We talked about a lot of technical stuff related to AI and machine learning in this episode.
In brief machine learning is the field of making learning robots. Human beings are still infinitely more adaptable than computers. The dream of AI creators is to make a machine that can truly learn on its own. We are nowhere near that and there are many different methods being tried.
Gareth mentioned reward functions, which is essentially a number in the computers program that tells it what to chase. For example, if you want a robot to complete an obstacle course then you set crossing the finish line as 100 points, and the score slowly increases as it gets closer to the goal.
The short story I mention is by Theodore Sturgeon, he is a really good classic sci-fi novel and short story writer. The story is called ‘When you care, when you love‘. I recommend his short stories, there is a lot of humanity in his writing. He wrote a bit for the original Star Trek as well.
The robot fails I mentioned were from the DARPA robotics competition. From 2012 to 2015 they ran a competition to test robots on a course that simulated a city or battlefield, with the idea to encourage the creation fo robots that can work in dangerous situations like emergency rescue. Opening doors is harder than you think.
The program developed to attack Iran’s nuclear program is Stuxnet. Nobody is 100% sure who made it, apart from the creators obviously.
Cut-up technique is a way of making poetry or lyrics but cutting up text and rearranging it. William S. Burroughs and David Bowie have used it, amongst others.
The viral video about loneliness that went viral.
Forgotify, the site that plays you tracks nobody else has listened to on Spotify.
The Turing Test is a theoretical test to see if an AI can pass as a human. The Voight-Kampff machine is from Blade Runner, where humans have to identify replicants that look just like humans, but don’t have the same emotional reactions.
Hopefully we passed both these tests and will continue to do so.