From: Richard Loosemore (email@example.com)
Date: Wed Aug 02 2006 - 07:42:18 MDT
For me, it is interesting to read press releases like this one from
"Imagination Engines, Incorporated", because there seems to be a very
definite category to which they all belong ... there is something about
the wording, the structure of the claims, etc., that trigger my BS Detector.
For example, take the references to their customers (NASA Langley, etc).
These big outfits (i.e. NASA) have people working for them who like to
get fired up by sexy-sounding stuff going on in little startups, and if
one of those people sets up a little contract, all of a sudden the
little startup has (wow!) NASA as a customer.
Happened at a company I worked for once: they *gave* one of their
machines to a NASA Langley guy, who then publicly enthused about it
wildly for three, maybe four years, whilst privately getting pissed off
with my company because the machine never actually did anything except
show promise. Looking at it from the outside, it would have seemed
amazingly significant that NASA was a customer, but in fact they were
not really a customer, they were a freebie recipient who did voluntary
hype generation for the startup.
I won't dismiss this particular company without knowing more, but I have
to say that from the tone of the press release, and from the content of
their website (which is *so* much like the website of the company I was
associated with, that it makes me want to laugh out loud), I am not sure
I would make the effort to look further.
Okay, so I did look further. The effect he is talking about is trivial.
If I trained a net to encode, say, the sound to spelling mapping in
English words (as indeed I did once, in a series of studies in the early
80s), and if then perturbed the network by introducing random levels of
noise, I would expect the net to spontaneously produce spellings ....
and - Hey Presto! - the spellings would tend to be consistent with
coherently spelled fragments of words and nonwords it had never seen
before. For example, if it was trained to spell "cat" and "mat" and
"art" and "fart" and "dart", then with a bit of random noise it might
spontaneoulsy generate... "cart" and "fat", but it wouldn't tend to
generate things like "ctdt" or "rft".
To an incredibly naive cognitive scientist (oh, there are plenty of
those around, I'll grant you), this effect might be hyped into a machine
that was spontaneously creative. To a bunch of idiotic defense
contractors who have more money that sense, it might look pretty cool.
To me it looks like garbage.
[Sorry: I don't suffer 'em gladly. Personality flaw I suppose].
H C wrote:
> "Summary - A radically new form of neural network based artificial
> intelligence has been conceived that in contrast to preceding forms of
> AI, builds itself and then enters into an intellectual bootstrapping
> process wherein it learns from its own mistakes and successes to create
> useful ideas and strategies. Not only is this technology capable of
> autonomously inventing and discovering new products, services, and
> procedures, as it has for numerous international corporations and
> government agencies, it has also devised several revolutionary neural
> network paradigms. For these reasons and more, AI visionaries and
> futurists, such as Dr. Dennis Bushnell, NASA Langley's chief scientist
> and visionary, have called this technology, known as the "Creativity
> Machine," AI's best bet at creating human to trans-human level
> intelligence in machines. ...This technology is produced and delivered
> exclusively by Imagination Engines, Incorporated as either customer
> tailored or mass produced products. "
> I've never even heard of this company before just now. I don't know how
> technically laughable or not so laughable this approach is to AGI, but
> they are obviously doing something pretty effectively given their
> customers, while at the same time making some pretty extraordinary claims.
> What do you guys think?
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT