DESIGN FOR A BRAIN 16/13 



that have an effect on the first reaction, and thus little loss in the 

 first adaptation. Thus the multistable system, without further ad 

 hoc modification, will tend to take advantage of the recurrent situation. 



16/13. It is of interest to notice that when two stimuli (or 

 parameter- values) are widely different, the multistable system 

 will tend to direct the activations to widely different sets of 

 step-mechanisms. It thus provides, without further ad hoc 

 modification, a functional equivalent of the gating mechanism 



r of s. 10/9. 



16/14. Conversely, as the two disturbances (or stimuli, or para- 

 meter-values) tend to equality, so will the overlap of the two 

 activated sets tend to increase. A large overlap in the step- 

 mechanisms will mean that the second set of trials will be severely 

 destructive to the first adaptation. Now the tendency for new 

 learning to upset old is by no means unknown in psychology ; and 

 an examination of the facts shows that the details are strikingly 

 similar to those that would be expected to occur if the nervous 

 system and its environment were multistable. In experimental psy- 

 chology 4 retroactive inhibition ' has long been recognised. The 

 evidence is well known and too extensive to be discussed here, so 

 I will give simply a typical example. Muller and Pilzecker found 

 that if a lesson were learned and then tested after a half -hour 

 interval, those who passed the half-hour idle recalled 56 per cent 

 of what they had learned, while those who filled the half-hour 

 with new learning recalled only 26 per cent. Hilgard and Marquis, 

 in fact, after reviewing the evidence, consider that the phenomenon 

 is sufficiently ubiquitous to justify its elevation to a 4 principle of 

 interference \ There can therefore be no doubt that the pheno- 

 menon is of common occurrence. New learning does tend to 

 destroy old. 



In a multistable system, the more the stimuli used in new learn- 

 ing resemble those used in previous learning, the more will the 

 new tend to upset the old; for, by the method of dispersion 

 assumed here, the more similar are two stimuli the greater is the 

 chance that the dispersion will lead them to common variables 

 and to common step-mechanisms. In psychological experiments 

 it has repeatedly been found that the more the new learning 

 resembled the old the more marked was the interference. Thus 



216 



