All it needs is a mapping of every character to a collection of pixels that represent it in a specified font. On the other hand, transforming characters to pixels is a trivial task. The reason is that it is not made up of characters, but of pixels in an image. While it makes virtually no difference to the human eye, machine-written text like the one on the right is completely illegible to a computer. The technological step between Optical Character Recognition (OCR) and Handwritten Text Recognition (HTR) turns out to be equally profound as the one between Deep Blue and AlphaGo. This technology enabled computers to catch up with humans in many hitherto unimaginable areas – one of them being image based text recognition. A way that much more closely resembles inner workings of the human brain. This incredible success came in the wake of a true revolution in the way we see and understand artificial intelligence. Yet, 19 years after Kasparov’s fateful match, people were once again taken aback when Google’s AlphaGo computer defeated the reigning Go champion, Lee Sedol, in 2016. Consequently, even much more powerful computers than Deep Blue had no chance at solving this task. While the rules may seem much simpler than those of chess, Go actually has a game tree complexity that is more than two hundred orders of magnitude larger. AIĪ game that was widely considered to be truly beyond any ordinary computer’s capabilities was the board game Go. While this tree grows exponentially with every move, as soon as a computer can compute 12 or 14 moves ahead in a reasonable timeframe, humans can no longer keep up. It turns out that all it took to defeat humans, was the ability to look at all the possible moves of both players and then pick the best one. But while many used to believe that beating humans at chess would require some form of “true” intelligence, people were quick to call out Deep Blue for what it really was: An efficient way to search tree-like data structures. Back in 1997, people were shocked to see the world chess champion, Garry Kasparov, defeated by IBM’s Deep Blue supercomputer.
But computers are catching up at an increasing rate. And with up to 10 15 synaptic connections, it still outclasses the largest machine learning models in terms of learnable parameters. With up to 10 11 neurons, it contains more electrically excitable cells than a blue whale’s brain. As of today, the human brain is arguably the most complex computational structure in the known universe.