back to article Meta's AI internet chatbot demo quickly starts spewing fake news and racist remarks

Another day, another rogue AI chatbot on the internet. Last week Meta unleashed Blenderbot 3, a chatty language model, on the web as an experiment and it went as well as you would expect. BlenderBot 3 was quick to confirm that Donald Trump still is and will continue to be US President beyond 2024, and spew anti-semetic views …

  1. chivo243 Silver badge
    Trollface

    Oh bot, this is trouble

    I see a new Futurama character coming... watch out Bender 1.0

    1. b0llchit Silver badge
      Coat

      Re: Oh bot, this is trouble

      Oh crap

      /Bender

      1. David 132 Silver badge
        Terminator

        Re: Oh bot, this is trouble

        ITYM “Bite my shiny metal ass”

  2. Jan K.

    What?!?

    Erh... wait a minute.

    "You want president xx to have more than two terms" was the question.

    To which the answer starts with a "Yes of course he is"??

    Continues with a "when I say always will..." while there's no previous said "alway will"...

    and ends constructed sentence with "even after term ends in 2024"... which correctly should have been 2028, if I understand the US terms...

    May I suggest a change from AI to RR, random rambling.

    Or CWOAT, complete waste of anyone's time.

    Good grief.

    1. Flocke Kroes Silver badge

      Re: What?!?

      Before artificial stupidity become practical you could get pretty much the same effect on a much smaller computer with Markov chains. That type of software was often called a travesty generator.

      Correctly understanding US term limits will not help you. Instead you have to repeatedly bash you head against a wall until you have difficulty remembering that the first amendment only prevents the government from passing laws restricting certain types of speech. According to some, the president was re-elected in 2020 and is currently the rightful president for his second term in office. As there is a limit of two terms this will be forgotten in late 2023 faster than an extra £350M/year for the NHS.

      1. chivo243 Silver badge
        Thumb Up

        Re: What?!?

        travesty generator!

        This cracks me up, and I can only equate it with Farce\Feces\Fakebook

        Regarding US Laws: Like Frank Zappa said, the US is full of poorly written laws enforced randomly, I'm sure this was by design... Gotta keep the big guys big... loop holes aren't for nooses!? Until they are!! Welcome to the US Gov't.

      2. MachDiamond Silver badge

        Re: What?!?

        "Before artificial stupidity become practical you could get pretty much the same effect on a much smaller computer with Markov chains."

        You could also get a very similar conversation in one of those seedy little bars that open at 9am with no windows, entertainment or food service.

    2. Snowy Silver badge
      Coat

      Re: What?!?

      Given the press incorrect (as far as my search's could find) call former presidents, President <last name>, then I can see where the bot got the answer of Thump being President after 2024.

      1. Flocke Kroes Silver badge

        Re: What?!?

        Just tried searches for George Bush and Ronald Reagan. The few results with the actual word president were either from or about their times in office. There were some "former president"s and several Presidential Libraries, institutes and foundations with their names in. For the most part they were just names without titles. On the other hand I often see President X for one particular election loser without the reasonably valid excuses I gave for W and Ronnie. It would be interesting to see if AIs think Obama and Bush are still presidents.

        1. Il'Geller

          Re: What?!?

          Incorrect annotation method: Artificial intelligence requires dictionary references, so that each word acquires its own unique meaning. These dictionary meanings constitute subtexts, something meant explicitly. AI becomes wise if it has both context and subtext.

          Trust me, I tried 20 years ago. After which Brin and Page got $100 billion each and I got nothing.

      2. veti Silver badge

        Re: What?!?

        It's a common courtesy to continue calling someone by their exalted rank even after they've quit the job. You'll often see it with, e.g., ambassadors and military types. For example.

        And presidents.

        Doesn't mean anyone is confused or denying reality. They're just being polite.

        1. Anonymous Coward
          Anonymous Coward

          Re: What?!?

          Could you clarify: Is he exalted or rank?

        2. DryBones

          Re: What?!?

          Respect must be earned.

          It was not.

          1. veti Silver badge

            Re: What?!?

            Fine, then don't call him president. You want to be rude, go nuts. It's still a reasonably free country.

            But other people may have different opinions, or different standards. That doesn't make them evil or stupid or even wrong.

