Short-Term Learning and Interocular Transfer in Detour Experiments with Octopuses

Octopuses will make detours in order to reach prey that is not directly accessible to them. Using the apparatus shown in Fig. 1 Wells (1964) trained octopuses to make detours to reach crabs shown in the feeding compartments flanking the central corridor. Eight out of twenty-nine animals tested completed detours successfully on entering the corridor for the first time at their first trial. The remaining animals nearly all entered the corridor but failed to complete detours until they were shown a crab at the far end of the corridor and were led round into the feeding compartment. There-after most of the octopuses made detours reliably without intermediate inducements. Thus in 220 trials carried out with eleven animals, each run for twenty trials beginning with its first completed unled detour, 197 correct responses were made. There were only ten errors (occasions at which the octopus turned into the wrong feeding compartment) and thirteen failures to run (octopus had not reached a feeding compartment within 10 min of the start of the trial).

The results from a variety of experiments (Wells, 1964) indicated that the detour was guided visually. Correct performance seemed to depend on the animals’ fixating the wall on the goal side when it went into the corridor and searching along this for a gap. Animals trained to run by means of a series of trials all on one side performed correctly first time when tested by showing them a crab in the other feeding compartment, so that there would appear to be no question of their having learned to follow a particular sequence of visual, olfactory or tactile cues in the maze.

Normally, a detour took 20 to 30 sec., but individuals varied and the best performers regularly completed their runs within 10 or 20 sec. of entering the corridor.

The account that follows deals with the effect of imposed delays, during which the animals were shut into the corridor in the course of their detours. It is also concerned with the effect of removing the vertical lobe from the brain. The vertical lobe is known to be important in visual learning (Young, 1961, 1964) and its removal prevents interocular transfer in some visual discrimination experiments (Muntz, 1961 a−c).

The experiments reported in this account were made during July and August 1964. Octopus vulgaris Lamarck from the Bay of Naples was used. Individuals weighing from 200 to about 450 g. were put into separate aquaria on arrival, and fed upon crabs. Only those that attacked crabs regularly and rapidly were used for the experiments, being transferred to homes in the detour apparatus about a week after capture. In most cases they were free to roam about the apparatus for 24 hr. or so before the experiments began, and during this time were fed once or twice at the end of the corridor close to the choice compartment.

The apparatus used is shown in Fig. 1. There were several sets of this apparatus and the corridor length varied somewhat, from 45 to 55 cm.; this variation made no detectable difference to the animals’ performance, and in general the larger individuals were used in the longer mazes. At either end of the corridor slots held transparent doors that could be lowered to delay the octopus at the ‘choice’ end of the corridor or to prevent its return home. The sequence of left and right trials was determined by tossing a coin, except during parts of the experiments including enforced delays when the normal pattern was left, right (both delayed), left, right (undelayed), left, right (delayed) and so on. Any octopuses that proved reluctant to make detours were trained to do so by showing them a crab, first in one of the feeding compartments and then in the corridor. If they still failed to run through to the feeding compartment they were led round in pursuit of the crab. About half of the animals had to be led round and trained this way, the rest completed detours at their first trial. For each animal, the trial sequence summarized in this account dates from the first occasion on which the octopus completed a detour—whether correctly, or incorrectly, into the wrong feeding compartment—without being shown a crab in the corridor. If, having once completed a detour, an octopus failed to do so within 5 min. of the beginning of a subsequent trial, this trial was recorded as ‘failed to run’. Following a few of these occasions animals were led through and fed, but there was normally no further training after the first successful unled run.

Failures to make detours were rarely due to failure to attack. In nearly every case the octopus attacked persistently and vigorously and in the great majority of instances also made several excursions into the corridor. But instead of completing the run down the corridor and through the choice compartment it would return home and to the direct attack, or remain wandering in the maze until the end of the 5 min. trial period. There were normally six to eight trials per day, with an interval of about an hour (to enable the octopus to eat its crab) between trials. Having completed detours, the animals were allowed to return home directly by raising the maze.

Operations on the brains of the animals were carried out under 3 % urethane anaesthesia. The lesions made were checked from serial sections at the end of the experiments. Removal of the vertical lobe produces no visible motor defects.

