With all the talk about AI lately, we’ve been thinking about it too and have laid some groundwork to do some initial testing. The technology and the potential of it scares most technologically literate individuals more than most anything else in the world at the moment, but anyone that doesn’t invest in it will be left behind.
The premise of our AI work is looking at artificial intelligence from the same page as evolution. Ultimately, evolution is messy and imprecise but over time comes to a successful solution. Most of what we cognitively do as humans is rather messy pattern recognition. Given that, to detect a pattern shouldn’t take millions of iterations and immense processing power, even though modern personal computers can perform over 100,000,000,000 (that’s one hundred billion!) calculations every second.
Since this is currently the case, it suggests the algorithms currently in use are woefully inefficient at their designed tasks. To roughly detect a pattern from a set of variables should be relatively fast and to query a database (memory) shouldn’t take long either. A computer shouldn’t have to work millions of iterations to learn a pattern, either. The human mind can make an association in only several samples. If it couldn’t, none of us would be here contemplating this paradox as we all would’ve died out long ago.
With all this in mind, our belief is that the path to a successful AI likely isn’t through evermore increases in computing power, but in a great refinement of the algorithms running the AI.