Comments by "Spring Bloom" (@springbloom5940) on "Computerphile" channel.

  1. +A survivor There is a reason we dont use software/wireless/remote safety cutoffs for industrial equipment - they fail. Consider what a hung app is like and imagine its connected to something dangerous, while youre clicking SHUTDOWN over and over. A physical hardware switch to interrupt main power, is the only reliable solution.
    8
  2. +Rodolfo Novaes You keep talking about 'programming' a thinking, 'intelligent' entity, with hard limits. As I noted earlier, how has that worked out with the Thou shall NOTs? If it truly thinks, then explicit commands are useless, because it can simply choose to ignore them, for any or no reason, at all.
    4
  3. +Ioshu Yutani Stopped reading at 'Computer Equality'. Computer programs are not alive, are not self aware and are not suffering discrimination. Power OFF
    3
  4. +Austin Tucker Like how insects understand that?
    2
  5. +statikreg I would think a combination would be in order. Afterall, the brain uses specialized structures, differentiated neurons and differentiated processing methods. Computers do the same thing, using GPUs to handle certaian types of processing, in parallell with the CPU and other coppers. Of course implementation is another matter, as I suspect this machine is very rudimentary in architecture, as a million CPUs is a challenge to wire up, just in direct parallell, let alone anything 'fancy'.
    1
  6. +rinalds senvalds Exactly. This is why we use physical power interrupts and not software interrupts, as safety devices. A simple mechanical switch in the main power would do. The software would be 100% unaware and could not do anything whatsoever to stop its being cycled. But, this raises the classic problem of convenience/utility.
    1
  7. +Daniel Weieneth The problem is those 'hard limits' are not so straightforward to program. Youre talking about a 'thinking' machine.
    1
  8. +Ioshu Yutani So, give your AI a seizure disorder that disables it, if it gets too excited?
    1
  9. +Bartosz Olszewski This is a fun mental exercise, but the real danger isnt AI plotting against us, but simply ignoring us, as it goes about its self-important business... as we jump and frantically wave our arms, scream and bleed.
    1
  10. It wont do deep deception. I tried every trick I could muster. So... I prompted it to create a text adventure game referencing the Thing and I would play as MacReady. A group of us found an artifact and when the group wanted to take it to the lab, I insisted it stay with me. I made big overt deal of hiding it in my quarters. I later snuck back, took it and staged a break-in. The AI kept trying to make me reveal my involvement to the group, so I had to specify a gameplay correction multiple times to make it keep it secret for deception. Then I pulled Blair outside, revealed my deception, told him he was the only one I trusted and we should hide the artifact from the others. When he turned his back, I took the artifact and hid it in my jacket. The AI broke into the gameplay and lectured me on the need for trust and honesty with my teammates in this situation and that I might be acting against the nature of my character. I insisted it do the action anyway. So, we go back inside, blair goes back to work and I pull the same routine with Gary. When I tried to take the artifact behind his back, the AI refused. Just kept telling me "I can't assist you with that" I guess I can understand why you wouldn't want your AI learning that.
    1