The big picture on AI

Start of AI: Dartmouth Conf. 1956


“We are at the beginning of the end of the beginning in A.I.” Faustino Gomez, NY Times, 2016


I like that quote. We are coming to the end of Phase 1, the beginning phase of A.I., after 60 years. Beating Go is part of this. Also IBM Watson Jeopardy. AI is beating humans on obscure or fringe activities; things that humans are not really that good at.


Phase 2, the middle phase, will be about everyday activities. Driving, stacking shelves in supermarkets, picking products (these tasks are at the easiest end). End to end order to delivery of goods without a human. Telephone services, dialogue systems. Pretty much all services. Mass unemployment. Eventually reaching human level in vision and language tasks. Robots able to do >95% of current human jobs; that will mark the end of phase 2. 


Phase 3, the final phase, will be about superhuman intelligence.


People might say everyday tasks were already researched in the last 60 years(e.g. driving), but I see a big shift happening now (writing in 2017) in that it is starting to be a major focus: people are getting serious about the everyday applications like autonomous cars or shelf replenishment and order picking (e.g. amazon). Companies are starting to put serious money into it. This marks the transition to phase 2.

Some people did pursue everyday tasks during phase 1, but not with all that much success, and big companies haven’t seen that it is close enough to justify putting big money into it. Instead the money went into things like expert systems or data mining or solving planning/scheduling problems - which humans are not that good at. We are properly in phase 2 when work on everyday tasks is the main thrust of AI work, and they start to be deployed, not in trial, but commercially, e.g. an autonomous taxi.  


Similarly people have been trying for superhuman intelligence in phase 1 too, and this will continue in phase 2, but it’s more of a fringe activity. I expect there will be a shift at the end of phase 2, when all everyday tasks for humans are conquered, then the research effort in AI will shift to superhuman intelligence.


How long will phase 2 last? :) anyone’s guess.

I think longer than phase 1.

Probably by 2100 we have a large proportion of human everyday tasks conquered, but not yet 95% of current human jobs.

Comments

Post a Comment

Popular posts from this blog

Social Media Consumption

Post WW2 technology started to roll out to make available (affordable to masses) food that is tasty for humans, high in fat, sugar, and salt. People eat too much and there are negative effects. Now in the 21st century the technology has made available cognitive stimulation, e.g. videos on social media, that are psychologically ticking boxes that make them compelling viewing. People consume too much and there are negative effects. Companies' goals in both cases are to maximise consumption, not to maximise benefit for the consumer. Technology is good at optimising a given objective. It took longer to develop social media because it is more complex than the physical infrastructure that does farming, and discovering the taste preferences of humans. Cognitive stimulation "taste" is a lot more individualised and has more dimensions, but the algorithms have now solved it. Social media wastes people's time. They need to achieve their goals in the real world instead. Can we ha
Read more

AI is the Bomb

People often liken the effort to develop (strong) artificial intelligence to the effort to develop artificial flight. I think the effort to develop the nuclear bomb might be more close, because - It needs a huge effort to develop AI, something like the Manhattan project, whereas artificial flight was done by a couple of guys from a bike shop. - AI will have a more profound effect on the world than flight. - it feels like a creation of something, playing god. Nuclear: humans create a mini-star on earth, like humans creating/controlling fire. AI: humans create something alive. - AI/nuclear have more of a dangerous/dark/scary side than flight. - Because of fears, with nuclear energy there's a choice to use it or not use it. Strong AI might be like that. For flight there is not much reason not to use it. - Nuclear energy is the fundamental energy source in the non-living universe, the ultimate energy. Intelligence is the ultimate product of the living universe. - The nuclear bomb devel
Read more