RE: Open Mind Common Sense project

From: Ben Goertzel (
Date: Sun Jan 06 2002 - 11:19:55 MST

> -----Original Message-----
> From: []On Behalf
> Of Eliezer S. Yudkowsky
> Sent: Sunday, January 06, 2002 10:32 AM
> To:
> Subject: Re: Open Mind Common Sense project
> Actually, my opinion on the Open Mind project remains as previously
> asserted: The gathered information of the Open Mind project is useless to
> an AI. Maybe a human research team will be able to do something
> interesting with it, like investigate clustering in human opinion space,
> but to an AI it is raw, opaque, ungrounded, unsemantic data. There is
> thus little point in debating what use the information would be to an AI;
> it only appears as information to *us* because we are already general
> intelligences, capable of human language comprehension, with a
> pre-existing, experientially grounded English lexicon.

My opinion on this is not as extreme as Eliezer's, but is in basically the
same vein.

I think it is possible there will be some stage of development of a "real
AI" at which
this kind of data will be useful.

The stage would be:

1) A decent but not perfect understanding of human NL

2) A strong cognitive capability, honed on domains where the AI has direct

For an AI at this phase of development, the Open Mind / mindpixel kind of
data could
potentially be informative.

The statement

> To an AI the Open
> Mind corpus is not information, it is noise.

is rather oversimplistic (and presumably was intended to for polemical

Of course, to an AI at the stage I described above,
it would *not* be "noise" in the strict sense, it would be a combination of
"noise" and
useful information.

> The entire Open Mind project
> is based on a fundamentally misguided approach to intelligence, Good
> Old-Fashioned AI at its worst,

When I talked to Push, this did not really seem to be the case. He did not
seem to have
any fixed philosophy on AI in mind. Rather, what seemed to be motivating
him was a respect
for *data*. He said that he considered all existing approaches to AI to be
obviously unsatisfactory,
and he hoped that, over time, playing with his data might be a part of the
process of arriving
at a satisfactory approach.

There is certainly a chicken-and-egg problem here, in that by the time it
can understand this commonsense
data, an AI system has got to have a lot of NLP knowledge, but how can it
gain this knowledge without some
commonsense data? But this just means that the OpenMind data is not a magic
pill for commonsense knowledge,
it doesn't imply that it couldn't be part of a valuable approach.

> and its only innovative aspect is that it
> has roped in a large audience of to-be-disappointed Internet volunteers.

Of course, roping in volunteers from the Net is not an innovation, and nor
disappointing them ;>


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT