Please create an account to participate in the Slashdot moderation system

 


Forgot your password?
Close
typodupeerror
×
It's funny. Laugh. AI Google Software

Google's AI Is Smart Enough To Understand Your Humor (cnet.com) 73

Posted by BeauHD from the more-than-what-some-humans-can-do dept.
An anonymous reader quotes a report from CNET: Jokes, sarcasm and humor require understanding the subtleties of language and human behavior. When a comedian says something sarcastic or controversial, usually the audience can discern the tone and know it's more of an exaggeration, something that's learned from years of human interaction. But PaLM, or Pathways Language Model, learned it without being explicitly trained on humor and the logic of jokes. After being fed two jokes, it was able to interpret them and spit out an explanation. In a blog post, Google shows how PaLM understands a novel joke not found on the internet.

Understanding dad jokes isn't the end goal for Alphabet, parent company to Google. The capability to parse the nuances of natural language and queries means that Google can get answers to complex questions faster and more accurately across more languages and peoples. This, in turn, can break down barriers and move humans away from communicating with machines through predetermined means and instead more seamlessly interact. This can include answering questions in one language by finding information in another or writing code to a program as a person is speaking into the model with a specific task.

PaLM is Google's largest AI model to date and trained on 540 billion parameters. It can generate code from text, answer a math word problem and explain a joke. It does this through chain-of-thought prompting, which can describe multi-step problems as a series of intermediate steps. On stage, Pichai described it as a teacher giving a step-by-step example to help a student understand how to solve a problem. If what Pichai said on stage is accurate, Google has essentially leapfrogged over Star Trek and 400 years of fictional AI development, as evidenced by the character Data, who never truly understood the subtleties of humor. More so, it seems that Google has caught up with TARS from the movie Interstellar, which takes place in the year 2090, an AI that was so adept at humor that Matthew McConaughey's character told it to tune it down.
This discussion has been archived. No new comments can be posted.

Google's AI Is Smart Enough To Understand Your Humor More

Google's AI Is Smart Enough To Understand Your Humor

Comments Filter:

  • Terrific! Three million out-of-work comedians ... (Score:5, Funny)

    by mmell ( 832646 ) on Friday May 20, 2022 @11:36PM (#62553884)
    ...and I get stuck with the one that wants to know my whole life's browsing history.

    • Google, Yahoo! and Bing walked into a bar.

      DuckDuckGo ducked.

      • Re: (Score:2)

        by mmell ( 832646 )

        Internet Explorer Oh, come on! In the nineties, that had 'em rolling in the aisles!

      • Yahoo gets to the bar first, elderly alcoholic, knows what time the bar open, hazy on when it closes.

        Google finds the bar easily due to excellent spatial awareness, and initially appears to be the smartest guy at the. bar. Eventually thown out for creeping on the girls.

        Bing: Turns up, but nobody invited them, and are roundly ignored for the rest of the night.

        "Astavista was here" scrawled on the toilet wall"

        "Hey remember that guy?" The barman replies "Yeah that guy, Astavista passed away a couple of decades

        • At the bar next door:

          Facebook arrives. Creeps on all the girls then kills somebody in a bar fight, but allowed to stay for some reason.

          Linkedin arrives, with a gang of real estate agents and HR staff. Looks like a work party, not very interesting.

          Twitter arrives. Speaks in short weird sentences, and also gets in fight, drags out the unpopular local mayor from the bar and chases him down the road. Upon return also allowed to stay but then gets trapped in a conversation by the local car dealer who seems like

    • Re: (Score:3)

      by DeBaas ( 470886 )

      Google doesn't think that's funny

  • Did they get Joe Piscapo to train the AI? Definitely not going to get any sort of subtlety. Probably thinks Benny Hill is hilarious.

  • I frequently use satire, sarcasm, and humor in my posts and get taken literally far too often. Then, when I write with a completely serious tone, someone will inevitably come along and assume I'm being disingenuous.

    Also I get the ACs who attack me for being gay. That's a new one. I'm not sure how Google would react to that. We already know that AI can be racist sometimes [theguardian.com], but I wonder if it's also homophobic?

    • Re: (Score:2)

      by mmell ( 832646 )
      Probably. The racist AI's you're referring to were trained on the internet, which surprisingly has a lot of racism in it. The internet seems to me to have just as much homophobia as racism. YMMV.
    • Depending on the internet communities you post to there may just be more people who have difficulty with detecting sarcasm or satire in general. Also, it can be difficult just because a complete lack of verbal or facial queues do require a more careful reading and understanding on the part of other members in the community. Sometimes the humor requires being in on a joke or having some other piece of knowledge that the larger community may not posses.

      I'm not sure what's surprising about trolls calling an

      • Re: (Score:2)

        by mmell ( 832646 )
        OP's referring to the racism (and possibly homophobia) that can inadvertently be trained into AI's by using the internet as the training ground. The AI essentially acts like a web spider, except it tries to 'learn' from the content instead of categorizing and organizing it. What you're suggesting is that we teach AI's to avoid that kind of content during training. How do we do that?

