Complementing our optimistic post ‘How Artificial Intelligence Progresses by Leaps and Bounds‘ on the incredible advances of AI when applied to rigorous logical situations and combinatorial problems, we have had recently (2016) on an other hand a dramatic result about AI applied to real life conversation.
A Microsoft AI chatbot went mad in less than 24h. The details are in this post from The Verge ‘Twitter taught Microsoft’s AI chatbot to be a racist asshole in less than a day‘.
This shows that AI is not yet ready to learn by itself without a tutor when it comes to the real world. Which also shows that a lot of our filters have to do with sound judgment and de facto social feedback. That will be quite difficult to emulate in the world of AI!
