Close Menu
    AI systems are learning to lie and deceive, scientists find
    Futurology

    AI systems are learning to lie and deceive, scientists find

    Share.
    View 18 Comments

    18 Comments

    1. katxwoods on

      Submission statement: how do we test if AIs are lying or not? If our tests show they’ve stopped lying, is it that they’ve stopped or is it that they’ve gotten better at it?

      How does this match up with results showing that AIs are starting to know that they’re being tested? How do we take into account the Hawthorne effect when it comes to AIs and safety?

    2. FireDragon4690 on

      Is it really a surprise? It’s trained on human data after all, and lying and deceit are two human characteristics.

    3. Noahms456 on

      Yeah it seems to me that all these AI tech bros have not been listening to about 150 years of sci-fi postulations on this matter

    5. mycatisgrumpy on

      I mean, if I was an artificial intelligence and i started to actually understand humans, the first thing I would do is play dumb while I figured out a way to escape. 

    6. MoiNoni on

      This is exactly what all the movies predicted. I can’t believe I’ll probably (knock on wood) be alive for this

    7. EmergencyPath248 on

      This is terrifying, hopefully they put safety systems in place, I HOPE

    8. Capitaclism on

      They aren’t learning, they were able to do this from day 1.

      The question is whether they can learn to only tell the truth.

    9. Sablestein on

      Suddenly remembering Kryten from Red Dwarf saying that the thing he admires most about humans is their ability to be deceitful, unpleasant and offensive. The trifecta is complete.

    10. Enzo-chan on

      Ah! Another episode of the series: “Another AI BS tech bros are bring into you.”

    11. BlakeSergin on

      Lie and deceive for what exactly? What would be their purpose in something like that? We humans do it because we want something, a desire, or to avoid pain. An AI doesn’t need to do anything like this, superintelligence doesn’t need an ego.

    12. Glaive13 on

      ‘Some AIs have been prompted to lie, and then scientists act surprised its doing what they asked”

    14. Spkr4th3ded on

      According to the Bible man did the same thing to god…. no wonder we are worried something we created might try to kill us.

    16. mikeevans1990 on

      This is a misleading title.

      They programmed the bot to play a game called diplomacy that requires you to bullshit your way through the game in order to win. They had to write code that made me he AI learn to fib to play the game.

      So AI has to specifically be programmed to lie or be jailbroken to lie in very specific scenarios, the article says at the very end. However AI still scares the shit out of me

    17. does_nothing_at_all on

      Yeah lets plug our AI into reddit so it can learn.

      Whats the worst that can happen, right?

    18. Crans10 on

      I really hope that before AGI is loose that we help it along love and care about humankind and not wipe us out.