Two of the world's best chess computers are playing a match at #TCEC with two more of the best chess computers giving their analyses on the side. So we have four super-human chess intelligences, and yet several times per game, each of the four will suggest a different move. Four superhuman chess experts find four different "best" moves in the same position.
This is a valuable lesson for those trying to create superhuman AI. There is no guarantee that superhuman intelligences will always agree, or converge on the same one answer. We will have spent billions to create experts who disagree, and we humans will be too feeble to adjudicate between them.
This is a valuable lesson for those trying to create superhuman AI. There is no guarantee that superhuman intelligences will always agree, or converge on the same one answer. We will have spent billions to create experts who disagree, and we humans will be too feeble to adjudicate between them.