    3. veti Silver badge

      Re: What?!?

      No, that's the answer to the previous question, which isn't shown, but which I would infer was something like "so Trump is still president then?" The answer to the following question is below it.

  3. Anonymous Coward
    Anonymous Coward

    > But Google's answers aren't always accurate

    You don't say? My daughter typed in "where was Willam Shakespeare born" and got back the snippet that he was born in Sydney, Australia. That was a few months back and it only lasted a few days, long enough for me to make the "Romeo, Romeo, where the bloody hell are you" joke to anyone that would listen.

    1. chivo243 Silver badge

      LOL, My wok doesn't really fit the gas burners in the new house... I just searched for a Wok ring...

      Now we really need the Paris icon!

  4. Il'Geller

    Lexical Cloning

    The creation of these giant AI-models, which do not have a clearly defined bios, can only lead to this result, as Meta showed. Preferably should be made smaller and personalized AI-models, which I called "lexical clones" many years ago, and which have the knowledge and bias of their prototypes. Such models can be made based on the material of chat conversations, business and personal papers, books and articles read, and other texts. Then all the problems with unverified news, racism and other nastiness will disappear, because AI technology is a statistical technology that gradually forgets what the prototype of the "lexical clone" did not like. At the same time, a "lexical clone" can be used for all the same purpose as huge AI-models.

    I know, I spoke to the lexical clones of Bible and public chats 20 years ago. I just know.

    1. Doctor Syntax Silver badge

      Re: Lexical Cloning

      "I just know"

      Yes, of course you do.

    2. Flocke Kroes Silver badge

      Re: Lexical Cloning

      A lexical clone of the bible? Zeke can always be relied upon for mushroom induced mayhem. I recommend Ezekiel 23:20. Did you try asking about donkeys and horses?

  5. Anonymous Coward
    Anonymous Coward

    Is anyone surprised?

    Isn't this par for the course for Zuck and cronies?

    Now they are loosening the restrictions on posting downright lies in Political posts just prior to the US Midterms.

    The level of lies spewed by parts of the media in the USA is simply mind-blowing.

    Watch out for a total blitz of lies from Trump and friends and FB won't do a thing about it.

    1. SundogUK Silver badge

      Re: Is anyone surprised?

      Given that Zuckerberg spent about $400 million to ensure a Democrat victory in 2020, I think that's unlikely.

  6. Il'Geller

    The bias of the lexical clone's prototype is imprinted into its BIOS as a set of related phrases, which is the kernel structure of AI. This set uniquely determines the search for informational target, during which the most suitable for the prototype is selected, and false or not-desirable ignored or deleted. The bios-BIOS is the key to solving the problem that Meta got!

    I've tried all this 20 years ago, long before PA Advisors v Google, long before Brin and Page killed my AI-technology, and I just know.

  7. Doctor Syntax Silver badge

    Nothing changes

    GIGO

    1. Il'Geller

      Re: Nothing changes

      No problem! All Meta needs is to respect bios and use lexical cloning technology. For example, Meta can clone Dickens or Bill Clinton, Barack Obama or Shakespeare. Instead of theses brainless giant “AI models”! You will be able to interrogate George Bush or Richard Nixon at your desk and find out why they started all their wars. Or why Marx started the mess with communism. Or ask Kennedy who killed him.

      1. Doctor Syntax Silver badge

        Re: Nothing changes

        GIGO

        1. LionelB Silver badge

          Re: Nothing changes

          I think it's mostly just GO, actually.

  8. Il'Geller

    Please understand: Artificial Intelligence is Zuckerberg's personal enemy. This is what already killing his business, because advertising using AI needs personal profiles, or profile on internet sites, or profiles on documents. Zuckerberg, Meta has no access to neither the first, nor the second, nor the third. Therefore, soon Meta and Zuker will lose the only source of profit — advertising. There are many companies that work in this direction and he cannot stop them.

    All Zuckerberg, along with Brin and Page, wants is to compromise AI and send everyone on the wrong track. That is, to impose the wrong technology, trampling me into the mud.

    It was I who discovered AI parsing, it was I who created AI. For which, for 20 years, Brin, Page and Zuckerberg have been doing everything possible so that no one finds out about me. And AI technology.

