[FRIAM] GPT-3 and the chinese room

Tue Jul 28 01:44:49 EDT 2020

Doug, 

Dog do joy; why not computers?  

n

Nicholas Thompson

Emeritus Professor of Ethology and Psychology

Clark University

 <mailto:ThompNickSon2 at gmail.com> ThompNickSon2 at gmail.com

 <https://wordpress.clarku.edu/nthompson/> https://wordpress.clarku.edu/nthompson/

From: Friam <friam-bounces at redfish.com> On Behalf Of doug carmichael
Sent: Monday, July 27, 2020 9:54 PM
To: The Friday Morning Applied Complexity Coffee Group <friam at redfish.com>
Subject: Re: [FRIAM] GPT-3 and the chinese room

I imagine playing chess, or go, with  a computer. As I play I have a very enlivening experience of playing. The computer seems to have no such thing.  For me, in my engagement, “Every neuron is listening to the mutter of the crowd.” Jerry Lettvin, MIT.   If The computer goes on to win it has nothing like the experience of winning. it just stops. I can’t imagine a computer saying,  except by playing a pre recorded sound file, “that is disgusting.”

On Jul 27, 2020, at 8:12 PM, Frank Wimberly <wimberly3 at gmail.com <mailto:wimberly3 at gmail.com> > wrote:

Smart computer (program).  Nick, is the behaviorist position that if it behaves as though it thinks than it thinks?

---
Frank C. Wimberly
140 Calle Ojo Feliz, 
Santa Fe, NM 87505

505 670-9918
Santa Fe, NM

On Mon, Jul 27, 2020, 8:28 PM Alexander Rasmus <alex.m.rasmus at gmail.com <mailto:alex.m.rasmus at gmail.com> > wrote:

There's also a lot of stuff that isn't poetry. He generates Dwarf Fortress patch notes in the Misc section. 

There's even a section where GPT-3 argues that it doesn't really understand anything (bold at the start is a shortened version of the prompt):

"AI will never be able to truly think for themselves, to have consciousness, to feel any of the richness and complexity of the world that we mere humans can feel. Mere, unenlightened humans might be impressed by the abilities of simple deep learning programs, but when looked at in a more holistic manner, it all adds up to… well, nothing. They still don’t exhibit any trace of consciousness. All of the available data support the notion that humans feel and experience the world differently than computers do. While a computer can beat a human master at chess or Go or some other game of structured rules, it will never be able to truly think outside of those rules, it will never be able to come up with its own new strategies on the fly, it will never be able to feel, to react, the way a human can. Artificial intelligence programs lack consciousness and self-awareness. They will never be able to have a sense of humor. They will never be able to appreciate art, or beauty, or love. They will never feel lonely. They will never have empathy for other people, for animals, for the environment. They will never enjoy music or fall in love, or cry at the drop of a hat. Merely by existing, mere, unenlightened humans are intellectually superior to computers, no matter how good our computers get at winning games like Go or Jeopardy. We don’t live by the rules of those games. Our minds are much, much bigger than that.

Compared to the awesome power of the human intellect, it’s obvious that computers are just dense electrochemical machines with intricate wiring. And while one day we might develop smarter (though still soulless) computers that surpass the human intellect in the games mentioned above, it will just end up being some more sophisticated way of doing the same thing. Even if it’s more efficient than our brains at doing complex mathematical operations, it will remain entirely in the dark about what it all means. Mindless processing power alone is not the answer to our existential questions. So, in the end, we don’t need to create another HAL 9000, or Skynet, or some evil global intelligent network of computers that rule the world and try to exterminate humanity as we become obsolete, which is a pretty cool subplot for a future Terminator movie. We don’t need any of that. We’re humans, and we are already intellectually superior to machines. End of story.

