Hi,
- what is the REAL reason highly talented AGI research groups keep pushing their deadlines back. E.g. Ben's announced imminent breakthrus several times ... the one fact he mentioned a few years back that made sense is the huge parameter space/degrees of freedom (you have at least 5 to 10 tunable parameters per module) but I wonder about the others he hasnt mentioned (barring the excuses) and even more so for other projects - newcomers might learn from concentrating their thinking on AGI aspects where current projects are weak.
Well, the real reason the Novamente Cognition Engine is taking so damn long to develop is that the design is big and the staff working on it are few.
The project was founded officially in 2001 but for much of the time between 2001 and 2004 there was NOBODY working on it full time. All of us founders had "day jobs", either actual jobs or AI consulting jobs, needed to pay the bills.
For the last couple years there were 2-3 people working on it full-time. Now there are 3 people working on it full-time, and a fourth just came on board, but hasn't come up to speed yet. Plus much-valued part-time efforts from a few others.
But 3-5 full-time people is not enough to make extremely rapid progress on a large-scale software system like Novamente. We need at least double that, just counting core AI stuff (not stuff like prettying up the AGISim sim world, system administration, etc.).
Now you may say: OK, that proves your AI design is too complex, so go make a simpler design that can be completed by a couple good computer scientists in a year or so. Well, I've tried. Novamente is the simplest thing I could come up with that has a prayer of working on networks of contemporary computers. Of the 10 or so major topics covered in the Novamente design document, there are 3-4 that haven't even been touched yet in terms of implementation, and a couple others that have only been handled on a prototype level. And even the aspects that have been implemented still have known shortcomings (relative to what the design specifies), that are being filled in.
Maybe, if we got the staff we need, we would then run into some OTHER obstacle. (Like the, "Oops, we built this whole huge cognitive system, and the theory says it should learn stuff, but in fact it's a complete moron" obstacle ;-) I can't rule that out, though I've certainly done a lot of theory to minimize the odds of it happening. But anyway, what I've said above is the actual reason why our progress has been slow.
-- 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
