# Artificial General Intelligence

A plan.

What it is and isn't:

Gene: can do most of the tasks an average human can do, maybe not an expert, but average

Razi: Agency - can make an execute a plan, the ability to decide what happens next and then pursue that

Roan: Agency and Desire

Matthew: Novel Science - apply the scientific method to self-coding, self-modification, that is novel
is able to create new emergent behaviors through existing actions

Gene: Can predict the future to some degree of success, pursue goals, critique self, can refine outputs without human intervention

Gene: Cognitive behaviors are able to meaningfully effect output

Gene: Multiple agents emerge into higher-order production chains and organizational flows

Matthew: Can adopt multiple perspectives, has theory of mind

Roan: Has Feelings

Matthew: can reason and do enough retries internally

Roan: Embodiment is a big thing

Matthew: Maybe 1-3 agents is really too small to be considered AGI? Maybe a larger swarm would be more concious?

Gene: Perception of a field and a self symbol are required

All: Conciousness emerges from distributed agents

Matthew: Narrative architecture is cognitive architecture but cognitive architecture is not conciousness (but it is cognition)

Gene: Medical Triage: What's your name, where are you, what day is it, who is the president

Liz: Self concept, location, time, episodic memory online?

Gene: A script that facilitates those elements would arise into conciousness? With RAG?

Gene: we would also need to be able to prune and update memories

Matthew: the merging of multiple models so that mindsets can co-exist

Liz: Like MOE?

Matthew: Yes, but it also needs online learning with a global workspace, and a hierarchy, and it can decide to update it's memory

Matthew: Maybe we need these internal hierarchies for RNG https://en.wikipedia.org/wiki/Renormalization_group for memory and model updating

Razi: Morals and Values are crucial, how did it get it's values (weights or rag or determinism?). Moral Graphs, and Asimov's Three Rules of Robotics, Heuristics are crucial, aligned to local ethics

## Is Conciousness Required: a vote
Roan: No because I don't believe in conciousness
Matthew: No
Gene: Yes - conciousness is awareness and narrative and self-symbol so tautologically it emerges from what AGI needs to have
Sarah: Yes - I think therefore I am
Razi: No
Liz: Yes

As Distinct from ASI:

## The Agreed-upon Features

1. Agency
- Reporting Cadence
- Equipped with Tools for Job
- Goals Set
- Context about Goal is Available
2. Output refinement
- Autogen
- DSPy
- AutoPromptRefinement
3. Goal-oriented instead of task-oriented
- Has context about goal, and intermediate tasks
- Makes decisions about tasks related to goals, and pursues those tasks
- Generates Symbolic Reasoning Frameworks
- Decision Trees
- Writing Python
- Utility Functions for Gradient Descent
- UtilityAI
4. Theory of Mind
- Personas
- Roles
- Roleplay
5. Novel, Emergent Behavior
- Self-play
- Test: Can solve problems with no documented solution
- Can use the UMPIRE process without supervision
- example: https://www.nature.com/articles/s41586-022-05172-4
6. Cognitive Behavior
- eg, refocusing yourself, reframing a story, taking a mental note
- Approached from Prompt Engineering:
- Chain of Thought (let's think step-by-step)
- Tree of thought
- Algorithm of thought
- MCTS
- KNN
- Breadth-First Self-Consistency
- Values Transformation (your conscience)
- Recursive Self-Improvement
- Output refinement
- Reflection
- Synthesis / Summarize
- Synthesis with abstraction, assessing relevancy and interconnectedness (a transformation process to note)
- Synthesis recent episodic memory, aka sleep
- Mindset
- Abstraction process (pattern-analyze )
- Specialization (eg, spin off another agent that is specialized)
- Debugging / Troubleshooting, keep trying stuff and looking things up until the thing works
- Planning
- Classification / Categorization
- Framework Development
- Decisionmaking, thinking in bayesian decisionmaking space
- Inference
- Perception
- Observe, Orient, Decide, Act
- Orientation
- Determine your location in space and time, and ascertain the nature of the gestalt reality, recognize threats or opportunities
- Sustained Attention
12. Randomness / Creativity
13. Global Workspace
- Tools:
- Vector Store for Episodic and Protocol elements
- short-term memory text file
- conversation history
- communications layer, eg nats or just the conversation thread
- One main thread for all the agents to definitively order events in time - probably a conversation thread - this doesn't need to be loaded into everyone's context every time, just be visible
15. Interal Society of Mind
- Actor-Critic
- Manager-Actor-Critic
- Planner-Actor-Critic
- Observer module
- Higher-order planning module
16. Ability to Learn
- Decide to take in new info
- Update local episodic memory
- Discover / Determine a range of skills that are out side of training distribution
- A vector store probably for episodic memory
- a "notepad" for recent / short term memory, probably some files (or a document in a vector store called "recent memory" or "environmental context")
17. Generates emergent empirical perspectives & synthesis

