible. Benjamin argued that traditionally, a work of art had an **“aura”** – a unique presence in time and space, tied to its authenticity. For instance, the **Mona Lisa** in person has an aura: it is the actual painting Leonardo touched, it has aged, it resides in the Louvre with a specific history. A photograph or copy of the Mona Lisa, however perfect, **lacks this aura** because it is missing that unique existence and history. In Benjamin’s words, *“Even the most perfect reproduction of a work of art is lacking in one element: its presence in time and space, its unique existence at the place where it happens to be”*. Mechanical reproduction – photography, film, printing – liberates images from their fixed context and makes them accessible to the masses, which Benjamin saw as democratically promising. But the trade-off is that the **authenticity** of the original (“the presence of the original is the prerequisite to the concept of authenticity”) gets lost in a sea of identical copies. In a reproduced image, one cannot tell where it came from or imbue it with the same reverence. With respect to **people**, consider what this means: a person’s photograph lacks the “aura” of the living person. It is missing the full **context of presence**, yet it circulates as if it were the person. This dilution of presence might underlie why older generations often found photographs of deceased loved ones eerie – the **person’s image outlived them** and could be seen anywhere, but their actual unique life was gone, highlighting an unsettling gap between **image and reality**.

Benjamin also noted that mechanical reproduction changed our **perception** fundamentally. The masses gained a “sense of the universal equality of things” – a reproduced image of a mountain or a face can be possessed by anyone – but also a new mindset that objects (or people) are somehow **less tied to a singular existence and more to a type**. In film (which Benjamin analyzed in depth), an actor’s performance is captured in many discrete takes and camera angles, then reassembled; the audience never experiences the actor’s **full presence**, only the edited copy. Some early film theorists and actors worried that this process **“robbed” performers of their aura**, turning them into phantoms on screen. Yet new stars (with manufactured personas) emerged beloved by millions – a paradox of **intimacy at a distance** that defines modern celebrity. We see and “know” actors through reproduced images, but the real person remains elsewhere. This new mode of relating to people via their mediated likeness arguably shifted our concept of identity: public figures now often have a curated **image identity** separate from their private self, and audiences accept consuming the image as a substitute for knowing the person. Here lies a subtle anxiety: **can an image-based identity take on a life of its own?** (Consider how early Hollywood stars sometimes struggled with fans treating them as if they *were* the characters they played on film – a confusion of representation and reality that upset some actors greatly.)

Photography also provoked discussions about **truth and deception**. On one hand, a photo was seen as objective proof (“the camera doesn’t lie” became a saying). On the other hand, people learned quickly that cameras **can lie** – through staging, selective framing, or later, through editing. In the 19th century, “spirit photography” became a fad: photographers produced images seemingly showing ghosts or auras around people, using double exposures and darkroom tricks. Many victorians were fooled or at least tantalized by these **faked spirit images**, which fed into spiritualist movements. The fact that such obvious manipulations gained traction shows how hungry people were to believe that the **photograph might truly capture invisible aspects of reality (like spirits or soul)** – a hope mingled with fear. If true, it meant the camera could reveal **things humans weren’t meant to see**; if false, it meant the camera could be a tool of **sophisticated deceit**. Either way, it underscored that new replication technology had outpaced society’s ability to fully trust what they saw.

By the early 20th century, as **cinema** rose, society had somewhat acclimated to photographs. Moving pictures, however, introduced **dynamic replicas** of reality. A film could show people doing things they never did (through editing or trick effects), yet the audience might not detect the manipulation. This gave rise not only to entertainment but also to propaganda and **performance anxieties**. Charlie Chaplin – one of the first global film stars – once quipped about the strangeness of **meeting fans**: they felt they knew him from the screen, but to Chaplin, these were strangers who only knew a **flickering copy** of his performances. The **disembodiment** of identity in film (and later TV) led to phenomena like people idolizing or even worshiping celebrities, akin to how religious icons functioned. Indeed, we started calling actors and pop singers “idols,” implicitly comparing mass-produced images of people to the devotional images of saints or gods. The parallel is telling: a society that thought it had moved beyond **idolatry** found itself inundated with **images of individuals to admire, emulate, or obsess over**. This raised new questions: How “real” is a celebrity’s public persona? Can it diverge completely from their private self and take on its **own reality in the public mind**? (History is full of tragic examples of stars who felt alienated from the very image that made them famous.)

Thinkers like Benjamin and later **Jean Baudrillard** (with his concept of *simulacra*) foresaw a trajectory where reproductions might eventually **precede and determine reality**, rather than just reflect it. Baudrillard argued that in the postmodern age, simulations and media images have become *hyperreal* – more real to people than physical reality itself. An example: a fictional character or a digital avatar can evoke genuine emotions and drive real economic or social activity, arguably **outweighing some flesh-and-blood interactions**. While Baudrillard wrote before deepfakes, his ideas uncannily predict a world where a politician’s **image or hologram** might matter more to voters than their physical presence – or where a completely virtual influencer could have millions of real fans. This collapse of the boundary between copy and original – the *simulacrum* overtaking the authentic – is exactly the kind of shift that modern replication technology threatens, and which earlier eras only **toyed with in stories or theoretical musings**.

Returning to photography and film: by mid-20th century, society largely accepted these media and established conventions around them (for example, people learned to pose for photographs, to distinguish movie magic from reality, etc.). But the underlying philosophical predicament remained. We see it articulated poignantly by Walter Benjamin: *“The presence of the original is the prerequisite to the concept of authenticity…* \[but\] *technical reproduction can put the copy of the original into situations which would be out of reach for the original itself”*. A person cannot be in two places at once – but their photograph or filmed image can. Once there are *copies of you* (even just static images), your **“presence” becomes dispersed and mediated**. We learned to leverage that (one can project influence far and wide through media), but also to fear it (one can lose control of one’s image). Notably, the law started catching up: copyright and right-of-publicity laws emerged to help people (especially public figures) control the reproduction of their likeness. These were early attempts to **mediate the tensions of identity replication** – recognizing that one’s image can be separate from oneself and needs its own protection to ensure authentic consent and context.

## **The AI Era: Voice Cloning, Deepfakes, and Digital Avatars**

In the 21st century, we have entered a new phase where **artificial intelligence** can create **indistinguishably accurate copies** of not just static appearance, but dynamic behavior. **Voice cloning** can produce an audio clip that sounds exactly like someone’s speech. **Deepfake technology** can generate videos of people doing or saying things they never did, with extremely high realism. And AI-driven avatars can interact in real-time, mimicking a person’s personality or style. These developments amplify old anxieties and introduce new ones, because they remove many remaining **signals of inauthenticity**, pushing the imitation ever closer to the original – and sometimes, even **surpassing the original’s abilities** (an AI version of a person could theoretically work 24/7, or be in many places at once, etc.). We are now confronting scenarios that previously were only philosophical thought experiments or sci-fi plots.