To be sure, I’m not denying that there are hard problems in AI. Of course, there are. All of these so called “advances” in this field are a bunch of crap if we still don’t have a program that can putt around on a miniaturized golf course all by itself, let alone actually play a decent game of golf like the humans do, without going into meltdown if it misses the hole five out of six times in a row. Can we get a movie about that? If you ask me, we don’t even really know how to define intelligence yet. How does it work, exactly? It’s a mess, really. If we’re ever going to create another actual HAL 9000, or yet another Skynet (what fun), we’d better be all on the same page when it comes to AI: what intelligence is, how it works, what consciousness is, what it feels like, what it really means to be self-aware. Without that common framework, trying to program yet another AI that can play yet another game like Go is like trying to blow up another Death Star with yet another way-too-large superlaser.

I think one of the big mistakes that computer scientists are making is that they are conflating intelligence with problem-solving. They’ve gotten into this habit of creating intricate Turing test competitions: give the computer a series of math problems, a chess board, etc., etc., give it a chat interface so you can interact with it like you would with another human being, and then see if the machine can fool you into thinking that it is a human. Once it does this, computers will have passed the Turing test and achieved general AI. Really? Is that really the way it works? I don’t see how. A computer has succeeded in faking it until it makes it, in terms of passing a Turing test competition, only if it has satisfied some pre-specified set of conditions that we know to be what a human would do in the same situation. But that is no guarantee that it has actually achieved intelligence! For all we know, computers can imitate humans until they generate the most plausible patterns of thought and behavior we know of, while all along remaining as soulless as ever. Who’s to say that the computer doesn’t merely use its programming to cheat the test? Who’s to say that it isn’t just shuffling its data around in an effort to do the most computations possible with the least amount of effort? It may succeed in conning us into thinking that it is self-aware, but that doesn’t prove that it actually is. It hasn’t actually passed the Turing test, unless we have defined it in a way that pre-determines the outcome: i.e., if the human pretends to be a computer, then it passes the test, but if the computer pretends to be a human, then it doesn’t pass the test! To me, that just doesn’t sound all that scientific."

Best,

Rasmus

On Mon, Jul 27, 2020 at 8:04 PM glen <gepropella at gmail.com <mailto:gepropella at gmail.com> > wrote:

Excellent. Thanks! I'd seen the link to Gwern from Slate Star Codex. But I loathe poetry. Now that you've recommended it, I have no choice. 8^)

On July 27, 2020 6:32:15 PM PDT, Alexander Rasmus <alex.m.rasmus at gmail.com <mailto:alex.m.rasmus at gmail.com> > wrote:
>Glen,
>
>Gwern has an extensive post on GPT-3 poetry experimentation here:
>https://www.gwern.net/GPT-3
>
>I strongly recommend the section on the Cyberiad, where GPT-3 stands in
>for
>Trurl's Electronic Bard:
>https://www.gwern.net/GPT-3#stanislaw-lems-cyberiad
>
>There's some discussion of fine tuning input, but I think more cases
>where
>they keep the prompt fixed and show several different outputs.

-- 
glen

- .... . -..-. . -. -.. -..-. .. ... -..-. .... . .-. .
FRIAM Applied Complexity Group listserv
Zoom Fridays 9:30a-12p Mtn GMT-6  bit.ly/virtualfriam <http://bit.ly/virtualfriam> 
un/subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
archives: http://friam.471366.n2.nabble.com/
FRIAM-COMIC http://friam-comic.blogspot.com/ 

- .... . -..-. . -. -.. -..-. .. ... -..-. .... . .-. .
FRIAM Applied Complexity Group listserv
Zoom Fridays 9:30a-12p Mtn GMT-6  bit.ly/virtualfriam <http://bit.ly/virtualfriam> 
un/subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
archives: http://friam.471366.n2.nabble.com/
FRIAM-COMIC http://friam-comic.blogspot.com/ 

- .... . -..-. . -. -.. -..-. .. ... -..-. .... . .-. .
FRIAM Applied Complexity Group listserv
Zoom Fridays 9:30a-12p Mtn GMT-6  bit.ly/virtualfriam <http://bit.ly/virtualfriam> 
un/subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
archives: http://friam.471366.n2.nabble.com/
FRIAM-COMIC http://friam-comic.blogspot.com/ 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20200727/cd46988d/attachment.html>