In this talk, Kevin Kau focuses on how to use AI to help researchers analyze and describe the data they collect. The lecture covered two parts: the first was a 50-minute presentation in which Kevin demonstrated how AI can help us analyze data, while the second was a workshop where students could try to use AI to analyze one of two datasets provided by Kevin.

One of the major challenges in conducting research is interpreting what the data we collect actually means, especially when its messy structure and overwhelming volume make this even more difficult. Today's AI tools allow researchers to understand their data much more easily. In fact, Kevin stated that the AI tools we use today are probably the worst AIs we will ever use. AIs will only get better and be integrated into more tools, which means we will have to learn how to best utilize them.

Before showing how AI can help us analyze data, Kevin first discussed the ethics of using AI in research. He emphasized that AI is only an assistant and not meant to replace the researcher. AI can help us perform certain tasks better, but it cannot do everything for us. The research community is mainly concerned about AI replacing researchers themselves, but the use of AI merely as a tool in data processing is accepted.

Kevin then introduced Julius AI, a specialized AI tool designed mainly for data analysis. Although schools may have concerns about the use of ChatGPT, the use of analysis tools such as Julius AI is encouraged. Not only is Julius AI more widely accepted by the research community, it also offers several advantages over ChatGPT, namely that it is easier to use, offers more options for data visualization, and can work with many different file types.

To demonstrate how Julius AI works, Kevin used a dataset on animal bites in the US, giving different prompts to see how it responds. Julius AI could construct different kinds of graphs in just seconds, be it bar graphs, violin plots, boxplots, or pie charts. Adjustments such as changing the color, font size, and scale of the graph can also be made. Beyond making fine adjustments, we can use Julius AI to help us answer specific questions. A dataset can have many different variables and researchers may be particularly interested in correlations between specific variables. For instance, which breeds of dogs bite people the most, or which body parts are most often bitten? We can ask the AI to sort the data for us and show only the variable we need by giving it specific prompts, such as "please do a bar chart of dog bites by breed." Not only is Julius AI practical to use, it also automatically provides the full Python code for everything it does.

The final step after creating all the figures we need is to describe the figures themselves. AI can be our writing assistant to help us formulate our research story, also known as data commentary: explaining what our data means, or what we can conclude from it. Kevin showed that you could download all of the figures created using Julius AI and then compile them into a single file. This file can then be uploaded to an AI, which we can then ask to help us describe the data. Here, Kevin again stressed the importance of using AI only as an assistant; thus, we should not ask the AI to write a full description of our data. Using ChatGPT as an example, Kevin showed that you could instead ask for an outline of noteworthy things about each graph. This outline can be used as a framework to guide us in crafting our research story, with our job being to decide which parts of the outline we should include in our writing.

In the workshop, Kevin provided two datasets for the students to choose from: a simpler dataset about student exam scores and a more complex New York City Squirrel Census. Students were given 30–40 minutes to use Julius AI to analyze the dataset, create figures from it, and see first-hand how the AI responds to different prompts.

In the last part of the talk, Kevin showed his own ideas for analyzing the two datasets, again using Julius AI to create figures, before uploading them to ChatGPT and asking for an outline of the data. Kevin finished his talk by reemphasizing his main message: AI is only an assistant, not the author. It is meant only to help us, not to replace us.