AI vs. uploading

Eliezer S. Yudkowsky (
Tue, 03 Aug 1999 03:13:44 -0500 wrote:
> "Eliezer S. Yudkowsky" wrote:
> <<But seriously, when you make a segue of that magnitude, please change the subject line. Or else try to maintain relevance to the topic; i.e. "And I think this demonstrates a disdain for all human beings that is the direct result of being an AI researcher, which is why I don't trust an AI Transcend.>>
> Yes! :-) And that is my point! I have yet to meet a
> serious 'hard' AI researcher who is not lacking in either
> ethics or social skills. So how can you possibly ask me to
> trust the 'sons-a-bitches' to create a benevolent AI?

Let me get this straight. You're using the lousy behavior of humans as an argument *against* AI?

> The
> assholes can barely behave themselves in the most controlled
> social settings, and you want me to hedge my bets and my
> life in the hands of this lot? Gimme a break!

I think you are perhaps overgeneralizing just a tad. In fact, you've just insulted... let's see, Douglas Hofstadter, Douglas Lenat, and... well, actually that exhausts my store of people I'd challenge you to a duel over, but I'm still a touch offended.

> Eli, you want to create an SI before nanotech destroys
> everything. People like Den Otter, Max More and myself want
> to IA ourselves to singularity before the SI's destroy us!

Why are you including Max More on the list? I have him down as an benovelent-Powers neutral in the IA-vs-AI debate, unless he's changed his mind lately.

As for you and den Otter... won't you feel like *such* idiots if the Powers would have been benevolent but you manage to kill yourself with grey goo instead? Well, no, but if that was the case, wouldn't you have felt like idiots if you'd been around to... this language doesn't handle nested subjunctives very well.

Anyway, a touch of IA and you guys will be following me into Externalist-land.

> What makes you think you can close the gap between
> assembler and SI any sooner than we can close the gap
> between SI and uploading? You said it yourself - we are
> fools to try to attempt to beat a 2020 CRNS technology with
> 2040 CRNS technology. But how are you any less the fool to
> try to beat a 2010 CRNS tech with a 2020 CRNS tech? And I'm
> using your CRNS estimates! :-)

Actually, it's 2015 vs. 2020, although Drexler's been talking about 2012, so I'd figured on 2010 for an attempted target date and 2005 for the earliest possible target date.

Am I less of a fool? Yes, I'm ten years less of a fool. I'm the absolute minimum of fool that I can possibly manage.

The whole uploading business is almost impossible to navigate because you aren't just trying to beat faster-developing technologies, you're trying to beat *prerequisite* technologies. Nanotechnology as a prerequisite for uploading, for example. And then with the nanotechnology for computing and finer neuroimaging for cognitive science, building an AI becomes much easier. (Assuming war goo doesn't kill you first, of course.) Anyway, the thing about AI is that between and, it's possible to accelerate it quite a bit. Accelerating uploading is a lot harder.

           Eliezer S. Yudkowsky
Running on BeOS           Typing in Dvorak          Programming with Patterns
Voting for Libertarians   Heading for Singularity   There Is A Better Way