In all, twenty-seven animals were run for a total of 1071 trials, at 883 of which they completed detours. Five of the octopuses had their vertical lobes removed before they were tested in the maze. The remaining twenty-two were run for ten or twenty trials as unoperated animals and ten of the most consistently successful of these were selected for operation and further testing after vertical lobe removal. Thepreoperational and postoperational performances of these ‘best 10’ animals can be compared. The animals even amongst the ‘best 10’ did not all have the same sequence of tests, and details of some typical individual sequences together with the scores made at each stage are given in Fig. 2. There appeared to be little or no improvement with practice after the first few trials (as in similar experiments reported in Wells (1964)) and the order in which the series of delayed and undelayed trials were carried out appeared to be irrelevant. A summary of all the trials made and the results obtained is given in Table 1.

The results show quite clearly that octopuses with the vertical lobes damaged or removed are at some disadvantage compared with unoperated animals in this sort of experiment. In a total of 338 undelayed trials, the fifteen animals with brain lesions made forty-nine errors, ending up in the wrong feeding compartment, and they failed to make detours at all on sixty-three occasions. Their score of 67 % correct responses is poor compared with the 82% made by twenty-two unoperated controls (ten of them the same animals, tested before operation) that together made only twenty-eight errors and failed to make detours fifty-three times in 448 trials.

The difference is even more marked when the results of delayed trials are considered. In all, the operated animals were subjected to 133 trials with enforced delays. In these they made twenty-seven errors and failed to make detours on forty-six occasions —only 45 % correct responses. In a similar series of 152 delayed trials the unoperated octopuses made twenty-two errors and failed to run twenty-six times—68 % correct.

The comparatively poor performance following vertical lobe removal could be due to defects in the memory system. One might suppose, for example, that the operated animals tend to forget where they are going before they arrive in the choice compartment. Or alternatively, that they are unable to compute the whereabouts of the crabs that they see relative to themselves so that they are lost once they pass into the corridor and out of sight of the goal. But before considering explanations of this type, there are some simpler possibilities that must be eliminated. One of these is the possibility that the operation slows the animals down, so that they run through the maze more slowly. It will be shown below that the chance of error rises with the time taken to make the detour, both in controls and in operated animals, so that any operation slowing locomotion may be expected to lead to an increase in mistakes for that reason alone.

A second possibility, since vertical lobe removal is known to interfere with interocular transfer under at least some conditions (Muntz, 1961b), is that the operated octopuses became confused on occasions when they changed the leading eye in the course of making a detour. Octopuses are potentially binocular, in that the visual fields of the two eyes overlap in front and behind. But they rarely if ever attack objects that they see using both eyes at once, and when moving from place to place at other times seem always to proceed with one side of the head facing forward. The ‘leading’ eye, here and subsequently in this account, means the eye facing in the direction in which the animal was moving at the time of the observation.

A more detailed analysis of the completed runs made by operated and unoperated animals, summarized below, indicates that neither slow running, nor eye changing can account for the poor performance of the operated octopuses.

Figure 3 a summarizes the performances of the twenty-two unoperated octopuses in 395 completed detours. It can be seen that the great majority of detours were completed within 30 sec. of the octopus entering the corridor and that there were very few errors in these quick runs, only three in 270 completed detours. But the animals did not always run so rapidly. In many instances they paused in the corridor or choice compartment, or wandered back and forth along the wall separating them from the crab before finally leaving the corridor through the exit gateway. In a number of instances an unusually long time was recorded because the animals, having arrived in the choice compartment, explored the walls and floor of this before finally passing into one or other of the feeding compartments (the crab seen from the home was removed as soon as the octopus went into the corridor, so there was no possibility of a choice based on direct observation). The chances of an incorrect choice of feeding compartment rose steadily with the increase in time spent in the detour. Runs completed within about 2 min. were likely to be correct. But an animal that spent longer than this in the maze was as likely to finish its detour in the wrong feeding compartment as in the correct one (Fig. 3a).

Again, the majority (169 out of 275) of the detours were completed within 30 sec. (Fig. 3b). This implies that animals with the vertical lobes removed can, and generally do, move through the apparatus as rapidly as unoperated octopuses in this type of experiment. As with controls, the proportion of errors made was relatively low in rapid detours, only eight out of the total of forty-nine errors occurring during the 169 detours completed within 30 sec. There was a progressive rise in the probability of error with slower runs (Fig. 3b).

