Detour Experiments With Octopuses

In 1949 the late P. H. Schiller published a paper entitled ‘Delayed detour response in the octopus’ in which he showed that this animal can be taught to make a detour through a passage to reach crabs previously seen from a ‘home ‘compartment at right angles to the passage (Fig. 1). Schiller was mainly interested in the length of time taken to make the detour and showed that at least one of the three individuals he used was still capable of making correct responses after being delayed in the corridor for as long as i min. He also noted that correct responses—i.e. the octopus turned into the correct feeding compartment on reaching the choice compartment at the end of the corridor—appeared to depend upon the maintenance of bodily orientation, or more specifically, on maintained contact between the lower surface of the octopus and the wall separating the animal from its goal. If the octopus was obliged to squeeze through a hole in a shutter at the exit end of the corridor, or left the separating wall during an enforced delay in the corridor (exit doot shut) it became disoriented and equally likely to turn left or right on entering the choice compartment.

These observations are interesting in the light of more recent experiments on the role of positional information in learning by octopuses. An octopus, it seems, cannot learn visual or tactile discriminations that depend on its taking into account the relative positions of parts of its own body or the position of the head and eyes relative to gravity (Wells, 1962, 1963a). Schiller’s observations would appear to confirm this finding, for if the animal is incapable of learning to recognize the changes in bodily position that it makes as it moves about, remaining in close contact with the surface separating it from its goal is one of the very few possibilities of correct performance left to it. The experiments reported below were begun simply as a check on Schiller’s results. They soon showed, however, that correct performance does not depend, as he believed, on the animal literally sticking to the intervening surface, but on the maintenance of visual contact with this surface. These results are discussed in relation to the organization of learned motor responses in flexible animals generally.

Octopus vulgaris Lamarck, from the bay of Naples was used. The animals weighed from 200 to 400 g. and were kept separately in ‘Eternité’ tanks measuring 100 × 60 × 30 cm. deep, being transferred to similar tanks with detour apparatus as soon as they were feeding regularly, or after some pretraining to take crabs out of glass jars.

The maze through which the animals had to make a detour in order to get a crab is shown in Fig. 1. Between trials the animals were normally confined to the ‘home’ compartment, being closed off from the corridor by a transparent shutter. The bottom of the experimental tanks was covered with a layer of coarse sand about 3 cm. deep.

For most of the experiments the following data were recorded: (1) The time spent attacking through the transparent wall of the home compartment before going into the corridor. (2) The time actually spent in the corridor. The octopus was recorded as having entered or left the corridor when a point on the head midway between the eyes had crossed the threshold of the entrance or exit doorway. In the text and figures, all times are given to the nearest 5 sec. (3) The orientation of the animal’s head and body and the surface to which most of the animals suckers were applied. And, latterly (4) which eye of the animal was leading in the direction it was going. There are difficulties in recording the orientation of an octopus because the animal is so flexible. The arms, for example, may shift from a vertical to a horizontal surface as the animal goes along, without appreciably altering the course or posture of the animal’s head, and for most purposes it is best to ignore the position of the arms altogether, recording simply the animal’s heading (defined as a direction at right angles forwards from the axis between the eyes), and track (the course taken by a point midway between the eyes). Heading and track are commonly at right angles, since the animal in general uses one eye at a time to see where it is going, even though it is potentially binocular (Muntz, 1961). Some of the trials were filmed, so that a more detailed analysis of the animal’s movements could be made. In describing the animal’s position in the maze, parts of the maze will be named in terms of their position relative to the octopus’s home. Thus ‘left-hand feeding compartment’ (LHFC) means the compartment on the left as seen from the brick home in which the octopus starts the experiment, ‘LH wall of the corridor’ is the wall separating this compartment from the passage, and so on. The trial sequence (Left and Right) was usually determined by tossing a coin.

Most of the animals were unoperated. Others had one or both of the statocysts removed, or were blinded in one eye by section of the optic nerves. Details of operative techniques have been given elsewhere (Wells, i960; Wells & Wells, 1956).

