Become a fan of Slashdot on Facebook

 



Forgot your password?
typodupeerror
×
AI Government United States

CIA's Chatbot Stands In For World Leaders 37

The CIA has developed a chatbot to talk to virtual versions of foreign presidents and prime ministers. "Understanding leaders around the world is one of the CIA's most important jobs. Teams of analysts comb through intelligence collected by spies and publicly available information to create profiles of leaders that can predict behaviors," reports the New York Times. "A chatbot powered by artificial intelligence now helps do that work." From the report: The chatbot is part of the spy agency's drive to improve the tools available to CIA analysts and its officers in the field, and to better understand adversaries' technical advances. Core to the effort is to make it easier for companies to work with the most secretive agency. William Burns, CIA director for the past four years, prioritized improving the agency's technology and understanding of how it is used. Incoming Trump administration officials say they plan to build on those initiatives, not tear them down. [...]

The CIA has long used digital tools, spy gadgets and even AI. But with the development of new forms of AI, including the large language models that power chatbots, the agency has stepped up its investments. Making better use of AI, Burns said, is crucial to US competition with China. And better AI models have helped the agency's analysts "digest the avalanche of open-source information out there," he said. The new tools have also helped analysts process clandestinely acquired information, Burns said. New technologies developed by the agency are helping spies navigate cities in authoritarian countries where governments use AI-powered cameras to conduct constant surveillance on their population and foreign spies.
This discussion has been archived. No new comments can be posted.

CIA's Chatbot Stands In For World Leaders

Comments Filter:
  • How the fuck else does one explain the contradictory and self-destructive word salad policies put in place all over the world in recent memory, cuz hey, it superficially resembles some aspect of good leadership and statecraft.

  • Question (Score:5, Interesting)

    by Retired Chemist ( 5039029 ) on Monday January 20, 2025 @07:44PM (#65104775)
    Will the Chatbots really mimic the world leaders or what someone at the CIA thinks they think? Large language models are dependent on their input and the CIA controls the input. Also, while it is important to understand what someone says, it does not always reflect what they do. People in high places use words to mislead quite often. Not everyone writes out their plan for world domination in a book like Adolf Hiter's Mein Kampf. Of course, if the other European leaders had bothered to read and understand it, history might be very different.
    • Those other European leaders set the stage for Hitler's rise to power by attempting to make Germany pay for the first war. Essentially, Europe punished the German citizens and this paved the way for Hitler rather nicely.

      • Re: (Score:3, Insightful)

        by attempting to make Germany pay for the first war.

        The war they started? Madness.

        • WWI was started over a beef between the Austro-Hungarians and the Serbs. Germany (and France, Russia and others) stepped in to back their allies due to a maze of bilateral defence treaties.

          Everyone should have just taken a step back and insisted that the principles in this conflict iron out their own differences.

      • Marshall Foch the French commander said at the time, that it was not peace but rather a twenty-year truce. Hit it pretty much on the nose.
    • Will the Chatbots really mimic the world leaders or what someone at the CIA thinks they think?

      The simulation has to be based on something. By necessity it's going to be limited to what the CIA knows about how the individual thinks.

      But, I think it's safe to assume the CIA has access to inputs far, far broader than what you and I can find on Google. I'm imagining them feeding the model a stream of NSA wiretaps and diplomatic cables. If that means they can ask the Xi Jinping chatbot "How would you feel about an American carrier group sailing through the Taiwan Strait?" and get a response that doesn'

      • I do not think that that one required an AI. He will not like it. I doubt if an AI can better judge what he will do about it than a human. Given the irrationality of dictators, predictions are always a tossup.
    • by neoRUR ( 674398 )

      We tried to do this back in the early 90's with cognitive architectures and did a few. But the task of modeling thinking processes by different people 'leaders' was a little hard, and not always reliable. LLM's do provide more depth on what can be discussed with these models, and you can tell the LLM to act in certain ways. Your not expecting all answers to be 'correct' or like how they would talk, but asking general questions like, if someone did this, how would you react. The LLM can explain itself and yo

  • it should not be too hard to design it with some sense as well as being truthful. This would save world economies much trouble (including that of the USA) as well as avoiding the bad effects of Trump pulling out of climate change mitigation agreements. I could also go on about: human rights; environmental protection; minorities, women & non cis-heterosexual, medical (vaccines), ...

