*This scripts covers exercise 2 of the "Opening up your communication" session to be delivered by TPS Turing team at CDT conference on November 15th 2023*

It is meant to accompany slides 13 - 23 of the [session slide deck](https://docs.google.com/presentation/d/1HGE7DEsRK97ukuV4Ps_qw19x5fjzNj8a/edit#slide=id.g1e6e5e35a94_2_0)

**00:00 - 05:00 | Why should you write a lay summary?**
Now, we're going to talk about lay summaries. A lay summary is a concise summary of a research proposal, usually written for the public rather than researchers or professionals. It's about making complex research comprehensible, and it's a great way to achieve a greater impact for your research.

**When to write it**
You might think that writing a summary is something to do once your PhD is done, but there are opportunities to write lay summaries throughout the research lifecycle. For example, when you come up with an idea, writing a summary can help gain clarity and initial feedback. When submitting a project proposal for funding, a lay summary can effectively communicate the impact of your research to grant reviewers who are often not domain-experts. If you do experimental work you might write a lay summary during participant recruitment. There a lay summary can motivate people to participate in your experiments. After publishing your paper, a lay summary can increase visibility of your research for example through a short blog post or tweetorial.

**05:00 - 07:00 | Create a hook for your lay summary**
Now, let's discuss how to write these summaries.
The first step is to craft a compelling hook. A hook can be the title or the first sentence of an article. It's the first thing people see and is crucial for capturing attention. If you want your research to have impact, you need a good hook for it. Even the most important research findings might be ignored if the writing doesn't engage people.

**07:00 - 10:00 | Warm-up Exercise**
Think about when you have been hooked last time? Did you maybe read a newspaper headline that triggered you to read the article or did you pick up a book just based on the cover? Or maybe you saw an advert somewhere that made you stop and look at it longer than you would've?
*Invite people to reflect on hooks they've encountered recently. What made it compelling?*

**10:00 - 15:00 | 4 Specific Hooks**
Here are 4 specific hooks that work particularly well for research content:

1. **The Credibility Hook:** Leverage your achievements and expertise to establish trust. For example by referring to your title, "I'm doing a PhD in area XY, and here's what I found out recently". But also emphasising the amount of work a piece took, can create credibility "This paper was 3 years in the making, finally we can share the results...."
2. **The Curiosity and Surprise Hook:** Pose an intriguing question that sparks the reader's curiosity.
3. **The Counter-Narrative Hook:** Challenge commonly held beliefs to grab attention. For example, "Tests that diagnose diseases are less reliable than you'd expect. Here's why...". This title challenges the idea that medical tests are reliable which counter-intuitive, so you want to keep reading to better understand why.
4. **The Identity Hook:** Resonate with a specific group by addressing shared experiences or interests. Example: "I recently had a newborn which is why I cared about this new vaccine...". Here the author identifies as a mother, which might trigger readers who are mothers as well to feel like this text was written just for them.

**15:00 - 20:00 | How to write a good lay summary?**
Let's get into a more technical exercise. When writing a lay summary, think about the contrast between academic writing and lay summary writing. In academic writing, you aim to be precise in describing your research, maintaining a neutral tone, and using specialized vocabulary. In lay summary writing, on the other hand, your goal is clarity. You want to write short paragraphs, use everyday words, and maintain a personal tone using active instead of passive voice. In contrast to the classical academic structure where results are described towards the end of your article a lay audience summary should lead with the findings and then address the five Ws: Who, What, Where, When, and Why (now?)

Why now? is an interesting question. It means you should think about why your research matters at this point in time. Can you relate it to recent events or current ideas or controversies? Embedding your research into current-day narratives can help increase the impact of your lay audience summary.

**20:00 - 35:00 | 15 minute Exercise**
You'll swap abstracts with your neighbors and spend 8-10 minutes identifying jargon, long sentences, and passive voice. Highlight or underscore them. Then, use the remaining 5 minutes to review the notes you got on your abstract and discuss in pairs what you have learned about your text.

