Is AI just search through a powerset of language?
2 comments
From a human perspective, I suppose you could say that. Keep in mind though that that still means an infinite search space given that (human) language is either context-free or context-sensitive and therefore capable of generating an infinite number of valid statements.
If you approach intelligence from a more general not-necessarily-human-like level you'd have to accommodate a much larger set of elements to search through. Think about those recent examples of AI Go players that apparently use approaches in their game tactics which often don't seem to make sense to human players.
AGI thinking might indeed appear very alien to us.
If you approach intelligence from a more general not-necessarily-human-like level you'd have to accommodate a much larger set of elements to search through. Think about those recent examples of AI Go players that apparently use approaches in their game tactics which often don't seem to make sense to human players.
AGI thinking might indeed appear very alien to us.
If one defines language as simply a symbol for some thing, than yes, A.I. is reducing the search space. But that's not exactly what happens in the mind. The mind is actually not making the search space smaller, it's remembering how searches are or are not effective. If it needs new knowledge, it remembers it, if it has the wetware to do so.
But here is the thing: when I read Socrates, I can connect with the knowledge that Plato transcribed in a way that is different than what A.I. does because I am human. If an A.I. is to truly be considered to have 'human' intelligence, it must be quasi-human. So I think it's important to discuss A.I. in terms that are closer to something that augments human intelligence rather than something that is independently intelligent. It's a little naive to think of it as a language search, because then you might say that Google's search algorithm is the most intelligent public A.I. agent on Earth. But Watson is way smarter than Google's algorithm, and I'd wager there are even smarter agents at work that we barely understand.
Likewise, language typically describes words that pertain to non science related events such as the President's farewell address. But, even though a linguist could rightly argue that his words are chosen with a certain scientific specificity, the whole of language is scientifically driven in many instances. So good A.I., were it to be as you posit, would reduce say, the amount of molecular activity in a couple of trillion cells to a single act, like walking, for instance, or riding a bike, or dialing a phone, or hammering a nail. All of these things are easily understood by the human mind, but are not so easy for A.I. because of the (importantly) complex activity involved in virtually everything a human does. In fact, I cannot think of anything a human does that isn't extremely complex in terms that would be fully comprehensible by an inorganic agent.
But I think you're fishing a little, and that's cool, because everyone wants to know what A.I. might or might not be, and we all have to learn as we go.