Octopuses in this group were subjected to a total of 152 trials that included enforced delays in the corridor. Delays (10 series of 10 at 30 sec., 4 series at 1 min., and 12 trials with waits of 45 sec −2 min., see Table 1) were imposed by lowering a shutter to close the exit from the corridor into the choice compartment. In most series, including all delays of 45 sec. and over, the shutter at the entrance to the corridor was also lowered as soon as the octopus had gone in. Delays were timed from the moment of entry into the corridor, the ‘exit’ shutter being raised at the end of the period of enforced delay. Such treatment ensured that the octopuses took at least 30 sec. (or whatever delay was imposed) to complete their detours, and thus eliminated the very fast runs typical of undelayed trials. Behaviour during the enforced delay varied considerably. The best scores were made by animals that moved round on to the wall separating them from the crab as they went into the corridor and proceeded along this to the exit doorway, waiting there until the shutter was raised. Octopuses that wandered about in the corridor during the enforced delay period were considerably less likely to complete the detour successfully when the shutter was raised.

Figure 4a summarizes the results of delayed trials with unoperated animals. There were ten errors in the 102 trials completed within 2 min. of entry into the corridor. Animals taking longer than this to complete their runs turned up in the wrong feeding compartment as often as not (24 runs, 12 errors). At twenty-six of the 152 delayed trials octopuses refused to complete detours at all, and at the end of the 5 min, trial had either returned home or were still trying to do so through the transparent shutter which had closed behind them as they went into the corridor.

One hundred and thirty-three trials were run with octopuses having no vertical lobes. Conditions were the same as in the delay experiments with unoperated octopuses, summarized above. The behaviour of the animals was also essentially the same, but the proportion of errors and failures to make detours was higher. In the 133 trials the operated octopuses failed to complete detours on forty-six occasions and together made twenty-seven errors (Fig. 4b). Again, the bulk of the errors were made in the slower runs; only four mistakes were made in the thirty-eight trials completed within 60 sec. In the trials taking longer than this the animals were equally likely to turn into either of the feeding compartments (49 trials, 23 errors).

The results in the delayed-detour experiments thus extend and confirm those obtained in undelayed detours. Octopuses lacking their vertical lobes turn right and left at random in detours taking more than about 1 min. Unoperated animals can do better than this, and will generally respond correctly in detours taking as long as 2 min. The situation is summarized in Fig. 5. In all but the quickest runs operated animals performed less accurately than controls; the longer the time taken, the worse their performance relative to that of controls. Their poor performance cannot be explained in terms of locomotor defects; they do not as a group run more slowly and they make more errors than controls even in trials taking the same length of time to complete.

The eyes of Octopus are on the sides of the head and although the two visual fields overlap slightly both behind and in front of the animal, the octopus rarely if ever makes a binocular attack (Muntz, 1961a). In nearly every instance movements from place to place, including attacks on objects seen, are clearly guided by one eye at a time; the other eye is facing away from the obvious target.

In the analysis that follows it has been assumed that the eye facing in the direction that the octopus is going is the leading eye, in the sense that it is the main source of the sensory information used to determine what the octopus does next. The leading eye can vary in the course of a detour if the animal changes the direction in which it is moving or its posture. In the great majority of the detours observed in the present series it did not (Fig. 6a). In most instances the eye destined to lead throughout the detour was determined during the initial direct attack through the window of the home compartment. The postures adopted while attacking through the window apparently make it more likely that the octopus will pass into the corridor on occasions when the eye on the corridor side is actually being used to direct the assault. At all events, completed detours in trials to the left are more often than not begun with an attack using the right eye, while detours to the right are mainly led using the left eye (Fig. 6a). In these cases the animal attacks, swings into the corridor and continues along the wall separating it from the crab, leading with the same eye throughout. In a minority of instances the octopus passes into the corridor following an attack directed by the eye on the side farthest from the corridor and, having swung round on to the wall separating it from the crab as it moves in, leads off up the passage with the other eye (Fig. 6b).

If a failure in interocular transfer were the cause of the increased proportion of errors made by octopuses lacking vertical lobes, one might expect them to make an unusually high proportion of errors in trials starting in this way. Indeed, if there were no interocular transfer, it is hard to see why they should continue to run at all after entering the corridor in this manner.

