Japan, despite a lot of interest back in 5th Generation computer days seems to have a difficult time innovating in advanced software. I am not sure why.
I talked recently, at an academic conference, with the guy who directs robotics research labs within ATR, the primary Japanese government research lab.
He said that at the moment the "powers that be" there are not interested in funding cognitive robotics.
So how do we get you and your team the necessary funding ASAP to complete your work? I don't know the legal issues involved but a bunch of very interested fans of Singularity could quite possibly put together the $5 million or so I think you last said you needed pretty quickly. This was brought up quite some time ago, by me at least, and at the time I think I recall you saying that the right structure wasn't in place to accept such funding. What is that structure and what is in the way of setting it up?
Well, $5M would be great and is a fair estimate of what I think it would take to create Singularity based on further developing the current Novamente technology and design.
However, it is quite likely sensible to take an incremental approach. For instance, if we were able to raise $500K right now, then during the course of a year we could develop rather impressive demonstrations of Novamente proto-AGI technology, which would make raising the rest of the money easier.
The structure is indeed in place to accept such funding: Novamente LLC, which is a Delaware corporation that owns the IP of the Novamente AI Engine, and is currently operating largely as an AI consulting company (with a handful of staff in Brazil, as well as me here in Maryland and Bruce Klein in San Francisco and Ari Heljakka in Finland). However, Novamente LLC is currently paying 2.5 programmers to work full-time toward AGI (not counting the portion of my time that is thus expended). But alas, this is not enough to get us there very fast...
If for some reason a major funding source preferred to fund an AGI project in a nonprofit context, we also have AGIRI, a Delaware nonprofit corporation. I am not committed to doing the Novamente AI Engine in a for-profit context, although that currently seems to me to be the most rational choice. My current feeling is that I would only be willing to take it nonprofit in the context of a very significant donation (say $3M+, not just $500K), because of a fear that follow-up significant nonprofit donations might be difficult to come by, but this attitude may be subject to change.
Bruce Klein has been leading a fundraising effort for nearly a year now with relatively success. To be honest, we are at the point of putting "raising funds explicitly for building AGI" on the backburner now, and focusing on "raising funds for commercial projects that will pay for the development of various components of the AGI, and if they succeed big-time will make us rich enough to pay for development of the AGI in a more direct and focused way." Which is rather frustrating, because if we had a decent amount of funding we could progress much more rapidly and directly toward the end goal of an ethically positive AGI system created based on the Novamente architecture.
The main issue that potential investors/donors seem to have may be summarized in the phrase "perceived technology risk." In other words: We have not been able to convince anyone with a lot of money that there is a reasonable chance we can actually succeed in creating an AGI in less than a couple decades. Potential investors/donors see that we are a team of very smart people with some very sophisticated and complex ideas about AGI, and a strong knowledge of the AI, computer and cognitive science fields -- but they cannot understand the details of the Novamente system (which is not surprising since new Novamente team members take at least 6 months to really "get it"), and thus cannot make any real assessment of our odds of success, so they just assume our odds of success are low.
As an example, in a conversation over dinner with a wealthy individual and potential investor in LA two weeks ago, I was asked:
Him: "But still, I can't understand why you haven't found investment money yet. I mean, it should be obvious to potential investors that, if you succeed, the potential rewards are incredible."
Me: "Yes, that's obvious to everyone."
"So the problem is that no one believes you can really do it."
"Yes. Their estimates of our odds of success are apparently verrrrry low."
"Well, how can I know if you yourself really believe that you can create an AGI in a feasible amount of time. You claim you can create a human-level AI in four years... but how can I believe you? How do I know you're not just making that up in order to get research money to play with?"
My reply was: "Well look, there are two aspects. There's engineering time, and then teaching time. Engineering time is easier to estimate. I'm quite confident that if I could just re-orient the Novamente LLC staff currently working on consulting projects to the AGI project, then we could finish engineering the Novamente system in 2-3 years time. It's complex, and there are still details to be worked out, but it's not THAT big of a software system. The question then is how long does it take to teach the child AI everything it needs to know. Obviously it takes many years to teach a human child enough to make it useful, but we have a lot of cheats at our disposal with Novamente, such as the capability to in some circumstances import large amounts of knowledge directly into its mind (once its conceptual framework is mature enough to accept the knowledge)...."
