Re:

Interesting approach ... sure, I can do that... may take a few days
for the answers to come back, of course ... I'll let you know once
they have...

On Tue, Apr 15, 2008 at 9:23 PM, J. Epstein jeeproject@yahoo.com wrote:

an interesting list.. what I suggest is that you contact the list and ask them to each to choose their all-star teams , no cheating . or cross talk. lets map it out

----- Original Message ----
From: Ben Goertzel ben@goertzel.org
To: J. Epstein jeeproject@yahoo.com
Sent: Tuesday, April 15, 2008 6:30:58 PM
Subject: Re:

On Tue, Apr 15, 2008 at 12:43 PM, J. Epstein jeeproject@yahoo.com wrote:

I want to do it.. the motivatin for the scientists. they are the ones that proposed tha accelerator,,

Well, there are a lot more scientists pumped about building a thinking
machine now than in 2002. Memes like "artificial general
intelligence" and "human level AI" are definitely gathering
momentum...

lets assume a brooklyn project, with no holds barred. how much, who and for how long?

Of course to answer that in detail would take me a few hours, rather
than just a casual email (and of course, I can put in those few hours,
if your interest is sufficiently serious to merit that). But for now
I'll give you a quickie answer, just to whet your appetite and assess
the extent to which we're on the same page.

Firstly, the answer to "how much, who and how long" depends on the
assumed organizational structure for the project.

Obviously a true "Manhattan Project" style structure isn't really
appropriate for this project because there's NOT a world war on, so
scientists aren't going to be motivated to abandon their daily lives
and ditch their families and move to an isolated facility in the
desert. (Well, some scientists could be persuaded to do this, but
those wouldn't necessarily be the best ones based on the other
relevant criteria....)

One reasonable structure (not the only one) would be as follows:

1-- a central physical location

2-- a full-time on-site staff of AI scientists

3-- a full-time on-site staff of software engineers

4-- a program giving funding to select academic AI labs to carry out
specific projects determined by the on-site staff

5-- a program funding academic and industry scientists to come for
sabbatical semesters and summers (or partial summers).

6-- an open-source codebase, so as to leverage the enthusiasm we've
built for OpenCog, and the efforts of a large number of scientists and
programmers around the world. This is also important for enthusing
academic AI labs about item 4 above

Points 4-6 would let us rope in some well-known AI scientists who
would be interested to contribute, but not willing to devote all
of their research time.

Regarding how many people. I would say that a staff of about

-- 7-10 full-time on-site scientists

-- 10-15 full-time on-site software engineers (some of these could be
outsourced to Brazil or elsewhere, but I guess that's a minor point)

-- 4-7 external academic labs

-- around 5 visiting scientists, and a few student interns, at any given time

would be about right. Anything substantially larger than that, would
probably degenerate into chaos. On a project like this, the team has
to be small enough that every contributing scientist understands the
essence of what every other contributing scientist is doing.

I guess you can extrapolate the costs from that about as well as I
can. $4M/year is a reasonable estimate if things are not done
exorbitantly, nor super-cheaply.

Regarding timing, I would envision it as a 10-year project to create
an adult human level AI. I would say that in 5 years or less we could
aim to have a human toddler level AI, which would be sufficiently
exciting that getting masses of ongoing funding and scientific support
for the project would be unproblematic.

It is possible things could go even faster than that. And it's of
course possible obstacles will be encountered to make things go more
slowly.

I would focus the project on creating virtual humanlike intelligent
agents in an online virtual world. This allows one to experiment with
embodiment without needing to muck with robotics, and also gives
really wizzy demos to show off...

What shocks me is that no project like this now exists. Not in
Google, not in Microsoft, not in the US government. No one is making
a serious, adequately-funded attempt to attack the AGI problem.