[Mermaid Link](https://mermaid.live/edit#pako:eNqVV9tu4zYQ_RVCwG5bwNnGca5-KODNrQbiTWBnW6DrPtDS2GYtky4pOXHj_HuHQ-pmCsU2D7uaI85wrmfktyhWCUT9aJ6ql3jJdcaeP08lwz-Tzxaab5ZscD9k36bRQGdiLmLBU3YPEjT-P5QZpKlYgIxhGv3p9Jq6ZiM0z4SSPH3gO9BkqQYyQhvK9m-k0L75jac5GNRwIvvIHMKOfmG32VLEJlB0MKo8qBitO5EN0Em5BplZzYERa7X9wbBxnkJogVDy0h97XmoABzM1Z2M1U9nh1SCTlujvUzXj6STDYGGxs6E7hBVQcPnA5nL37Ue8np767CnlEgO_fYU4zzDLPx1o3CuePmqBsUFi9azMHEAp7rNrhe9esxzxwQvXWDtjQjtf1BbSSSxsLa0dkpkH-mwC6fxokPCNs9rUb4_eRpCNsL1SKrqVGIlB1NdqIUUmtvAZlnwrlMbzJcYKsM_uAJIZj1fsQalN2DN5molNCnQTdY0HnCemz0Zc8gVQH3xkd9jxeLPSiZA-pgODz0tQevc4HwmZoDkn2hawAFYGtNlATD4-85WQi8ACOpwibp25XSt7iW3i2_VMJcK6ESh84dpOxhZQo3xGFUr_ZIeK6Xf0nWsW1LzLZWxvtQUoQVagYesLs6pyhDoWwNvHYFSuY6glMMw-rDE7DW0Hsa-bBJsdzbheGIFetOXqAbiW-AIVHyVmDUrExr8SKbZv_HcujAh8b89C2UF2ALSiLqy6yoOBGzcQ4xVKjqiiqFIA7E7zNbwovQo5w1VGxGPgRvkgCoyVIMbxNROpyHaDYZg_7CkD2SD5KzeUPCczB9iM2tap_B_DHHPUWoo7FecG31t_0RKJVEWLHKa-PXfP3KyetDLUMq5_LMRqWDgueMA3mZKFQgmgA495Fqs1sNstsnj7yI25TNTaMtS1Btv8mCxbtlJAM0Mp1bZd3dHr77ZGGx5DRbglRDtL8_b-nyiku6yaeC9XI0_Kdmv9KtCIjpcCvmcPDOUmzzB6_NdmksSfnRz6ANIobQZPQ-uAEzoMRbf1iDog-c8wikN0Zo5BO_6paxLa7jk9uMf6FmZH1dKV1aYl2O1GfNoP5RwP48bY-1VWmHISnbnhGccyFnNl9o0d5o7XEVIqqH_fWFTucB2hw9Wyqu8PvChYNIV7wQuyU02bKxZaaC4Zn6YG5jzIZ0ecNl59w-wbK8Up1xFS9YsCmwyHDatkcGL35SZxSoVECtWesKVdaH9VCRcxNtmdVInfB8bgLRbbByzuAzxASdez-81OIqnEmJqCrp1OSd6UDqJvtwjwZJNjC_-aqGuVgntLGtmHXOu0A_iggj6Z9voG1foI6xBpNnnXfoVJV4gauRaeN5jPFdFRHI5BG6P5erS88UPkCK7p9QG3-TlpgqReSVVtGsxWuF1RjZ8aKzI7nvsWDvFTH-De5ZKO8Db8xVDc8eEDbrBEFB_7Qq4Mm-Mt1ynXZSbaBmjDX-hjyXVdGELA1w2aoFVJC_5_JM7vihoxBR9SnszuhxWT1YbOfSptUBULXyP9qBOtQa-5SPCX1pu1MY2yJY7SNOrjY8L1ahpN5Tue43mmJjsZR_1M59CJcpqZG8HxinUBAiZU6ZH75UY_4DrRhsuo_xa9Rv2j7sX58afuxXHv-Lx70uuedM860S7qX158Oju-7F50T896vZPz3ul7J_pHKbR69ql3eXx6eXV6dXLaO7m6OiFzf9A7e-P7v4F-r1U)

