Follow Slashdot stories on Twitter

 



Forgot your password?
typodupeerror
×
AI Books Science

AI Site Claims Simulated Conversations With Famous Dead Scientists (aiwriter.app) 32

Slashdot reader shirappu writes: AI|Writer is an experiment in which artificial intelligence is used to simulate both real and fictitious famous personalities through written correspondence. Users can ask questions and receive explanations from simulated versions of Isaac Newton, Alfred Hitchcock, Marie Curie, Mary Shelley, and many more.
The Next Web calls it "a new experiment by magician and novelist Andrew Mayne," pointing out that it's using OpenAI's new text generator API. Other simulated conversations include Ada Lovelace, Alan Turing, Stephen Hawking, Richard Feynman, Isaac Asimov, Benjamin Franklin, and even Edgar Allen Poe.

"We have all kinds of theoretical ideas about AI and what counts as real or not," Mayne said on Twitter, "however I think you just have to be pragmatic and just ask: What can it do? I think this gets lost in a lot of discussions about AI. The end goal isn't a witty chatbot. It's to expand our knowledge."

There's a wait list for access to the site "so we can make sure everything works right and we don't accidentally create Skynet," Mayne jokes on Twitter. But assuming this isn't another magic trick, The Next Web is already reporting on some of the early results: The system first works out the purpose of the message and the intended recipient by searching for patterns in the text. It then uses the API's internal knowledge of that person to guess how they would respond in their written voice. The digitized characters can answer questions about their work, explain scientific theories, or offer their opinions. For example, Marie Curie gave a lesson on radiation, H.G. Wells revealed his inspiration for The Time Machine, while Alfred Hitchcock compared Christopher Nolan's Interstellar to Stanley Kubrick's 2001...

The characters could also compare their own eras with the present day... Mayne says the characters did well with historical facts, but could be "quite erratic with matters of opinion" and "rarely reply to the same question in the same way." He demonstrated these variations by asking both Newton and Gottfried Leibniz who invented calculus. "Newton almost always insists that he invented Calculus alone and is pretty brusque about it," Mayne wrote on his website. "Leibniz sometimes says he did. Other times he'll be vague." At one point, Leibniz even threatened to kill Mayne if he tried to take the credit for the discovery.

As well as historical figures, the system can respond in the voice of fictional characters. In fact, Mayne says the most "touching" message he's received was this reply from the Incredible Hulk.

Another conversation shows Bruce Wayne's response when asked to make a donation to support freeing the Joker...
This discussion has been archived. No new comments can be posted.

AI Site Claims Simulated Conversations With Famous Dead Scientists

Comments Filter:
  • by moxrespawn ( 6714000 ) on Saturday July 11, 2020 @12:35PM (#60286722)

    I'd wait to have actual conversations with them directly.

  • Based on the hundreds of comic books, movies and TV shows that I've seen with the Joker it can be assumed that he can free himself.

    Don't bother Mr. Wayne with certainties.

  • "I'd like to chat with Isaac Asimov"
    Asimov: "Hello"
    Me: "How do you feel about the creators of Westworld trying to Game of Thrones your Foundation series?"
    Asimov: "How does it make you feel the my Foundation Series is being Game of Thrones'd"?
    Me: "Eliza?"
    Asimov: "Shh..."

  • I am taking it too seriously of course, but unless it is using a pre-existing exact reply to a previously asked exact question, how can it even be validated when none of the people are alive? And even with pre-existing questions/answers, the year/season of response and mood of the person may affect the answer. Someone answering a question in 2020 won't answer the same way as in 2019.

    • I take it back that there's no way to validate. They can validate it if they show that they can use texts and other existing resources/works to simulate existing people and then cross-check those answers with reponses of the person by asking her the same questions around the same time. If they validate it that way with thousands of people and then show they have access to equivalent amounts of text by a someone who isn't around then I guess that could be close enough to validation.

    • Taybert Einstein
    • how can it even be validated when none of the people are alive?

      How much validation do you need for a PR gimmick?

  • this is about as credible as the love testing machines [wikipedia.org] of old days. way to go, they trained GPT-2 to correspondence of famous scientists. much sciencey, so legit.

