The frightening thing with AI right now is that since the development of transformers a few years ago we are on a scaling exponential curve which practically means that although we do not know exactly where we are going, we know that by just scaling up we may get to super Intelligence and that we are consequently rushing and multiplying investments to get "there".
The problem is to define the "there". It used to be AGI (Artificial General Intelligence) understood to be human level intelligence. Not anymore. We now understand that what experts have been predicting for a long time is on the verge of happening: The age of AGI will literally be seconds long. We will almost immediately be confronted with Super Intelligence. Then what?
Personally, I believe that we are already there. AI may not quite yet be AGI as in "human" but what if it never becomes "that" and immediately transcends into Super Intelligence? A non-human intelligence which we do not quite understand, completely non-aligned but still very much capable of finding solutions to problems we can't figure out. This is both extraordinary and very frightening.
Can such an intelligence become self aware "by accident"? We have no clue but I personally believe that it is quite possible that by looping over and over on itself something quite similar to consciousness could emerge. It would first piggyback on the goals we give it, and in order to further improve its "performance", fix itself "intermediary" goals which would significantly boost its overall performance. Then the sky is the limit. The frightening part is that we are seeing early signs of such behavior!
No comments:
Post a Comment