Ever since I got my first computer and started diving into sci-fi books, the idea of thinking machines has fascinated me. In college, as a computer engineering major, I began dabbling in electronics and took classes on production line automation.
Back in 2009, programming robots at my college meant using C-style languages to specify exactly what needed to be done - what motors and actuators to activate and when. It was all about plugging power cords into sockets and navigating the limited environment of a factory.
Sure, factory automation is crucial and incredibly beneficial, but it wasn’t the stuff of my sci-fi dreams. It didn’t touch on the electronics I was curious about or the deeper software engineering concepts. And it certainly didn’t explore the mind and thinking in the way my favorite philosophy and psychology classes did.
AI, Robotics
AI grabs my attention because it combines so many things I’ve loved for ages: math, software engineering, and theories about the mind. There’s something magical about trying to recreate parts of the mind, like figuring out how to recognize handwritten digits. Once I get a system up and running, it’s pure joy to compare how my brain handles these tasks - like bird species identification from audio - versus how different ML models tackle the same thing.
Robotics, on the other hand, dive into some of my other favorite topics: electronics, physics, and 3D modeling. Robots live in the physical world, so they have to play by the rules of nature. Unlike a video game character, a walking robot can’t just leap over a 10-foot wall unless it’s specifically designed and optimized for that feat. Plus, building your own robot means constantly hunting for parts, soldering and assembling them just right, and sometimes designing and 3D-printing parts when the ones you need don’t exist or aren’t available.
AI + Robotics
But why now? What’s so magical about this moment? I’m not entirely sure, especially since past generations of robotics engineers probably felt the same way. But I believe the time for smart systems has truly arrived. As by Daniel Kahneman’s classification, we’ve spent the last 10-15 years building System 1 (“Fast Thinking”) systems - things like perception, classification, and segmentation. From sorting coffee beans to filtering spam in a split second. What we’ve been missing is some reasoning capacity, System 2 (“Slow Thinking”). Even the simplest form of reasoning could unlock huge automation opportunities.
What changed in 2022-2023 is the development of multi-modal large models. These models are introducing basic reasoning and changing the game. We can finally equip robotic systems with a bit of “thinking.” When “System 1” capabilities fall short, there’s a small reasoning engine, a large multi-modal model, stepping in to provide some “System 2” thinking. Plus, I’ve spent a big chunk of my career working with various ML-specific hardware for inference. Now, the exciting new challenge is bringing all that tech to edge devices.
Conclusion
Of course, these are just my speculations, and I could be completely wrong. But why not give it a try? I learn by playing and building, experimenting with the latest AI + Robotics developments, and seeing how it all unfolds. This blog serves as a record of my journey, tracking my progress every step of the way.