[Recording Start]
Scott Douglas Jacobsen: So, you’ve been thinking about the Turing test since we last talked. It’s been a little bit. I’ve been at competition level equestrianism for show jumping and eventing.
Rick Rosner: Yeah, you’ve been tending to fancy horses and people.
Jacobsen: Correct, this is a whole other side of things we can talk about another time. So, you’ve been thinking about the Turing test in the interim.
Rosner: I just quick looked it up, it comes from 1950 and World War II was the Heyday for pragmatic genius; all its glory and also ignominy. You’ve got these guys who are great theoreticians Feynman, Oppenheimer, and Turing who had to become super pragmatic to solve the problems of World War II. Turing basically won the war for the Allies by decoding the German Enigma machine which is a great feat of practical code breaking of figuring out what technology could be brought to bear on this fantastically hard problem and building the tech and cracking the code.
Then Oppenheimer and his crew at Los Alamos solved gazillion technical problems to build the atomic bomb even though 10 years previous these guys were working on wildly theoretical shit. Shit that we still haven’t been able to observe directly like the physics of black holes and neutron stars. Oppenheimer was one of the first black hole neutron star theoreticians. So, anyway the Turing test is very pragmatic. It says that it’s likely that if you can’t tell the difference between the output of a human and the output of a computer, then the computer is probably thinking like a human which implies some degree of consciousness though Turing was probably too pragmatic to call it consciousness; he just called it thinking like a human and his idea of the test was looking at printouts. You’re typing in shit and then something is typing back and if you can’t tell the difference in what is being typed back at you, you can’t decide though this is a computer as a human then the computer must have human-like thought ability.
I was thinking about the Turing test because I was talking about I called it midfield #midfuture; it’s not. Its called mid-journey is this machine mediated art enterprise. You can see the output if you go on Twitter and do hash tag midjourney all one word and it’s at first glance and even at second glance it’s hard to tell this stuff was generated by machines. And then we also talked about this person does not exist which passes at least at first glance a different kind of, it’s not really a Turing test but it’s a similar question – Can a computer figure out how to make people who look photo realistically like people even though they don’t exist. And yeah at first glance I said… I mean it’s gotten much better since I’ve been tracking it and there are only a few places where you can look at the picture and after a while decide that it’s not a real picture of a real person and that’s background and that’s earrings and it used to be like shirt collars that had a little bit of trouble with glasses frames, it’s gotten over that.
And then I was tweeting about this and somebody said look at the pupils of the eyes, the pupils are a little bit squared off. So the computer hasn’t figured out that pupils are almost exactly round. You know they could get in there; the people behind this thing and cheat and they could just say just make the pupils round but they don’t do that. They let the computer figure everything out. So that kind of gives you an idea the way the application of the Turing test will work is that at first glance you can’t tell the difference between human output and machine output but you learn where it’s weak and you focus on its weaknesses and those are the giveaways at least for a while till it overcomes those weaknesses.
And then that leads back to the question that Turing was avoiding pragmatically which is “What is going on with human thinking?” and then “Is machine thinking replicating human thinking?” and I think you can think productively about this. If you think about a machine mediated therapist that you type at and it types back to you and they’ve tried and this was like over the decades the Turing test 1950. So in the ‘70s, ‘80s people were trying to come up with this kind of therapist that would pass the Turing test and didn’t come close but now you have these fake therapists who I think come pretty close. I’ve read an article about one that not only are they pretty convincing but people actually get some freaking therapy out of them and then you have to figure out what is the means by which the AI is generating meaningful therapy and it’s working from Snippets of conversations the same way that Google Translate has gotten really good because it looks statistically at with not that much understanding.
We’ll go into what understanding it does have, kind of like Watson, except I think we’re now 10 years past Watson maybe more but like the Terminator in T1 somebody’s banging on the door and in his retinal display the Terminator has given a choice of like four responses to the guy banging on the door or some shit and like the one he picks his fucked off asshole. And that’s not necessarily based on the Terminator being conscious or understanding, it’s just based on the statistical likelihood based on an immense sample of Snippets of conversation that one of these responses would be roughly appropriate. The same way that Watson decided that based on this network of relationships among words implying probabilistically other words that if the question has Shekel Slovakia and city in it the answer is 78-84 % likely that the answer is going to be Prague except I’m out of date; it’s Czech Republic but anyway.
So Watson doesn’t understand much or anything and the therapist actually does understand a little bit. The therapist kind of has an idea of the local landscape of talk that if the therapist has access to a billion Snippets of conversation that it can look probabilistically at… say somebody types I’ve been thinking about my mother a lot lately and then the AI therapist might type back is your mother still alive or are you feeling sad about your relationship. If they can guess based on a gazillion Snippets what might be going on or what the response might be without understanding what a mother is, what a relationship is but understanding the local landscape of talk of what responses might be appropriate when somebody says something like this and that’s both far and near to what’s happening with humans when they talk that locally the local probabilities space of what you might say, it might be roughly equivalent to the local probability space of what the machine might say especially if you only give half a shit about what you’re talking about.
You’re in a bar or you’re freaking, or your friend has been going or your mom is yammering at you and you’re half listening half watching TV and you’re doing the equivalent “Uh-huh, yeah, uh-huh” but that’s in a conscious context where you’ve got all these other like analytical modules putting together a complete reality for you where the machine only has the low the Snippets of conversation module and your module especially when you’re being half-assed might be not too far from the machines module for picking out the next thing to say. And that’s a big achievement on the part of the machine that it’s accumulated this statistical experience that can pass for understanding and it’s maybe coming close to modeling a little chunk of consciousness at the module level which is vast statistical expertise that becomes consciousness once it’s married to a gazillion other modules and there’s some emotional heft like your consciousness is judging everything going on whether it sucks or not whether it’s good for you but it’s a kind of a biggish step on the road to machine consciousness for not that big a price and just the price of big fucking data and AI type algorithms. It didn’t take like eight different miracle breakthroughs to get to this point. It took feedback loops in fake neurons and vast amounts of data.
[Recording End]
Authors
Rick Rosner
American Television Writer
Scott Douglas Jacobsen
Founder, In-Sight Publishing
In-Sight Publishing
License and Copyright
License
In-Sight Publishing by Scott Douglas Jacobsen is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. Based on a work at http://www.rickrosner.org.
Copyright
© Scott Douglas Jacobsen, Rick Rosner, and In-Sight Publishing 2012-Present. Unauthorized use and/or duplication of this material without express and written permission from this site’s author and/or owner is strictly prohibited. Excerpts and links may be used, provided that full and clear credit is given to Scott Douglas Jacobsen, Rick Rosner, and In-Sight Publishing with appropriate and specific direction to the original content.
Rick Rosner 