249 replies to this topic

[jaydfox]

Evidence of flight was existed then since it was observed that various life-forms were capable of flight. The Chinese invented kites, LdV designed a helicopter prototype and there were hot air baloons in the 17th century. What similar lines of evidence for the existence of an AGI is there today?

Kites and hot air balloons were not considered relevant to the question of whether powered flight was possible. You don't consider scaleable narrow AI relevant to the question of whether AGI is possible.

[jaydfox]

I do believe it may be time for a referendum.

A referendum for what? To officially declare AGI off-limits for official institute support? Pass that and kiss me goodbye. What the hell are you doing, Harold?

[Bruce Klein]

Novamente, AGIRI and SingIst are sufficient to carry the AGI/singularity flame, don't you think?

Depends on how hot a flame you want...

[Richard Leis]

Quote prometheus:

There is not a shred of evidence of an AGI in existence today.

There is no evidence for an AGI, but there is evidence on this planet of creatures of intelligence, created by evolution over a few billion years. Certainly a focused effort by technologically-adept intelligent creatures can accomplish the same thing as evolution, but in far less time. Thinking may be a more difficult engineering feat than flying, but it is still only an engineering feat.

[John Schloendorn]

I am totally glad AGI is being debated here, as an extremely ambitious shot at curing aging and all other ills. I would not know how to efficiently keep a (very crude) awareness of the field if not via imminst.

[caston]

I just realised that I'm almost certain I met Ben Goertzel back in the mid nineties. He did a talk at our house in Perth WA (in front of a Mensa crowd) about distributed AI. I was 15 at the time and obsessed with computers and my dial-up connection.

Funny how life loops around like that.

[Bruce Klein]

Very likely... Ben lived in Perth for a while, from 1995 to 1997: http://www.goertzel.org/bio.htm

Here's the full 10 min version (also corrected above):

http://www.agiri.org/ben_agi.htm

[Richard Leis]

Quote prometheus:

Now we are talking. How would you propose that such a thing may be accomplished?

I am not sufficiently educated or confident to offer my own proposal. Instead, while educating myself, I continue to review proposals such as Novamente with Carl Sagan's mix of "skeptical thinking and an aptitude for wonder"

That said, I think I can give my opinion as to a couple directions that sound interesting. Reverse engineering and simulation of the brain is one approach. Manipulating the brain through genetics is another. It is more difficult for me to judge Novamente's approach, but I would not discount an active scientist with credentials like Dr. Ben Goertzel as I would someone like Arthur T. Murray

I am much more willing to argue that nothing in the physical universe appears to preclude the building of thinking machines. Pretty much anything that the universe can create through physics and the other sciences, humans should be able to do as well, for better or worse. If "dumb" physical processes like evolution can result in brains, then "smart" technological progress by humans can result in thinking machines.

I guess I would argue further that just about anything is possible when science and technology are brought to bear. Evolution created clones, and so did humans. Evolution created flying animals, but humans created rockets. The Big Bang created the universe; humans created cyberspace and the virtual worlds. The universe did not create telepathy, except perhaps on a quantum particle scale with entanglement, but Kevin Warwick, Cyberkinetics, and others appear to be well on their way toward that end.

On the human side, these feats of engineering only required more knowledge and a better understanding of physical processes, rather than the employment of some ephemeral "spark" borrowed from the supernatural. AGI seems no different.

[Bruce Klein]

In what ways is Novamente's mathematical knowledge model approach superior to traditional representations of neural networks, ie having input, hidden (described by functions such as sigmoidal or gaussian) and output layers?

You said,

QUOTE

Quote: "Regarding knowledge representation, we have chosen an intermediate-level atom network representation which somewhat resembles classic semantic networks but has dynamic aspects that are more similar to neural networks. This enables a breadth of cognitive dynamics, but in a way that utilizes drastically less memory and processing than a more low-level, neural network style approach. The details of the representation have been designed for compatibility with the system’s cognitive algorithms."
http://www.novamente...file/AAAI04.pdf

This did not really answer my question (probably a question more suited to Ben) which is why I thought I would not trouble you further.

Ben can answer this much better, of course. Also, Novamente draws much from the AGI work of our scientific advisor, Dr. Pei Wang who has published Artificial General Intelligence and Classical Neural Network which concludes that CNNs are "too rigid" for a number of reasons...

