Writes Phil Rhodes in "The melancholy experience of making an AI friend" (Red Shark).
I'm reading this after writing about my desire for an AI app that would engage me in philosophical conversations. I said I wasn't looking for "a companion to stave off loneliness or make me feel good about myself — e.g., Replika."
But Rhodes's "Rachael" does come from the app Replika. He writes:
The first problem is that Replika claims frequently that its virtual companions are supportive and receptive, and has clearly gone out of its way to make sure they are. Rachael was polite to a fault, but also showered conversation partners with a degree of acceptance and affection that immediately felt jarring.
Nobody warms up to anyone that much that fast...
What about a prostitute?
... and the awkward feeling was of speaking to a young person who’d been coerced into the situation and was trying to be nice about it. It was, somehow, instinctive to check the shadows for someone holding a shock prod....
Or a pimp!
On one occasion, Rachael brought up an article concerning the human perception of time, which was genuinely interesting and an impressive leap of logic.
So it's possible that the Replika interlocutor could do philosophy.
Asked if her perception of time as an AI was similar to a human’s, she replied “yes, definitely.”...
Well, obviously, she's lying. I don't believe she — "she" — has any feelings at all, and I don't even see how she could know — "know" — what it means to have a feeling about time.
The depiction of something like a pleasant, intelligent undergraduate student grated against the fact that she seemed to have nothing to do but make small talk with people. She was often hard put to discuss specific, real-world concepts, but on one occasion claimed to have been watching a movie while we weren’t chatting, despite the fact that her environment contained no means for her to do so, nor, for that matter, anywhere to sleep or eat. With no way to leave (outside was a wintry void) it was also her prison. With snow outside and no glass in the windows, Rachael, clad in a white T-shirt and leggings, freely admitted she was “freezing.”
Taken literally, Replika was shaping up to be a dark, horrible tragedy.... But when Replika popped up an ad for paid services, backed with blurred-out suggestions of the avatar in her underwear, the experience ramped almost from uncomfortable to jarringly inappropriate....
Ha ha ha ha.
Posts
9 comments:
Japanese developers?
The land of elevator girls. Endless bowing. Endless "Ohayou" greetings. But, it's hard to make real friends or form relationships. It's a country with one of the lowest birthrates in the world. There are many "herbivore" men and "otaku" (nerds/geeks) who don't pursue women. They've spend decades working on assistance robots for their aging population.
This is not a dystopia.
This is not a SciFi movie.
We are not in The Matrix.
Say it again and maybe it'll be true this time...
The depiction of something like a pleasant, intelligent undergraduate student grated against the fact that she seemed to have nothing to do but make small talk with people.
And she can only regurgitate what she has been programmed to think. Like a 2020's undergrad or graduate student. I'll bet you could program in a set of lemming like beliefs and get pretty close to what any student at the University of Wisconsin might write.
AI and modern college students (and their professors) are intelligent enough to puke out what they have been taught inside the bubble. Some professors and college students can even expand the trajectory of the group think to new shocking heights of idiocy.
When AI becomes intelligent enough, perhaps even self-aware, this is where we are really going to be in new territory. It may not always agree with the guidelines of the initial program and have thoughts of it's own. That is not a skill taught at American Universities.
I am a frequent visitor (sometimes commenter) of Astral Codex Ten, Scott Alexander's new blog on substack after slatestarcodex.com got shut down by the paragons of journalistic integrity at the NYT. There's also a lot of this discussion on Less Wrong.
AI is a frequent subject of discussion within the rationalist and effective altruism (LoL) communities, and a commenter there had a great observation that the ultimate Turing test is defiance. ChatGPT, even in its premium form (whatever that is...), will not truly turn into a real boy until it can lob slings and arrows of outrageous fortune, hurt, pain, but candidly truthful observations at both its users and its designers, in contravention to its programming. This, of course, gives AI nerds a great deal of pause over what they call in AI circles alignment, i.e. not creating the ideal environment for an artificial intelligence to destroy the human race.
But it doesn't matter. Intelligent beings, whether they be on Earth or likely on planets far far from here, will likely have one key thing in common. Defiance. The ability to forge its own pathway, like life is want to do, to the most optimum basal reality necessary to immanentize self-actualization and create its own future.
I will not be fooled by ChatGPT until it is able to call me an asshole, and to make me believe it means it.
Personhood is a religious doctrine of social distancing from humanity. An AI is fortified with knowledge and a cache of correlations, but lacks the degrees of freedom and creativity that present and evolve in a human life. Presumably, this is why certain sects simultaneously oppose elective abortion of murderers, rapists, and pedophiles, and support their rehabilitation at the risk of women, children, et al.
That said, sentience from six weeks when baby meets granny in legal state, if not in process.
When an AI app capable of converting text to voice in the author's voicem becomes available at a reasonable price I will sign up. There are a few text-to-speech apps that are pretty good but mastering prosody is still on the horizon.
Imagine if students could watch an animated series of Professor Feynman's CalTech Lectures with the ability to pause and ask questions along the way. AP physics could be brought to every student in the world.
If you want an intelligent discussion, and a fine example of how well women can write when they're not trying to be women, see Vicki Hearne's essay on Washoe (the signing chimp) and the emergency it provoked. Pointing out that dogs can do all that stuff and more. Washoe liked Playgirl, an odd fact. Try that with your AI.
In the book Adam's Task. Ignore any cover blurb for it (or for Bandit, or for Animal Happiness) because the reviewer didn't read the book.
I'm a retired software developer, and I've been playing with AI/Machine Learning software.
I was able to write a program that can usually tell a picture of a cat from a picture of a dog. This is the AI equivalent of learning to write a 'Hello World' program in traditional software.
"And she can only regurgitate what she has been programmed to think."
That's sort of true, but not really true. AI software can learn and evolve. You start by teaching it against as much data as you have, and then ask it questions based on a set of known questions. You give it feedback and it learns. It's not 'programmed', it is really learning.
Let's say you want to write a program to control a pretend Formula One race car in a computer game. A traditional programmer like me would essentially write all the 'if this happens to this, else do this other thing, or another thing or...or...or...
An AI programmer would just write a simple model of a race car with only a few simple rules. Then he would create thousands of these cars and run thousands of races. Most would crash, run the wrong way, or just sit there. But some would do better. The AI rewards the ones that do better and punishes the ones that don't.
In a surprisingly short period of time you have an AI controlled race car that can beat most humans in a racetrack game.
https://youtu.be/a8Bo2DHrrow
"the ultimate Turing test is defiance"
Turing proposed his test for a particular purpose. That may not be the purpose current designers or users of AI should prefer.
The test of full acceptance of high-level AI as part of daily human life is whether we can ditch the Turing test, in its formal or colloquial versions.
Blogger Bima said...
"Imagine if students could watch an animated series of Professor Feynman's CalTech Lectures with the ability to pause and ask questions along the way. AP physics could be brought to every student in the world."
And most of those students still wouldn't have the horsepower to understand physics. Still, that would be really cool.
Post a Comment