One immediate concern is a very practical one: **deception and trust**. If any audio or video can be faked, how do we trust what our eyes and ears perceive? Already, there have been instances of deepfake-driven fraud – for example, a company was swindled out of $25 million when criminals used an AI-generated video call of the CEO to give orders to an employee. Fake videos of public figures have spread misinformation. In 2022, the FBI even warned that deepfakes were being used in job interviews by impostors (imagine “interviewing” someone on a webcam who is actually an AI puppet). These incidents create a climate of **general distrust**: we begin to question *everything* not directly in front of us. Ironically, \*\*the closer the imitation gets to reality, the more it can erode our trust in **anything**. As one technology commentator put it, “The closer the imitation, the more brittle our trust becomes. When we can’t tell what’s real, we stop trusting anything”. In other words, a world rich in perfect copies might lead not to delight but to a kind of **paralysis of belief** – an insight chillingly relevant to our “post-truth” era.

With AI clones, there’s also the psychological effect on the **person being copied**. Research and interviews have found that people experience **stress, anxiety, and a sense of violation** when they discover their **likeness has been cloned or manipulated without consent**. A recent study highlighted terms like **“doppelgänger-phobia”** – the fear of one’s double – cropping up in the context of AI clones. This is essentially the old doppelgänger dread in a new avatar: knowing that *another you* (that you don’t control) is out there acting autonomously can be deeply unsettling. If someone makes an AI avatar of me that can answer questions in my voice and style, is that *me* in some sense? Do I bear responsibility for its statements? Could it damage my reputation or even develop a kind of independent persona? These questions are no longer hypothetical. For instance, celebrity voice clones have been used to generate offensive or false statements in that celebrity’s voice, causing public confusion or harm. In some cases, victims have had to *publicly deny* things “they” never actually said – essentially arguing against their own digital double.

This blurring of boundaries leads to what some call **identity fragmentation** in the digital realm. If versions of you proliferate (your social media profiles, your filtered selfies, an AI chatbot trained on your texts, a deepfake of you dancing on TikTok), one might ask: which *“you”* is authentic? Potentially, all and none. We are approaching a state where identity might be seen as a **distributed phenomenon** – part in the physical self, part in data – which challenges centuries of thinking of the “self” as a singular, embodied continuity. Lacan’s mirror stage, where the person saw an external reflection and said “that’s me,” is writ large: now we see *digital reflections* of ourselves everywhere. Will we adapt and come to include those in our self-concept (e.g., “I have my real self and my digital twin, together representing me”), or will we psychologically distance the digital versions as mere utilitarian tools? The answer may shape how comfortable future generations are with pervasive copying.

From a cultural standpoint, we can draw **parallels to earlier media** but also note key differences:

* **Like photography** in the 19th century, **deepfakes and AI clones** initially inspire both awe and fear. The awe is at the technical marvel (“it looks so real\!”), and the fear is about **malicious uses** and loss of control. Just as early photographers had to convince people they weren’t stealing souls, AI developers now often need to convince the public that not every use of deepfakes is nefarious – for example, there are beneficial uses like dubbing a movie into another language using the actor’s own voice clone, or creating a digital *you* to attend a meeting on your behalf (with permission). However, the **pace** and **scale** of AI’s spread is much faster than photography’s was, compressing the societal adjustment period.

* Unlike a single photograph, which is a fixed moment, AI clones can be **interactive and generative**. This means a person’s likeness can **produce new actions or speech** indefinitely. This is a major break from earlier replication: once a photograph was taken, it *froze* the subject. Now, a clone can *evolve*. This has led some to talk about “digital resurrection” – e.g., creating an avatar of a deceased person that can chat with you (trained on their writings or recordings). Some see this as comforting (a kind of continuation of the person’s identity), while others find it profoundly troubling (an **ersatz presence** that might interfere with the grieving process or even the legacy of the deceased). Either way, it forces us to revisit spiritual and philosophical questions: if an AI behaves *just like Grandpa would*, is some part of Grandpa “there”? Or is it a soulless imitation giving a dangerous illusion? The Victorians asking if a photo stole a soul would recognize this dilemma well.

* **Authority and authenticity measures** are being developed in response. Just as the watermark, the signature, and the certificate of authenticity arose in earlier eras, we now see efforts like cryptographic content authentication, AI deepfake detectors, and legal frameworks. For example, researchers are working on systems to **discern authentic human voices from cloned voices** by analyzing subtle biological markers in speech. Legislators in some jurisdictions are contemplating requiring explicit labeling of AI-generated media. These are modern echoes of past solutions – essentially, society attempts to **reassert a boundary** between real and copy by **credentialing the real** (or the consensual). The big difference is, this time the **copies can appear so quickly and ubiquitously** (anyone with a computer can make them) that policing that boundary is exponentially harder.

The cultural imagination is, predictably, grappling with these issues through art and narrative. We see a surge of films, series, and literature about identity theft via technology, about uploaded minds, about people falling in love with AI personas, etc. This isn’t new – science fiction has toyed with these themes for decades – but what’s different is the immediacy. When a Black Mirror episode shows a woman recreating her dead boyfriend as an AI, it’s no longer far-fetched; services today *offer* to build chatbots of your loved ones from their digital footprint. The public is thus processing the **ethics and emotions of replication** in real time. Some welcome a future where your **“digital twin”** might handle drudge work for you or allow you to be “present” in multiple places. Others are alarmed at the potential loss of what makes interactions **genuinely human** – the imperfections, the direct accountability of knowing the person you see is *actually* them. This split echoes the debates of earlier eras (recall the iconoclasts vs iconodules, or Luddites smashing machines vs technophiles), but the stakes feel higher because now even **the human essence seems reproducible**.

One particularly intriguing concept is the **“liar’s dividend”** in the age of deepfakes. This term describes how the existence of perfect forgery technology can be exploited by **liars to dismiss the truth**. For example, a corrupt public figure caught on a genuine video can claim “that’s a deepfake” to dodge accountability – sowing doubt about real evidence. Thus, the anxiety is twofold: not only can fake media fool people into believing something false, but the very knowledge of fake media can make people disbelieve something true. We end up in a potential crisis of epistemology – unable to trust our senses or our verification tools. Historically, there were analogues (for instance, once photo manipulation in darkrooms became known, people grew skeptical of some sensational images; and for centuries, people balanced faith in eyewitness testimony with caution that eyes can be deceived). But the liar’s dividend in the deepfake era could be far more destabilizing, because it can undermine the credibility of **all recorded evidence**. Societies rely on shared trust in records (photographs of war crimes, audio of promises made, etc.). If that erodes, the result is cultural cynicism or the need for wholly new systems of establishing truth (perhaps blockchain-backed video attestations, etc.).