As for who, it's easy to make a list of scientists currently doing
serious work toward human-level AGI, and whose work is sufficiently
inter-compatible to make their participation in a shared project
sensible. The big question is which of these would be interested to
participate in such a project full-time, versus part-time via having
their labs funded. While funding academic labs to work on particular
tasks can be an important ingredient of the project, I feel there has
to be a very strong centralized component if we're really going to
solve the problem and build a human level thinking machine.

Some strong prospects (whom I know personally) would be:

-- Nick Cassimatis, who runs the Human-Level AI Lab at RPI

-- Stan Franklin, a grand old man of AI, who is nearing retirement,
but might well be psyched to spend his retirement as an emeritus
contributor to the Brooklyn Project...

-- Moshe Looks, a former collaborator of mine and a pioneer of
automated program learning, currently at Google

-- Georges Harik, the guy who created Google AdSense (Google's ad
placement engine), who would be psyched but would possibly not leave
Silicon Valley for anything... not sure... [Moshe and George are
unusual in being gifted AI guys and also great programmers and
software engineers...]

-- Martin Pelikan, leader of the MEDAL lab at University of Missouri
St. Louis (who was on Moshe's thesis committee)

-- Eric Baum, author of "What is Thought?", who is a very great AI
researcher, living near NY (mildly independently wealthy so not
currently employed at any university... but he came out of Princeton
and has worked at NEC...)

-- Richard Hudson, the inventor of Word Grammar, one of the most
AI-friendly approaches to computational language understanding

-- William Schuler, a computational linguist from U. Wisconsin, who
has done some really cool work on embodied language learning

-- Deb Roy from MIT if we could get him ... again he's done great
stuff with symbol grounding ... language, robotics, etc..... But he's
got a great gig at MIT so he would need some serious convincing...

-- Wlodek Duch, a Polish neural net guru who is co-organizing a
session on thinking machines with me at the WCCI conference in Hong
Kong this June. I wouldn't take a neural net approach but Wlodek is
brilliant and would bring great insights.

-- Alexei Samsonovich from George Mason U, who led a team on the DARPA
BICA (Biologically Inspired Computing Architectures) program a couple
years back

-- Sam Adams, from IBM, whose Joshua Blue AI architecture bears some
resemblance to my own ... and whose AI work is neglected at IBM...

Well you get the idea. There are others who presented at AGI-08 who
would also be great in this role.... All these folks are gifted AI
scientists who are super-psyched about really building a thinking
machine ... and have ideas close enough that collaborative work on a
common system should be possible.

A couple others who would be great advisors but whose approaches are
too idiosyncratic for them to want to join a collaborative Brooklyn
Project would be

-- Selmer Bringsjord at RPI
-- John Laird at CMU, the creator of SOAR
-- Pei Wang

Roping in Tom Mitchell from CMU as an advisor would be great and
likely, but he seems too married to CMU to actually join such a
project full-time.

Wow, that took about 45 minutes to think/type out ... more than I
intended to spend.

BTW if you want to situate this project at a university, and you're
open to a non-NY location, one possibility is the Krasnow Complex
Systems Institute at George Mason U in the DC area ... where Alexei
Samsonovich is located. Being focused on complex systems they are a
bit more open-minded than most AI departments.

But, New York wouldn't kill me either ... I could be there 3
days/week, it's not a terrible train ride from DC. I'm kinda trapped
in DC for the next 6 years due to sharing child-custody w/ my ex-wife,
alas....

Anyway, this is a fun conversation... plz think it over and decide
how/if you'd like to proceed. If you're sufficiently serious we could
likely find a way to gather a significant subset of the above
individuals (and some others) to discuss the possible Brooklyn Project
with you. For instance, if you could pay airfare/accomodations,
folks would probably fly in for a couple days to have a focused
workshop discussing the Brooklyn Project possibility. It would
certainly be an interesting weekend ;-)

Yours,
Ben

--
Ben Goertzel, PhD
CEO, Novamente LLC and Biomind LLC
Director of Research, SIAI
ben@goertzel.org

"If men cease to believe that they will one day become gods then they
will surely become worms."
-- Henry Miller