**35:00 - 40:00 | Helpful Tools**
Finally, here are two helpful tools for writing summaries: the Hemingway app and First Word. First Word provides a readability score indicating the text's accessibility to a broad audience. The Hemingway app highlights issues in the text, such as hard-to-read sentences, complex words, and passive voice.
*Briefly show the two in web browser to make people curious about it*

We'll share the slides so you can check out these tools later. Now, let's move on to the next exercise.

---

## Materials for exercise:

### Kangaroos
[Campbell, Nora E., and Terry J. Ord. "Wild kangaroos become more social when caring for young and may maintain long-term affiliations with popular individuals." Animal Behaviour (2023).](https://www.sciencedirect.com/science/article/pii/S0003347223001999)

Kangaroos are an iconic group of Australian fauna. Despite considerable research on kangaroo behaviour, key gaps remain in our understanding of their social organization in the wild. In particular, it remains largely unknown whether kangaroos form long-term social bonds and what factors might prompt individuals to associate or dissociate from one another. Over 6 years, we monitored the social affiliations of individually identified eastern grey kangaroos, Macropus giganteus, in a large wild population. We investigated the short-term and long-term relationships of kangaroos and the extent those relationships varied with age, sex and reproductive state. We found evidence that long-term relationships among eastern grey kangaroos are possible, especially between adult females. Those individuals that were more sociable within years were also more likely to establish affiliations across years. Contrary to previous studies, we observed females actively associating with other mothers in the years in which they had young. These data suggest that the fission–fusion dynamics of eastern grey kangaroo social behaviour allow females to modulate their social position with conspecifics according to their current reproductive state. We highlight the adaptive implications of the formation of long-term bonds and the changes in social behaviour observed in females.

Scicomm article ["The social lives of kangaroos are more complex than we thought"](https://theconversation.com/the-social-lives-of-kangaroos-are-more-complex-than-we-thought-213770)

### Large Language Models
[Golan, Tal, et al. "Testing the limits of natural language models for predicting human language judgements." Nature Machine Intelligence (2023): 1-13.](https://www.nature.com/articles/s42256-023-00718-1)

Neural network language models appear to be increasingly aligned with how humans process and generate language, but identifying their weaknesses through adversarial examples is challenging due to the discrete nature of language and the complexity of human language perception. We bypass these limitations by turning the models against each other. We generate controversial sentence pairs where two language models disagree about which sentence is more likely to occur. Considering nine language models (including n-gram, recurrent neural networks and transformers), we created hundreds of controversial sentence pairs through synthetic optimization or by selecting sentences from a corpus. Controversial sentence pairs proved highly effective at revealing model failures and identifying models that aligned most closely with human judgements of which sentence is more likely. The most human-consistent model tested was GPT-2, although experiments also revealed substantial shortcomings in its alignment with human perception.

Scicomm article ["AI models struggle to identify nonsense, says study"](https://phys.org/news/2023-09-ai-struggle-nonsense.html)

### Reinforcement Learning
[Dong, Shi, Benjamin Van Roy, and Zhengyuan Zhou. "Simple agent, complex environment: Efficient reinforcement learning with agent states." The Journal of Machine Learning Research 23.1 (2022): 11627-11680.](https://dl.acm.org/doi/abs/10.5555/3586589.3586844)

We design a simple reinforcement learning (RL) agent that implements an optimistic version of Q-learning and establish through regret analysis that this agent can operate with some level of competence in any environment. While we leverage concepts from the literature on provably efficient RL, we consider a general agent-environment interface and provide a novel agent design and analysis. This level of generality positions our results to inform the design of future agents for operation in complex real environments. We establish that, as time progresses, our agent performs competitively relative to policies that require longer times to evaluate. The time it takes to approach asymptotic performance is polynomial in the complexity of the agent's state representation and the time required to evaluate the best policy that the agent can represent. Notably, there is no dependence on the complexity of the environment. The ultimate per-period performance loss of the agent is bounded by a constant multiple of a measure of distortion introduced by the agent's state representation. This work is the first to establish that an algorithm approaches this asymptotic condition within a tractable time frame.