Chain of distraction

[fanf]

I was catching up with Dave Farber's Interesting People list this evening when I read a brief post about Ray Kurzweil's graph of a 100-year generalization of Moore's law. This caught my interest, so I googled for "Kurzweil" to see if I could find what it was referring to. I found the graph as an illustration of the Wikipedia article about "accelerating change" which in turn led me to Kurzweil's long essay on the subject. He says that Moore's Law is just a particularly quantitative instance of a general rule of exponential progress, encompassing not just technology in general, but also biological evolution, and leading to an AI singularity that he expects before 2050. The Kurzweil AI site has lots of interesting articles on the subject, one of which is a log of an on-line chat between Kurzweil and Vernor Vinge, in which VV plugs a collection of his short stories that was published a few years ago, of which I did not remember having a copy. In fact, it has been languishing on my Amazon wish list because it isn't available new in the UK. However, it is available in the US, and the exchange rate is currently very favourable. So I ordered the book along with a few others, including a copy of the Oxford Companion to the Year, which at $80 is two thirds of its price over here, and a history of the atomic clock which is similarly cut-price. They should go well on my shelves next to The Measurement of Time and Calendrical Calculations (aargh, now there's a third edition!). I was lucky enough to be able to buy the latter two books from the Cambridge University Press bookshop which gives a 20% discount to University Card holders.

Comments

[ronebofh.livejournal.com]

I find any time-specific prediction of an AI singularity highly comedic. We don't understand our own intelligence and we're going to come up with the conditions that will allow an artificial intelligence that surpasses ours to arise? That's about as sensible as Creationist clowns who refute evolution by pointing out that life has not spontaneously occurred in a jar of peanut butter.

[fanf]

I'm less sure. One of the problems of AI is that it is continually redefined as the things we don't yet know how to do. Thus, things that were previously the exclusive domain of humans and the topic of AI research become every day engineering. Text-to-speech, speech-to-text, OCR, text classification, translation, navigating and driving a car, playing chess, spotting faces in pictures, etc. etc. In Kurzweil's 40 year timeframe, if trends continue as they have for the last 120 years, we'll be able to buy more artificial computing power than the totality of human brain power for less than $1000 in today's money. At the same time mechanical miniaturization and sensor technologies are also following exponential trends, so even if traditional Google-style approaches to AI don't pay off, it'll be feasible to upload (destructively or not) and emulate a brain.