My blogs reporting quantitative financial analysis, artificial intelligence for stock investment & trading, and latest progress in signal processing and machine learning

Friday, June 17, 2011

Scientific American: How Simple Photos Could Be Used as a Test for a Conscious Machine

The latest issue of Scientific American has an article: How Simple Photos Could Be Used as a Test for a Conscious Machine by  Christof Koch and Giulio Tononi (you need to access the issue to see the full-text. The website just provides you an introduction to the article). It is very interesting. If I have time this summer, I will attend the competition. In fact, I have several ideas to cheat their smart computer. But I don't want to say much right now. However, I'd to say, the picture that the authors provide (see below) is easily recognized by computer algorithms as a unreasonable picture. Their algorithm just needs to do object recognization and then do some semantic reasoning, then it can know this picture is not reasonable. So, don't be misled by the authors' picture.


(The picture's credit to Scientific American)

6 comments:

  1. Trying to imitate conciousness with current computer technology have serious limitations. One needs a mechanism with an architecture similiar to a neurological system to even get close.

    ReplyDelete
  2. Thank you for your comments.

    I agree your first sentence. But I want to say, AI and ML(machine learning) are fast evolving fields. Currently they may not achieve good performance in terms of mimicking human wisdom, but in the future they may. For example, in image understanding, Objective Recognition + Space Allocation + Semantic Reasoning + sufficient memory storage storing priori knowledge is powerful to understand challenging “new” photos/pictures if we have good algorithms in each part. By “new”, I mean there are no similar photos/pictures storing in the memory or in the training set.

    For your second sentence, I do have several concerns. First, I don’t know what’s you mean by “mechanism with an architecture similar to a neurological system”. I am curious about such architectures. I don’t keep my eyes on the latest findings on such architectures. But based on my knowledge, most existing architectures are more like a computer technology (ML/AI algorithms). The basic reason is that we are not clear about what are the communication/information processing mechanisms inter/intra groups that consists of hundreds of neurons, needless to say groups consisting of tens of thousands neurons or more. Even for single neurons, such mechanism is not much clear. So we do not have such architecture governed by some mechanisms that close to the real mechanisms in neurological systems.

    Based on my knowledge, the existing architectures (or artificial neural networks) are more based on people's simplest hypothesis (e.g. sparsity, economical energy, optimal information flow) and mathematical convenience. Let's take the neural coding in early vision as an example. One of the popular methods is sparse coding via ICA (or Boltzman Machine, Deep Belief Network, etc). And we did find they can provide the similar phenomena as simple cells and complex cells, and we also did use these techniques to do face recognition with good recognition rates. However, such artificial neural networks per se are far from the real neural systems in the brain.

    So, I don’t see the strict bound between current AI/ML algorithms and the architecture with similar mechanisms as in neurological systems

    My second concern is whether a fancy architecture using mechanisms similar to a neurological system can win an AI/ML algorithms? People in computational neuroscience tend to ignore the importance of reasoning when they build some architectures. But the semantic reasoning is very important in a system that mimics human wisdom.

    ReplyDelete
  3. Hi Zhilin,
    What am thinking about is that to be able to really model consciousness you need integrating knowledge from different layers. In my opinion current instruction based CPU architecture, where you scan pixels using for loops, is doing something else other than modeling consciousness (still very useful of course). Their was an article in IEEE spectrum last year suggesting a possible alternate computer design , but unfortunately I can't find it.

    ReplyDelete
  4. Hi, Hasan,

    Do you mean the attention is an important difference between current computer techniques (modeling consciousness) and neurological systems?

    I'll look for that paper. Hope I am lucky finding it.

    ReplyDelete
  5. This is an example of the new hardward I was describing, just announced by IBM:
    http://www-03.ibm.com/press/us/en/pressrelease/35251.wss

    ReplyDelete
  6. Thank you very much. The news is very interesting.

    ReplyDelete

Note: Only a member of this blog may post a comment.