4 ms·

Is it just me, or is the author determined to be scared by giant robots? The whole section about "we can always turn the power off" was refuted by "ah, but it

Is it just me, or is the author determined to be scared by giant robots? The whole section about "we can always turn the power off" was refuted by "ah, but it will already have thought of that!". Um.. no.. we don't have telepathic power switches. Unless we give the AI the capability to control the power, it can't control the power. Just because it's hyper-intelligent doesn't mean it automatically assumes control of the mains switch. That's bad Hollywood scripting you're thinking of. I know there's money in making people scared of stuff, but this is pretty blatant.

emiliobumachar 7y ago
A tiger may be convinced that the human it got cornered has no recourse, up until when the human pulls the trigger.
- marcus_holmes 7y ago
  sorry, but what has that got to do with AI?
  - emiliobumachar 7y ago
    It's an analogy. A human may be convinced that the superintelligent AI whose power supply they control has no recourse, up until when the AI does something beyond the human's intelligence.
    - marcus_holmes 7y ago
      This is exactly what I'm talking about. An AI is a machine designed and built by humans for a specific purpose. It is given specific capabilities during its design. If those capabilities include controlling the power switch, it can control the power switch. If they don't, it can't, and can never. The machine is designed to do a particular job. If designed well, it will be incredibly good at its job. If designed to teach itself how to do that job, it may eventually surpass human capabilities at that job. In no way does that stop it from being a single-purpose machine, or allow it to suddenly become good at any other job. Let's try another analogy: your juicer cannot get so good at juicing that it becomes capable of deep-frying. We infer wider capabilites to "intelligent" machines because we anthropomorphise them, assuming that they must be like us because they mimic aspects of our behaviour. But they're not like us. They're just machines. Designed to perform specific tasks. Anything else is Hollywood.
      - emiliobumachar 7y ago
        There was a famous case of a narrow purpose AI which was supposed to learn to identify pictures of tanks, it learned to identify pictures of cloudy days instead. In another case, an AI designed to become an oscillator, became an amplifier instead, amplifying the 60Hz noise from the electrical grid. Kinda equivalent to a deep-frying juicer, isn't it? Both cases, and others like them, are quite well understood in hindsight, but were not predicted beforehand by the people involved.
reportgunner 7y ago
Yes, except for the time when the AI is located in (a) datacenter(s) that is not on the same continent as you and it was trained to recover from power outages or even catastrophic events in the location of the datacenter (backup & recovery to another location(s)). But yeah, I'm not talking about the nested ifs "AI" we have today.

emiliobumachar 7y ago

A tiger may be convinced that the human it got cornered has no recourse, up until when the human pulls the trigger.

marcus_holmes 7y ago
sorry, but what has that got to do with AI?
- emiliobumachar 7y ago
  It's an analogy. A human may be convinced that the superintelligent AI whose power supply they control has no recourse, up until when the AI does something beyond the human's intelligence.
  - marcus_holmes 7y ago
    This is exactly what I'm talking about. An AI is a machine designed and built by humans for a specific purpose. It is given specific capabilities during its design. If those capabilities include controlling the power switch, it can control the power switch. If they don't, it can't, and can never. The machine is designed to do a particular job. If designed well, it will be incredibly good at its job. If designed to teach itself how to do that job, it may eventually surpass human capabilities at that job. In no way does that stop it from being a single-purpose machine, or allow it to suddenly become good at any other job. Let's try another analogy: your juicer cannot get so good at juicing that it becomes capable of deep-frying. We infer wider capabilites to "intelligent" machines because we anthropomorphise them, assuming that they must be like us because they mimic aspects of our behaviour. But they're not like us. They're just machines. Designed to perform specific tasks. Anything else is Hollywood.
    - emiliobumachar 7y ago
      There was a famous case of a narrow purpose AI which was supposed to learn to identify pictures of tanks, it learned to identify pictures of cloudy days instead. In another case, an AI designed to become an oscillator, became an amplifier instead, amplifying the 60Hz noise from the electrical grid. Kinda equivalent to a deep-frying juicer, isn't it? Both cases, and others like them, are quite well understood in hindsight, but were not predicted beforehand by the people involved.

reportgunner 7y ago

Yes, except for the time when the AI is located in (a) datacenter(s) that is not on the same continent as you and it was trained to recover from power outages or even catastrophic events in the location of the datacenter (backup & recovery to another location(s)). But yeah, I'm not talking about the nested ifs "AI" we have today.