>Do you then think the notion of AI risk is ridiculous, sufficiently ridiculous that anyone who raises it as an issue has declared their own autism?
Yes. Read an actual textbook on AI and see how the algorithms work. AI is nothing more than heuristic decision making. They're not going to have "thoughts" any time soon in the next couple of centuries.
As I hope you realize, the whole notion of AI risk is based on what AI *could* someday do according to our current understanding of what the inherent limitations are and are not. It has *nothing whatsoever* to do with the abilities of present-day techniques.
>makes a confident prediction when analysis of AI predictions by field experts has literally no consensus.
>Claims AI is nothing more than 'heuristic decision making' - a phrase of nearly equal generality to 'problem optimisation' or 'information processing'
>Claims that heuristic decision making is all it takes without acknowledging that Eurisko, Cyc and Neural Nets have all been high-confidence AI projects utilising those principles that ran into limited/no success.
>Touts heuristic decision making without acknowledgement that decision theory is currently incomplete and thoroughly non-computable apart from its most basic elements.
>Claims to have read an AGI text book but still refers to it by the misnomer of AI
>Implies that EY's logical error is to impute AGI's anthropomorphically with "thoughts" when this could not be further from the truth
>Considers two centuries a buffer zone for the solution of exotic problems in AI Alignment so spacious that he thinks it justifies his dismissal of EY as 'dumb'
>Uses the phrase "AI is nothing more than"
Wow, you are so little threat to humanity that its not even funny. There are lead fume breathing famine victims in the wastelands of the Sudan who stand more of a chance at mocking up an AI than you do. I hope you don't get paid for anything to do with Computers, mate. You are helpdesk material at best.
>our current understanding of what the inherent limitations are and are not
We have no idea what thinking even is let alone how to make computers think.
>>Touts heuristic decision making without acknowledgement that decision theory is currently incomplete and thoroughly non-computable apart from its most basic elements.
Oh great, it's another kid that spends all his time reading Wikipedia articles who only know [math]of[/math] things. Decision theory has nothing to do with making decisions. Read a ToC textbook too.
No, YOU have no idea what thinking even is let alone how to make computers think. That doesn't mean other people don't. There is a lot of AI theory research, developing theories of intelligence in general, possibilities and limitations, mathematical background, the specific workings of *human* intelligence, mathematical analyses of existing AI techniques and the limitations this implies, and how similar topics. This is quite a different field from what you'll find in typical AI textbooks, which focus on the workings of current techniques and what you can do with them. Notions of AI risks are based on the results of this AI theory.
Decision theory has to do with robustly predicting the outcomes of decision making agents. Hence why merely giving your AGI the ability to weigh alternatives according to an heuristic as opposed to framing its decision making process in the architecture of a computable decision theory is the difference between Friendly AGI and getting wiped out of existence.
I would point out your the unjustified nature of your calling me a know-nothing kid considering that you have volunteered next to zero domain content in this argument, but ad hominem seems to be kind of your whole deal.
This
wikipedia page "machine learning"
Disregard the brain fart, I was thinking of Decidability.