Notes From "Downloading Consciousness: Connectomics and Computer Science"


  • Event: CMU's BrainHub Panel Discussion on Oct 29, 2015
  • Topic:
    Downloading Consciousness: Connectomics and Computer Science

Panel Introduction

  1. Sandra Kuhlman (Assistant Professor of Biological Sciences)
  • The Brain is composed of different cells of which each has different functions.
  • There are also different types of synaptic connections between these neurons (e.g. strong, weak).
  • Given the complexity, pessimistic about being able to learn enough about the brain within the current generation.
  • Need various computational/algorithmic tools for accurate simulation of the brain.
  • Simulation Example: Digital Reconstruction of Neocortical Microcircuitry
    • Simulates 300,000 neurons.
    • Was able to observe different patterns.
    • Was able to somehow represent results from Vevo experiments.
  1. Wayne Wu (Associate Director of the Center for the Neural Basis of Cognition, Adjunct Faculty Member in the Department of Philosophy)
  • The question: How do we get from brain activity to consciousness? What generates consciousness?
  • Answer example: Brain and Mind (Dehaene 2011)
    • Compares to theoretical models of conscious processing, such as the Global
      Neuronal Workspace model.
    • Not all brain activity becomes consciousness (e.g. detected sound vs. non-detected sound).
  • Answer example: Computing Machinery and Intelligence (Mind, Turing 1950)
    • Turing Test: a suggested test to measure how successful you are at creating a machine that thinks.
  1. David Touretzky (Research Professor of Computer Science)
  • AI is like flight. There have been lots and lots of (unsuccessful) trials.
  • AI is a misnomer. No one has ever created artificial "intelligence" yet.
  • But some fake intelligence does accomplish specific tasks.
  • Fake Intelligence Example: IBM Watson
    • Really good at answering formatted questions.
    • But humans can't have a conversation with it.
    • This is due to lack of real understanding.
  • Fake Intelligence Example: Loebner Prize winning chatbots
    • Basically using tricks. It pretends to pass turing test with various hacks (e.g. repeating words, evading details, etc.)
    • However, the machine from the example dialog in Turing's paper doesn't evade at all. Depicts what should be "real" intelligence.
  1. Anind Dey (Charles M. Geschke Director of the Human-Computer Interaction Institute)
  • Imitating the real might be able to be conceived as a representation of consciousness.
  • Imitation example: New Dimensions in Testimony Classroom Concept
    • The representation is only a 3d image.
    • People can ask questions to it, and it can reply with physical gestures.
  • Imitation example: SimSensei
    • Virtual therapist which uses lots of technology (e.g. detecting user's body pose, etc.).
    • Reacts in real time.

Questions From Moderator

Q: It was brought up that even for a machine that seems to imitate humans well, it is still regarded as AI rather than real intelligence. At least how well are we doing at making these AI?

  • David
    • It depends on what task you want it to perform.
    • The interaction has to feel natural for people to accept it.
  • Anind
    • Even achieving a level of going beyond being perceived as fake and diminish disbelief to have people start interacting with them is a great accomplishment.
    • For certain tasks, this imitation might have a place.
  • Wayne
    • Limiting the goals from the task could help, but how much to limit needs negotiation. The Turing Test's expectations are too high.

Q: Lots of big money projects are happening now. Do we need to know the biological part at all? How will computer science people use this knowledge for their algorithms?

  • Sandra
    • This depends on brain task again.
    • There needs to be an agreement on the brain task.
    • Currently, there is no brain task that can be well-simulated yet, but research for doing so is continuing:
      • The nature of multifunctional neurons have been found.
      • Need to record cell activities for human behaviors.

Q: What are your thoughts about the article from the New York Times about attempting to preserve the brain for the future? How would people conceive something that doesn't look like a person but contain memories of the person that you knew before?

  • David
    • It is impossible to preserve the brain with current technology.
      • As in the article, fixation only got to about 20% after a MRI check.
      • This case is more interesting in that it was trying to preserve a broken brain. In order to fix a broken brain to a healthy brain, you need to know how the brain works (which is centuries off).
  • Sandra
    • It usually might be easier to start out new rather than to fix something.
    • But brains also can fix itself from chunks taken out, so it might be easier than that expected to fix it.

Q: Even if you end up with the circuit, how is it different from the mind?

  • Wayne
    • The same questions will arrive about how successful it is of taking place of the mind.
      • If two things are physically the same, is it really the same?
    • There is currently no easy and clean connection from the physical state to the mental state.

Q: What are we learning from biology (e.g. small circuits) that helps scientists from other fields?

  • Sandra
    • Physiological Data -> Computational Simulation -> AI
    • We are getting clues from these research:
      • Brain patterns (from various states, such as sleep)
      • Biological clock (e.g. brain activity is very active at night if under blue light)
  • David
    • Currently, most flows happen from computer science to neuroscience
      • Computer science tries all crazy stuff. Some of those become useful. Then they wonder if that exists in the brain.
    • This flow will eventually turn corner, where neuroscientists will tell computer scientists what are actually in the brain.

Q: What would be the next best thing we could get based on the understanding of the brain?

  • Sandra
    • Contribution to translational research.
  • David
    • Achieving good theories. An example of a good theory is attractor neural networks (something that was found to really exist).

Q: How necessary is the mind? Can we do without it?

  • Wayne
    • Lets say that even if replication succeeds, we still might feel awkward towards it.
    • Need to figure out what the value of the mind is.
    • A possibly hardest Turing Test: having to interact with loved ones.
    • These are hard questions that push us to ethics.
Open Comments