No, Facebook Did Not Panic and Shut Down an AI Program That Was Getting Dangerously Smart - Gizmodo
If you look up a word in the dictionary, you’ll find it is described using other words. The mathematically-inclined among us (and machines are very much mathematically inclined) might notice the relationship between this property of language and the ability to describe vectors in terms of other vectors. Such linear algebra doesn’t even rise to the level of artificial intelligence, and instead follows a long tradition of developing mathematical analogues for physical phenomenon like the orbits of the planets or the transverse lengths of rectangular corn fields. Algebraic projections of language, or “embeddings,” are merely the first step in building a translation or natural language system.
It should be unsurprising then that AIs “invent” new languages. Their quantitative nature basically requires this. But it goes further. It might be hard for a person to realize that I’ve written word word word word word. You might need to go back and count them carefully to realize I wrote five words. Machines are bad at free-association but good at things like counting. How might a machine project an abstract understanding of language into a stream of words (a “serialization”) in a way that makes use of its own unique cognitive properties? It might negotiate a trade of hats for balls with sentences like “have a ball to me to me to me to me to me,” a phrase that might grate against human sensibilities but be readily understood by a machine as an offer of 5 balls in response to a request for a number of hats.
Most of intelligence is built this way, a pragmatic projection of abstract notions into functional techniques that can survive evolutionary pressure. The greatest insight to be gained from artificial intelligence is not the vast and frightening capabilities of machines but the prosaic origins and workable solutions at the heart of what we call intelligence. That we find AI’s initial stumbles disconcerting rather than endearing suggests we are not yet ready to give up the pedestal on which we have placed our own minds.