grasped through numerous examples, but which we do not yet know how to characterize precisely. The difference feels, purely metaphorically, like the difference between the direct sum of two matrices and their Kronecker product; or one might feel that one structure is somehow built out of ingredients bearing the mark of another, in the way perhaps that a quotient group tells how one group is built up out of another. When the infant sucks the nipple, his hands may open and close in the same rhythm as his lips; and guided by a host of similar occurrences we may say that the schema of grasping and taking in is extended to movements of the hand, and comes ultimately to permeate more and more movements. This is one reason why living organisms, in contrast to machines, can perform the same purposive and skilled action with any available set of effectors.

In present design of learning devices a typical elemental mechanism is the association of elements of information. In biological thought and action we typically find, in addition, that objects and processes become incorporated into ever-enlarging schemata in the way hinted at above.

Present machine design may lead to a single device that transforms all patterns of a large class to size- or orientation-invariant form. Biological systems may in addition develop such transformations in a piecemeal way. For instance, a child learns some of his spatial abilities by looking at and grasping various objects here, then there, then another place. This is much less efficient for any particular ability than the former approach (which is also common in living organisms), but it can lead to highly integrated total patterns of behavior as the grasping schemata gradually permeate increasingly varied and extensive schemata.

In present design the receptor system is generally separate from the effector system. In living organisms it is a corollary of the discussion of the piecemeal elaboration of perceptual schemata through the piece by piece integration of schemata of action, that a machine that really can be said to perceive will not just perceive, but will have potentialities for performing skilled actions, at least in rudimentary form.

Now what techniques might be used to implement these principles? Are they just as elusive and ill-defined? Without trying to be at all precise or complete, we believe that in the living organism they involve such things as some sort of activating impulses, scanning, servomechanical processes which tend to conserve useful or interesting states, and hierarchical organizations of all the above principles and structures. These are all conceptually familiar to the engineer. In other words, if we can clarify the meaning of principles of behavioral organization, such as the five which I have mentioned, we have tools now to implement them.

If we think only in terms of the customary trends of emphasis, we shall continue to deal brilliantly with the mechanical manipulations of symbols for our ideas, but the mechanical synthesis of the thought itself will seem supernatural. I believe that appreciation of some of the above-mentioned biological patterns of emphasis makes this synthesis seem remarkable, but not supernatural. Although some of these points may seem like describing the same things in two different ways, it can be shown that the two points of view actually can lead to different research objectives.

To employ this biological pattern of emphasis would be merely silly in designing machines with typical present-day goals; and to abandon these goals and the much more efficient machine approach to them would be unthinkable. To aim at duplicating living organisms would be quixotic. But somewhere in between I believe there is a level of great flexibility and versatility of perception and skilled action which engineers are going to achieve through the use of these biological methods. They are altogether compatible with all present techniques, and in fact would depend upon them. Their rudiments may be clearly seen in a number of programs which are in working order at present. I hope that more people of varied experience will become enthusiastic about keeping these goals in mind as they go about doing the kinds of things reported at this symposium.

A FINAL QUESTION

Some years ago some bright young people
set up
atomic
an
chain reaction. Now
some of them regret that they gave away
their new toy too readily. Do you think
that something comparable will arise in
bionics?''

Answer (Dr. McCulloch): The problem is much more serious than you realize. Whatever any scientist produces that is bona fide science will, as you have just heard, be used, and whether it is used for good or for evil depends not on the scientist but on mankind. If we decide to blow ourselves to hell we will probably do so. If we decide, instead to turn into an ant heap we will probably do so, but as long as we have the stuff of life in us, it will be those who are alive at that time who will determine whether it is used for good or for evil. There are only some things I have to make clear to you. I have no doubt that mankind will booby trap the world. I have every desire to see to it that his information handling is sufficiently reliable so that the boobytrap is not sprung by accident, by enemy intervention, by the "nut" factor, or by something else with which I am less familiar. The necessary good here exceeds the potential evil.