The Artificial Future

Around 200 years ago, humanity experienced a dramatic shift in the nature of living quite unlike anything in the millennia before. We went from using simple machines to amplify human work in specific tasks, to creating complex, powered machines to replace the effort of many people at once. These machines have a big problem: they are stupid. We're about to fix that.

Computers are the modern form of simple machines. They can perform a huge number of tasks - performing those tasks better and faster than humans - but it's nothing compared to what's coming next. Computer scientists, engineers, and programmers have been working toward a true artificial intelligence for decades, though the sheer magnitude of the problem was embarrassingly underestimated in the earliest years of computer research. Even very recently it's been anticipated that a true A.I. will only be achieved 30 to 100 years in the future, with some people convinced it's an impossible task. I feel there are good reasons to expect it's on the nearer end of that spectrum.

It's well known in my field that there's a large divide between two broad classes of problems - the things "computers are good at" and the things "people are good at." For example, looking up an entry in a list of billions of entries is easy for a computer, but identifying a bird is hard. Very hard. However, last year a new algorithm was developed that provides computers with a robust set of tools necessary for processing exactly the sorts of problems "people are good at." It's called Deep Learning, and it's an example of the sorts of progress being made to bridge the gap between computer processing and intelligent thinking.

If you have the time for it, I strongly recommend watching the developer's TED talk on the algorithm, so we're all on the same page as to exactly how far we've come and how conceptually little distance there is left to cover. It is precisely these sorts of rapid developments that are likely to see us changing from calling computers dumb, to calling them smart, to calling ourselves dumb. For computers to outdo humans in a field, it needs to be trained once and have its brethren uploaded with the trained data. There are entire fields where a computer would be able to learn more, faster, and more accurately than any human ever could.

I want to leave you with an example that might demonstrate exactly how significant this is. When you want to learn to cook something new, you might read a recipe, or watch a Youtube video. A group of researchers used deep learning to teach a robot to cook, and it also learned how by watching Youtube videos.

Computers are still stupid, just chugging merrily along like we tell them. That might not be the case for much longer.