I would build on "The Argument From Wooly Definitions" that the typical premise for superintelligence explosion begins from an AI with (1) "human level" intelligence (2) Machine level attributes like hyperfast computation and a large working memory
I would argue that a "human level" intelligence that can think thousands or millions of times faster than a human with a vastly larger working memory (and, sometimes it is argued, can be trivially scaled to work faster / more efficiently given more hardware) is in fact already a super-intelligence.
If you want to argue for intelligence explosion given a "human level" intelligence, I say start with an AI black box wherein if you give a basic algebra problem to an ordinary human at the same time as you give it to the black box, they will think for about 30 seconds and then give an answer that is half likely to be wrong.
If you want to argue from a "human level" intelligence, you need to acknowledge that throwing more hardware doesn't inherently make it smarter. You can throw five ordinary humans in a room and give them a simple algebra problem, it doesn't mean they will solve it five times faster. Likely they will quickly decide amongst themselves who is best and math, and that one person will then give their answer to the algebra problem, which should be slightly more likely to be the correct answer (and possibly a slightly slower response) than if you just tried with one ordinary human.
I would argue that a "human level" intelligence that can think thousands or millions of times faster than a human with a vastly larger working memory (and, sometimes it is argued, can be trivially scaled to work faster / more efficiently given more hardware) is in fact already a super-intelligence.
If you want to argue for intelligence explosion given a "human level" intelligence, I say start with an AI black box wherein if you give a basic algebra problem to an ordinary human at the same time as you give it to the black box, they will think for about 30 seconds and then give an answer that is half likely to be wrong.
If you want to argue from a "human level" intelligence, you need to acknowledge that throwing more hardware doesn't inherently make it smarter. You can throw five ordinary humans in a room and give them a simple algebra problem, it doesn't mean they will solve it five times faster. Likely they will quickly decide amongst themselves who is best and math, and that one person will then give their answer to the algebra problem, which should be slightly more likely to be the correct answer (and possibly a slightly slower response) than if you just tried with one ordinary human.