AI or HI: An ML Coursera Adventure

EDIT (2024): My perspective here has obviously evolved in the past 11 years, but I'm leaving this largely unedited since it is worth a good chuckle.

The Before

Since I realized my formal computer science education is practically over, I decided to learn a little something extra this summer. After surveying the many options on Coursera, I settled on Machine Learning with Professor Andrew Ng. And I can’t wait.

I’ve always been fascinated by artificial intelligence, computer vision, natural language processing, and other machine learning applications. Afterall, using computer vision libraries to help me automate video game tasks was what got me into programming! But even then, how could a programmer possibly codify all of the possible paths these problems could take? What does it even mean for a machine to "learn"? Well hopefully some insights to these questions will come over the next couple weeks.

The After

After completing this machine learning course from Stanford on Coursera, I'm even more deeply fascinated and inspired by artificial intelligence, and its implications for the future of software engineering, but I also feel somewhat cheated. From my limited viewpoint thus far, I can overly simplify the current state of machine learning as human intelligence with computer automation, which really isn't fundamentally different from other types of software systems. (That is what information systems are all about anyway). Let me elaborate.

The majority of the course covered various types of linear regression and classification problems. These I had seen before from my studies in statistics and the only interesting programmatic portion of these was the massive amount of data that can be obtained and classified. Here the machine learning portion is really just a pseudonym for 'can acquire lots of information' and doesn't really constitute what I would have previously considered real learning. I claim this type of artificial intelligence really isn't artificial intelligence at all as the onus is entirely on the engineer to do all of the thinking for the algorithm ahead of time, the algo simply solves a fixed equation. People in tech these days frequently throw around terms like 'big data' and 'machine learning' to obfuscate the real value of their platform more often than to justify it, and I can see why. I used to place enormous weight on the term machine learning, and it conjured awestruck visions for the magical possibilities, but the shroud of mystery has been lifted, and I'm sometimes left with cheap (yet still very clever and powerful!) tricks.

Later in the course was where I became amazed at the power and implications of what had just happened. Professor Ng introduced the concept of neural network learning to us, a class of machine learning methods frequently referred to as 'deep learning'. In short, this type of machine learning attempts to mimic the way the human brain learns, and for a relative machine learning newbie, it was mind blowing the types of results you could get. It also had a startling implication for me in that once processing power became fast enough to fully mimic the human brain, why would I even need to program it? What would my job be as a software engineer, if I can just dump all of the data in the world to this thinking machine, and it makes all the necessary connections, algorithmic decisions, and logical realizations on its own?

Then, it dawned on me. Perhaps I didn't really want machines to truly learn at all! I love programming and a world where there's nothing left for me to do is not one I am excited about. Thankfully, there still seems to be an enormous amount of data manipulation, problem constraint, and objective definition to be done by us humans, so until the day comes when Artificial Intelligence simply becomes Human Intelligence, I suppose I'll just keep trying to improve my own.