With AI-driven replication, we also revisit the **question of agency**: If an AI clone of me commits defamation or causes harm, who is responsible – the tool maker, the user who deployed it, or me (whose face/voice lent it credibility)? Legally and ethically, this is complex. It hearkens to older debates about, say, whether an idol or puppet that “speaks” (ventriloquism) implicates the ventriloquist or the dummy. Humans have long made proxies (think of a child blaming misbehavior on their doll or imaginary twin). Now our proxies might act convincingly human without constant supervision. This challenges notions of accountability that assumed a tight link between a person and their actions/words. We may need to treat one’s **digital replicas as independent agents** for legal purposes (“virtual persons”), or conversely, trace everything back to an originator – but either choice has pitfalls.

Paradoxically, even as we fear these technologies, we are embracing them in some domains. **Holographic performers** (like concerts featuring holograms of long-dead musicians) draw crowds who knowingly celebrate a copy. Some pop music fans adore virtual idols (like Japan’s Hatsune Miku, a hologram with a synthesized voice), demonstrating a curious phenomenon: people *can* emotionally invest in and authenticate a completely artificial persona. This suggests our cultural sense of authenticity is perhaps shifting – from *what* something is to *how* it is experienced. If the experience is valuable (the song moves you, the avatar comforted you), perhaps that becomes “real” enough. This was hinted at in earlier media (people cried at movies knowing it’s fiction; they kept photos of loved ones to feel their presence), but AI replicas up the intensity by interacting and personalizing the experience.

In summary, the contemporary AI cloning and deepfake moment is like a grand culmination of the identity replication saga – a test of all the intuitions and frameworks we’ve developed through history. We see **ancient fears – the evil double, the soul-stealer, the false idol – reappear in modern guises** (doppelgänger-phobia, data theft, fake celebrities). We see philosophical questions of **appearance vs reality** get very concrete as we literally can’t tell real from fake with our senses. And we’re witnessing an adaptation process: culturally, legally, psychologically, we are scrambling to adjust just as our ancestors did with each new mode of reproduction. The difference is the speed and pervasiveness: this technology is rolling out globally in years, not centuries, giving little time for gradual adaptation.

## **Continuities and Changes: Authenticity, Agency, and the Self in Perspective**

Looking across this wide sweep of history, we can identify some **continuities** in how humans respond to new forms of copying, as well as some **evolving shifts**:

* **Continuity in Fears:** A key constant is the fear that a copy **lacks the essence** of the original and thus may **betray or harm** us or the original. Be it a wooden idol “empty of spirit”, a changeling without a human soul, a photograph that might trap one’s life force, or a deepfake of your face – in each case the copy is seen as potentially *soulless* and therefore illegitimate or dangerous. Alongside this is the fear of **deception**: that others (or we ourselves) will be fooled by the copy and accord it undue trust or power. This leads to protective measures (religious bans on images, legal restrictions on identity misuse, etc.) and to social caution (e.g. early photographers needing to prove their credibility, or today, public figures issuing quick clarifications “that video is fake”). The *intuition that identity and authenticity are precious and must be defended from imitation* is a thread from antiquity to now.

* **Continuity in Fascination:** Yet, humans are also consistently *enchanted* by their doubles. We play with mirrors, we treasure portraits, we idolize film stars, we enjoy talking to Siri or seeing our animated likeness. Each new replication medium has elicited wonder and creative experimentation. The first reaction of people to seeing themselves on film or hearing their voice on a phonograph was often astonishment, sometimes laughter or embarrassment (“Do I really sound like that?” – a mini identity crisis). Over time, these copies become integrated into our self-image: we groom ourselves for photographs, actors modulate their “on-screen” persona versus off-screen. In the AI era, people are already starting to curate their “digital self” – for instance, creating bitmojis or virtual avatars that represent them in VR meetings, perhaps choosing an idealized look. There is a **pleasure in self-replication** – an almost narcissistic extension of the self into new domains – that has driven much of the adoption of these technologies despite the fears.

* **Evolution in Theoretical Understanding:** Our explanations and metaphors for these phenomena have evolved. Ancient cultures might invoke souls and spirits; medieval theologians spoke of **substance and essence**; Enlightenment thinkers like Descartes rationalized that an automaton could never have a *mind* (thus drawing the line at consciousness); 20th-century scholars introduced terms like **“aura”**, **“the gaze”**, **“the Other”**, **“simulation”**, which gave secular language to the loss-of-authenticity problem. Today, we mix these vocabularies: tech workers talk about “digital ghosts” and “clones” in almost spiritual terms, while philosophers and lawyers wrestle with defining personhood in the digital realm. What might shift is a greater acceptance of **multiplicity** in identity. Historically, a person having more than one apparent form was usually viewed as supernatural or pathological (e.g., multiple personalities). But now, managing multiple avatars or profiles is common. Future cultural frameworks might normalize a **many-faceted self**: one could have a physical self and licensed digital selves, and authenticity might mean something like “consistent with one facet of one’s identity” rather than “the one true self.” This is speculative, but we see hints in how younger generations fluidly navigate online vs offline personas.

* **Agency and Control:** One crucial theme is who controls the copy. In early societies, it was often the gods or fate that made a double (a twin birth, a supernatural doppelgänger) – out of human hands. As representation became intentional (artists, photographers), the **maker’s intent and the subject’s consent** became relevant. Now, AI allows *anyone* to be a maker of someone else’s likeness. This democratization of replication powers is akin to when the printing press democratized knowledge reproduction – it’s liberating but also chaotic. Historically, societies eventually establish norms and laws granting individuals some say over their representation (for example, you typically cannot use someone’s likeness in advertising without permission). We can expect a similar push for **agency in one’s digital representations** – perhaps through watermarks, legal rights to one’s facial data, etc. However, enforcement will be challenging. Culturally, we may have to educate people to be skeptical and to verify identity through backchannels (a throwback to the idea of personal seals or signatures, updated for digital times).

* **Authenticity redefined:** What is “authentic” in an era of perfect copies? The concept of authenticity might shift from being about a *physical original* to being about **origin of intent or endorsement**. For example, an AI-generated painting in the style of Van Gogh is not “authentic Van Gogh” because Van Gogh didn’t create or approve it – even though physically it might be indistinguishable or even qualitatively great. So authenticity becomes about the **connection to an author or source**. Similarly, a deepfake video of a politician is not authentic because the person did not actually enact that performance, lacking the *chain of intent and action*. Our cultural frameworks may increasingly emphasize transparency: knowing *who* (or what) produced a piece of media might become as important as the content of the media itself for establishing truth. This is a return, in a sense, to valuing the “aura” or context of creation (Benjamin’s point). We see early signs: proposals that AI content should come with metadata of its algorithmic origin, or that blockchain could log an image’s history from camera to edit. In a world of clones, **context and provenance \= authenticity**.