But the operated animals did continue to make detours under these conditions and, as Table 2 shows, the proportion of errors made following an eye change was scarcely greater than in runs made with no changes in the leading eye. In twenty-nine trials beginning with an eye change the operated animals made seven errors (76 % correct responses). In 197 trials with no eye change the same animals made forty-two mistakes (79 % correct responses). Controls made fewer mistakes but again the proportion of correct responses was unaffected by eye changes made on entering the corridor.

Since there was a possibility that animals making successful runs beginning with an eye change had in fact managed to observe the crab using both eyes during the period before moving into the corridor, a check was made on all the eye changes made by the operated animals during the direct attack periods preceding their successful runs. Records for each trial started from the moment that the octopus first approached the window into the feeding compartment. In twenty-eight out of twenty-nine trials at which successful ‘eye change’ detours were made by octopuses lacking their vertical lobes the same eye was used from the moment of the first approach to the window until after the octopus had passed into the corridor for its final run. In the remaining trial the animal attacked using the right eye (crab in the left-hand side feeding compartment), went into the corridor, returned to the home compartment and attacked again, this time using the left eye, before running into the corridor for the second time and successfully completing a detour guided by the right eye. In this one instance there is no need to invoke interocular transfer to explain the continued detour; this octopus could have been running on a basis of information collected through either eye. In the other trials the octopuses had no opportunity to see the crab with the eye subsequently used to guide the detour. The external walls of the aquarium were rough, grey and nonreflecting.

On this evidence it would appear that the relatively poor performance of octopuses with their vertical lobes removed cannot be attributed to a failure in interocular transfer.

Schiller (1949) carried out detour experiments with octopuses, and the apparatus used for the present series of tests is essentially a copy of his. He used octopuses of about the same size and mazes of similar dimensions.

In all Schiller ran three octopuses, one of which was a female, induced to detour by placing the eggs that she was brooding in the ‘feeding’ compartment. He used two sizes of maze, first training the animals to make detours through a corridor considerably shorter than those used in the present series of experiments and then extending the corridor to a comparable length. His over-all result of 72, 75 and 82 % correct responses in a total of 109 trials with the three octopuses is clearly comparable with those summarized above, obtained using unoperated octopuses in undelayed trials. Schiller found, moreover, that the mean times taken on successful runs (measured, as here, from the moment of entry into the corridor) were shorter than the mean times taken to complete unsuccessful detours; 25 sec. (short maze) and 38 sec. (long maze) as against 47 and 74 sec. for runs ending in mistakes.

The female with eggs was also used for enforced delay experiments, with a shutter covering the exit from the corridor. In ten 1 min. delays this octopus made eight correct responses, one error and one ‘ambiguous’ response. The mean time for the completed delayed detours was 85 sec.

Schiller concluded that octopuses are capable of completing detours after delays of up to i min. He compares this figure with some of his own previous results using minnows, Phoxinus laevis, which, in a similar apparatus, were unable to complete detours if delayed for more than a few seconds (Schiller, 1948). The result with Octopus is similar to those obtained in the present longer series of experiments.

The relatively poor performance of animals lacking the vertical lobes cannot be attributed to slower running or to failures in interocular transfer. The latter finding is of particular interest because previous experiments have shown quite clearly that vertical lobe removal sometimes prevents interocular transfer. Muntz (1961a) trained octopuses to make visual discriminations by presenting the figures to be distinguished one at a time and always to one side of the octopus. He found that animals trained in this way performed almost equally well when tested by presenting the same objects in the field of the untrained eye. Having shown that interocular transfer is normally complete or very nearly so (the degree of transfer is somewhat less for difficult discriminations and in the early phases of learning easy ones) Muntz (1961 b) went on to show that it could be prevented by removing the vertical lobe. After this operation, animals could still be trained on one side, but did not discriminate when subsequently tested on the other. The results of the present experiments appear to be contrary to this; so far as making detours is concerned it seems that removal of the vertical lobe does not prevent interocular transfer. An octopus making a detour in order to reach a crab seen through one eye a few seconds before is able to lead down the corridor with either eye; there appears to be no question of one side not knowing what the other side has just learned.

