Re: longevity vs singularity

Bryan Moss (bryan.moss@dial.pipex.com)
Sun, 25 Jul 1999 00:36:38 +0100

Billy Brown wrote:
> As I've pointed out before, uploading is much, much more difficult than
> most of its proponents seem to realize. You are better off concentrating
> on neural interfaces and cybernetic brain enhancments. [...]

I can imagine an argument between a software engineer and a neuroscientist, the software engineer claiming that simulating a brain would be much more difficult than neural interfaces, the neuroscientist responding that we know so little about the brain that neural interfaces would be far more difficult than a mere simulation. Then you have the AI-researcher claiming he just programmed a brain in LISP - it counts the usage of words in a piece of text, compares them to a database, and tells you the subject of the text - it shouldn't take more than a weeks work to program a fully-functional art critic.

den Otter replied:
> I wasn't claiming that uploading would somehow be "simple" though, quite
> on the contrary: exactly because it's (presumably) harder than creating a
> conscious AI, we should put extra effort into it, while at the same time
> strongly discouraging attempts to make conscious machines.

An upload would be a simulation of a brain. In this sense some connectionist systems could be seen as very primitive uploads since they're (apparently) based on actual brain structure. The field of research that is AI (rather than the all-encompassing science fiction concept that is AI) seeks to model thought processes using computers. Thus we have two approaches, simulation of brain structure and the creation of programs that model thought processes. The latter suffers from ambiguity unbound in both the definition of 'thought process' (or mental model, or what have you) and it's ultimate goal. I personally don't buy the idea of a computer scientist sitting down and creating a 'mental model' while paying little or no attention to brain structure; it reminds me of monkeys and typewriters. An unhealthy majority of AI-research is pseudo-science plain and simple. Don't get me wrong, we've seen a few useful algorithms come out of this, but if AI-researchers do find the Holy Grail of intelligent-thought it will be coincidence, like the monkeys and Shakespeare.

> By "uploading technology" I meant the entire spectrum of neural
> interfaces/enhancements etc. which could make our brains more efficient
> and less organic, btw, not (just) a single, "instantaneous" procedure
> that would transfer one's thought into a computer. I'm strongly in favor
> of the gradual approach, as opposed to the scanning method where
> essentially a "copy" is being uploaded.

In my opinion the technology will happen in the following order, with some overlap:

Neural Interfaces
Uploading
Artificial Intelligence

First we start simulating brains, not for uploading but for research purposes. The simulations get increasingly more detailed until at some point we can run them in real time. There are many software and hardware bottlenecks to overcome before this is possible; we're not just waiting on Moore's Law. This is still a long way off uploading. For the brain to communicate with the outside world we need neural interface technology, you can't just hook speech output into a speaker. It may be possible to do a non-destructive brain scan and simulate a particular brain although this would not be as simple as it sounds. All this time we'd be learning how and if we can quantize the brains into more flexible software, so eventually it may be possible to write a 'brain' from scratch although I pity the fool who tries. Making pure uploading feasible will involve putting many brains inside a simulated world, this will take time, and it will move us into the realm of less rigorous fields like psychology. Eventually we'll be able to see if brains can be made smarter, made to run faster, et cetera, but this won't happen anytime soon. Smarter and faster have problems, the first requires a detailed understanding of the brain and the second warrants a discussion on how brains interact. It may turn out that smarter brains are easy and nature simply hit a bottleneck (or a pelvis) but I doubt we should be so lucky.

BM