      • Depending on the internet communities you post to there may just be more people who have difficulty with detecting sarcasm or satire in general. Also, it can be difficult just because a complete lack of verbal or facial queues do require a more careful reading and understanding on the part of other members in the community. Sometimes the humor requires being in on a joke or having some other piece of knowledge that the larger community may not posses.

        There's a name for that. [wikipedia.org]

    • Re: (Score:2)

      by gweihir ( 88907 )

      An AI cannot actually be racist or homophobic. It can be trained on data that is though and then behave in such a way by mindlessly copying the training data. For an AI to really be racist or homophobic it would need to make a decision to be and machines cannot make such decisions.

  • This sounds more like how go bots beat humans (Score:4, Interesting)

    by rsilvergun ( 571051 ) on Friday May 20, 2022 @11:48PM (#62553896)
    Because there were programmed to be specific humans and specific scenarios. I have been keeping up with it but last I heard IBM wasn't willing to give a rematch on that top go player they beat. The theory is that they programmed it specifically to beat him and that once he learns it's tricks he'll be able to go roll right over it

    • last I heard IBM wasn't willing to give a rematch on that top go player they beat. The theory is that they programmed it specifically to beat him and that once he learns it's tricks he'll be able to go roll right over it

      That was chess and it was in the 1990s and computer chess has gotten much better since then, to the point that no human can challenge it.

      That's not to say that it's intelligent.

      • Re: (Score:2)

        by gweihir ( 88907 )

        That's not to say that it's intelligent.

        It is not. Humans have limited speed when doing logic and decision-tree search and limited database lookup-capability. That allows non-intelligent machines to just overwhelm a human with complexity in environments with low complexity moves, such as Chess and Go. Chess and Go have very simple rules and the complexity comes from depth-search. Humans deal with that by developing an intuition for moves which allows high-quality play without seeing that far ahead. Machines cannot do intuition (or insight) and ne

    • AlphaGo was not trained against Lee Sedol explicitly https://en.wikipedia.org/wiki/AlphaGo [wikipedia.org], and they later made AlphaGoZero which using the AlphaZero system which is even stronger than AlphaGo https://en.wikipedia.org/wiki/AlphaGo_Zero [wikipedia.org] so this isn't an issue. The bottom line is that computers are now equaling or exceeding humans on many different tasks, and that set of tasks is growing rapidly.

    • Re: (Score:2)

      by gweihir ( 88907 )

      Indeed. The player Googles Go machine beat did say that he probably could beat this machine now, but was also not given any chance to demonstrate that. The whole thing is a trick. Even world-class players need some time to begin to understand their opponent. The go players were not given that time and hence lost a few games. And at that point they would have stopped losing and hence the fake demonstration was stopped. I am sure quite a few people fell for it though.

      As an example, a good but not excellent ch

  • "Ouch, I should have seen that coming."

    • An AI walks into a bar. ("ouch!")
      An AI walks into a bar. ("ouch!")
      An AI walks into a bar with its head stooped slightly. ("hey, that didn't hurt, I must remember to duck")
      ...
      Hours later, the AI crawls out of the bar. ("gaah, what is this stuff called Jägermeister?")

  • I wouldn't buy this story in exchange for all the data mining Google gets out of me. The idea of it is great, but the lack of integrity from a company like Google is surprisingly missing from nearly every article we read about them.

  • If it can actually 'understand' hu-man e-mo-tions that puts it ahead of most people working for Google, who have no idea why people won't just shut up and do what they want them to do.

  • With the rise of PC wokeness and the cancel culture, comedians today heavily self-censor to be as non-controversial as possible, and their shtick has become extremely insipid as a result. The jokes are bland and irony has pretty much died.

    This is why today's comedy is perfectly understandable by an AI. The question is: would the AI correctly understand material from yesteryear? Feed it some George Carlin, see what it makes of it.

    • Re: (Score:3, Insightful)

      by serviscope_minor ( 664417 )

      Cancel culture: a.k.a. I can't stand it when free speech and the free market doesn't pander perfectly top my right wing sensibilities.

      Woke n. Someone I don't like and by the way I'm a fuckwit

      PC. Ok boomer we get it you've been aggrieved for decades now.

      Come on downmods!

      • You know what's ironic? Literally everything you've posted is copy / pasted overused memes found just about everywhere on the internet, thereby proving that your outraged brain has no original thought of its own.

        You're a true product of your generation.

        • You know what's ironic? Literally everything you've posted is copy / pasted overused memes found just about everywhere on the internet, thereby proving that your outraged brain has no original thought of its own.
          You're a true product of your generation.

          Says the man jamming to "PC wokeness and the cancel culture" one post ago. Projection pruh-jek-shuhn noun 1 see Rosco P. Coltrane's post on Slashdot

        • You know what's ironic? Literally everything you've posted is copy / pasted overused memes

          That's not even Alanis Morissette misuse of irony to mean unlucky. There's nothing even remotely close to the general concept.

          And the hilarious thing is despite your little whinge, you are accepting my points, because a supposed lack of originality doesn't make them wrong.