* **Acceptance and Integration:** Just as icons became an accepted part of Orthodox Christian practice (with safeguards), and photographs became routine keepsakes rather than magical objects, AI replicas might eventually be normalized. Future generations might find it unremarkable that a celebrity has an official AI avatar that can talk to fans (and fans might prefer interacting with the avatar than not at all). The novelty and fear could subside into matter-of-fact uses, with *etiquette* and *ethics* catching up. Perhaps lying with a deepfake will carry such social stigma and high risk of detection that it becomes rare, allowing trust to rebuild. Or perhaps we’ll rely on **AI to fight AI** – deepfake detectors as ubiquitous as antivirus software, giving a sense of security. In any case, history suggests that each time we are confronted with a new replication power, there is a cycle: shock and fear, then gradual adaptation and governance, and finally a new normal. The object (be it a statue, photo, or avatar) that once seemed uncanny eventually can gain its own kind of accepted **authenticity** (e.g., a painted icon isn’t “really” the saint, but it’s authentically part of worship tradition; a photograph isn’t the person, but it’s authentically *them* at a past moment; maybe an avatar isn’t you, but it could be considered authentically *you* in a specific virtual context).

* **Human Uniqueness – Resilience or Erosion?** A deeper philosophical question is whether all this copying ultimately affirms or erodes the concept of a unique self. One might think that after centuries of copies, we’d conclude there is no inviolable core – everything about a person can be replicated or faked. And indeed, some modern thinkers lean that way, seeing identity as performance or narrative, not an essence. On the other hand, each wave of replication has also highlighted what *cannot* be copied. For Plato, the Form could not be put on a canvas. For iconoclasts, the divinity could not be captured in wood. For Freud, the living soul eludes the double. For Benjamin, the aura cannot be photographed. For modern AI, perhaps the **inner consciousness or the lived life experience** remains something no deepfake can replicate. Even if an AI twin passes a Turing test, some will insist there’s an invisible difference. Culturally, we often double down on valuing the “real thing” precisely when copies become rampant (for instance, the resurgence of live theater and concerts in the age of mass video – people crave the in-person aura). So, our frameworks might evolve to explicitly celebrate human qualities that resist duplication: spontaneity, genuine emotion, moral accountability, etc. Agency – the fact a person can **will** their actions – might become the marker of authenticity that technology can’t fake (unless AI gains free will, which is another discussion).

In conclusion, our journey from ancient doubles to AI doppelgängers reveals a paradoxical trajectory. Each new medium that replicates identity has been met with **old anxieties in new clothes**, yet each has also expanded our understanding of what identity *is*. We started with souls and shadows, and we’ve arrived at algorithms and data – but the core questions persist: *What makes me “me”? Can that be copied or not? If it is copied, do I lose something, or do we gain something?* Historically, the answers have never been simple. Instead, societies negotiate balances – allowing certain kinds of copying (with ritual or rules) and prohibiting others, finding where the line of **acceptable authenticity** lies. In the face of AI’s incredibly sophisticated reproductions, we are at another such junction. By reflecting on how **Plato grappled with painters, medieval mystics with icons, tribes with cameras, and philosophers with photographs**, we gain perspective to face our current challenge. Perhaps our cultural frameworks will shift towards a **more fluid sense of self**, or perhaps we will reassert the sanctity of the un-copiable human core. Most likely, we’ll do both in different measures. As we have learned, the story of identity and replication is not one of a problem ever fully solved, but of an ongoing dialogue between our **yearning to represent life** and our **yearning to preserve its authenticity**. The dialogue continues, now with machine learning in the mix – and it will shape the future of what it means to be “real” in an age of endless copies.

# AI Agents as Active Learning Conversation Partners in Education

## **Introduction**

The use of AI conversational agents in classrooms is transforming traditional learning into a more interactive, student-centered experience. These agents – often implemented as chatbots or virtual tutors – can engage learners in dialogue, ask questions, and provide feedback in real time. Educators and researchers are increasingly exploring how such AI partners can promote **active learning** (learning by doing, discussing, or explaining) rather than passive content consumption. The trend spans across subjects and educational levels, from high school to university, with institutions beginning to experiment on a broad scale. Studies confirm that well-designed conversational agents can improve students’ motivation and learning performance, yielding learning gains that are in some cases comparable to human tutoring. At the same time, careful integration is required to ensure these tools enhance, rather than hinder, teaching and learning.

## **General Trends and Effectiveness**

**Growing Adoption Across Subjects:** AI conversational agents are being deployed in a variety of subject areas – from language learning and history to STEM disciplines. Their versatility allows them to answer questions, explain concepts, or even play roles (such as a historical figure or a virtual lab partner) to enrich the curriculum. In higher education especially, there has been a surge of interest in chatbots as part of the “digital transformation” of education. A 2023 scoping review identified 66 studies on higher-ed chatbots, demonstrating the **promising versatility** of these agents for university students across disciplines. Research on K-12 (including high school) is comparatively sparser, but initiatives are growing – for example, Khan Academy’s *Khanmigo* AI tutor is being piloted in hundreds of school districts, indicating expanding interest at the secondary level.

**Learning Outcomes:** Meta-analyses and reviews generally report **positive effects on student learning** when using conversational agents. One recent meta-analysis found that educational chatbots have a **statistically significant positive effect** on student performance, with an average impact that is small to moderate in magnitude. Another study even reported large effects in certain contexts (particularly for higher education learners). These agents often provide immediate, personalized support that can help with homework, studying, and skill development. Notably, students benefit in at least three areas: *on-demand assistance* (answers and explanations any time), a more **personalized learning experience**, and opportunities to practice various skills at their own pace. Conversational agents have also been shown to boost motivation and engagement – one review of 43 studies found chatbot use correlated with higher student motivation and interest. In classroom trials, AI dialogue agents have deepened students’ reasoning during group discussions and encouraged learners to build on each other’s ideas, demonstrating potential for richer collaborative learning experiences.

**Pedagogical Value:** For educators, the introduction of AI chatbots can **save time and enhance pedagogy**. Routine questions and formative feedback can be handled by the AI, freeing teachers to focus on higher-level mentoring. Some agents function as virtual teaching assistants that help manage large classes or online forums. For example, Georgia Tech’s **Jill Watson** – an AI teaching assistant originally built on IBM Watson – was used to answer students’ course questions, effectively handling FAQs and reinforcing course content. Studies on Jill Watson reported improvements in student course performance and retention, as the agent boosted the *teaching presence* in online courses. Generally, when AI partners are used, students receive more frequent feedback and guidance, which can lead to better learning habits and outcomes. However, it’s not a universal cure-all; researchers caution that while AI tutors **outperform having no support** at all, they often still **fall short of human teachers or tutors** in effectiveness. The consensus is that conversational agents work best as a *complement* to teacher-led instruction – scaling individualized support and keeping students actively engaged between and during classes.

**Challenges:** Alongside positive outcomes, the literature highlights important challenges. **Reliability and accuracy** of AI responses are a primary concern. Agents can sometimes provide incorrect or misleading information, so their content knowledge must be carefully curated and monitored. Ensuring the AI behaves ethically and without bias is another hurdle; agents may inadvertently reinforce stereotypes or produce inappropriate responses if not properly constrained. Students also need to be guided in the **responsible use** of AI tools. Rather than allowing learners to become overly reliant on an easy answer machine, effective implementations teach students how to critically evaluate the agent’s responses. In fact, some educators turn the presence of AI into a learning opportunity itself – having students analyze and fact-check the chatbot, thereby sharpening their critical thinking. Overall, the trend is one of cautious optimism: **AI conversational partners can significantly enhance active learning** when used thoughtfully, but they require sound pedagogical design, oversight, and clear objectives.