The difference in result in the two types of experiment is presumably related to the difference in time scale of the memories that must be established in the course of performing the two tasks. In Muntz’s reward-and-punishment training experiments, discrimination took several days to establish and must therefore have involved the laying down of relatively permanent memory traces. In the present type of detour experiment the importance of long-term learning is minimal. There is, indeed, some indication that performance improves over the first few days of tests in that as trials continue slightly less time is spent battering against the windows of the feeding compartments and in running through the maze once a detour has begun (Wells, 1964). But that is all; there is clearly no advantage in the octopus remembering the direction that it took at the last trial and there is no evidence that the animals learn to recognize the appearance of crabs in the left- and right-feeding compartments as distinct situations requiring different courses of action. Animals trained to run always to the right, or always to the left perform perfectly on the first trial to the opposite side (Wells, 1964). It would appear from the present analysis that each trial is independent in that the sight of the crab in the feeding compartment initiates a response guided only by this particular experience in the immediate past. Unlike the discrimination experiments, this is a short-term learning situation. The fact that interocular transfer occurs in the short-term situation but not in long-term learning is interesting as a possible clue to the nature of the traces established under the two conditions.

The results of detour experiments indicate that an immediate result of showing a crab to an octopus is the establishment of a representation within the octopus that for a while helps to determine what the animal does.

The same conclusion has been reached from delayed-response discrimination experiments. Thus Dilly (1963) showed octopuses two identical rectangles at the far end of their tanks while the animals were held back behind a transparent shutter. A crab was shown next to the ‘correct’ shape and then removed. The octopus was released up to 30 sec. later and rewarded if it went at once to the correct rectangle. The correct rectangle varied in a random manner from trial to trial so that the octopus could only know which was the correct shape from its immediately past experience of the crab. The experimental animals very nearly always went to the correct shape, even when their orientation was disturbed by poking them about during the delay period following removal of the crab. This result can only be explained by supposing that a representation, recording the presence and location of the crab, persists within the octopus for at least 30 sec. after the crab has been removed.

In Dilly’s experiments six octopuses with the vertical or superior frontal lobes removed performed as well as controls. This shows that the representation set up by exposing the octopus to the crab cannot be within the vertical/superior frontal lobe system. A similar conclusion must be drawn from the detour experiments where again removal of these parts does not prevent correct performance. Damage to the inferior frontal system in the more anterior part of the brain does not detectably interfere with visual learning (Wells, 1961) and there is no evidence that the basal lobes, underlying the superior frontal and vertical lobes, play any part in learning; these are higher motor centres concerned in the organization of walking and swimming (Young, 1964). This leaves the optic lobes as the likely site for the representations set up as a result of visual stimuli and, although the matter cannot be proved since removal of the optic lobe cuts off all visual input, it will be assumed for present purposes that these lobes are indeed the location of visual memory traces.

Delayed-response experiments and detour experiments indicate that representations can persist for a couple of minutes as a result of a single exposure to a visual stimulus. With repeated exposure this period can be enormously extended. Octopuses trained to make visual discriminations, using a reward-and-punishment technique, will still perform when tested weeks later (Sutherland, 1957). The process of building up these more permanent representations is much slower in animals lacking the vertical lobes (Young, 1961, 1964) and it is believed that one of the functions of the superior frontal/vertical lobe system is the prolongation of the effects of each individual visual experience so that the total sensory input required to establish a more permanent memory trace is reduced (Young, 1965). This view is compatible with the results of detour experiments where intact octopuses apparently remember individual events (seeing the crabs) for longer than animals with their vertical lobes removed.

In a further series of visual discrimination experiments Muntz (1961 b) showed that long-lasting representations are set up in both sides of the brain when one side of an octopus is trained. Removal of the optic lobe from the trained side did not prevent discrimination in tests made subsequently using the eye on the untrained side of the animal. A longitudinal vertical cut dividing the superior frontal/vertical tract into two halves prevented interocular transfer if made before training, but not if made afterwards. It seems from these experiments that a bilateral trace cannot be set up in the absence of a pathway from one side to the other through the vertical lobe. If this pathway is cut a ‘permanent’ trace is established on the trained side of the brain alone. This trace is not available to regulate actions taken in response to stimuli received through the untrained side.

Anatomically, this is a surprising result. The optic lobes of the two sides are connected by massive commissures that lie well below regions at all likely to be disturbed by vertical lobe removal. We know that in the short-term learning situation both sides ‘know’ what one side has just learned, because the animal can make a detour towards a crab seen through one eye using either eye to guide the response. It seems only reasonable to suppose that the optic commissures are concerned in this, since animals can change eyes and complete detours successfully even after the vertical lobe has been removed.

