15/3 DESIGN FOR A BRAIN 



It will be seen, therefore, that in a complex system, every 

 group of stimuli will have a holistic quality, in that the response 

 to the whole group will not be predictable from the responses 

 to the separate stimuli, or even to sub-groups. The dog that 



salivated to each of two stimuli but not to the two together is 

 therefore behaving in no way surprisingly, and such behaviour 

 is no evidence of any ' supra-mechanistic ' power. In complex 

 systems such non-additive compoundings are to be expected. 



15/4. Another variation in stimulus-giving occurs when a 

 pattern is varied in some mode of presentation without the 

 pattern itself being changed, as when an equilateral triangle is 

 shown both erect and inverted. The same argument as before 

 prevents us from expecting any necessary relation between the 

 two evoked responses. 



In some cases the two evoked responses are found to be the 

 same, and to be characteristic of the particular pattern even 

 though its presentation may have been much changed : an 

 object may be recognised though its image falls on a part of 

 the retina never before stimulated by it. This power of Gestalt- 

 recognition was also sometimes thought to demand ' supra- 

 mechanistic ' powers. But in 1947 Pitts and McCulloch showed 

 that any mechanism can show such recognition provided it can 

 form an invariant over the group of equivalent patterns. As 

 the formation of such invariants demands nothing that cannot 

 be supplied by ordinary mechanism, the subject need not be 

 discussed further here. 



To sum up, these examples have shown that no matter how 

 small the difference between stimuli, or initial states, we can, 

 in general, if the system is complex, put no limits to the differ- 

 ence that may occur between the subsequent lines of behaviour. 

 From this we may deduce that if the system is one with many 



168 