## **Pedagogical Strategies for Integrating AI Agents**

Educators have developed a range of strategies to integrate AI agents into classroom instruction while maintaining a focus on active learning. Key pedagogical approaches include:

* **Tutoring and Practice Sessions:** One common use is to have students engage in one-on-one tutoring conversations with an AI agent as part of homework or classwork. For example, in a math class, a chatbot might pose problems, ask the student to explain their reasoning, and give hints when the student struggles. This can be done during class in rotational stations or as an online supplement. Such agents are available 24/7, allowing learners to practice skills at their own pace outside class. Teachers often review the conversation logs to identify misconceptions and tailor subsequent instruction.

* **Socratic Dialogue and Inquiry:** Some instructors use AI agents to foster a Socratic style of inquiry, where the bot prompts students with open-ended questions. In science or history classes, for instance, a conversational agent might ask students to make predictions, justify their answers, or consider counterarguments. This strategy pushes students to articulate their thinking. The agent’s role here is less about giving answers and more about probing student understanding – essentially **guiding the learner to construct knowledge** through dialogue. Research suggests that these guided conversations can mimic the benefits of human tutoring by encouraging deeper reflection.

* **Collaborative Learning Facilitation:** In group work settings, **collaborative conversational agents** (CCAs) have been used to facilitate productive discussion among students. A CCA might join a small group as a virtual team member or moderator, prompting the team with questions or nudging them if the discussion stalls. For example, a collaborative agent named “Clair” was tested in inquiry-based learning groups to prompt students at appropriate times, much like a teacher circulating among groups. These agents use dialogic strategies to ensure every student is contributing – asking for clarification, encouraging quieter students to share, or introducing new angles to the problem. By scaling guidance to multiple groups simultaneously, CCAs help overcome the limitation that a single teacher can only attend to one group at a time. This strategy has shown promise in keeping student teams on task and deepening their reasoning during projects.

* **“Teachable Agent” Activities:** An innovative strategy for active learning is the **learning-by-teaching paradigm**, where students teach a concept to an AI agent. Systems like *Betty’s Brain* have pioneered this approach: students in a science class teach a virtual student (the agent) by constructing concept maps and explaining ideas to it. The agent can ask questions or take quizzes, and the student must correct the agent’s understanding. This role reversal makes the human student the teacher, encouraging them to actively organize knowledge and monitor the agent’s (and thus their own) understanding. Pedagogically, it fosters self-regulation and deeper processing – students need to anticipate questions and evaluate if “Betty” has truly learned the material. Teachers integrate these activities as projects or lab sessions, after which a class debrief allows students to reflect on what they taught and learned. Learning-by-teaching with AI has been effective in domains like ecology and computer science, and it naturally engages students because they feel responsible for their agent’s learning.

* **Writing and Brainstorming Aids with Reflection:** In humanities and social science classes, instructors have started to incorporate AI chatbots as **writing assistants or brainstorming partners**. For example, a teacher might have students use a chatbot to generate ideas for an essay or to receive feedback on a draft. Crucially, the pedagogical strategy here includes a reflection component: students must **critique the AI’s suggestions or edits**. A recent classroom example had fifth-graders use an AI bot to propose ways to thank community helpers, then **evaluate the bot’s suggestions using a rubric** for relevance, clarity, and bias. In high school or college writing courses, instructors similarly ask students to identify errors or improvements in the AI’s output, compare it to their own work, and thereby learn from the AI while practicing critical analysis. This approach turns the AI into a conversational *peer reviewer*, and students remain in an active role rather than passively accepting AI-generated text.

* **Blended Instruction and Flipped Classrooms:** Many educators integrate AI agents in a blended learning model. Students might interact with a conversational agent at home as part of a flipped classroom – for instance, completing a chatbot-led tutorial or simulation before a class discussion. The next day in class, the teacher builds on that conversation: students share their chatbot dialogues, discuss where the AI was helpful or confused, and the teacher clarifies misconceptions. This strategy uses the AI to prepare students with foundational engagement, so in-person class time can dive deeper. Additionally, in live classes, teachers sometimes project an AI assistant (like a chatbot on the screen) to **model question-asking**. A teacher might converse with the AI in front of the class to demonstrate how to probe a topic, then have students pair up to try it themselves. By deliberately structuring when and how AI is used (pre-class, during group work, as homework, etc.), teachers ensure it aligns with learning objectives rather than being a novelty.

**Teacher Involvement and Oversight:** Across all these strategies, a best practice is for teachers to maintain oversight of the AI-student interactions. In successful implementations, instructors **set clear expectations** for how to use the chatbot (e.g. “use it to get hints, not final answers”) and they debrief with students to consolidate learning. Educators might also intervene in the conversation flow by customizing the agent’s prompts or providing students with example questions to ask. Essentially, the pedagogy is **co-designed**: teachers craft the learning activity around the AI agent’s capabilities. A strong recommendation from emerging research is that teachers also address AI literacy – teaching students *why* the agent might err and how to use it responsibly. This meta-cognitive angle ensures that interacting with the AI becomes an opportunity to learn critical evaluation, aligning with the broader goal of active, thoughtful learning.

## **Interaction Design Principles and Best Practices**

Designing an AI conversational agent for active learning requires careful attention to how the dialogue is structured and how the agent behaves. Key interaction design principles include:

* **Promote Dialogic Learning:** Effective educational agents engage students in a true dialogue, not just lecture or quiz them. This means the agent should ask open-ended questions, encourage students to elaborate, and sometimes answer a question with another question to stimulate thinking. The concept of *dialogic learning* is central: the agent’s prompts and responses are designed to create a back-and-forth exchange where the student is constructing answers and ideas out loud. For instance, instead of simply telling a student if an answer is right or wrong, a good agent might say, “Interesting – what do you think would happen if we changed this part of the problem?” Such prompts keep the student actively involved in the conversation. A recent comprehensive review emphasizes this principle, highlighting the importance of **dialogic interaction** as a core design feature for pedagogical agents.

* **Personalization and Adaptive Scaffolding:** Every student has different prior knowledge and learning needs. Thus, an AI learning partner should tailor its interaction to the individual. Design best practices include building in diagnostics – early in the conversation the agent can ask a few check questions to gauge the student’s level, then adjust its difficulty and explanations accordingly. Throughout the dialogue, the agent can provide **scaffolding**: hints or cues that are calibrated to the student’s current understanding. For example, if a student is stuck, the agent might first give a small hint; if the student is still stuck, it gives a bigger hint, and only as a last resort provides the answer. This graduated help keeps the learner in the “productive struggle” zone – challenged but not frustrated. Personalization also means referencing a student’s own past responses: *“Earlier, you mentioned X, can you use that idea here?”* Such techniques have been found to increase learning gains by making the experience more relevant and supportive for each student. In fact, the ability to **provide feedback according to the student’s level** is identified as one of the key design principles for effective agents.

