[ExI] The present and future of AI

[Jason Resch]

On Fri, Jun 12, 2020 at 2:45 PM BillK via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

> On Fri, 12 Jun 2020 at 19:50, Jason Resch via extropy-chat
> <extropy-chat at lists.extropy.org> wrote:
> >
> > I wrote an analysis of the current capabilities of modern AI systems:
> > https://alwaysasking.com/when-will-ai-take-over/
> >
> > I found what some systems are capable of, especially in terms of
> computational creativity, and general purpose learning algorithms, to be
> shocking.
> >
> > For instance, DeepMind has a single AI that has learned to play every
> Atari game better than any human.  GANs have been made that produce
> original art pieces and can take a single photo and envision it as a
> talking head. Google's Duplex can make phone calls and hold conversations
> so well the person at the other end of the line doesn't realize they're
> talking to a machine.
> >
> > I am curious how close members on this list feel we are from achieving
> AGI. Most estimates seem to put it around 2040 plus or minus 10 years. Is
> that the general consensus? Could it happen much sooner? Does anyone here
> believe it will happen much later than that?
> >
> > Jason
> > _______________________________________________
>
>
> Brian Wang has a recent article considering AGI development.
> <
> https://www.nextbigfuture.com/2020/05/time-for-iterations-and-costs-going-from-inferior-agi-to-superhuman-agi.html
> >
>
> Estimates depend on whether you believe that progress will be
> exponential or linear.
> Also whether financial / political / climate / environment / epidemic
> etc. disasters might derail the train of progress.
>
> The military and politicians are beginning to think of superhuman AGI
> in the same way that they think of nuclear weapons. So researchers
> will not be left alone when they appear to be nearing successful
> development.
>
>
>
The difference is it took a meaningful percentage of national electricity
to refine uranium. The tools of AI are open source, and the necessary
equipment available for anyone to purchase on Amazon's or Google's clouds.
The algorithms behind learning, creativity, and understanding are already
known. I think the only thing standing in the way of superhuman AI is the
compute resources and perhaps the data sets necessary for training.

Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20200612/8a2a73e7/attachment.htm>