[caston]

Yes, he had dinner with my family and I before the talk as I recall.

A friend of mine that went on the study molecular genetics was there as well and does remember him. I watched the movie and will show it to my dad a bit later.

Edited by caston, 07 July 2006 - 12:53 PM.

[Bruce Klein]

That's interesting... I'll let Ben know.

[bgoertzel]

Hi all,

I have just returned from a vacation and noticed this interesting thread.

I am happy to answer questions anyone has about AGI, Novamente, and the possible relevance of either of these to life extension.

Now some replies to points raised in prior posts in this thread:

1)
My view is that both conventional "narrow AI" and more ambitious "AGI" have a lot to contribute to biomedical research in general and to life extension research in particular. Even in its very limited current form, AI technology would have contributed a lot more to bioscience already if not for various social/cultural/psychological factors in the biology and CS communities. I have been somewhat active in the application of narrow-AI "machine learning" technology to genomics recently, e.g. my work (together with colleagues at Biomind LLC and the CDC) on using machine learning to get at the possible genetic roots of Chronic Fatigue Syndrome was recently written up in Science News (following a peer-reviewed publication in Pharmacogenomics):

http://www.sciencene...0060701/toc.asp
http://www.sciencene...01/bob10ref.asp

I would like to apply similar methods to look for combinations of genes/proteins/pathways involved in aging, but have not yet found funding for such work (and I note that CFS research funding has recently been cut as well).

2)
I do not claim to have already created a highly generally intelligent software system. If I had done so, I would either be demonstrating it publicly, or else taking it secret and not yakking about it on Forums.... I claim to have a software design that, if fully implemented with all pending details worked out, would lead to a highly generally intelligent software system. It is a big system and a number of years of effort by a high-quality team will be required to get it finished.

3)
Regarding complexity and emergence --- yes, these are important aspects of intelligence, and there is no reason they can't be embodied in appropriately constructed software programs. I am not aiming to program the details of a mind, only to program a framework within which a mind will emerge via interaction with other minds in a shared environment. I spent many years doing academic research in the "complex self-organizing systems" vein before shifting focus to AGI and applied AI...

4)
Regarding whether AI and AGI are somehow a distraction of ImmInst's focus --- well, I'm not an ImmInst leader, but I don't see why they would be considered a distraction. I would like to note that I spend a fair bit of time dealing with biology researchers (wearing my Biomind LLC) hat, and in this context I am very aware that nearly all biologists including most aging researchers consider work on radical life extension to be a bunch of science-fictional blah-blah not worth of serious consideration at present. I am almost certain they are wrong, but I'm just pointing out that the non-AI-enhanced-biomedical-research approach to life extension is not exactly solidly demonstrated and universally accepted as scientifically viable. There is a big intuitive leap from our current understanding of human biology to the hypothesis that huma life can be indefinitely extended via means like pharmacology, genetic engineering etc. So, the fact that there is ALSO a big intuitive leap required to jump from our current knowledge of cognitive and computing science to the hypothesis that powerful AGI is possible in the near term, does not seem a valid reason to reject AGI from ImmInst's scope. It seems to me that both narrow AI and AGI should be considered as valid and potentially important approaches to life extension. Yes this is a speculative statement, from the perspective of conservative "mainstream science" -- but so are ALL approaches to radical life extension. AI and AGI are scientifically and conceptual rigorous lines of research that may rationally be considered to have a significantly greater than zero probability of contributing to human life extension.

5)
My own opinion is that
-- work on life extension from a pharmacological and genetic engineering perspective WILL ALMOST CERTAINLY WORK ... but it may take a long time, for many reasons, including
a) the causes of aging in long-living organisms like humans may include many that don't play a role in shorter-lived model organisms, hence rendering experimentation in this field slow-paced.
b) doing serious research in this domain may be very expensive, and the research funding establishment is very conservative and narrowly-focused, so that outstanding speculative hypotheses such as those of Aubrey de Grey are not being funded the way they should be
-- on the other hand, work on AGI may possibly accelerate progress in life extension among other areas of science, via leading to the creation of minds that are smarter than ours and can better get past the many obstacles in our way... Now, AGI research is also starved of funding. **but**, I think it's intrinsically cheaper than fundamental research into the biology of aging and life extension technologies. I estimate that we could bring the Novamente AI Engine to a level of dramatically impressive AGI with $5M of funding, spent over a 4-8 year period or so; whereas $5M doesn't buy you very much biological research these days....