Under these circumstances one must suppose either that a representation is present in both sides of the brain or that the ‘untrained’ side reads out from a representation in the side that has recently seen the crab. The former seems more the probable explanation, since we know that long-lasting representations are established in both sides of the brain when a normal animal is trained. Why then is no permanent trace established in the untrained side when the vertical lobe is absent?

A possible explanation is summarized in Fig. 7. Ex hypothesi any patterned sensory input sets up a pattern of activity in the cells of the optic lobe on the stimulated side. The activated cells evoke motor responses that depend upon the animal’s past experience of such patterns. The optic commissures connect cells in the stimulated side with corresponding neurones in the unstimulated side, so that electrical activity in the stimulated side affects the state of homologous units in the untrained half of the brain (Fig. 7a). The altered state of the neurones on the untrained side presets them so that when the ‘learned’ pattern is presented to the untrained side, the resulting activity is channelled down pathways homologous with those already active on the trained side. This ‘preferred’ route for any output resulting from sensory inflow on the untrained side means that when a pattern of stimulation shifts from one visual field to the other, the animal will continue with activities already initiated in response.

It should be noted that in this model the neurones on the untrained side are prepared—partly depolarized perhaps, so that some pathways will facilitate more readily than others—but inactive. There is no representation of the sort supposed for the trained side where patterns of neurones have been caused to fire, establishing chains of activity as a result of the original stimulus.

The state of preparedness on the unstimulated side lasts only for as long as nerve impulses continue to arrive down the optic commissures; it dies away as soon as electrical activity ceases on the stimulated side. Once the effect has faded, there is no reason why inputs on the untrained side should be channelled into producing a ‘trained’ pattern of response, and they may evoke quite different patterns of activity (Fig. 7c).

A variety of experiments on vertical lobe function (see above) has indicated that one property of this region is to prolong the neural activity caused by visual stimulation. This has generally (see Young, 1964) been seen as a result of self re-exciting chains of neurones repeatedly reactivating the pattern originally activated by the visual input. Repeated representation from within is seen as leading to growth changes resulting from repeated use (Fig. 7a, b).

If one further supposes that the vertical lobe system relays its input to the inactive side as well as representing it to the side of the brain from which it came, a possible explanation of the effect of its removal becomes apparent. The pattern relayed to the unstimulated side, because it mimics the original sensory input, will set off the pattern of activity preset but as yet inactivated in the untrained side (Fig. 7d). This in turn activates the vertical-optic lobe cycle on that side and a long-lasting trace may be established by growth changes. In this event subsequent testing after any electrical activity has died down will evoke appropriate responses whichever eye, trained or untrained, is stimulated (Fig. 7e). Subsequent splitting of the vertical lobe system, or removal of the trained-side optic lobe will not eliminate the response.

The above hypothesis assumes that the transient trace responsible for interocular transfer in the detour experiments is electrical in nature, and that the ‘permanent’ trace is structural. There is some evidence for these contentions in Octopus. We know that long-term memories survive anaesthetics and faradic stimulation of the octopus brain (Boycott & Young, 1955), and we have evidence from tactile experiments where lateral transfer is not immediate as it is in visual learning (Wells, 1959), that there are processes in octopus learning that take an hour or more. What we do not yet know is whether disruption of electrical activity immediately after training prevents the establishment of more permanent traces in octopuses as it does in mammals. And we have no evidence as yet to suggest whether, for example, blocks to protein synthesis do the same, as seems to be the case with fish and mammals (see Agranoff, et al. 1965). There is clearly a whole range of experiments yet to be done. The present account indicates that detour experiments are a valuable means of investigating learning in that they provide a short-term learning situation that produces results apparently at variance with some of conclusions drawn from the discrimination training techniques hitherto used for work on the effects of brain lesions in cephalopods.

This work was supported by a grant from the Science Research Council of Great Britain. ‘Table’ facilities at Naples were given by Cambridge University. The author would like to thank the Director and staff of the Stazione Zoologica di Napoli for their hospitality, and Prof. J. Z. Young, F.R.S., for reading and criticizing an early draft of this manuscript.