> 4) So, the question is not whether DARPA, M$ or Google will "enter the
> AI race" -- they are there. The question is whether they will adopt a
> workable approach and put money behind it. History shows that large
> organizations often fail to do so, even when workable approaches exist,
> allowing the disruptive innovations to be made by smaller organizations
> that are oriented toward taking bigger risks.
History may also show that the deep-pocket guys often steal the
innovations and overtake the original inventors. In fact, it always
happen when they *can*.
If you think more about it, such an important technology as AGI will
not easily fall into your hands without some nasty competition. As
soon as Novamente or some AGI projects show a little bit of real
promise, the AGI approach will almost certainly be copied. M$ has
shown time and again that they always wait for others to do the
exploratory work and then enter the market when the tech is mature.
IBM also stole the PC idea -- they actually had a secret operation to
design their PC.
If we succeed at creating the first AGI, it will not be because anything
"fell into our hands." It will be because we
a) put in the many years of hard thinking to create a working AGI design
b) put in the many years of hard, often tedious, work to implement, tune
and test it
at a time in history slightly BEFORE the point when many others felt
that a) and b) were worthwhile ways to spend their lives.
> 5) I don't want to get into arguments about my own personality and
> motivations, but I don't think anyone who knows me F2F would consider me
> "complacent" ;-) .... In fact I am frustrated at the Novamente
> project's relatively slow progress, and actively trying to solve this
> problem via bringing in funding to hire more staff. I am please to
> observe that our progress is exponentially accelerating, but frustrated
> that the exponent is not larger! I may be "complacent" in the sense
> that I think the Novamente AGI design is workable and doesn't need
> fundamental rethinking, though, if that's what you mean.
I think the Novamente design has many strong points (eg your uncertain
logic seems to be more mathematically rigorous) but I think you still
need to do something special to prepare for the competition scenario.
"Failing to plan is planning to fail" -- Roman proverb. IMO you got
to do something special that makes NM more likely to emerge as the
winner in such a scenario.
What we have done that is special is to articulate, in detail, a
workable AGI design ;-)
> 6) I agree with you that there is more than one workable AGI design.
> But I still think that coming up with a workable AGI design is **hard
> problem**. It sure took me a long time. Once you get beyond the
> various conceptual mistakes that are endemic to the AI and cognitive
> science fields and really understand the nature of the problem, the
> hardest issues are computational resource efficiency, and complexity of
> parametric dependencies. Novamente is not ideal in these regards but
> it's much better than anything I came up with before, and IMO better
> than anything else I've read about.... But of course I don't know the
> details of other proprietary AGI designs.
Your AGI design may be at a more advanced stage than the others, but
this can be overtaken later. That's why I point out that the critical
thing is to recruit and utilize talented people in your organization
-- which is even more important than developing *your* particular
version of AGI.
Let me stress again that there is nothing inherently wrong with your
AGI. I'm talking about an organizational problem.
YKY, firstly, you don't know enough about the Novamente design to know
if there is anything wrong with it ;-)
Secondly: Regarding the Novamente technical organization (which you also
know little about ;-) ... IMO the only problem issue we have is that the
team is not large enough, given the magnitude of the task.
By far the easiest way to recruit MORE talented people to work on
Novamente would be to raise money to pay them! Toward this end we have
been developing a business model for Novamente, oriented toward using
the Novamente AI system to control virtual agents in simulation worlds
(e.g. Second Life, MMOGs, and training simulations such as are used by
government agencies among other potential customers). If we can raise
investment $$ to build a commercial agents-control product based on the
Novamente system, this will enable us to hire more talented people so as
to accelerate our progress. What I like about this business direction
is that it does not entail distraction from the end goal of AGI: what is
needed to provide better and better embodied agents control in sim
worlds, is very close to what is needed for moving toward AGI anyway.
Aside from acquiring $$ to pay people, we of course are open to talented
volunteers who are able to join the NM project without being paid.
Years of experience, however, shows that such people are rather hard to
come by. A deep and difficult project like NM is hard for people to
contribute to on a part-time basis; and not that many people are in a
life-situation enabling them to spend full-time on an AI project without
pay. (Yes, if we were demonstrably just a few small steps away from a
human-level AGI, plenty of people would be willing to quit their jobs,
move back home with mom and dad, and devote 100 hours a week to helping
us finish Novamente. But if we were at that stage, funding would also
be easy to raise from a variety of sources.)
I realize you are in favor of us open-sourcing the Novamente design, but
I continue to have reservations about that approach, for medium-term
AGI-safety related reasons that we have discussed before. I am also
unconvinced that this would lead to dramatically faster progress. As we
have found with the AGISim simulation world project, simply
open-sourcing something does not magically attract talented and
dedicated people to spend a lot of time on it. We have been fortunate
to get some really good volunteers to help with AGISim, but they are
busy people with other fish frying and progress on AGISim has been
pretty slow in spite of these excellent part-time helpers.
-- Ben G
-----
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/?list_id=303