The AI in Schools Program
Introducing high school students to programming, data science, and AI
Last year I teamed up with teachers from Wimbledon High School in London to develop a new program to teach year 8 students more about data science and machine learning. This year we delivered the first iteration of the AI in Schools Program we developed. In this blog post, I will outline what the program involved, what I learned, and what I plan to improve on for our next delivery. Click here to read the school’s report on the day.
The program was delivered over several weeks, and the girls conducted the project in teams.
The program began with an AI Build Day during which the girls assembled and programmed Arduino sensors to collect data on temperature, light and movement patterns across the school. Each group was responsible for one data type from one device. After the build day, the devices were placed at strategic locations around the school and began collecting data to be analyzed later.
Next, the girls had three lessons during class time on modelling time series data. These lessons would introduce the key methods they would use to visualize and analyze their data on the final day but used data on London bike hires instead.
The program finished with an AI Code Day, which I attended in person. This was the day when all the collected data and programming skills the girls learned would come together. Below is an outline of what the day involved.
Career journeys into data science
At the start of the day, I gave an introductory talk where I described my career journey from biochemistry into programming. I didn’t initially choose to study programming because I had no experience of it from high school and was unclear what jobs in IT would be like. I chose to study biochemistry at university and was introduced to programming as part of a summer research project.
I taught myself to program using online resources and was supported by my PhD supervisor and the lab I was working in part-time. There was a big shortage of programmers in my field while I was training, so it was ok that I made slow progress initially. I chose to do a PhD in bioinformatics because I loved the freedom and flexibility that came with computational research and wanted to analyse some of the big genomic datasets that were beginning to be produced.
Illustrating how real people move into programming is valuable for helping students picture their own future career trajectory, but expectations need to be managed. Today, many school leavers expect to learn to program on the job or self-teach, but more employers expecting new hires to already be proficient at coding. As a result, stories of career journeys like mine can be counterproductive if they give students an unrealistic expectation of what entry-level programming jobs will look like. Students today have a big advantage if they can gain advanced digital skills in school, and risk falling behind their peers if they rely on teaching themselves later in their careers.
The AI Code Day
The main task of the day was to get the girls to analyze Energy Sparks data the school had collected and do some time-series forecasting of energy use. We also wanted to see the impact of the additional data the girls had gathered on the accuracy of the forecasts. We used Kaggle notebooks to take the girls through three levels of modelling the data:
Level 1: visualizing and modelling the Energy Sparks data using Facebook’s Prophet package.
Level 2: including the Arduino data as a variable in the model and comparing the accuracy of the predictions with the simpler model from Level 1.
Level 3: incorporating a ‘big data’ source — energy usage data from households across London.
Each step invites the students to think critically about the data they’re using and the outputs from the model and aims not just to show the potential of big data and AI but also to highlight some of their shortcomings.
At the end of the day, each team filmed a Flipgrid video explaining what they had found in their investigation and what their recommendations were to the school on how they could cut energy consumption. I outline some of our findings from this work in another blog post:
Logistics
Because we delivered the program during the Covid pandemic, we needed to work out how to run the activity while maintaining appropriate social distancing measures. We ran the event out of the school’s main hall and had each team sit around their own table. I sat at the front of the room during the day and girls contacted me via Microsoft Teams and shared their screens if they got stuck on a programming challenge and their teachers couldn’t fix it quickly.
It was a full day coding challenge, so we broke up the day with talks on careers in data science, project management, and climate change, as well as periodic Mentimeter polls to check how the girls were progressing.
Student feedback
We used Mentimeter, post-activity interviews with some of the girls and Flipgrid videos to assess how effective the program had been, what the girls had learned and how they felt about the program.
- The students learned a lot about programming in Python, how to plot data, and how to interpret their graphs. The program didn’t just teach programming skills. It also taught students how to work as a group, manage time and play to team members’ strengths.
- Their understanding of what AI is changed over the program. For example, some students interviewed had previously thought that AI was computers being able to understand the world. After the course, they now see AI as computers analyzing a lot of data and learning from many different scenarios to try to predict what will happen in the future.
- The students gained realised how important humans are in the design, training and evaluation of AI. The choices people make in all of these steps can make a big difference in the predictions an AI might make.
- The students were surprised by how much energy the school used even when it wasn’t open
- The exercises were difficult, but when students persisted and worked as a team, they overcame these challenges and felt proud about what they had achieved. They were more motivated to do the work because it was working with real data and real-world issues.
What worked well
- Spacing the program out over a series of weeks allowed time for the key concepts to sink in. Even after the final code day, the girls were still integrating what they had learned, and groups that filmed their videos a few days later had developed more insights about their results.
- Kaggle was a great platform for sharing the data and code for the activity with each student group. It also gives the girls their own notebooks for showcasing what they have learned in the future to potential employers.
- Having the students work in groups helped them solve problems amongst themselves and created the opportunity to develop better teamwork and project management skills.
Things to improve for next time
Based on these findings, my priorities moving forward with this engagement work will be to:
- Explain the benefits of careers using advanced digital skills and what makes this career cool — ability to work from anywhere in the world, good job prospects, ability to apply the same skills to many different issues the students might be interested in
- Emphasize that some of my job titles didn’t exist or were very rare when I was in high school. We are likely preparing students for jobs that don’t exist yet, so transferable skills like programming set students up well to pursue their interests and get good jobs in the future
- Describe and introduce more examples of women working in careers using advanced digital skills and give more details of what these jobs involve
- Include a survey of students’ perceptions of careers in data science before and after the program, to see if we have shifted interest in the subject or careers involving these skills
- More clearly segment the notebooks to highlight the skills the girls are learning and where else they could be applied (e.g. “in this section you are looking at the raw data, often called exploratory data analysis. This is useful for gaining initial insights about the data before doing any modelling”)
Things to explore
Getting a group of four students to work on code together can be a bit tricky. I am exploring the use of Glitch to allow the students to work on the code simultaneously on separate laptops and also allow me to jump into their project and read/edit the code live with the students.
LinkedIn has also launched Greykite, which is their own version of Facebook’s Prophet package. It looks like it may do a better job of modelling time series data out of the box, so I will be looking into substituting this for the next iteration.
Next steps
The first iteration of the program received great feedback, but we aim to improve the program even more for the next school year. We also want to scale the program up for other programmers to deliver in high schools across the UK and develop teaching materials that can be delivered without needing a programmer on hand to troubleshoot.
Any more ideas? Please leave them in the comments! 😊
