Hi,
Later (few years), once the background science issues are sorted out, I'll
do the detailed planning/costing for likely route to the first AGI is as
follows:

0) It is big science directed at one big result. It is a 'moon-shot' or
'CERN Collider' or a 'WEBB Telescope' or a 'Human Brain-Project' ... that
kind of size.
1) Duration: 10-15 years Following a 3 year (?) setup process and
preliminary science.
2) Goal: Dog-sized robot with the brain the size of a bee (~500K neurons
equivalent).
3) Testing: Two identical robots. Subject (a) with a traditional
computer-physics chip brain (neuromorphic or Von Neumann). Subject (b) with
a brain-physics chip (not a computer, no software). The differences and
similarities will be compared/contrasted in the same autonomous learning
context. More details of test regime available on request.
4) Project requirements: (i) A chip foundry with 2 parallel prototype
production lines dedicated to making the brain of subject (b). (ii) A robot
fabrication facility to develop and refine, on an ongoing basis, the
embodiment of both subjects. (ii) A large and novel AGI test/certification
facility and its international oversight. Double-blind testing done without
direct human involvement in the test. The test subjects themselves deliver
the result inside the test facility. All humans can do is watch.
5) Process: Software for (a) developed in standard ways (possibly run by a
cloud?). Human assisted evolution of the brain of subject (b) over the
years.
6) Estimated Cost: US$2-$3Billion over the whole period. I'd allow
$5Billion if I was running it and not tell anyone.

The new chips are 3D and large. They can be made in slices by current
foundry tech. Stacked slices clamped to become the final chip. The project
could start after a US $50million preparation phase. The result AGI at
BEE-level certified. There are a lot of other implications but I'll leave
it there.

This is doable, right now. We are actually over-prepared because we've been
getting the science itself wrong for decades and the tech has actually
outstripped the needs of the project to some extent.

It's also what I call 'intrinsically safe' AGI. Because it comes implicitly
'boxed' by hardware limits, like natural general intelligence. By 'safe' I
mean 'as safe as it gets'. Nobody's ever properly considered the real
practical safety landscape of an AGI/Robot that does not use a computer or
software for a brain. It's a very different proposition.

I won't say any more about this. if you are compiling a list of ways ahead:
this is my offering.

A few hundred million a year for a decade? Cheaper than UBER, and you
actually get a truckload of real IP at the end.

Colin



>
> We don't have the social & political structures to even start an
> ambitious single software project with 10000 computer scientists working
> for several decades. And we might not even know what organizational
> structure could make such a project possible.
>
>
> So I don't expect AGI to come soon.
>
>
> --
> Basile STARYNKEVITCH   == http://starynkevitch.net/Basile
> opinions are mine only - les opinions sont seulement miennes
> Bourg La Reine, France
>
>

------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T87761d322a3126b1-M7c580f6cbfe44525d9078961
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to