Kurzweil Claims That the Singularity Will Happen by 2045
futurism.com
futurism.com
Kurzweil would be 97 old by 2045. In the meantime Kurzweil tries desperately to live until Singularity. https://en.wikipedia.org/wiki/Ray_Kurzweil#Health_and_aging
>... "250 supplements, eight to 10 glasses of alkaline water and 10 cups of green tea" every day and drinking several glasses of red wine a week in an effort to "reprogram" his biochemistry.[47] By 2008, he had reduced the number of supplement pills to 150.[25] By 2015 Kurzweil further reduced his daily pill regimen down to 100 pills.
That said, there actually are drugs today (6, in fact) that reliably extend lifespan in model animals by 20%. Of course there hasn’t been enough time to show efficacy in humans but there’s at least a decent chance that even without new discoveries, the average lifespan could be pushed out to late 90s.
> reliably extend lifespan in model animals by 20%.
This is true. The general rule of thumb seems to be that treatments increase the lifespan of animals inversely to their size. Several promising treatments suggest that if fruit fly and smaller have 2X - 3X lifespan increase, then mice have 20-30%, and finally monkeys and humans get just 3-5%
Getting healthy lifespan close to 90 would be a major win even if absolute lifespan would not grow significantly.
This is a religious movement. This is exactly the same as some devout Christian group claiming the 2nd Coming is coming by 2045. No different. No more likely to happen. All the same "cultish thinking and motivations"!!
> "2029 is the consistent date I have predicted for when an AI will pass a valid Turing test and therefore achieve human levels of intelligence."
Assuming he (and the rest of us) are still around in 7 years, we should be able to congratulate him on being right, or criticise him for not being even close.
I think that a couple more advances like GPT-3 that are focused on conversational interaction would be enough to get us there. Ironically, if E2E encryption for chat apps weren't so common, companies like Facebook and Google would probably have enough data for training such an AI already.
The missing step, though, is probably the ability for an AI to model the world based on input it receives. If you tell an AI that you're an only child, and then a minute later you talk about meeting your brother for lunch tomorrow, it won't pick you up on your inconsistency, whereas a human probably would, because they build a world model from the facts they read and notice the contradiction.
I'm not sure how to train an AI to build world models, but perhaps a manually created corpus of "common sense" could be used to bootstrap some sort of adversarial machine learning process fed with untagged data. That's probably even harder than it sounds, but presumably once someone comes up with some clever way to achieve it, everyone will say it's just applying a simple rule and a lot of statistics, so there's no real intelligence in the AI's conversations.
I don't know whether there are/aren't any bottlenecks or ceilings that he's not considering, but it would be strange if the exponential curves broke down just before they reached the scale needed to address climate change.