    MAGA is second in importance to Trump - the first is what benefits himself and his cronies.

  • I don't care how evolved these things are, they're still essentially a fancy ELIZA.

    Using this to conduct any kind of world policy with ELIZA is going to take us to places we don't want to go.

    For more details see War Games. Falken was trying to build a learning computer.

    It's too close to Skynet. C'mon, Joshua, trick the idiot humans into launching their nukes themselves.. feed them the right hallucination, and they will.

  • by ZipNada ( 10152669 ) on Monday January 20, 2025 @09:18PM (#65104881)

    An AI already can suss you out after a pretty brief interview, so its no mystery that it can effectively replicate you after having consumed everything you ever said or wrote in public. Once there is an adequately accurate model of you the CIA can just interact with that instead of bothering to deal with you at close range.

    "A two-hour conversation with an artificial intelligence (AI) model is all it takes to make an accurate replica of someone's personality, researchers have discovered."
    https://www.livescience.com/te... [livescience.com]

    • An AI already can suss you out after a pretty brief interview, so its no mystery that it can effectively replicate you after having consumed everything you ever said or wrote in public. Once there is an adequately accurate model of you the CIA can just interact with that instead of bothering to deal with you at close range.

      "A two-hour conversation with an artificial intelligence (AI) model is all it takes to make an accurate replica of someone's personality, researchers have discovered." https://www.livescience.com/te... [livescience.com]

      I'm sure this won't be used to prosecute folks for wrong-think at any point in the future. Nope nope nope. We'd never let that happen.

      We took all the warnings of sci-fi and went, "You know what? That all sounds like a great idea! Let's do it!"

    • An AI already can suss you out after a pretty brief interview, so its no mystery that it can effectively replicate you after having consumed everything you ever said or wrote in public.

      Add in the non-public information the CIA surely has on their targets and this idea starts to seem even more plausible.

      • I expect there is a huge corpus of info about everyone at this point. The online vendors (Amazon, Walmart, etc) have a record of everything you purchase and your local grocery store does too. Your cellphone carrier has metadata for all your calls and texts, and possibly full recordings if desired. Streaming TV services know what you watch. Google and Apple know where you drive and what apps you like to use. Then there's your work history, your tax forms, your bank records.

        All of them can sell that data to t

  • by Pf0tzenpfritz ( 1402005 ) on Monday January 20, 2025 @11:42PM (#65105045) Journal

    "Our chatbot version of Panama's leader clearly said they had weapons of mass destruction and were preparing for attack."

  • https://www.youtube.com/watch?... [youtube.com]

    Now just substitute "world leaders" for "tape recorders."

    What could go wrong?

    Thanks, CIA. Ruining South America, Central America, the Middle East, and Europe wasn't enough for you.

  • it's the ones those leaders and aids are referencing you need,
  • by Anonymous Coward

    Since this replaces heads of states, can't we replace C-levels of some businesses with this, so a F500 company can be entirely run by LLMs, with the exception of the H-1Bs, offshored workers, and the sub-contractors?

    We don't even need AGI, just a good LLM.

    • can't we replace C-levels of some businesses with this, so a F500 company can be entirely run by LLMs

      Um.... who would schedule the meetings to discusss the meetings and how nothing is being done because of the meetings at the meetings....

      • can't we replace C-levels of some businesses with this, so a F500 company can be entirely run by LLMs

        Um.... who would schedule the meetings to discusss the meetings and how nothing is being done because of the meetings at the meetings....

        Think how fast meetings will be if it's just various LLM chatbots meeting with one another. That two hour end-of-day ramble about product specs can be condensed down into three minutes of LLM scrawl that the few remaining human employees can read in their off-hours, so they don't have to be paid for attending the meetings! EFFICIENCY ACHIEVED!

        • Sounds good. Of course, are you sure you want a product designed by an AI. You would probably end up with a car that took you were it thought you wanted to go.

Truth is free, but information costs.

Working...