Superhistory, not superintelligence

AI can be seen as “living in superhistory” instead of “embodying superintelligence.” This view argues that the occasional superhuman performance of ML models is caused by enormous amounts of accumulated experience in a short amount of time. AlphaGo was trained on millennia of gameplay, so its “data age” is orders of magnitude larger than its absolute age. By exapting latent representations, people might be able to also reap the benefits, and learn at a rate of decades per week.



There are no notes linking to this note.