Jeff Hawkins, the guy behind Palm Computing and mobiles such as PalmPilot and Treo, argues in his book, On Intelligence, that the current way of thinking about intelligence and how the brain works is flawed and therefore we cannot create intelligent machines based on this knowledge. He is so convinced of this that he founded his own research institute, the Redwood Neuroscience Institute (RNI), now permanently relocated to the UC Berkeley campus as Redwood Center for Theoretical Neuroscience.
In his book, Jeff Hawkins states that we currently have a lot of data about the brain but not much as regards theories of how it really works. In fact, the Redwood institute task is to ”to use mathematical and physical principles to understand the nature of coding, dynamics, circuitry and plasticity in nervous systems.”
Being a brain geek and proud owner of a brain myself ;), just like Jeff, although definitely with less knowledge on this topic than him, from the books I have read so far on the brain and how it functions I agree with him. An excellent short introduction to the brain is the book, The Brain: A Very Short Introduction, by Michael O’Shea. For a more visual and less technical book I suggest any book by Richard Restak, but specifically The Secret Life of the Brain.
The more you read about the brain to more you realise that a lot of research is being done on how the brain functions, from an electro-chemical point of view, and a lot of data has been gathered which is cool, but when it comes to the really hard questions, such as how is a simple picture like a shape encoded and stored in the neurons, you come empty handed. Does a single neuron store all the information about a shape? If it does, how does it do it? I do not think that a single neuron stores information about a simple shape, but rather a group of neurons interact together to store that information. As in all things in nature I think that the brain tries to economise the use of its resources, and so if it has a neuron with the colour yellow encoded in it, it will reuse it in any group of neurons that need to encode data about anything yellow, for example, a yellow ball.
The above of course is just my idea of how the brain might encode information, but that is the kind of real brain knowledge we need to get before we can create intelligent machines. No wonder that with the current knowledge no computer is capable to recognise simple characters and digits that are slightly morphed or covered with noise, yet humans can do it with ease. This type of problem is called the invariance problem, in that a human being can identify a familiar object, say a tree, no matter from which angle or under which lighting conditions it is presented, variances.
Presently, this is good news for people in the security sector or website owners who exploit this known algorithmic difficulty, through captchas, to defeat bots employed by spammers from flooding websites with comments or doing other malicious stuff.
Some would argue, but is it wise to try and develop really intelligent machines, citing the possibility of these machines doing harm to humans. Jeff Hawkins argues against the idea of intelligent machines ever becoming dangerous to human beings, since he says it will be easier to build machines that can think better than us in higher level thought such as physics and mathematics, than to build walking robots. He is right about creating walking agile robots like Terminator. Just have a look at this video from Boston Dynamics of their Big Dog robot. I studied electronics myself and I can tell you that robot is as advanced as they can be as of now, to handle the rough, uneven, moving, sliding terrain, and the sideways pushing. Yet, it still is light years away from approaching the fluidity of movement and adaptability of a human being.
However, I disagree with Jeff that intelligent machines cannot become evil or used for malicious purposes, since he is assuming that for an intelligent machine to do harm it must move in real life 3D space. In an ever wired world, moving around the globe and causing havoc can easily be done in the not so virtual world of bits and bytes. I mean, look at today’s primitive bots. They still manage to wreak havoc on the internet. Now imagine a really intelligent bot, at least capable of reading captchas, what do you think it would be capable of?
Well, the good news is with intelligent machines there will be both good and evil bots roaming the internet, so the continuous war between white and black hat has no end in sight. Interesting times ahead.
Let me know what you think