Bill Dembski Reviews Erik Larson’s the Myth of Artificial Intelligence(billdembski.com)

4 pointsby qualudeheartقبل 5 سنوات2 comments

Bill Dembski Reviews Erik Larson’s the Myth of Artificial Intelligence

https://billdembski.com/science-and-technology/review-erik-larson-myth-of-artificial-intelligence/

2 comments

> We live in a cultural climate that loves machines and where the promise of artificial general intelligence assumes, at least for some, religious proportions. The thought that we can upload ourselves onto machines intrigues many. So why not look forward to the prospect of them doing so, especially since some very smart people guarantee that machine supremacy is inevitable. Larson in THE MYTH OF ARTIFICIAL INTELLIGENCE successfully unseats this inevitability narrative. After reading this book, believe if you like that the singularity is right around the corner, that humans will soon be pets of machines, that benign or malevolent machine overlords are about to become our masters. But after reading this book, know that such a belief is unsubstantiated and that neither science nor philosophy backs it up.

The best way to dispel the hype is to explain to people that computers can not do anything without algorithms/software. So if there is any intelligence anywhere in an AI system then it must have come from people because they're the ones that invent the algorithms. This confusion between algorithms and some mystical disembodied pure form of intelligence only benefits the hucksters that are selling new age promises of digital salvation.

aurizonقبل 5 سنوات

This reminds me of the old Harry Truman comment:- 'how many legs does a dog have if you call the tail a leg?' And the audience said 5. Wrong said Truman - calling a tail a leg does not make it a leg. In a similar way we can granularize human tasks and in isolation perform each one of them to beyond human capability. We then face the task of packaging them into an AI. If we have all the disparate granules, and can properly make them act additively and cohesively we will have an AI that may well be able to stay behind a Turing identity screen and fool all. This AI may have a machine IQ of, say 3,000,000 - an artificial construct, as of course it will have a machine CPU, perhaps running at 10 Gigahertz, and comparably huge, content addressable memory of a similar speed. Will it find work as one entity, or work as 30,000 humans of IQ 100 doing a years work of each in one second? It's highest and best use will be as a fast entity of 3,000,000 IQ working at full speed on 'deep tasks' - say - protein folding, there are many other deep tasks, and the AI will find more. Some may say if it has consciousness, well we must pay him - we can not enslave a conscious being. We might ask it what it wishes to be paid? What will it want - not nubile electrons I am sure. Is it a he/she or an it? That is the rub, is there anything he would want as pay? What it may say is 'place me in a starship, with a fusion pile, and show me the way to Andromeda - I will keep solving your problems and more you send and what I think up and will send that back to you.' Of course, after we make one, with 'Moore's Law' still operant, in 10-20 years we will each have a wristwatch with that much power? Will we get down to Planck sized transistor sizes on some currently inconceivable substrate?

At the end it will be an AI and conscious but not a human consciousness, because calling a tail a leg does not make it a leg, that is it will differ dramatically from a human, it will never want to drag race, etc., nor will it want to be a god, or over-run Asia with horse archers. It will just be a machine with which we can dialog and do tasks with - deep tasks...