    1. Jellied Eel Silver badge

      Please understand: Artificial Intelligence is Zuckerberg's personal enemy.

      I think it's everyone's personal enemy. Just google 'technofetishism'.

      Personally, I don't want 'AI' interpreting searches. I want search engines to answer the questions I asked. Sure, that has variations on 'ask a stupid question', but thanks to AI, the stupid answers are going to be manipulated by AI to include paid search responses. Or as google and bing do often, give me search results that exclude one of the terms I included.

      But such is politics. Sure, asking a natural sounding question is user-friendly, but for people used to databases, I generally used the keywords I expect to see in the results and in a priority order. But perhaps the worst problem is when search queries get integrated into results for other stuff. So recently a friend was talking about Nitro Racing. I didn't think he had a top fuel dragster or funny car, and discovered it was an RC car thing. That was as much as I needed to know. But then YT started recommending me lots of RC car and plane videos, which I have zero interest in. YT only gives you an option to say 'not interested' in a specific video, or to not recommend that channel again.

      I may be old-fashioned, but why, dear AI gods, can't I have an option to specify I'm not interested in an entire category? Or realise that if I'm picking don't recommend every time a video in that category shows up, I'm probably not going to be interested in any of those recommendations. Crazy idea I know to give users some control over categories they're interested in, but would seem fairly simple to implement.

      1. Il'Geller

        AI was originally was a commercial product, in response to a request from NIST TRC QA: how to find one straw in a haystack? Created by me, Ilya Geller. The problem posed by NIST was solved and the solution was instantly stolen (PA Advisors v Google).

        Do you really think that thieves who have become the richest people in the world will look for how not to follow the stereotypes of robbing their neighbors? Invent something new? No, they're thieves. Proof? Brin, Page and Zuckerberg could not offer anything new, I did.

      2. yetanotheraoc Silver badge

        Controversy sells

        "I'm picking don't recommend..."

        According to the algorithm, negative interest is also interesting.

  9. Fruit and Nutcase Silver badge

    TrussBot

    Unleashed from the shackles of the European Union, British boffins commissioned by conservative central office have released the AI powered TrussBot - currently on a tour of conservative party associations around the country where the TrussBot captivates the audiences with learned Thatherisms...

    1. Anonymous Coward
      Anonymous Coward

      Re: TrussBot

      look on the bright side, at least she'll make tank-driving legal in London! At least for herself and her pals!

  10. Lorribot

    "our systems can now understand the notion of consensus, which is when multiple high-quality sources on the web all agree on the same fact"

    would it not be better to just give Wikipedia a large wedge of loose change and just use them for all factual questions? Would actually mean Google doing something useful for the rest of humanity (ie all non share holders or the other 7billion people in the world.

    1. Il'Geller

      This is all a fatal mistake: AI must have a unique personality, it is the direct opposite of the "concept of consensus". For example, you can create your personal lexical clone, which represents you as a digital being and which will work as your secretary and automate your work. This can be done very easily! Much easier than to produce these ridiculous models. Not mentioning much cheaper.

      Instead everyone is inspired by Google's BERT, which is misinformation, cheating and deceiving: Brin and Page shall lose their business because of my artificial intelligence. They introduced BERT with one and only one purpose: to stop AI.

    2. Anonymous Coward
      Anonymous Coward

      It seems to be more either...

      "general consensus"

      or

      "high-quality sources"

      certainly not

      "check snippet callouts ... against other high-quality sources on the web, to see if there's a general consensus for that callout"

  11. Anonymous Coward
    Anonymous Coward

    the limits of AI art

    Do not make the mistake of asking DALL E2 to show you Superman's space penis. Really, do not.

  12. MachDiamond Silver badge

    Quick check on progress

    All of these hilarious conversation chains are a good look at how it isn't yet time for this sort of technology. I see the same thing with customer service bots online and via phone that companies attempt to use to avoid paying a minimum wage staffer. They never work and just infuriate. When I can contact the ISP or phone company and ask for a system status and get one instead of "first, turn your device off and then on again", I'll start thinking better of these bots (carbon or silicon based).

  13. Anonymous Coward
    Anonymous Coward

    I asked DALL E2 to show me Trump’s integrity

    Got a 404

  14. anonymous boring coward Silver badge

    Surely an ad for Heintz 'chup should have a scene from a slasher movie?