When an octopus sees a crab through a transparent wall it attacks vigorously and there is at first no attempt to go round the obstacle. The suckers are mainly applied to the wall, with the eye used to direct the attack close to it. From time to time the animal will shift position and change eyes as it struggles to get through the partition. In the apparatus used for the experiments this direct attack may last for several minutes at the first trial (mean 60 sec. for 29 animals). Sooner or later, however, the animal’s movements carry it to the door into the corridor (usually one or more arms extend into the corridor first, but this is not always so) and it passes in and out of sight of the crab, generally turning to face the wall separating it from the crab as it does so. The octopus may then continue down the separating wall, or return to continue the attack from the home compartment. Of 29 animals tested 8 completed the detour into the correct feeding compartment after entering the corridor for the first time, having had no previous training to run through the apparatus. Four of the eight, and seven other unoperated octopuses were subsequently used in runs of 20 or more trials, generally at a rate of 4 trials per day. The 7 animals that did not complete a detour on entering the corridor for the first time were subsequently trained in various ways, generally by showing the octopus a crab in the choice compartment when it went into the corridor, and luring it through to be fed in the appropriate feeding compartment. The number of these training runs was variable, individuals being allowed to complete the detour ‘unaided’ as soon as they appeared likely to do so by continuing down the corridor without pause on entering it. Details of training are given in Table 1.

The following samples from the performance records of a twelfth octopus, K 16, are typical. This animal was first tested 3 days after being placed in the apparatus. During these 3 days it was free to wander in the maze, and was fed 5 times, attacking down the corridor to take a crab from a jam-jar placed in the choice compartment.

Crab in jar in RHS feeding compartment (FC). Octopus attacks at once. Into passage 1 min. 45 sec. arms pressed to the RHS of the corridor. Returns to attack, into passage again 2 min. 30 sec., about halfway along, returns to attack. Into corridor again, 3 min. 50 sec., stays about 10 sec. with arms extending as far as and into the choice compartment. Home, and attacks again, into corridor 4 min. 45 sec. (always along RH wall). This time almost into the choice compartment, but returns to attack again. Into corridor 8 min. out, and home. Attacks again almost at once, into corridor 9 min. 45 sec., out and home. No detour in 10 min. Led round into the feeding compartment by a crab on a string. After ca. 5 min., animal detours home, having first tried to return directly, through the wall of the home compartment.

(2$12$ hr. later.) Crab in jar in LHS FC. Attack and struggles against the glass. Into passage 1 min. 10 sec., back 10 sec. later. Again into passage and out at 2 min. 45 sec. Into passage 3 min. 45 sec., this time went through and into the choice compartment (no time recorded). Embracing the doorway into the feeding compartment. Detaches, wanders in LHS of the choice compartment. Suddenly seems to see the crab and dashes into FC (5 min. 15 sec.). As before tried to take crab home direct, then detours home (ca. 3 min.). First successful run.

(2 hr. after trial 2.) Crab in RHS FC. Into corridor 10 sec. moves slowly along facing the RH wall, but not closely applied to it. Into FC 40 sec., having seen the crab from the choice compartment (octopus darkened-chromatophores expanded). Tries to return home through the wall until 2 min. 20 sec., then detours home.

After the 10th successful run (having once done it, this octopus never failed) bricks were put into the feeding compartments to screen the view of the crab from the choice compartment. From then on there was no possibility of the octopus making a choice as a result of seeing the crab in the feeding compartment, and the decision about which feeding compartment to enter had to be made in the absence of direct clues. The animal continued to perform correctly nevertheless.

Trial 12. (3 days after the start of the experiments.) Crab in RHS FC. Attack, into corridor 15 sec., into choice compartment 50 sec., turns into RHS FC, quickly skirts the brick and gets crab. Animal with suckers applied to the RH wall all along until the exit, when it spreads out onto the floor of the choice compartment before turning right.

The performance of K 16 is summarized in Fig. 2. Both the time spent attacking before going into the corridor and the number of abortive entries into the corridor (octopus returns to attack without completing the detour) fell progressively as the trials continued. The time actually spent in the corridor altered little. This animal in-variably crept along heading towards the wall separating it from its goal, with the arms turned towards and generally applied to the wall. The run down the corridor took it a minimum of about 10 sec., longer times being in general due to the animal pausing halfway along the passage, or near the exit while it felt around the exit door.

