TED Conversations

Jennifer Duffy

This conversation is closed. Start a new conversation
or join one »

If we can produce perfect AI do we fully understand the brain?

Imagine the following hypothetical scenario:

From the collaboration of scientists and engineers around the world, a robot has been built that has the exact same motor, cognitive and perceptual capacities of the average human. It is able to learn facts, faces, patterns, and languages, can navigate complex environments, recognize handwriting, discriminate different visual inputs, self-reflect, etc. For any given input, it responds exactly like you or I might. This is a truly high-tech device!
In fact, there are some that claim that the mere existence of such a machine (without looking inside or knowing how it works) is sufficient evidence to say that we understand how the human brain works. There are others that disagree.


Where would you stand on this issue? Do you believe that such an assessment is fair to make? Why or why not? What do you think it means to “understand” how the brain works?

Related Talks:
0
Share:
progress indicator
  • thumb

    Ken brown 30+

    • 0
    Jan 15 2013: We are close to decoding what the brain actually see's through our eye's.
  • thumb

    Mathew Naismith 10+

    • 0
    Jan 14 2013: G'day Jennifer

    I'm a spiritually aware person which says to me that without a soul one can never be truly human, it’s the soul that makes us what we are thought & all. You could also take past life experiences & karma into consideration as well. The brain is subjected to more than what one experiences & knows in one life time I believe.

    Love
    Mathew
  • thumb

    Krisztián Pintér 200+

    • +1
    TED Translator
    Jan 14 2013: not necessarily. to show why, let me call my first witness, the so called "neural networks". neural network is a computational method that involves a rudimentary model inspired by brain cells. they are composed of stupid little automatons that can only sum the total of their many inputs using some coefficients, and output some function of that sum on their output. we then arrange these mindless little items in a huge network. there are mathematical methods to tweak the coefficients in a way that the network can solve very hard problems that are otherwise not easily approachable. we do that by training the network with a large set of examples and their expected results. after a time, the network learns the patterns. as an example, one can train such a network to recognize a hand written numeric digit from its camera recorded image. the network takes the individual pixels of the image as inputs, and "light up" one of the ten outputs as an answer.

    the point is: nobody has a clue what the actual logic is, just by examining the coefficients or the layout of the network. the computer can answer questions, but we don't know how. just as we don't know how the brain solves the same problem. there is an algorithm somewhere, embedded in the thousands of parameters of the network. but having this network helps us in no way understanding it. if you set yourself the task to develop a method of recognizing handwriting, or you just want to understand how the brain does it, this neural network would not be helpful to you.

    • John Smith 30+

      • 0
      Jan 14 2013: Yes, neural networks are an excellent example of us building stuff that's effective without knowing the precise algorithm. Many more examples exist in the field of machine learning and they will be used in any decent AI.
  • thumb

    Allan Macdougall 30+

    • +1
    Jan 14 2013: An intelligence, in the most comprehensive sense of the word, can never fully understand itself. There will always be an impermeable existential/cognitive barrier that only a separate, higher intelligence would have any chance of surpassing.

    Therefore, AI, or anything else our own brain creates will always be level with (at best), or several notches below, our own capabilities - no matter how big the collaboration of scientists and engineers.

    Another point is that self-understanding is absolutely NOT just down to a collaboration of science and engineering. Unless those two disciplines can get to grips with everything that constitutes "Multiple Intelligences" (including, dare I say it, "spiritual intelligence"), then we will remain light-years away from a true understanding:

    http://www.infed.org/thinkers/gardner.htm

    To use a rather silly analogy, a vacuum cleaner could never fully suck itself up, but a bigger, more powerful one potentially could...
  • thumb

    Lawren Jones 10+

    • 0
    Jan 14 2013: That's one of the fundamental principles of science, Jennifer. In order to say we understand something, we must be able to model it successfully (If only mathematically, perhaps). If we can make a perfect AI, then we would say that we fully understand the human brain.

  • John Smith 30+

    • 0
    Jan 14 2013: Such an AI is likely based on knowledge of the human brain and will emulate many information processing techniques of the human brain, yet it will probably operate differently from the human brain in many areas as well. It is also possible an AI is created without its creators really understanding why this particular computer has gained consciousness (through some unexpected and complex interaction) when earlier, more primitive prototypes did not. So no, creating an AI does not guarantee complete understanding of the human brain.
Showing all 7 comments