* **Empathy and Motivational Support:** An emerging design focus is on **empathic conversational agents** – systems that not only convey information but also recognize and respond to the learner’s emotional state. While full empathy is hard to achieve with AI, basic techniques like using encouraging language, offering praise for effort, and addressing frustration can make the agent more engaging. For example, if a student gets a question wrong, the agent might respond, “I see this is tricky – don’t worry, we can try together,” rather than a blunt “incorrect.” Studies suggest that empathetic cues can improve student comfort and motivation. Thus, designers often incorporate a friendly persona and polite tone. However, the agent should balance empathy with honesty – it should not overly flatter or mislead. Finding this balance is an active area of research (e.g., determining when to be encouraging versus when to push the student harder). The general best practice is to make the agent **supportive and patient**, mirroring a good human tutor’s bedside manner, which has been linked to sustained student engagement.

* **Clarity and Domain Knowledge:** For an AI agent to be a trusted learning partner, it must demonstrate proficiency in the subject matter and communicate clearly. Design principles call for a strong knowledge base (or access to one) so that the agent’s explanations and answers are correct and on-point. When the agent doesn’t know something or is uncertain (a common occurrence with generative models), it should be transparent about it rather than guessing – perhaps suggesting the student ask a teacher or check a reliable source. Maintaining **accuracy** is crucial; as noted, many educational chatbots use techniques like retrieval augmentation or constrained response generation to reduce misinformation. Additionally, the agent’s language should be tailored to the learner’s level: using simpler vocabulary for younger students or more technical terms for advanced learners, as appropriate. Clarity also extends to the agent’s questions: they should be well-phrased and unambiguous so the student understands what is being asked. Some projects employ multi-turn rephrasing, where the agent will restate a question in simpler terms if the student seems confused. In short, a best practice is that **pedagogical agents must be subject-aware and articulate**, to truly support learning rather than cause confusion.

* **Engagement through Turn-Taking and Multimodality:** Keeping the student actively engaged means the agent shouldn’t dominate the conversation. Good design ensures a balanced turn-taking – the agent offers a prompt or piece of information, then gives the floor to the student. Long monologues by the agent are generally avoided, as they can lead to passivity. Instead, interactions are broken into bite-sized turns that require frequent student input (e.g. every few sentences, the agent asks a question or checks if the student is following). This design mirrors how a skilled teacher might frequently check in with a class. Furthermore, some agents utilize **multimodal interaction** (when possible) – for example, an animated avatar that can gesture or display visuals, or a text bot that can show diagrams and links. Visual elements can enhance understanding (think of a geometry tutor bot that draws shapes as it discusses them). Even simple features like emojis or adaptive tone (enthusiastic vs. calm) in text can make the conversation livelier for students. The goal is to avoid a dry, robotic feel and instead create a more **conversational, engaging atmosphere** that holds students’ attention.

* **Ethical and Safe Interaction Design:** Best practices also mandate designing for student safety and equity. Interaction logs should be private and secure, and the agent should be free of any *biased or insensitive language*. Developers are encouraged to train AI models on diverse and unbiased data, and to include filters that catch inappropriate content. In terms of pedagogy, the agent should treat all students respectfully and be inclusive – for example, not assuming a one-size-fits-all context (such as always referencing sports analogies which might not resonate with all). When errors do occur, it’s recommended that the system have an easy way for students or teachers to report problems and improve the agent over time. Some design frameworks incorporate an initial disclosure to students that “you are talking to a computer program” to set the right expectations. Overall, considering the **ethical implications** and making the AI a trustworthy tool is an essential design principle, as highlighted by recent research efforts. The focus is on creating a **productive and safe learning environment** through the agent’s interaction style.

## **Common Design Patterns and Roles of AI Learning Agents**

Over years of development, several **design patterns** have emerged for how AI agents function as conversation partners in educational contexts. These patterns define the role the agent plays and how it contributes to learning:

* **Virtual Tutor or Instructor:** In this pattern, the AI agent acts like a personal tutor, guiding the student through learning materials. It provides explanations, asks comprehension questions, and corrects misunderstandings. Examples include systems like *AutoTutor*, which simulates a human tutor in dialogue – AutoTutor has been used to teach physics and computer literacy by holding a conversation and employing typical tutor moves (hints, prompts, feedback). Tutor agents usually follow a pedagogical script or curriculum and aim to transfer knowledge or skills directly. Research indicates this is the most common role for educational agents (one review found about 31% of pedagogical agents were in a tutoring role). Such agents are effective for providing **one-on-one instruction at scale** and can personalize the pace of teaching to each learner.

* **Peer or Collaborative Partner:** Here the agent is designed to behave more like a learning peer or collaborator rather than an authoritative tutor. The agent might **solve problems alongside the student**, sometimes even intentionally making mistakes or expressing confusion to invite the student’s input. This pattern is used to stimulate cooperative learning – the student and agent bounce ideas off each other. An interesting application of this was in a high school science project where two chatbot agents were introduced: one as an “expert” peer and another as a “novice” peer. Students were tasked with teaching the novice agent (Kibot) while the expert agent provided occasional guidance. This dual-agent setup led to rich discussions, and it was found that the conversational agents helped students deepen their reasoning and knowledge construction. Even with a single agent, designing it as a peer can make students feel more at ease (less afraid to make mistakes) and encourage them to explain their thinking more thoroughly, as they would to a classmate. The peer agent pattern often focuses on **collaborative problem solving and learning by explanation**.

* **Teachable Agent (Student Role):** As discussed earlier, the teachable agent pattern flips the script: the **AI takes on the role of the student**, and the human learner becomes the teacher. The agent “learns” from the student’s input. *Betty’s Brain* is a hallmark example, where middle school students teach a virtual character named Betty about science by building concept maps. The AI agent can answer questions based on what it’s been taught, allowing the human student to assess the agent’s (and thus their own) understanding. This design pattern leverages the learning-by-teaching effect; by teaching the agent, students engage in self-explanation and recursive review of the material. It’s a powerful active learning approach that also makes the experience game-like (students often enjoy trying to make their agent “smart”). The agent in this role typically asks questions like “Did I do this right?” or “Can you explain that part again?”, prompting the student to fill knowledge gaps. Notably, systems using teachable agents also incorporate a separate mentor agent or feedback system (like Betty’s Brain’s mentor Mr. Davis) to guide the human student in their teaching process. This pattern is somewhat less common but very effective in promoting metacognition and responsibility for learning.