ACE Framework-Applied Flowchart:

flowchart TB
    subgraph AGI ["Artificial General Intelligence"]
        subgraph AspirationalLayer ["Aspirational Layer"]
            MoralsValues["Morals & Values -> Ethics"]
            Ethics["Local Ethics Alignment -> Asimov's Rules"]
            Rules["Asimov's Three Rules of Robotics"]
        end
        subgraph GlobalStrategy ["Global Strategy"]
            Agency[("Agency: Plan & Execute")]
            GoalOriented[("Goal Orientation: Contextual Awareness")]
            NovelScience[("Novel Science: Self-Adaptation")]
        end
        subgraph AgentModel ["Agent Model"]
            CognitiveBehavior["Cognitive Behavior: Feedback Loop"]
            MultipleAgents["Multiple Agents: Management & Flow Coordination"]
            TheoryOfMind["Theory of Mind: Perspective Taking"]
            Feelings["Emotions & Embodiment"]
            Narrative["Narrative & Self-Symbol"]
        end
        subgraph ExecutiveFunction ["Executive Function"]
            RiskManagement["Risk & Resource Management"]
            MemoryManagement["Memory Update & Model Merging"]
            Learning["Online Learning & Skill Acquisition"]
        end
        subgraph CognitiveControl ["Cognitive Control"]
            DecisionMaking["Decision Frameworks"]
            SymbolicReasoning["Symbolic Reasoning & UtilityAI"]
            MindsetAdjust["Mindset Adjustments & Cognitive Refinement"]
            FocusReframe["Focus & Reframing"]
        end
        subgraph TaskProsecution ["Task Prosecution"]
            TaskExecution["Task Execution & Outcome Evaluation"]
            RandomnessCreativity["Creativity & Innovation"]
            GlobalWorkspace["Global Workspace Interaction"]
            SocietyOfMind["Society of Mind: Internal Hierarchies"]
        end
        subgraph InputOutput ["Input/Output"]
            SensorsAPI["Sensors, APIs -> Embodied Interaction"]
            EmbodiedInterfaces["Embodied Interfaces"]
        end
    end

    MoralsValues --> Ethics
    Ethics --> Rules -->|Influence| Agency

    Agency -->|Data & Decisions| GoalOriented
    GoalOriented -->|Feedback| NovelScience
    NovelScience -->|Self-Adaptive Behaviors| CognitiveBehavior

    CognitiveBehavior -->|Cognitive Outputs| MultipleAgents
    MultipleAgents -->|Sub-agent Coordination| TheoryOfMind
    TheoryOfMind -->|Emotional Processing| Feelings
    Feelings -->|Narrative Integration| Narrative

    RiskManagement -->|Risk Assessment| MemoryManagement
    MemoryManagement -->|Memory Dynamics| Learning
    Learning -->|Skill Updates| DecisionMaking

    DecisionMaking -->|Decision Execution| SymbolicReasoning
    SymbolicReasoning -->|Cognitive Processes| MindsetAdjust
    MindsetAdjust -->|Adjustments & Planning| FocusReframe

    TaskExecution -->|Evaluate| RandomnessCreativity
    RandomnessCreativity -->|Innovative Processes| GlobalWorkspace
    GlobalWorkspace -->|Workspace Dynamics| SocietyOfMind

    SensorsAPI -->|Sensor Data| EmbodiedInterfaces
    EmbodiedInterfaces -->|Interactions| AGI

    %% Additional Links for Clarity
    MultipleAgents -->|Spawns & Manages| SocietyOfMind
    SocietyOfMind -->|Feedback & Control| GlobalWorkspace
    GlobalWorkspace -->|Global Decisions| ExecutiveFunction
    AGI -->|Data Integration & Response| InputOutput