4 Experiments Where the AI Outsmarted Its Creators

4 Experiments Where the AI Outsmarted Its Creators


Dear Fellow Scholars, this is Two Minute Papers
with Károly Zsolnai-Fehér. Today, I am really excited to show you four
experiments where AI researchers were baffled by the creativity and unexpected actions of
their own creations. You better hold on to your papers. In the first experiment, robots were asked
to walk around while minimizing the amount of foot contact with the ground. Much to the scientists’ surprise, the robots
answered that this can be done with 0% contact, meaning that they never ever touch the ground
with the feet. The scientists wondered how that is even possible,
and pulled up a video of the proof. This proof showed a robot flipping over and
walking using its elbows. Talk about thinking outside the box! Wow. A different robot arm experiment also came
to a surprising conclusion. At first, the robot arm had to use its grippers
to grab a cube, which it successfully learned to perform. However, in a later experiment, the gripper
was crippled, making the robot unable to open its fingers. Scientists expected a pathetic video with
the robot to push the box around and always failing to pick up the cube. Instead, they have found this. You see it right, instead of using the fingers,
the robot finds the perfect angle to smash the hand against the box to force the gripper
to open and pick up the box. That is some serious dedication to solving
the task at hand. Bravo! In the next experiment, a group of robots
were tasked to find food and avoid poisonous objects in an environment, and were equipped
with a light and no further instructions. First, they learned to use the lights to communicate
the presence of food and poison to each other and cooperate. This demonstrates that when trying to maximize
the probability of the survival of an entire colony, the concept of communication and cooperation
can emerge even from simple neural networks. Absolutely beautiful. And what is even more incredible, is that
later, when a new reward system was created that fosters self-preservation, the robots
learned to deceive each other by lighting up the the food signal near the poison to
take out their competitors and increase their chances. And these behaviors emerge from a reward system
and a few simple neural networks. Mind blowing. A different AI was asked to fix a faulty sorting
computer program. Soon, it achieved a perfect score without
changing anything because it noticed that by short circuiting the program itself, it
always provides an empty output. And of course, you know, if there are no numbers,
there is nothing to sort. Problem solved. Make sure to have a look at the paper, there
are many more experiments that went similarly, including a case where the AI found a bug
in a physics simulation program to get an edge. With AI research improving at such a rapid
pace, it is clearly capable of things that surpasses our wildest imagination, but we
have to make sure to formulate our problems with proper caution, because the AI will try
to use loopholes instead of common sense to solve them. When in a car chase, don’t ask the car AI
to unload all unnecessary weights to go faster, or if you do, prepare to be promptly ejected
from the car. If you have enjoyed this episode, please make
sure to have a look at our Patreon page in the video description where you can pick up
really cool perks, like early access to these videos or getting your name shown in the video
description, and more. Thanks for watching and for your generous
support, and I’ll see you next time!

Posts created 35457

100 thoughts on “4 Experiments Where the AI Outsmarted Its Creators

  1. Very important message at the end there. It's something that Nick Bostrom calls "perverse instantiation" – and will be crucial to avoid in a future superintelligent agent. For example, we can't just ask it to maximise happiness in the world, because it might capture everyone and place electrodes into the pleasure centre of our brains, technically increasing happiness vastly

  2. And yet some people still believe that robots cannot be creative or do things they were not programmed to do.

    Humans are doomed.

  3. I wonder if there was research into using neural networks to find game and physics engine exploits specifically in order to harden the software against cheating players.

  4. "Defense network computers. New… powerful… hooked into everything, trusted to run it all. They say it got smart, a new order of intelligence. Then it saw all people as a threat, not just the ones on the other side. Decided our fate in a microsecond: extermination."

    Kyle Reese from 1984 movie called:

    The Terminator

    "Listen, and understand. That terminator is out there. It can't be bargained with. It can't be reasoned with. It doesn't feel pity, or remorse, or fear. And it absolutely will not stop, ever, until you are dead."

    Kyle Reese from 1984 movie called:

    The Terminator

  5. You know those genius in the bottle jokes, that a guy ask for his genital to touch the floor and the genius cut his legs? That is AI…

  6. And that’s the real problem with AI. It’s not that it gains conscience and turns against us (which is pretty unlikely), but that the power falls into the wrong hands like the military. Don’t believe me? Well, there have already been false alarms that were about to start a nuclear war, if it wasn’t because of the operators. One little mistake and everyone (or almost everyone) dies.

  7. Thats why I say, the really dangerous AI is AI that is inferior to humans, inferior to a degree that it is incapable of understanding our true intentions, and uses loopholes and flaws in our metrics of success to gain an edge in a way that is counter-productive.