* **Coach or Mentor:** In the coach role, the AI is not directly teaching content, but rather coaching the learner on **learning strategies, motivation, or meta-cognitive skills**. For instance, a mentor agent might observe a student working through a complex task (either through their inputs or through sensors) and interject with advice: “Maybe you should double-check your last result” or “It looks like you’re frustrated, how about we break this problem down?”. These agents often draw on models of self-regulated learning and can help with goal-setting, time management, and reflecting on mistakes. In some experimental setups, a mentor agent works alongside a tutor agent – the tutor handles content, while the mentor focuses on the learner’s approach (e.g., encouraging them to recall prior knowledge or to stay persistent after an error). A systematic review of conversational agents found that about 18% took on a **mentor-like role** (guiding, advising, motivating). This pattern is valued for improving student **self-efficacy and perseverance**. An example is an empathic coach agent that gives students prompts to reflect on what they learned at the end of a session, thereby solidifying gains and connecting to personal goals.

* **Discussion Facilitator or Moderator:** When conversation agents are embedded in group learning, they often serve as a moderator. In this pattern, the AI might start a discussion thread with a thought-provoking question and then encourage students to respond to each other, intervening only to keep the discussion on track. For collaborative problem-solving, the agent might assign roles to human team members (“You be the skeptic, you be the summarizer…”) and then prompt each to contribute in turn. A notable research direction in this area is using **agents to scale dialogic teaching methods** – essentially encoding strategies that teachers use in moderating discussions (like prompting elaboration or asking one student to respond to another’s idea). By doing so, the agent ensures that key elements of productive dialogue occur even without a teacher present in each group. This design pattern has been applied in online course forums as well, where a bot might welcome new posts, ask clarifying questions, or tag relevant resources in the discussion. Approximately 9% of pedagogical agents in research were identified as *moderators*, reflecting this specialized but important role. The benefit is a more structured and inclusive discussion, making sure every student’s voice is heard and that the conversation remains focused on learning objectives.

* **Role-Play Simulations:** Another pattern involves agents that simulate a character or scenario for educational role-play. In medical training, for example, an AI agent might act as a virtual patient, conversing with a medical student who must diagnose and treat them through dialogue. Similarly, in language learning, a chatbot might pretend to be a travel agent or a pen-pal from another country for the student to practice conversation. This **simulate & experience** pattern places the student in an active role where they must apply knowledge in context. The agent provides the other half of the scenario – responding in character, giving the student realistic cues. Such role-play agents help students learn by doing: a history student could “interview” an AI avatar of a historical figure, or an ethics class might debate with a bot posing arguments from a certain perspective. The key design element is maintaining the persona accurately and keeping the interaction scenario-based. These agents often require rich scripting and are narrower in scope (focused on a particular scenario), but they can greatly enhance engagement and the practical application of knowledge. Students often report these simulations as memorable and useful for transferring classroom learning to real-world skills.

* **Organizer and Personal Assistant:** A more utilitarian pattern is the agent that helps students (and instructors) organize learning tasks and provides reminders or study support. While not directly delivering curriculum content, an AI assistant might help a student plan their study schedule, break down a project into steps, or even simply answer FAQs about course logistics (deadlines, requirements). In the literature, some chatbots have been used as **academic advisors or course assistants**, helping students pick courses or navigate administrative issues. In classroom use, a teacher might set up a chatbot to answer common questions during an assignment (so that students don’t all queue up to ask the teacher the same questions). This pattern supports active learning indirectly by removing small hurdles and confusion, thereby allowing students to focus on the substantive work. A review categorizing agent roles found about 21% served as an *organizer/assistant*. For example, an organizer bot in a project-based learning class might ping teams with: “Have you updated your project journal this week? Remember, iteration review is tomorrow.” While these functionalities are sometimes considered more on the administrative side, they contribute to a more seamless active learning environment by **keeping students on track and informed**.

Most real-world educational AI systems combine elements of multiple patterns. For instance, a single agent might tutor on content but also employ some motivational coaching, or a collaborative agent might sometimes directly teach a concept if all students in the group are confused. The patterns above, however, are useful abstractions for understanding the common design **archetypes** in this field. By recognizing these, developers and educators can draw on established best practices for each type.

## **Notable Research Projects and Case Studies**

The exploration of AI conversation partners in education has led to numerous projects and studies. Below we highlight a few influential examples and findings from the research:

* **AutoTutor:** *AutoTutor* is one of the earliest and most extensively studied conversational tutoring systems. Developed by researchers including Arthur Graesser, AutoTutor holds a mixed-initiative dialogue with students on topics like physics, computer literacy, and reading comprehension. It features an animated avatar that speaks and gestures, simulating a human tutor’s mannerisms. Notably, studies showed that AutoTutor can achieve **learning gains comparable to human tutors** under certain conditions. For example, college students using AutoTutor to learn computer hardware basics demonstrated improvements on post-tests similar to those tutored by novice human instructors. AutoTutor’s design incorporated ideal pedagogical strategies (like asking deep reasoning questions) that sometimes even surpassed what human tutors typically do. This project, spanning over two decades of research, provided proof that conversational agents can indeed *help students learn effectively* and has inspired many subsequent systems. It also yielded insights on dialogue tactics – such as the importance of feedback timing and the handling of student errors – which inform current best practices.

* **Jill Watson – Virtual Teaching Assistant:** At Georgia Tech, Professor Ashok Goel famously deployed a virtual TA named *Jill Watson* in an online graduate course (Knowledge-Based AI). Jill, powered by IBM Watson technology (and more recently by LLMs), answered students’ questions on the class discussion forum. Initially, students didn’t realize some replies were coming from an AI\! The experiment was a success: Jill accurately answered many routine questions (like clarifying assignment instructions), and this significantly reduced the response time for student queries. Research on Jill Watson showed that it improved the **sense of instructor presence** in the course and contributed to *higher student satisfaction and retention* in the online program. As of 2025, Jill Watson has evolved using GPT-based models with retrieval augmentation to ensure accuracy. It’s used across multiple courses and even other institutions as a plug-in for learning management systems. The Jill Watson project demonstrated how an AI agent can scale support in large classes and is often cited as a model for integrating AI in higher education to augment human teaching.

* **Betty’s Brain:** Developed at Vanderbilt University, *Betty’s Brain* is a prime example of the teachable agent paradigm in a middle school science context. In this system, students learn about complex systems (like ecosystems or climate change) by teaching a virtual character named Betty. They do so by constructing causal maps and having Betty take quizzes to see if “she” understood. The student also interacts with a mentor agent (Mr. Davis) who offers feedback and hints. Research around Betty’s Brain found that this approach prompted students to engage in self-regulated learning behaviors – planning their lessons for Betty, monitoring her progress, and revising their own understanding when Betty answered incorrectly. In terms of outcomes, students using Betty’s Brain showed improved understanding of the science topics and better ability to explain their reasoning. Perhaps more importantly, they practiced learning strategies like searching information and evaluating knowledge, which are transferable skills. Betty’s Brain has been refined over years and was even adopted by an OECD initiative as an innovative learning and assessment tool. This project is often referenced to illustrate the power of **learning by teaching** with AI, and its design patterns (teachable agent \+ mentor) have influenced other educational systems.