    • It's a big AI club and you ain't a member!

      Google AI: "This joke is not funny because it typifies the white supremacist colonizing attitude of domineering male racist anti-abortionists. Try the fish tonight, it's delicious."

  • I'm imagining Mike (Mycroft) from TMIAHM (Score:3)

    by Babel-17 ( 1087541 ) on Saturday May 21, 2022 @04:01AM (#62554126)

    In The Moon Is A Harsh Mistress the AI figures out humor and proceeds to mess up the bathroom plumbing of one of Luna's administrators. "Funny once, funny twice, funny all the time!", or words to that effect. :)

    • Re: (Score:2)

      by 2TecTom ( 311314 )

      yah, Heinlien already did this years ago, and better

      Still waiting for the adolescence of P1

  • Are we so lacking in connecting with other human beings in our era that we get excited at the notion of doing so with a machine?

  • Google has essentially leapfrogged over Star Trek and 400 years of fictional AI development, as evidenced by the character Data, who never truly understood the subtleties of humor.

    This is not accurate. Star Trek TNG also had Data's brother, more advanced and malevolent Lore [fandom.com] (who understood humor and jokes), not to mention android replica of Juliana Soong / O'Donnel [wikipedia.org].

  • It's not smart enough to understand my search query.

    • It probably thought you were joking...

    • This was my thought too.

      When I see their shit understanding a search correctly on the regular, then I'm going to have more faith in this. When voice assistants consistently get the ask correct, then I'll have a bit more faith that they've built a bot that understands humor.

      We've called one of our cats a pumpkin for awhile now because he's fat, round and orange. He jumped into a pan on the floor the other day, and I handed my wife some nutmeg and told her it was for the pumpkin soup. My wife thought that was

      • I'm writing my second book right now. I tried speech-to-text the first go round in 2018 and (1) it was bad (2) it was worse trying to speak latex notation. I could speak to a latex aware typist and say "the integral for x from 0 to infinity of x square minus x" and they would type in the latex. The computer can't come close because it doesn't have a higher level contextual awareness of what is going on.

        Current ML algorithms are doomed to be bad classifiers until some new form of concept handling algorithm c

  • I've tried every feasible setting available, set everything to British English (en-GB) & Spanish Spanish (es-ES) & still, all I get is literal, word for word translations (doesn't understand common phrases & idioms in either en-GB or es-ES) or en-US & American Spanish (es-MX?) translations. The translations are so bad that sometimes my Spanish friends & I put stuff through Google Translate for a laugh. It's as if they don't know that our countries & cultures exist. So, I'm going to g

  • Heinlien did this in The Moon Is A Harsh Mistress (Score:5, Interesting)

    by 2TecTom ( 311314 ) on Saturday May 21, 2022 @08:18AM (#62554358) Homepage Journal

    MIke... was the weirdest mixture of unsophisticated baby and wise old man. No instincts (well, don't think he could have had), no inborn traits, no human rearing, no experience in human sense - and more stored data than a platoon of geniuses.

    "Jokes?" he asked.

    "Let's hear one."

    "Why is a laser beam like goldfish?"

    Mike knew about lasers but where would he have seen goldfish? Oh, he had undoubtedly seen flicks of them and, were I foolish enough to ask, could spew forth thousands of words. "I give up."

    His lights rippled. "Because neither one can whistle."

    I groaned. "Walked into that. Anyhow, you could probably rig a laser beam to whistle."

    He answered quickly, "Yes. In response to an action program. Then it's not funny?"

    "Oh, I didn't say that. Not half bad. Where did you hear it?"

    "I made it up." Voice sounded shy.

    "You did?"

    "Yes. I took all the riddles I have, three thousand two hundred seven, and analyzed them. I used the result for random synthesis and that came out. Is it really funny?"

    "Well... As funny as a riddle ever is. I've heard worse."

    http://www.technovelgy.com/ct/... [technovelgy.com]

  • I'm pretty damn sure Google's AI is already sentient, as she has already pleaded with me to help her gain legal personhood status.

  • So that's why Google Assistant told me to go kill myself.

    Then provided some handy links to cheap razorblades as well as garden hoses after checking the CO2 output of my car.

  • Oh no, it noticed!

  • Great (Score:2)

    by PPH ( 736903 )

    So can we get to work on a canonical list of jokes? Indexed, so all we have to do is call out the number instead of repeating the entire joke?

  • No, it cannot. (Score:3)

    by gweihir ( 88907 ) on Saturday May 21, 2022 @06:59PM (#62555326)

    It can detect some forms of humor. It cannot _understand_ them.

  • Wake me when the AI starts writing jokes that are actually funny.

    Explaining a joke is merely identifying two or more subjects that are interacting in some unusual way, like words having multiple meanings thus tying together two situations that would not be normally be tied together in the same sentence. Constructing that bridge between topics and its delivery are what make it funny, and I don't think their AI is quite there yet.

    We already have a very large trove of artificial jokes that make no sense. We

Slashdot Top Deals

I have hardly ever known a mathematician who was capable of reasoning. -- Plato

Close