For this vehicle, Braitenberg's design innovation is to adjust the thresholds of units on the basis of the total number that are active. If too many processors are active, then thresholds will be raised, and if too few are active, thresholds will be lowered. The point of this innovation is to permit "the possibility of sustaining long successions of distinct brain states for the purpose of exploring knowledge already incorporated in the brain.
The immediate motivator for the innovation comes from a "remark on pathology" -- namely, with the kinds of modifications discussed so far, at some point processing connections will be come strengthened to the point that *all* processors will turn on, and will never turn off. This is a "mental locking", and is the kind of problem that has been raised in discussion of early semantic network models that work via spreading activation.
To solve this problem, one needs the capability of adjusting the thresholds of the vehicle's processing units, in order to turn them off when too many become active. "So if we watch the operation of the brain -- and in particular the total amount of activity in it -- we can always prevent an attack of epilepsy by raising all the thresholds. If there is not much activity, we can lower all the thresholds and thereby encourage the circulation of activity through the brain." This defines a kind of global negative feedback on overall internal processing of the vehicle.
Importantly, this "kludge" to solve a vehicular pathology has some interesting unanticipated effects. For example, Braitenberg envisages how this procedure will produce a focusing of attention. For instance, when an object that triggers an internal concept (i.e., a cell assembly) is present, the activity of the assembly will (of course) increase. This in turn, though, will cause thresholds throughout the system to raise, which will effectively turn off all other processors except those detecting the object (because the presence of the object will ensure that they remain active.
This focusing of attention has a further implication, in the sense that the only associations that can be learned in the system are those that are sufficiently strong to survive this "dampening" of the system. This will be true for both simultaneous and contiguous associations. These latter are of particular interest:
"The threshold control, alarmed by all this growth of activity, will quickly raise the thresholds, smothering most of the activity and leaving only the most resistant group of elements activated. As we have already seen, this will be the group with the strongest reciprocal connections. In terms of concepts we may put it this way: The next concept, among all the concepts that are possible consequences of the present one, will be the most consistent or familiar one -- the one most strongly established by experience."
With this then, the prediction is that as these familiar contiguous associations grow, the threshold mechanism will produce a number of sequential state changes in the internal workings of the vehicle. It will progress through a chain of associations. "The upshot is something very much akin to thinking, to that process so familiar to our introspection, where images appear in succession according to rules reflecting the relations between the things they stand for." This chaining of internal states achieves the intended goal for the vehicle, extended periods of "thinking."
Again, with respect to the major theme of this book, an observer watching a Vehicle 12 without knowing its internal structure will find the behavior very difficult to predict. Indeed, the observer might attribute free will to this kind of machine because of this. (For Braitenberg, free will amounts to indeterminacy. If we can't even predict how many processors will be active from one time to the next, then "this is proof of FREE WILL in Vehicle 12.")
Pearl Street | Vehicles Home Page | Dawson Home Page |