4 replies to this topic

[Richard Leis]

I was not sure if here, the blog itself, AGIRI, or elsewhere was a better place for these questions and comments. I just read - and reread - the blog post "A new direction for Novamente" as well as the product sheets. Hopefully Ben or Bruce can confirm that the change in direction is in how Novamente intends to reach the ultimate goal of AGI. The ultimate goal itself has not changed. Illustrating below what I *think* I learned from the post:

Old plan: AI Consulting Work --> invest profits in R&D to develop true AGI

New plan: Intelligent Virtual Agents --> invest profits in R&D to develop true AGI, but sooner

Questions:

1. Ben has been open in his writing about past work and results, including business lessons learned. Does this change in direction indicate a new lesson learned, or is it unrelated to difficulties faced with previous incarnations of the business?

2. Are there - today - working virtual pets and agents in Novamente's "laboratory" that match the capabilities in the product sheets, or are the years provided in the sheets a guess at when development will reach those goals?

3. How will the virtual pets be better and/or different from the bestselling "Nintendogs"?

4. I think Bruce and Ben made a good case for why AGI is important to life extension. Are intelligent virtual agents, in and of themselves, important to life extension, or are they important only because they are a stepping stone to AGI?

[bgoertzel]

Hi Richard,

I'll try to answer your questions both relatively briefly and accurately ;-)

1. About "lessons learned" and business....

As I wrote about a while ago, in the company Webmind Inc. I co-founded in the late 90's, we made some rather profound business mistakes (due to inexperience, and the general confusingness of the dot-com-era business world). In Novamente LLC and Biomind LLC on the other hand, I don't think we've done anything stupid. We've been feeding ourselves, delivering AI-based products and services to customers, and progressing slowly towards general AI. The problem is we just have not been progressing fast enough. And from a business perspective, though we've been putting food on the table and from time to time generating enough excess profit to fund some AGI R&D, we haven't succeeded in scaling up to the point where we can fund AGI R&D on the scale I believe is going to be necessary to get to general AI quickly.

The reason we have resisted throwing ourselves into a vertical-market-focused product-development direction previously, is that it always seemed like it would be a HUGE distraction from our deeper AGI goals.

I didn't mind the idea of throwing my time into applying AI to life extension research because, even if it **is** a distraction from the short path to AGI, at least it's of dramatic value in itself. But this proved infeasible because we failed to attract either adequate investor interest in Biomind LLC, or adequate customer interest in Biomind LLC's machine learning technology for analyzing biological data. Biomind is a successfully operating concern but at a small scale, without enough staff or profit to fund really ambitious AI-for-life-extension work at this stage.

The virtual agents biz model is the first one I've found that doesn't seem to constitute a major distraction from the "short path to powerful AGI." That is what excites me about it, as well as its apparent strong viability as a business play.

2. Do we have virtual pets and agents matching the capability in the product sheets.

Nope. We have virtual agents in the AGISim simulation world, but they don't do everything the product sheets describe. We'll need to put in a bunch of work to get to that point. Actually, our virtual agent R&D work so far has been differently focused -- we've been focusing on starting Novamente agents from **zero knowledge** and teaching them basic stuff. Whereas for the virtual agents product line, we will focus more on having the system improvise and improve on knowledge that it's been supplied with, as a head start...

3. Nintendogs ... Neopets ... etc

These things are not true minds, they don't have long-term memories, and they don't really learn either (though they give a limited appearance of it). Variation in personality between different Nintendogs is slight...

4. Are IVA's important to life extension or just important indirectly as a stepping stone to AGI?

Mainly they are important as a stepping stone to AGI, which will then be critical to life extension.

As a direct example of this, consider bio-NLP. Embodied virtual agents can use their embodiment to help them understand English language better (via "symbol grounding"), which can help them read biomedical texts better, which can help them make new discoveries...

For those like Martine Rothblatt who believe that creating "digital twins" living in virtual worlds is a valid form of "uploading" then of course IVA's are a direct form of life extension. However, I have some quibbles with Martine about "continuity of consciousness" and all that, and am less bullish than her that a digital twin that acts and looks like me would actually be me, unless it was created by a quite precise imitation of my brain-state rather than simply by imitating how I control an avatar..

-- Ben G

[Live Forever]

Is there a general timeframe that you guys are looking at in terms of the coming years and the point in development you would like to be?

(I understand if you are not able to answer due to trying to not give out too much info on your business model or whatever else.)

[Bruce Klein]

Hi Nate,

On time line, as posted back in June 06, this position generally holds true today.

[Live Forever]

Hi Nate,

On time line, as posted back in June 06, this position generally holds true today.

Jeez, I even posted in that thread. How quickly I forget. ) Thanks man.