A comment


There is a site called Less Wrong that I visit (here) because occasionally there is an outstanding post there. I do not comment on the posts as a rule because it is something of a boys club of AI guys and I don’t feel that I belong. But last week there was a post that got me a little worked up and I commented. My efforts lost me some karma but never mind, I didn’t know I had been playing the Less Wrong game. Here is the comment:

“The local worldview reduces everything to some combination of physics, mathematics, and computer science, with the exact combination depending on the person. I think it is manifestly the case that this does not work for consciousness.”

No it doesn’t work because you have left out BIOLOGY. You cannot just jump from physics and algorithms to how brains function.

Here is the outline of a possible path:

  1. We know that consciousness has an important function because it consumes a great deal of energy – that’s how evolution works.

  2. Animals move – therefore they must have a model of where they are, where they are going etc. – like the old Swedish joke, ‘I cant yump when I got no place to stood’.

  3. To make a model, animals need to sense the environment and translate the info into elements of the model (perception).

  4. In order to use the model to plan and monitor motor action, they have to also model themselves – so the model is of the animal-in-the-world – the tree is not the real tree in reality but the modeled tree and the me in the model is not the real me in reality but the modeled me.

  5. In order to make a good model that was useful it would have to be a unified global model of the animal in the world – all the parts of the model have to be brought together in order to create the best fit scenario and in order for various functions to use the information.

  6. In order to make a good model that could be used to plan and valuate actions it would have to model the needs of the animal such as goals, motivations, emotions etc – the model has to have a theory of mind for the animal – so my thoughts in the model are not my real thoughts in reality but the modeled mind. When we introspect we are aware of our model of ourselves but not of ourselves in reality. Definitions can be a problem here – do we use the word ‘mind’ for cognition or for awareness? For we have trouble if we confuse these two things.

  7. To make the model more useful it should be predictive to overcome the time it takes to construct the model – so if ‘now’ is t, then the model would be created from the information the brain has at t-x used to predict what reality will be after x duration where x is the time it takes to construct the model – this allows errors in motor actions to be monitored and corrected because the sensory data coming it does not match the model prediction – even the ‘now’ is a modeled now and not the now in reality.

  8. So the biological criteria for a good model are unity, speed, accuracy and predictive power. The elements used to create the model must be easily manipulated in order to achieve these goals and must also be capable of being stored as memories, imagined, communicated etc. The qualia of the model will be anything and everything that is biologically possible and makes a good model. We have the data that the sense organs can measure and some effective ways of representing that information in the model.

So the question “Why red?” can be answered with “Why not – it works.” And the question “Where is the red?” can be answered by “In the structural elements of the model”. If someone has a better way to model the frequency of light, I have never heard of it.

If you cannot envisage this modeling as a sequential computer program that is because it isn’t one. It is a massively parallel assembly of overlapping feedback loops that involve most of the cortex, the thalamus, the basal ganglia and even points in the brain stem. It has more in common with analogue computers then digital ones.

Leave a Reply

Your email address will not be published. Required fields are marked *