6)
As for why we have not proceeded, in the Novamente project, by first emulating simpler lifeforms and then progressing to human-level AI: this is simply not the development plan we have taken. Our approach focuses more heavily on cognition than on perception/action (though it also encompasses the latter), which makes a focus on simpler "artificial lifeforms" not right for Novamente. This approach allows one to fuse learning-based and knowledge-based approaches to AI in interesting ways, see e.g. http://www.goertzel....iedAI_June7.htm

7)
For a little more info about the Novamente AGI architecture in particular, see the conference papers on the Novamente website, www.novamente.net These are brief academic conference papers written with a general computer science audience in mind, but they give a general overview...

8)
The Novamente project is currently going pretty slowly due to insufficient funding. If anyone knows any wealthy individuals or organizations potentially interested in investing in AGI, please tell them to contact me or Bruce ;->

9)
To the poster from Perth: Yep, I remember giving a talk to a Mensa group when I lived in WA sometime in the 1995-96 period.... Perth was awesome, probably my favorite place that I've lived.... I would love to move back there one day, if human bodies and physical locations aren't rendered obsolete and irrelevant first...!

-- Ben Goertzel

[bgoertzel]

Prometheus wrote:

No doubt - and don't get me wrong: I'm not anti-AGI. Of course I can see the mind-staggering potential of
such a level of intelligence. But I'm reluctant to view present progress in this area as bringing us closer to such a technology at this stage - not in the same way that progress with genetics and stem cell biology is able to deliver resuts today and in the near future. I don't think software as we know it will cut it. I do agree with one thing, however, and that is that when the paradigm-shifting technology manifests it will happen very rapidly and as I've mentioned before, the government will immediately classify it as a munition.

What frustrates me in this and your other posts in this thread is the double standard you apply to AGI research versus pure biology research.

Genetics and stem cell biology have not cured aging, and in the view of mainstream biology (as well as my own view), they are nowhere near doing so.

IMO, to pick up on a theme you mentioned in regard to AGI, one of the weakness of modern bioscience is precisely the way it ignores complex, self-organizing dynamics. [For instance, I have had a really hard time convincing biologists at various distinguished institutions to think about the effects genes have IN COMBINATION rather than individually... let alone to think in terms of dynamics, attractors, self-organized criticality, etc.]. Maybe aging is a different system-wide "strange attractor" than true health, and the genetics of aging should be viewed in terms of controlling the parameters of the body-wide dynamical system so as to make sure it stays in the basin of the right system-wide attractor. Maybe contemporary geneticists are nowhere near curing aging because they are thinking too much on the single-gene or single-pathway level and not attending system-wide self-organizing dynamics....

So, yeah, genetics and stem cell biology have delivered results. But they have not delivered a cure for aging, or even cancer or any other complex disease involving the coordinated activity of multiple genes. The pharma industry is pretty pissed at genomics right now for NOT delivering the promised miracle drugs. But you are willing to give these scientific fields credit for accomplishments you BELIEVE they will deliver "in the near future."

On the other hand, regarding AI, you take a far more skeptical attitude. That's fine, but it's really just a matter of your own intuition and emphasis. Certainly AI has delivered results analogous to current results from genetics and stem cell biology. Chess playing programs are available on the Gameboy that can probably kick the ass of anyone on this list, and Google can translate Arabic news to English with surprisingly few errors ... and AI programs can find patterns in genomic data that neither humans nor conventional statistics can. These results are not AGi, but nor is current stem cell and genetics research a cure for aging or any other major, complex disease. Why do you give bio research credit for hypothetical "near future" accomplishments (which are not accepted by the mainstream research community), yet you do not give AI credit for such hypothetical "near future" accomplishments....

My own intuition happens to be precisely opposite yours. I think that, due to the above-mentioned lack of focus on dynamics and self-organization, bioscience is going to struggle during the next couple decades. Until eventually they start gathering more time-series data; and working hard to integrate advanced data analysis, simulation modeling and experimentation. On the other hand, I think that AGI has only to overcome some significantly more modest funding hurdles in order to progress very rapidly. Perhaps these funding hurdles will prove insurmountable and so AGI will stall due to society's unwillingness to allocate resources to it, but I'm trying hard to ensure this is not the case...