The performance of the 11 other unoperated animals was similar to that of K 16. Once having made a successful detour in this apparatus an octopus rarely failed to perform correctly on subsequent occasions. In 220 trials there were only 13 failures to make a detour and 10 errors through animals going into the wrong feeding compartments. At all other trials, the octopuses made successful detours in the right direction. Evidence that the performance of octopuses in this situation improves (i.e. that the animal learns anything from successful trials) must therefore come, not from any reduction in the number of errors made, but from reductions in the time spent attacking through the wall of the home compartment and in the number of abortive entries into the corridor. In Fig. 3 the performance of the 11 octopuses is summarized in these terms. In the first half of their 20-trial experiments, these animals together made 47 abortive entries into the corridor; in the second half they made 20. Over the same trials the number of animals entering the corridor within 10 sec. of the start of an experiment rose from 21 to 31 and there was a corresponding drop in the number of animals taking a very long time to enter the corridor.

There is, then, some improvement in performance during the 20 trials following the first successful run by an octopus. Most of the improvement, however, clearly occurs right at the start of the experiment, when the majority of the animals have to be led through the maze to get them to run at all (see Table 1). Making detours is clearly a part of the normal behaviour of octopuses but it improves with practice ; the animals learn to make appropriate responses.

One of the points of interest in any detour experiment is that the animal making the detour maintains a course of action in the absence of direct stimulation from the goal ; it is a test of nervous retention. This can only be certain in the present case if the stimulus is wholly visual. The octopus passes out of sight of the crab when it goes into the corridor but the high proportion of correct responses could in principle be due to the animal picking up other stimuli en route. A crab presumably smells, and we know that the octopus has a very acutely developed chemotactile sense (Wells, 19636). Or the octopus might conceivably receive vibratory stimuli from the crab’s movements, and home in on the prey in this way. These possibilities have been eliminated in the present experiments in two ways: (1) by removing the crab as soon as the octopus had gone into the corridor, (ca. 50 trials; this was the normal technique in some of the experiments, the octopus being fed on fish when it arrived in the correct feeding compartment), and (2) by enclosing the crab in a heavy glass jar with a screw top before offering it as a stimulus (21 trials, no errors). Since neither of these alterations in technique produced any detectable changes in behaviour on the part of the octopuses, it can be assumed that the effective stimulus is entirely visual.

Schiller (1949) has stated that correct performance in this type of detour appears to depend on the arms remaining in continuous contact with the wall separating the octopus from its goal; that the animal crawls along the intervening surface. His animals, like those reported here, hardly ever failed by going into the wrong feeding compartment.

Fig. 4 shows tracings from a cine film of a typical run by an unoperated octopus, K 45. Schiller’s account is clearly an adequate description of what happens in this instance, since the animal remained with most of the arms applied to the intervening wall throughout most of the detour ; K 45 lost contact with the wall only on entering the feeding compartment. It was, however, noted that maintained tactile contact with the wall separating the octopus from its goal was by no means invariable. At a considerable proportion of trials octopuses hardly touched either wall of the corridor, and commonly lost contact with these altogether on entering the choice compartment. They nevertheless performed correctly. If continuous tactile contact with the intervening wall is not the means by which an octopus maintains its orientation towards a goal in these experiments, what is?

Fig. 4 shows that although the position of the arms may vary, the orientation of the head does not. The animal continues to head towards the correct side throughout the detour. The rare errors made by unoperated animals nearly all followed loss of this orientation, commonly (5 cases out of the 10 errors in 220 trials) because the octopuses turned to face the end wall of the choice compartment on leaving the corridor.

