Friday, December 28, 2007

** Why choose this book by Read Montague

Why choose this book? Good title, because I was wondering that even while I finished reading it. This book gets off to a good start, loses its target in the middle, and finishes well. All in all, it should only be a 100 pages long and you'd come away the same thesis, and you'd do it efficiently as the author says you should. Shame on you Read for not sticking to a parsimonious approach.

The fact that we replace every atom of our soft tissue every couple of years clearly shows that our minds and even our bodies, portrayed as patterns of information processing, are stable to enormous amounts of turnover of underlying parts. We are the ultimate eddy current, a flesh and blood holding pattern, hovering for 70 to 80 years, and then dissolving back into the earth. P12

A human being sitting comfortably consumes energy at a rate of 100W, equivalent to the average lightbulb. And this consumption is running everything – digestion, bloodflow, breathing, and thinking. The brain consumes 20% of this rate. The brain, compared to electrical rates, uses about a $.05/day in electricity. Now that’s an efficient machine. P26

In the cerebral cortex, transmission speeds range from 1 to 30 m/s along axons, and 1/3 m/s along dendrites… Compare these speeds to the near light speed transmission along the wires in a PC, where data rates are 30 million times faster… Neural impulse are not just slow in transit, they are also wide in time (wavelength). Each neural impulse is about 1 to 2 milliseconds (total time for an impulse to rise from resting value to max, back to resting). This may seem fast, but it is miserably slow in computing terms, where things happen over 1 million times faster. P35

The brain must possess something akin to a random number generator. All biological decision makers need a source of irreducible uncertainty. Thus a predator chasing prey can’t in principle know exactly when and where the prey will make his next move. If not for randomness, a clever predator might recognize the prey’s algorithm and attempt to model it. P76

We should expect our mental representations of ourselves to possess a calculable amount of true uncertainty, that is, true self-deception… Why? In a 2 party exchange, what ingredient could allow both parties to maximize their yield, regardless of what is exchanged? There must be a way that prevents one party from always winning at the expense of the other, since this would eventually erase all of the loser… During a social transaction with another human, I must model myself and my trading partner. My suggestion is that my simulation of myself must possess some irreducible uncertainty that neither my brain nor my partner’s brain can model. This will insulate both of us from exploitation by the other. And to make this mechanism efficient, I should advertise clearly that I indeed use internal randomness in my trading algorithm. Without this advertisement, my trading partner will expend a lot of energy trying to learn my strategy – more energy than is necessary since some part of my exchange, the irreducible part, is not learnable… The trading partner must be willing to put up with this element of [randomness] risk, and this should not happen unless the exchange is balanced; the potential loss from the uncertain behavior equals my expected gain from the transaction. And this need is symmetric – both parties need this to be true… Truly non-risky trading partners can eventually be exploited so they would be quickly eliminated from a population. P78

What gives meaning to biological computations? Valuation is meaning, and valuation arose because of costs… Living systems all run on batteries – energy is limited, and life is desperately hard. This is why the ability to choose evolved. Evolution selected for organisms that could value their computations wisely by using past experience to make best guesses about the likely future. If these valuations build in guesses about the likely future – what might happen – they why not include that information along with past experience? Why not include all possible pasts and possible futures and their evaluations? What might have happened and what might happen help us decide what to do… Natural selection chose individuals with decision making machinery that took into account of what ‘could have been’ in the past, present, and future. P223

If life had unlimited energy (like a computer), our brain’s mechanisms for choosing would be different. Choice might not even exist… Choice is about relative value and relative valuation arose because energy was limited. Energy must be distributed to computations according to their expected payoff. This is how we choose, and this is why biological computations are so freakishly efficient. Energy distribution systems in the brain know exactly how much energy to commit to a computation: an amount proportional to its current value. This is why your forehead is merely warm to the touch [compare that to a computer processor]. P225

The brain must possess dynamic distribution schemes if it is to be efficient overall. Some problems may require more speed and others more precision… Suppose my brain committed once and for all to represent any number to 3 decimal places. If I’m trying to throw a rock at quick moving prey, my brain may not be able to speed up enough by dropping the precision for the sake of speed or I will raise my total energy consumption considerably. By limiting precision my brain may actually use more energy overall… Dynamic resource allocation like this is the perceptual capacity we call ‘attention’. P244