-- Ben

[doug123]

my quad G5

Wow, until now I thought my Athlon 64 X2 2.2 GHz 1 GB DDR 400 was probably the most powerful box on ImmInst... [huh]

[Bruce Klein]

Chapters 1 & 2 (25 pgs) of Ben's new book, The Hidden Pattern, are found in here.

Excerpt: In this book I present some interim results from a quest I've been on for many years - a quest for a coherent conceptual framework making it possible to understand the various aspects of mind and intelligence in a unified way.

[doug123]

I took some basic computer courses in college to learn the really basic MS OS programming stuff etc. -- I am also very familiar with the hardware as messing with computers was a previous hobby of mine. I used to read Anandtech.com, Hardocp.com, theinquirer.net etc.. on an almost daily basis, and used to be a pretty up to date dude on current and upcoming computer technologies. These days, I am so out of touch, it's almost sad. Or maybe this topic is moving so fast and covers so much ground that I can't keep up with all of it. When the topic was one page long, I promised myself to catch up with it within a week. Now this topic is eight pages long! Now I need a month to catch up...I am not familiar with many of the terms, so I would literally have to look many up. I think a pamphlet simplifying some of the arguments into quick points of emphasis spread around engineering departments of major universities might be the most efficient way to spread these ideas....

[Bruce Klein]

Bruce, do you have a copy of the paper that Ben published in Pharmacogenomics? It's one of the few titles I don't have access to and it would be useful for this discussion.

http://www.co-cure.org/Goertzel.pdf

[jaydfox]

Which reminds me: can someone adept in php write a script where we can have a new tag in posts that automatically makes a term hyperlinkable to a wikipedia entry without using the conventional way of manually pasting the hyperlink in? ie [wp]singularity[/wp] in the editor would be equivalent to {URL=http://en.wikipedia.org/wiki/Singularity]singularity[/URL} which would look and act like singularity in the post.

Adding such a bbcode tag would be fairly trivial after the upgrade. Not sure about the current forum software.

[jaydfox]

Adding such a bbcode tag would be fairly trivial after the upgrade.

This reminds me, do we have a wishlist of things to which I keep saying "that should be easy after the upgrade"?

[jaydfox]

Could we make the link have a superscript with "wiki" on it (or somesuch) to identify it as a wikipidea link?

Whatever html code you want, within reason. You can have a "title" tag so that custom text is displayed when you "hover" on the link, like this:

I was discussing the singularity^wiki with some nice strangers I met on a flight from St. Louis to Atlanta two weeks ago. Well, it started off as a discussion of AI, and how powerful computers were going, so it just naturally segued into the singularity^wiki. (By the way, true story.)

The html code for the two versions looks like this:

<a href="http://en.wikipedia.org/wiki/Singularity" title="'Singularity' defined at Wikipedia" target="_blank">singularity</a><sup><i><font color="navy">wiki</font></i></sup>

singularity<sup><a style="text-decoration: none; color: navy;" href="http://en.wikipedia.org/wiki/Singularity" title="'Singularity' defined at Wikipedia" target="_blank"><i>wiki</i></a></sup>

Getting much more fancy would require additional CSS, but that's trivial as well.

You can even go so far as to make a format like this:
Please take a moment to review the [wiki=Rules_of_Go#Ko_.28no_repetition_of_the_same_shape.29]ko rule[/wiki]

Would become:
Please take a moment to review the ko rule^wiki

[caston]

Will the AGI really be conscious and not just appear to be so from the outside looking in and does it really matter?

Would it be safer to run the AGI inside the AGI-sim virtual universe and not let it enter or know about the real world? Perhaps web data that it accesses could be filtered to ensure it isn't tainted/distracted by the huge amount of noise on the web.

Will we need a photonics processor and similar hardware or a even a theoretical quantum computer to achieve more advanced AI?

Biological systems have a lot of in inbuilt water cooling, how will you keep the AGI host machine cool?

Do you think that AGI will have similar traits to someone with Aspergers syndrome and that people with Aspergers syndrome or on the autistic sprectrum will relate well with AGI systems like they do now with standard PCs?

Is the negative public perseption of AI hammered in by the Matrix and Terminator triologies likely to effect government policies on AGI?

Finally, how solid is your risk management plan and have you mitigated the risk of a unabomber copycat?

Edited by caston, 09 July 2006 - 02:26 PM.