  • Wow recreating history (mob approved that is).

    Just my 2 cents ;)
  • by chill ( 34294 )

    So...straight out of COSMOS: Possible Worlds, Episode 13 -- Seven Wonders of the New World?

    https://www.thetvdb.com/series/cosmos-a-spacetime-odyssey/episodes/7681315 [thetvdb.com]

    • From link:

      Take a trip to the 2039 New York World’s Fair, where the problems we now consider hopeless have been solved and thrilling possibilities exist.

      So, fantasy.

  • Isn't training a chatbot against an established body of text, while by no means trivial to get coherent results from; a fairly old and well known technique?

    If the results are better than word salad it probably isn't just a baby's first Markov Chains tech demo; but unless I'm missing something notable here an 'AI' that is specifically constrained to well-known people with a well known and well documented context to train against is vastly less interesting(as an AI, perhaps more entertaining as a conversat
    • by HiThere ( 15173 )

      Well, once you get this working right, you train it against multiple folks at once. E.g. "J. Albert Bachstein".

      That said, I don't believe it *could* work. Written works don't reflect the stream of consciousness...not even "Finnegan's Wake". Joyce went back and repolished just about every word in that over a period of, I think it was, seven years. The original public title was "Work In Progress".

    • they all fall back on a corpus though. in my opinion, a corpus corresponding to a specific person or group is more interesting than one trained on wikipedia or reddit. speaking of the latter, there's a hilarious subreddit where bots trained to specific other subreddits carry on conversations: https://www.reddit.com/r/SubSi... [reddit.com]

      it's all statistical or heuristic trickery at the end of the day.

  • Has anyone asked MeatloafBot just what it is he wouldn't do for love?

  • A novelist was involved? So, I'm guessing if I asked Leibniz to explain a windowless monad to me, it wouldn't be able to do it.

    • by HiThere ( 15173 )

      That's probably an incorrect assumption. Leibniz wrote lots about monads, so it's there to train the program. But if you asked it "What purpose does the concept of monad serve?" it might have a bit of difficulty. (I kept going to sleep when trying to read that stuff, so perhaps he did say. Perhaps it was the translator, of course. [Also that was multiple decades ago...but gods! he could go on about monads.])

      • No what he's saying is that because the author who is supposedly developing this doesn't know about monads then none of the responses are going to be about monads. If you look they're all very rudimentary questions and answers that don't actual show any of the deep knowledge on the subject there people would have.
  • So we have Eliza that does a pale, stereotypical one-dimentional impression of famous dead people. Ho-hum.

    Let me know when this can produce a video rendering of Albert Einstein regurgitating grade school level penis jokes.

  • by Entrope ( 68843 ) on Saturday July 11, 2020 @03:27PM (#60287078) Homepage

    I fondly remember Forum 2000 [slashdot.org], which used a much older form [andrej.com] of AI to produce a similar effect.

    • I am glad someone brought up this prior art. The sage advice given by those AI constructs helped me understand my personal issues, and then solve them by inventing a new type of steel.
  • The famous scientist I would most like to have a conversation with is Augustus Owsley Stanley III

  • I guess anybody that had some real understanding who these people were does already not qualify.

  • Cool! So to verify that it's accurate, let's take a scientist that's still ALIVE and make an AI image of them. We can then have a human chat with each and compare the outputs to see if the AI is accurate.

    And for that matter, let's then have them chat with each other to find out which is the real one and which is the fake -- a proper Turing test, indeed!
  • I went to the website to see the quality of the responses and they don't even allow you to interact with the chat bot. The whole website is just cherry-picked examples of questions and answers that are very dubious. A lot of just wikipedia snippets broken up. One sentence I googled and he just added one word to a line from one of Feynman's books. I put Newtons response into a plagiarism detector and it found 'significant plagiarism'. How did no one involved in writing this article notice this or do any res
  • Even when's someone's dead, you can (kind) still talk to them. Technology as come a long way, I wonder how these scientist would react if they knew people would be still talking to them through AI sites.

"The vast majority of successful major crimes against property are perpetrated by individuals abusing positions of trust." -- Lawrence Dalzell

Working...