DESIGN FOR A BRAIN 8/6 



8/6. We can now appreciate how different an ultrastable system 

 is from a simple stable system when the conditions allow the 

 difference to show clearly. 



The difference can best be shown by an example. The auto- 

 matic pilot is a device which, amongst other actions, keeps the 

 aeroplane horizontal. It must therefore be connected to the 

 ailerons in such a way that when the plane rolls to the right, its 

 output must act on them so as to roll the plane to the left. If 

 properly joined, the whole system is stable and self-correcting: it 

 can now fly safely through turbulent air, for though it will roll 

 frequently, it will always come back to the level. The Homeostat, 

 if joined in this way, would tend to do the same. (Though not 

 well suited, it would, in principle, if given a gyroscope, be able to 

 correct roll.) 



So far they show no difference; but connect the ailerons in 

 reverse and compare them. The automatic pilot would act, after 

 a small disturbance, to increase the roll, and would persist in its 

 wrong action to the very end. The Homeostat, however, would 

 persist in its wrong action only until the increasing deviation 

 made the step-mechanisms start changing. On the occurrence 

 of the first suitable new value, the Homeostat would act to stabilise 

 instead of to overthrow ; it would return the plane to the horizontal ; 

 and it would then be ordinarily self-correcting for disturbances. 



There is therefore some justification for the name ' ultrastable '; 

 for if the main variables are assembled so as to make their field 

 unstable, the ultrastable system will change this field till it is 

 stable. The degree of stability shown is therefore of an order 

 higher than that of the system with a single field. 



8/7. The experiments of Marina and Sperry provide an excellent 

 introduction because they are conceptually so simple. Some- 

 times a simple experiment on adaptation may need a little thought 

 before we can identify the essential features. Thus Mowrer put 

 a rat into a box with a grilled metal floor. The grill could be 

 electrified so as to give shocks to the rat's paws. Inside the box 

 was a pedal which, if depressed, at once stopped the shocks. 



When a rat was put into the box and the electric stimulation 

 started, the rat would produce various undirected activities such 

 as jumping, running, squealing, biting at the grill, and random 

 thrashing about. Sooner or later it would depress the pedal and 



108 



