On July 8th, five high school kids showed up to a place in Silicon Valley. I was one of them. We all came from different areas and different backgrounds, but we were united by one mission: to learn about and understand the inner workings of AI. There was Ary who has an avid interest in physics, Clement who introduces himself with “I do ETL and like to write scripts”, Muti whose height is definitely the first thing you notice about him, Thomas who has extreme determination, and me. And there were also Michael and Ying, the camp instructors.
One of the first things that Michael, aka Dr. Z, taught us was the seven step process to creating an AI product. We learned that AI is not just complex algorithms. For example, if you do not evaluate your model, you will not know how you should improve it. Therefore, you can not deploy your model, and then what’s the point if you can’t share what you made with others? Aside from this, the amount of data that you have to collect and label to train a neural network is insane. That part is by far the most tedious and time consuming part of creating an AI product. Here’s some fun statistics: Ary labeled 2892 images, Clement labeled 2377 images, I labeled 2309 images, Thomas labeled 1224 images, Muti labelled 787 images, Michael labeled 764 images, and Ying labeled 201 images.
On the first day, all five of us successfully were able to get an object recognition neural network to run on our computers. We used an online tutorial and an open source neural network called YOLO and within an hour, they managed to get a neural network running. This was a critical moment in their multi-step process. It’s amazing nowadays that anyone who has access to the internet can learn whatever they wish. Information, tutorials, and funny fail videos are just one Google search away. But here’s another thing. You cannot just read information about coding, for example, and expect it to go in your head and stay there. You have to implement it, and practice, because you learn code the best by doing.
One thing that Michael constantly repeated and pretty much engraved in everyone’s minds was to find the deepest ocean. What does this mean, you might ask. You do not learn to swim by reading books about swimming. You find an ocean and figure out how to swim while you are in the ocean. Once you’ve figured out the basics of swimming, only then you might go read a book about swim technique. It’s this idea that you have to put yourself if challenging situations and learn as much as you can from them until that ocean of knowledge does not seem so deep anymore because you have learned what there is about that ocean. Then it is time to find a deeper ocean.
Throughout our time at AI camp, we had the opportunity to visit various companies and speak to Michael’s buddies from Cal Tech and Stanford. All of the people we spoke to were very experienced in their jobs and life. A common theme was that forming good relationships with good people is extremely important. Rather than calling it the buzzword “networking”, everyone at AI Camp calls it what Clement calls it: “friendship is magic”. We also learned that forming strong foundations of trust among the people you work with is critical. But more than just establishing trust, establishing an environment of vulnerability is key. Voicing out concerns and troubles, and asking questions is a necessary part of learning, which brings me to the next point. Similar to the advice Michael gave us about finding the deepest ocean, all of the speakers talked about learning. They said that it is important to find a job that is challenging so that you are constantly learning. And it is important to constantly learn, using the plethora of resources available, because the mindset that you know everything and that there is nothing more to learn is sure to lead to your demise.
Another lesson from the speakers is establishing fundamentals and developing a deep sense of understanding for everything that you learn. Michael started by teaching us basic fundamentals of python: if else statements, loops, functions, etc. By the second week we were evaluating data, manipulating it, and visualizing it with more advanced code. By the third week, we were actually understanding how parts of a neural network works. With code, yes, you can find code on GitHub and just copy and paste it. But you don’t actually learn anything from that. You have to go read line by line to understand what each part of the code is doing so that you can tweak it to your needs. And once you understand lots of code, you will develop an intuition for what to do when there is a problem so that problems you once thought were hard are now easy as pie.
Although I wrote about many different things that I learned at AI Camp, the main and most important one, I think, is that I learned how to learn. I’ve always been a kinesthetic learner, and I think that’s why I’ve always been so drawn to science––because you can do experiments yourself––but the advice from Michael of diving into the deepest ocean is a piece of advice that I will now carry with me for the rest of my life.
If you are curious about what we built at camp, visit this website: https://emotion.ai-camp.org