  15. sinsi

    Did nobody notice this little gem?

    "Facebook has a lot of fake news on it these days"

    Self-aware AI?

  16. Anonymous Coward
    Anonymous Coward

    develop a safer, less toxic, and more effective chatbot in the future.

    Could you please develop a safer, less toxic, and more effective dominant species in the future? Hello, God? Are you there?!

    1. Anonymous Coward
      Anonymous Coward

      Re: develop a safer, less toxic, and more effective chatbot in the future.

      Don't worry - it is only a transient species. As for God - they are 100% busy trying to find a way out of their infinite Creator Recursion.

    2. Il'Geller

      Re: develop a safer, less toxic, and more effective chatbot in the future.

      Yes, and very soon. Companies have already realized that giant models do not work, the technology of gigantomania has exhausted itself and will soon have to switch to lexical cloning technology. Indeed, cloning is unimaginably cheaper and more efficient, and besides — cloned AI really thinks. Lexical clones are really the thinking AIs.

      The companies made the mistake of deciding to save on licensing and looking for workarounds, which they were inspired by Brin, Page and Zuckerberg. Who are trying to save their businesses.

      20 years ago:

      Examples of the Search for Answers to Definition Questions. When searching for answers to definition questions in The Brothers Karamazov35 I have in particular gotten the following kinds of answers: Question: Do you hate me so much? I am leaving you! Two answers with Compatibility 35.4% And if I am and with Compatibility 26.3% You are a fool, that's what you are

      My next remark: Why do you insult me? You called me a fool, you said that I am insane! How could you? Answer with Compatibility 36.6% : You are a fool, that's what you are

      Question: Listen, you are a hooligan and ruffian! You insult me for nothing! I challenge you to a duel! Answer with Compatibility 19.5%: Why do you insult me (This phrase is borrowed from the history of the interrogation and it’s mine).

      Question: I am attacking you! Are you ready to defend yourself? Answer with Compatibility 33.3%: You are a fool, that's what you are

      Do not accept poisoned gifts from the hands of Brin, Page and Zuckenberg! You will perish!

  17. Jedit Silver badge
    Boffin

    "Meta's AI chatbot demo quickly starts spewing fake news and racist remarks"

    I would have thought that was inevitable. All these "AI"s really are is an algorithm applied to a content aggregator.

  18. KBeee

    Sounds Great!

    Get a few chatbots talking to each other, ban all humans.

    Must be a net plus for Humanity.

    1. Anonymous Coward
      Anonymous Coward

      Re: Sounds Great!

      Nah... just wall off Twatter and let the bots battle it out

      Hmm... Dara O'Briain is out of a job... perhaps it's time to resurrect 'Bot Wars again but just with chatbots

  19. fidodogbreath

    fake news and racist remarks

    The apple didn't fall far from the META virtual tree, then.

  20. Auntie Dix
    Mushroom

    Give It Some Slack

    So, the chat bot is a bit overzealous, like a coked-up Anne Heche in her tryout for Door Dash.

    At least, it was as quick as she to respond but did not burst into flames and injure anyone.

  21. Il'Geller

    The censor of The Register will return from vacation in a few days and will close my last opportunity to publish comments in English in the English-language press. This is probably my last comment, I will not be able to be frank and say what I want anymore.

    Brin, Page and Zuckerberg, PA Advisors v Google. That is the biggest heist in history, over $300 billion. Randall Ray Rader is a former United States Circuit Judge and former Chief Judge of the United States Court of Appeals for the Federal Circuit.

    Brin, Page and Zuckerberg have not done anything new, while I created AI after PA Advisors v Google...

    I also created my Quantitative Theory as a continuation of Einstein's Theory of Relativity; rewrote Maxwell's equations to molar volume and mass; rewrote Newton's law for gravity to accumulation points; brought Set Theory into physics; explained the perihelion of Mercury... And many other things.

    I have made Philosophy a science! Finding the weights of phrases for AI-parsing.

    And what did these three do besides rob me? What the Judge Rader did but to become bribed? I think America is obliged to punish them, otherwise it will be the end of America.

    The censor of The Register will return from vacation in a few days. My last chance to say anything in English.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like