"Ok, so you really don't know how long it will take!!"
"Well, as I said, I can estimate how long completing the engineering and the initial tuning of the AI algorithms and software infrastructure. I can't really estimate the time required to teach the system, because frankly, teaching a baby AGI is something that has never been done before. There isn't much experience to go on in making precise time estimates."
"So it could take 30 years."
"That seems very unlikely. If it takes 30 years, or even 12 or 15 years, that means something is badly wrong in our theory of how to create AGI. Which is possible of course -- as a scientist I can't claim absolute knowledge -- but in fact I am highly confident in our AI approach. We have been refining the theory and doing preliminary experiments with various software components for a number of years now. We know what we're doing."
"Soo... you claim you will have an artificial baby or child within four years, if you get funding."
"Yes, exactly."
"And how sure are you of this? Will you chop off your left hand if you get the funding and then fail?"
"Sure!" -- grin -- "I've always wanted the chance to experiment with cutting-edge prosthetics...."
"Would you really?"
"Yes, I am very confident we can complete the engineering and make an artificial child and start teaching it. And furthermore -- by the time we're at that point, if the teaching winds up to take longer than hoped, I can assure you raising more funding won't be problematic. At that stage we will clearly have what is by far the most advanced AI system ever created in the world, and plenty of people will be begging to fund us and buy parts of our company...."
"So you'd cut off your hand..."
"I won't offer to cut off my son's head, though...."
"No, of course not.... "
This particular potential investor is "still thinking about it" ... he's currently on vacation and will discuss further when he gets back. Of course this was an unusual conversation due to the amputation theme (and the amount of wine being consumed during the conversation, as it was over dinner rather than in an office setting), but other than that it was pretty standard. Skepticism about AI runs reallllllly deep, it seems.
A number of trusted advisers have suggested to Bruce and I that we give up on trying to raise money explicitly for AGI, and just seek business funding and then play the difficult game of using some of the funding for AGI R&D and some for product dev, and coordinating the two sides to keep everything happy and productive. We may well take this path but obvious it is much more difficult and frustrating than being funded to work directly toward AGI. One thing I have learned during the last 5 years of working on narrow-AI consulting projects is how difficult it is to harmonize AGI R&D with narrow-AI based product development. Such harmonization is not impossible but is a significant challenge unto itself, on top of the already tough challenge of creating AGI...
One business area we have been thinking of is "natural language question answering." PowerSet recently raised $10M in venture funding to create a product in this area. But I doubt their system will be nearly as functional as what could be done with fairly primitive Novamente-based technology, let alone full-fledged Novamente AGI. This is a domain where there is some reasonable hope of harmonizing AGI R&D with product-oriented development, because the desired product functionality genuinely seems to be unachievable with adequate quality by narrow-AI means.
Oops, that was longer than I meant to type, and perhaps not adequately organized. But I suppose you get the idea ;=-)
Anyway, Samantha, if you happen to know any wealthy individuals interested in funding either
* a project to build the world's first AGI, by a team of experienced computer scientists and software engineers, based on a very fully fleshed out design
or
* a project to create the world's best natural language question answering software, via utilizing AGI technology, and incrementally improve the QA software's capability as the AGI gets smarter
then let me know...
I don't want to come across as pessimistic: we WILL build Novamente, eventually, one way or another ... unless of course someone else with more money manages to create their own AGI first, which I doubt will happen but can't be ruled out.... But the process is proving incredibly long, slow and painful compared to what it could be if just ONE wealthy individual or institution would decide that it's worth taking a risk on funding the project...
One thing I am doing now is working on a financial trading project which, if successful, may in a couple years generate enough $$ for me that I can fund Novamente AGI work myself. But this of course is a speculation and also a time-sink ... it would be vastly better if I and the others on the Novamente team could simply focus on completing and then teaching our incipient AGI system...
-- Ben
This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