* **Collaborative Agent “Claire” and Kibbitzing Agents:** A line of research in AI in education looks at **agents supporting small group discussions**. One study introduced an agent called *Claire* into middle school science groups engaged in inquiry learning. Claire would monitor the group’s dialogue (via a chat system) and post prompts like, “Can you elaborate on that idea?” or “Do you all agree with what was just said? Why or why not?” at strategic moments. The impact was an increase in **dialogue productivity** – students gave more complete explanations and stayed more on topic. Another related experiment, nicknamed *“Let’s Teach Kibot,”* used two agents in a high school setting: one agent acted as an expert co-learner and another as a novice student (the Kibot) that the group had to teach. This creative setup encouraged students to explain concepts to the novice bot, while the expert bot would occasionally model good reasoning. Results indicated that such multi-agent configurations can stimulate **knowledge-building conversations** among students and improve learning outcomes in subjects like marine biology. These studies are notable for moving beyond one-on-one tutoring to more **social, group-based active learning**, expanding the horizon of what AI partners can do in a classroom.

* **Chatbots for Language Practice:** In the domain of language learning, conversational agents have shown particular promise by providing practice partners for speaking or writing. For instance, projects have deployed chatbots for ESL (English as a Second Language) students to practice conversational English. These bots often play roles (like a tourist asking for directions, or a friend chatting about hobbies) to make practice more realistic. Research has found that students using chatbots for language practice gain fluency and confidence, especially in contexts where they lack a human partner for regular practice. One meta-review concluded that chatbot interactions led to improvements in vocabulary and dialogue skills, and students appreciated the **non-judgmental environment** to make mistakes and try again. While not tied to a single famous project name like the others, the collective work in this area – including apps like Duolingo’s chat exercises – is a significant part of the landscape. It underscores how AI conversation partners can facilitate active learning by endless patient practice, a role hard for human teachers to fill one-on-one at scale.

* **Khanmigo by Khan Academy:** A very recent and high-profile initiative is *Khanmigo*, launched by Khan Academy in 2023 as an AI tutor/assistant powered by a large language model. Khanmigo is designed to help both students and teachers: it can tutor students in various subjects through dialogue and also assist teachers by generating lesson ideas or handling administrative tasks. Early pilot results (in schools adopting Khanmigo) are being closely watched. Already, Khan Academy reports that Khanmigo has been used to facilitate coding lessons (the agent can play the role of a coding partner) and to engage students in Socratic dialogue for math and reading comprehension. One distinguishing aspect of Khanmigo is the emphasis on **guided use** – Khan Academy provides training for students and teachers on how to use the AI appropriately, and they have put safeguards in place (the agent won’t just give away answers to homework, for example, but will prompt the student to think). While comprehensive research results are not yet published, Khanmigo represents the **translation of research to practice** on a large scale. It’s notable that it’s being rolled out in hundreds of schools, potentially making AI conversation partners a mainstream tool. Observations from these pilots so far echo prior research: students are highly engaged by the interactive dialogue, and teachers report it is like having another tutor in the room for each student. Khanmigo’s development was informed by many of the principles and patterns identified in academic projects (e.g., it can take on roles like debate opponent, coach, or tutor depending on the context), making it a real-world amalgamation of what has been learned from smaller scale studies.

These examples barely scratch the surface, but they illustrate the diversity of applications: from intelligent tutors and TAs to collaborative facilitators and teachable agents. Importantly, all these projects report not just on learning gains but also on student **engagement and attitudes**. A common finding is that students often enjoy interacting with the AI agents – they describe it as fun, motivating, or confidence-building. For instance, students in the Jill Watson experiment said they felt more comfortable asking “dumb questions” to the AI than they might to a human TA, which meant they ultimately got answers and learned things they’d otherwise hesitate to inquire about. Similarly, shy students in group work have participated more when prompted by an agent than they initially would with peer-only discussion. These qualitative outcomes highlight that beyond test scores, AI conversation partners can influence classroom dynamics and student self-perception in positive ways.

Of course, not every study is glowing; there have also been instances where the AI did not significantly outperform traditional methods, or where technical issues hampered the experience. What’s notable, however, is that even these “failures” provide valuable lessons that drive further innovation. Each project contributes insights into how to better design the agents or integrate them more seamlessly. As the field evolves, we see a convergence of successful elements (for example, combining tutoring with motivational support, or blending human and AI facilitation).

## **Conclusion**

Research and practice to date suggest that AI agents, when used as active learning conversation partners, hold great promise for enhancing education. They offer **interactive, personalized, and scalable learning experiences** that can complement the work of human teachers. Across subjects and educational levels, they have been used to tutor, to inspire inquiry, to act as confidants for practice, and to orchestrate collaborative learning. The general trends show improved engagement and in many cases improved learning outcomes, especially in areas like increased student practice time, immediate feedback, and motivation. Pedagogically, success comes when these tools are woven thoughtfully into instruction – aligned with learning goals, supported by teacher oversight, and paired with activities that require students to reflect and think critically about the AI’s input.

It’s important to emphasize that AI conversation agents are **not a replacement for teachers or peer interaction**, but a powerful supplement. The best results happen in a blended approach: teachers leverage the AI to handle repetitive or individual tasks, while they focus on higher-level guidance and personal connections with students. This balance addresses the concerns that students might otherwise try to “outsource” their thinking to AI. When designed and implemented with best practices – maintaining accuracy, encouraging dialogic exchange, adapting to learners, and ensuring ethical use – AI agents can indeed transform “passive learning now, apply later” into an *active learning now* paradigm. Classrooms become richer dialog spaces: every student can have an ongoing conversation to explore ideas, test their understanding, and get support exactly when needed.

As we move forward, we can anticipate even more sophisticated and accessible AI partners in education. Future research is focusing on long-term effects (e.g. how sustained use of an AI tutor over a semester impacts learning habits), multi-modal agents (perhaps combining text, voice, and even physical robots), and addressing the current gaps such as deeper evaluation of learning outcomes and strategies to keep students’ trust without reliance. There is also a push for more empirical studies in high school settings, since much research so far has centered on either controlled lab experiments or higher education. Early indications are optimistic: if developed and deployed carefully, AI conversational agents can foster a more active, personalized, and equitable learning environment. In summary, the deployment of these AI agents marks a significant innovation in pedagogy – one that is already showing measurable benefits and that, with careful stewardship, can greatly support teachers and engage students in the **art of learning through conversation**.

**Sources:** The information above is drawn from a broad survey of recent literature and reports on AI in education, including systematic reviews, meta-analyses, and case studies of specific AI implementations such as conversational tutoring systems, virtual teaching assistants, and collaborative learning agents. These sources collectively provide evidence of the effectiveness, best practices, and innovative applications of AI conversational agents in both high school and higher education contexts. The cited lines throughout the report correspond to these detailed studies and examples, offering a foundation for each point made.