[FRIAM] GPT-3 and the chinese room

[uǝlƃ ↙↙↙]

Much of this is begging for, as Nick pointed out, one's cryptic definition of "computer". Lee posted a nice definition awhile back, one which I think is flawed. But it was nice anyway:

On 3/11/15 3:41 PM, lrudolph at meganet.net posted:> A computation is a process whereby we proceed from
> initially given objects, called inputs, according to a fixed
> set of rules, called a program, procedure, or algorithm,
> through a series of steps and arrive at the end of these
> steps with a final result, called the output. The algorithm,
> as a set of rules proceeding from inputs to output, must
> be precise and definite, with each successive step clearly
> determined. (Soare, 1996, p. 286; definitional emphases
> in the original)

I think it's fairly obvious that a dog is not a computer according to this definition. My primary objection to the definition is the *definite* requirement. And that objection, then, percolates out to the concepts of successive *steps* and any well-foundedness and fixedness of the rules. But none of it really relies on finiteness. I think infinite states are commensurate with this definition. It might be tempting to claim "the end of these steps with a final result" (typically part of the concept of an algorithm) conflicts with infinite states. But I don't think so. As long as we have a way to choose a value (definite or not) from that infinite set, we're good to go ... [ahem] I mean we're good to stop.

I think what we see with things like Transformers, including BERT, is a challenge to the definiteness of computation more than to any stopping or finiteness.


On 7/28/20 9:38 AM, thompnickson2 at gmail.com wrote:
> On 7/28/20 8:20 AM, doug carmichael wrote:
>> dog is highly interconnected - hormones, nerves, senses, and environment. neurons are not binary . every synapse is an infinite state variable. 
> 
> These points might serve as an explanation for why dogs can and computers cannot exhibit joy – but only once we had agreed, up front, what it would be for a computer to exhibit joy.    For my part, I guess, I would say that to exhibit joy, a computer would have to be “embodied” – i.e., be a robot acting in an environment, probably a social environment – and that robot would have to behave joyously.  Or perhaps it could instruct an icon, in a screen environment, to behavior joyously.  But I assume any one of a dozen of the people on this list could design such a robot, or icon, once you and I had done the hard work of defining “joyous.”
> 
> Programmers do this with games, etc., all the time.
> 
> Heider and Simmel did it with a time-lapse camera and a few felt icons on a glass draft deflector.
> 
> Lee Rudolph, if he is still amongst us, can send you a program in netlogo where an icon exhibits joy.


-- 
↙↙↙ uǝlƃ