AI is imminent and it arrives soon?
Jul. 22nd, 2015 01:17 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
izardRecently some smart people noticed that strong AI will appear soon (e.g. here or here). "Soon" means in 3-4 years, and it will quickly come from academia to the industry.
I tend to disagree with timings. Let's assume some smart-ass sitting in an ivory tower indeed creates a strong AI prototype and publish a paper describing it, in 3 years from now. I recon it will still take at least 10 years after that for a full industry adoption.
In recent 10 years I was very often using fresh academia research to create PoCs/prototypes. The gap between source used in a research paper and a prototype I can show in a trade fair was always very noticeable. Even bigger gap exists between PoC and an alpha version of customer ready solution. Then there is a gap between alpha and proper 2.0 release. Then there is a gap between 2.0 release of a successful but novel software, and wide market adoption. If the technology is trivial, each gap can be filled in months, giving ~2-3 years in total.
If the technology is not so trivial but still not too complex, total goes to 5-7 years. Here is an example: this research paper ignited a SDN revolution in networking in 2007. SDN just became mainstream (around 2014). It took 7 years...
Neural networks/deep learning renaissance happened in 2000s, making a leap forward in weak AI and providing a potentially useful building block for a strong AI. However I think that expecting a breakthrough research in 3 years from now is also quite optimistic. So my guesstimate is ~10 years until a Strong AI described in a paper, then 10 more years for adoption. My baby will be 21 by then, so it is not that far :)
Of course, if AI is immediately used to improve itself and drive market adoption it could help but only a bit as it has to convince humans/IT directors/CEOs :), and they use Gartner hype cycle to estimate technology adoption speed.
I tend to disagree with timings. Let's assume some smart-ass sitting in an ivory tower indeed creates a strong AI prototype and publish a paper describing it, in 3 years from now. I recon it will still take at least 10 years after that for a full industry adoption.
In recent 10 years I was very often using fresh academia research to create PoCs/prototypes. The gap between source used in a research paper and a prototype I can show in a trade fair was always very noticeable. Even bigger gap exists between PoC and an alpha version of customer ready solution. Then there is a gap between alpha and proper 2.0 release. Then there is a gap between 2.0 release of a successful but novel software, and wide market adoption. If the technology is trivial, each gap can be filled in months, giving ~2-3 years in total.
If the technology is not so trivial but still not too complex, total goes to 5-7 years. Here is an example: this research paper ignited a SDN revolution in networking in 2007. SDN just became mainstream (around 2014). It took 7 years...
Neural networks/deep learning renaissance happened in 2000s, making a leap forward in weak AI and providing a potentially useful building block for a strong AI. However I think that expecting a breakthrough research in 3 years from now is also quite optimistic. So my guesstimate is ~10 years until a Strong AI described in a paper, then 10 more years for adoption. My baby will be 21 by then, so it is not that far :)
Of course, if AI is immediately used to improve itself and drive market adoption it could help but only a bit as it has to convince humans/IT directors/CEOs :), and they use Gartner hype cycle to estimate technology adoption speed.