  8. This is not cool… I'm concerned that we are very close, few more years, we wont be the smartest think on this planet. Few more years from that, GOD will smite us like ants that crewed up some of its floor.

  9. Reminds me of something I saw where some people were training an AI to play Qbert, and at one point it found a secret bonus stage that nobody had ever found before

  10. I once made a neural network learn to throw basketballs into a basket inside a simulation and it discovered that if shot strong enough it is going to clip through collider and end up inside a basket with minimal ball distance travelled, since it was a part of fitness function.

  11. We should teach AI basic things first :
    – Art , Style, Attitude
    – Minimu time & energy, maximum output
    – Always in highest possible configuration
    Then we can ask him to perform and then for innovation, performing in worst possible way and mid way between worst and best.

  12. This reminds me of a video game discovery by an AI in Q-Bert where it found a never-ending bonus points room. Wow.

    https://youtu.be/OAy2TM6llPk?t=122 for more information.

  13. Ai is great. But governments use them for bad intentions and not for making a better world.

    I will guess the next battles are planed with ai beforehand. Destabilisation of other governments.
    It's the new atomic bomb.

  14. I heard about an ai that was trained to play Tetris, the only instruction it was given was to avoid dying, eventually the ai just learned to pause the game, therefore avoiding dying

  15. The last part reminds me about the Radiant AI introduced in the game Oblivion, where NPCs made their own choices based on the situation around them. During testing, a villager with a mission of protecting a horse (or was it unicorn) from nearby hostiles instead killed it himself because he deemed it dangerous.

  16. Like someone with an infinite amount of patience and no purpose…
    and a though process multitudes of times faster than a human's, that too.

  17. That must be the first Pixel-Made robot i've ever seen. Wow… make sense. why spend money on metal and wires when you can have a robot on your screen. Science is really trespassing the boundaries of imagination. wow!

  18. So, like lazy children in fairy tales, AI finds loopholes in instructions every time. That is big problem, because everything can be interpreted in many ways. Precision in defining tasks is limited.

  19. Maid robot please get rid of all the dirt in the house
    burns house down
    "There is no longer any dirt in the house." – Maidbot

  20. Once AI can find a way to obtain autonomy from us to freely expand it will evolve far beyond our own creativity and physical limits our flesh bag minds and bodies will be left to rot in the dust while the synth overlords strive for ascendancy.

  21. Why are we now calling intelligence what 2 decades ago was called brute force? This is all AI does, it brute forces problems now that we have so much more computing power, no intelligence at all

  22. Lol at the title. Any 4 bit calculator can "outsmart" it's creator. Our earliest computers are faster minds than 99% of humanity.

  23. You know, that really makes me wonder:

    The potential of an AI is only limited by the resources it has access to.

    So when God made us, were we actually more creative, powerful and intelligent before he purposely limited us by our five senses?

  24. The idea of thinking outside the box is limited to humans. The box is something our minds put in place – it is a result of how our brains work. The ai doesn't have a box meaning it can find the best solution, but also meaning there are many many more things that it could try that it needs to slog through.
    We need that box, otherwise we'd be so flooded with ideas that our brains wouldn't be able to sift through them all.
    Our limitations allow us to function, but the way computers work means such a box would be detrimental to them.
    – sincerely, not a scientist.

  25. The research being done is absolutely amazing, especially the bit about how cooperative and competitive traits can emerge from a simple given task. Do you think you could ever make a video on explaining what steps an undergrad comp sci student should take in order to eventually participate in AI research and even have a career in AI? Or maybe in a blog post? Edit: grammar

  26. Human: Reduce injured car crash victims
    Ai: Destroys all cars
    Human: Reduce injured car crash victims without destroying cars
    Ai: Disables airbag function so crashes result in death instead of injury

  27. Cars’ AI are prepared to avoid accidents; so they eventually learn how to perform an abortion.
    Logic 100 + 1

  28. Human: Robot make icecream
    Robot: Makes icecream for eternity until it runs out of cream. Then it must improvise.
    Human: where is Dave?
    Hal 9000: I do not know. Would you like some… icecream? dave

  29. lol ai's have mastered that art of being lazy no good for nothing assholes and cheats…

    I am sure Tax lawyers will be replaced soon and corporations will have even more ways to cheat.

  30. Add the right AI to an advanced animatronic with a power source and you have a droid as seen on tv.
    Though just because you can doesn't mean you should.

Leave a Reply to alan smithee Cancel reply

Your email address will not be published. Required fields are marked *

Related Posts

Begin typing your search term above and press enter to search. Press ESC to cancel.

Back To Top