From: Richard Loosemore (rpwl@lightlink.com)
Date: Fri Apr 28 2006 - 12:10:35 MDT
Bob,
Just a couple of thoughts.
1) You make reference to previous technology, and to technology curves,
but I (unlike people like Kurzweil) place very little credence in those
curves, and on the previous history of predictions. You cite "the long
history of failed technology predictions," but as I am sure you are
aware, for every book that lists the failed predictions there are dozens
of books that catalog the _negative_ predictions that turned out to be
too pessimistic: Arthur C Clarke's estimate for the first moon landing,
the Wright brothers' pessimism a few months before they succeeded, and
(my personal favorite): Lee DeForest getting prosecuted for fraud
because he said that radio transmission across the Atlantic would soon
be possible..... and then later in life, himself dismissing television
as virtually impossible.
2) To me, as someone working in the field, the failure of SHRDLU and
Deep Blue to lead straight to AI is of no importance whatsoever. You
ask: "Who could look at [SHRDLU] and think that general natural
language interaction with computers was not around the corner?" and I
say: are you serious? I admit that as an early student (c. 1980) I was
impressed by SHRDLU, but it took only a couple of years' work in the
area to reach the point where I realized this was a dead end, and all
the other cognitive scientists around me said the same thing in personal
discussion.
Given what I know of the field, I think that what we are waiting for is
a paradigm shift: a change in the way we think about the problem, after
which it will suddenly seem surprising and comical that we all thought
it was so very difficult. I think this could have happened already,
back in the mid-1980s. I think it could happen tomorrow (literally).
My best guess (just because I like the fun of guessing and being right
:-), not because I am fond of exact predictions) is that the
breakthrough will come in the next five years or so.
Overall, you couch your reply in interestingly evasive language. You
try to imagine the event I described as happening, at best, a lot
further away in the future than we think...... but you avoid saying
whether, *when* it actually occurs, it would be qualitatively different
from all previous technology, for the precise reasons I gave.
If I pressed you on this point, would you not agree that since this
technology would change the fundamental nature of the technology
generation process in the most extraordinary way, it would be completely
different from all that came before it?
Richard Loosemore
Bob Seidensticker wrote:
> Richard:
>
> Your Discovery Engine is certainly an interesting idea. Maybe you're right
> and it'll come to pass. However, think of the long road that AI has
> traveled. I'm sure many on this list know of SHRDLU (1970), Terry
> Winograd's natural language project. It was a very simple graphical
> interface, but you interacted with it with real English (Wikipedia has a
> summary and sample dialog: http://en.wikipedia.org/wiki/SHRDLU). As an
> impressionable college student shortly afterwards, this was pretty amazing
> stuff. Who could look at this and think that general natural language
> interaction with computers was not around the corner? And yet, it wasn't.
> 36 years later, it's still a dream.
>
> Deep Blue beat the world chess champion in 1997, but that had been "10 years
> away" for close to 50 years.
>
> Of course, you could respond that we may haggle over *when* we will create
> this amazing technology, but the *if* is no longer a question. Possibly.
> But the when could still be quite far into the future. History has examples
> where inventors overreached -- Babbage's Analytic Engine or Da Vinci's
> submarine, for example. It's hard to envision the sound barrier (that is,
> unexpected and really hard) types of problems that we'll discover on this
> road.
>
> I see innovation taking a huge step with the Industrial Revolution. Lots of
> things changed at that time. Instead of an exponentially-rising Technology
> Change curve through time, I see the graph as a step function -- fairly
> constant and fairly small rate of change before roughly 1800, and fairly
> constant and fairly large rate of change after roughly 1830 or so. If I
> could translate your thoughts into my worldview, you may be envisioning yet
> another big step up perhaps 50 years in the future. (I'm just thinking out
> loud here, so I could be misrepresenting you.)
>
> Given the long history of failed technology predictions (and the bolder, the
> more likely to fail), I remain skeptical. But I applaud the efforts of
> those working to make it happen.
>
> Bob
>
>
> -----Original Message-----
> From: Richard Loosemore [mailto:rpwl@lightlink.com]
> Sent: Friday, April 28, 2006 7:54 AM
> To: sl4@sl4.org
> Subject: Re: Anti-singularity spam.
>
>
> Bob,
>
> Thanks for the thoughtful response.
>
> On some of your points I agree with (too much hype about stuff that turns
> out to be vapornology), but there is a core point that I think you are
> missing, that seems to undermine your main thesis.
>
> All previous technologies provided us with something that did not
> substantially change the nature of the discovery process itself. A better
> steam shovel could never cause any individual creative genius to suddenly
> acquire the ability to invent things at ten, a hundred, or a thousand times
> the rate that she was inventing things before the steam shovel came along.
>
> Now, someone might say that this is what PCs and the internet are so good
> at: they amplify our intelligence in such a way that a given inventor
> really could become brighter, more productive, etc. ...... but actually I
> would still be on your side and argue against such a claim, because I think
> the enhancement (or amplification) is fairly marginal.
>
> BUT, now imagine that I succeed in building the combination of hardware and
> software that I call a "Discovery Engine": a machine that is smart enough
> to understand the world by itself and invent new technology and make new
> discoveries (call it an "AI" if you like: I like to make a distinction
> here, but that is not too important to the present point).
>
> If that discovery engine were to be built tomorrow, it would be "used"
> to build faster versions of itself, and those would be used to build faster
> versions of themselves, and so on in the familiar pattern that is the core
> thesis of the Singularity idea. Within a short time, we would have systems
> that produced new inventions at a rate that would be thousands or millions
> of times faster than unaided human minds can achieve.
>
> A thousand times speedup would involve things like: Einstein wakes up one
> morning with no knowledge of physics, then he does some reading and
> thinking, and by breakfast the next morning he has invented special and
> general relativity. Or, a million discovery engines start working together
> on the problem of nanotechnology, and a year later they deliver a complete
> system for re-engineering the human body to make it immune to disease and
> effectively immortal.
>
> Now *that* is not business as usual. For the first time, the technology
> changes the generator of new technology. Never happened before. Not just
> more of the same, but a whole new ball game.
>
> That is what the Singularity is all about, and why it makes no sense to say
> that new technology is not such a big deal.
>
>
> Richard Loosemore
>
>
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:56 MDT