Re: Steve Harris, "Return of the Krell Machine" (Singularity Article)

From: Dan Clemmensen (
Date: Sat Jan 19 2002 - 14:07:33 MST

Ben Goertzel wrote:

> Hi,
> I think we largely agree conceptually, we just differ on estimates of
> various quantitative amounts (*how much* improvement can be achieved through
> groupware and non-AI search tools, *how hard* will it be to get to real AI
> without substantially improved such tools, etc.)...

Agreed. It's refreshing to rapidly converge on our differences without
a lot of diversionary rhetoric. Is it worthwhile to attempt to quantify
these estimates on both our parts? I'll try here, but I realize that
I'm operating from little more than gut feeling, and I'll understand it
if you don't wish to play this game..

I've been a systems architect/programmer for 30 years. I've seen only a
modest improvement in software development productivity in this time,
with most improvement at the lowest ( i.e., lexical and syntactical)
levels (better editors, compilers, and languages.) attempts at
higher level improvements (CASE tools, etc.) have mostly failed. However
I find that my productivity has begun improving over the last few years
for two main reasons: the avauilability of open-source software, and the
availability of search engines. Neither of these is actually designed to
enhance software development productivity. What if we now explicitly
try to build tools to build on the open-source base by using tuned
search engines? There are several scattered efforts, but I know of no
integrated effort with this as an explicit goal. I feel that
productivity could easily triple, and that this could lead to (another)
qualitative change in the nature of software development. If this gut
feeling is correct, then I believe rapid feedback will commence as we
apply the techniques to enhancing the techniques.

I feel that this could happen at any time and is likely sometime in the
next four years. ("Likely" means 90% probability.)

> [Dan wrote:]
>>My first little paper on this subject ("Paths to the Singularity," 1996) is
>>lost, darn it. At the time, I thought that we needed new peripherals,
>>but I changed my mind shortly thereafter. A good computer display can
>>present information as fast as a human can accept it, and a
>>joystick/game controller can accept decisions as fast as a human can
>>convey them: If you don't believe this, watch a kid playing a video
>>game. We would all like to believe we think faster than this, but we
>>don't. In any event I don't think the human/computer interface is an
>>important constraint on the human/computer SI. It's a software problem,
>>mostly in the data presentation.
> If I believe you on this, then I'm going to downgrade my estimate of how far
> SI can be pushed via collaborative technologies.
> Because I still suspect there's a pretty low upper limit on how deep
> collaboration can be made via current tech.
> The best we're going to get without very good AI or radical new interfaces,
> in my view, is:
> 1) conferencing tech that lets globally distributed teams work as if they
> were local
> 2) collaborative-filtering-enhanced search tech that does a better job of
> returning info you want, not based on understanding of content, but based on
> clever use of ratings and user info, together with minimal non-AI
> understanding of content...
> I hope I'm pleasantly surprised...
> I'm thinking quite practically & personally about the tech we're
> discussing...
> Personally I would love to see 1) come about, especially if it worked on the
> narrow pipe to Brazil, since that's where most of my collaborators are. But
> I don't believe 1) would double the efficiency of my globally distributed
> team. A factor of 1.5 in productivity would be my guess (of course, we're
> very productive already..). As for 2), I'm not sure how useful it would be
> for obscure topics; I think the main win would be in presenting popular
> topics to new audiences (e.g. I'm a CS guy who wants to learn about genetic
> regulatory networks -- where have other CS guys looked before when wanting
> to learn about such things?). At this point we don't do that much library
> or Net research for our work, so this would speed up our process by an even
> smaller factor....
> Another point is that 1) would make it easier to recruit new people to the
> team....
> All in all I'm feeling maybe a 2x increase in productivity due to advances
> in groupware and search and related technology, for the specific "real AI"
> oriented work I'm involved in. Of course, the increase in productivity
> could be far greater for people working in different domains, e.g. domains
> focused on looking up the research of others, or contacting others widely.
> Our work is based on a team of people with a common idea working closely
> together... we already understand each other about as well as can be done
> without looking into each others' brains... and we use collaborative tools
> (e-mail, phone, CVS, documents with particular naming conventions, etc.)
> that we all are comfortable with. Could things be improved? Yeah, of
> course.... But a factor of 10?? I dunno...
> ben g

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