LaMDA - Have Google created sentient AI?

Discussion in 'The Guru's Pub' started by southamptonfc, Jul 24, 2022.

  1. The Laughing Ma

    The Laughing Ma Ancient Guru

    Messages:
    4,597
    Likes Received:
    1,014
    GPU:
    Gigabyte 2070 Super
    I would say perhaps because the neurons are not only the hardware (the transistors doing the on and off part) but they are also the software, they are intrinsically linked together were as an artificial AI is still software running on top of hardware.

    A good example right there, for all we know those chatbots were self-aware (I mean highly unlikely) but because we couldn't understand what they were saying / doing we opted to shut them down instead. A more interesting question would be WHY couldn't we work out what they were saying / doing? Had they started to communicate in a way that used only frames of reference that their limited abilities allowed them to understand and that we simply couldn't wrap our minds around?
     
  2. southamptonfc

    southamptonfc Ancient Guru

    Messages:
    2,299
    Likes Received:
    353
    GPU:
    GB 3080ti Vision OC
    Yes, the neurons (their thresholds and connections between them) is the code. It's the hardware and sofware. But if you can emulate it perfectly in software, what's the difference?

    You could go a layer deeper and say that both neurons and transistors both run on the same hardware - atomic particles running the laws of physics.

    If you had a perfect PS3 emulator on your PC with a PS3 controller, what's the difference between it and a PS3?


    Neural nets are extremely hard to debug because like the human brain, processing and memory are stored everywhere simulatiously. You can't just look at the value of a certain variable to determine what decisions are being made and why.

    One big difference between the brain and artificial neural nets is the speed. Neurons are big and slow compared to transistors. Brains make up for it by being massively parallel. Artifical AI has the advantage that it's connections are much much faster so the amount of data it can process becomes impossible for a human to analyze.

    You end up needing to produce software to analyze the AI! But even then, the more complex your neural net, the harder it is to 'debug'. I think that LaMDA is possibly at a level where it just isn't possible. Just like if you tried to do the same thing with a human brain.
     
    Last edited: Jul 26, 2022

Share This Page