September 2023

29 28 Q: So where do you see it really helping in the future? AI is already helping us to discover molecules faster. We may be able to skip a lot of the research and development. The difficulty is starting from scratch. This is where AI is promising. Imagine being able to start with an informal request such as, “I want to make chemical X. I have access to these chemicals at these prices, so make me an initial conceptual design and give me many different ideas.” Once you have something, then we already have very rigorous techniques to optimize things. Q: How did you start your AI project? At Imperial College, we have 3 months to pursue a research project. You are supposed to choose from set projects, but a friend and I wanted to do our own. We built an interdisciplinary proposal. Initially, we were discouraged from doing a project such as this, but we were quite committed to it. Fortunately, there was a professor in the AI group in Chemical Engineering who knew us from previous work done with him. We spoke to him. We said, “Hey, if we specify all of our options as your projects then, if we get you, can we just do our own thing?” Luckily, he said yes. We were very grateful for that. Q; How did you find working in independent research? It was quite an adventure. We became completely obsessed about it for those three months. We called it the “mind virus” because we could not think of anything else during that time. We contacted a bunch of researchers in the field, and we went over to Cambridge to talk to people. That was quite exciting. We knew a bit about machine learning by then, but we learned a lot more by completing the project. Learning how to implement complicated maths involves a lot of failures, headscratching, and searching for answers. But I'm happy about that because then we actually learned how to learn stuff. That was one of the great things about that project. Q: What initially sparked your interest in doing a deep dive into AI? When I was in high school, and I came across a YouTube video channel called 3Blue1Brown by Grant Sanderson. These videos explain what a neural network is and give a beautiful visual explanation. I encourage high school students to watch this themselves. However, it is important to watch his linear algebra series first though. But I would caution them that you can leave that being naively convinced that you know everything. You won’t, but naivete is very useful to make you go further. It’s very exciting and gives you enough energy to keep moving forward in your learning. Q: What are your excitements and concerns for our future with AI? I think that the future is in our hands, and it is bimodal. What we do has great potential in both directions. I am very excited, but I am concerned about the use of AI in mass weapons. I’m also afraid of large language models being used for controlling sentiment on political policies and even simpler models controlling public sentiment on ideas. On the other hand, it brings extreme power to the individual. What used to take 100 people now takes 10, and that group of 10 people can be very agile. Q: Final Thoughts? The real value of these tools is knowing where to use them and when to use them. A lot of these techniques have been around for ages, and they’re being repackaged. Deep learning is sort of a subset of those techniques that has advanced significantly. If you are interested in AI, then learn by doing a project that you're interested in. Then you're going to be much more willing to learn those things. I think that this is what high schoolers need to do. Start with “why” and then that leads into the “how” and the “what”. In many classrooms right now we'll have students learning calculus or precalculus and they're like “why am I doing this and when am I ever going to use this?” If you start by seeing something cool and want to learn how to do that cool thing, then you are going to move through your learning quickly and thoroughly. You will have a reason to learn. Right now, you can learn almost anything with ChatGPT. Suddenly the barriers to learning are way lower because you have a personal tutor. It’s exciting. Here is Sam’s full project on AI for Chemical Process Development. Right now, you can learn almost anything with Chat GPT. Suddenly the barriers to learning are way lower because you have a personal tutor. It’s exciting.