Some attempts were made to upset the orientation of two of the experimental animals by inserting a shutter with a hole in it at the exit to the corridor, in series of trials following the initial 20 already summarized. In all 17 such trials were made, 7 with one animal (K 10), 10 with the other (K 33). K 10, the smaller of the two octopuses, made no mistakes. The other, K 33, made 4 errors in 10 trials, having made only one in the previous 20, without the shutter. Detailed records of the performance of K 10 show that this octopus was hardly inconvenienced by the shutter, and completed all 7 of the trials made under these conditions without once losing its orientation, continuing to head towards the correct side even while passing sideways through the hole. K 33, a larger animal, had more difficulty, and emerged from the hole heading directly towards the end wall of the choice compartment (head orientation lost) on 5 out of 10 occasions. At 3 of these trials it subsequently went into the wrong feeding compartment and on one occasion it returned home through the hole without having gone into either. At 4 of the 5 trials at which K 33 succeeded in squeezing through the hole without losing its head orientation, it moved swiftly into the correct feeding compartment; at the fifth trial it spent an unusually long time wandering in the choice compartment (45 sec.) before finally going into the wrong side. Schiller (1949) reports similar results in the same situation (10 trials, 4 correct, 4 wrong, 2 failures to make the detour in 10 min.), but gives no details of his animals’ orientation, beyond stating that the orientation of the lower surface of the arms towards the goal was upset.

These experiments suggest that correct performance depends on the maintainance of head orientation, so that the octopus heads towards the side of the tank containing its goal throughout the detour.

The paired statocysts of Octopus are in the head, embedded in the cartilage that surrounds the central part of the brain. Each statocyst contains a macular gravity receptor and a crista with flaps arranged in three planes at right angles (Young, i960). The statocysts are clearly capable of giving information about turning movements, and there is physiological evidence that the system can differentiate left from right-handed rotations (Dijkgraaf, i960; Maturana & Sperling, 1963). After removal of the statocysts octopuses no longer respond to rotation by making compensatory movements with the eyes (Dijkgraaf, i960; Packard, 1964). Since it was possible that the octopuses in the present series of detour experiments were, for example, learning to make all possible clockwise rotations after observing a crab in the right-hand compartment and vice versa, the effect of removing these rotation receptors was investigated.

Three animals, already trained in runs of 20 trials before the operation, had the statocyst removed on one side only and a further 10 trials (5 right, 5 left) carried out with each. In these 30 trials one animal made 2 errors (1 R, 1 L, LHS statocyst removed), the others none. There were four failures to make a detour, all the rest of the responses being correct. On 6 out of the 26 occasions when a detour was made, entry to the corridor was within 10 sec. of the start of the trial. Having got there, the operated octopuses averaged 15 sec. in the corridor. (This figure excludes the two occasions on which errors were made. On both these 1 min. 40 sec. was spent in the corridor.) Neither figure indicates a performance in any way abnormal, and there were no detectable differences in trials involving rotation to or away from the operated side. This confirms previous work (Boycott, i960; Wells, i960) reporting an absence of behavioural defects after unilateral statocyst removal.

With bilateral lesions, it is difficult to get octopuses to make detours at all. After removal of both statocysts an octopus moves unsteadily with the head rocking to and fro. The animal may roll over onto its side or back and appears to trip over its own arms if it attempts to move at all rapidly by means of jets from the funnel (Boycott, i960; Wells, i960). Under these conditions, and especially in the first few days after operation, it is difficult to persuade octopuses to come out of their homes and attack. When they do so they are liable to lose sight of the crab as they lurch about, and entries into the corridor are at first more often than not followed by a return home without completing the detour.

In the present series of experiments 4 animals had both statocysts removed, 2 of them octopuses already tested after unilateral operations. In all, these animals eventually completed 39 detours (20 R, 19 L), making only 6 errors. Three of the errors were made by an animal (octopus K 65), which was also blinded in one eye, an operation that itself tends to produce errors (see next section). Twenty-seven of the detours, including the other 3 errors, were made by one octopus, K 99. This animal had no pre-operational training to run the maze. It first made a successful detour 3 days after the operation, and completed the remaining runs in the following 7 days. Destruction of the statocysts was made certain by removing both the macula and the crista from each side, and the absence of regeneration checked subsequently from serial sections of the brain. Details of the performance of K 99 are given in Fig. 6.

The times taken to enter and pass down the corridor by animals with both statocysts removed were well within the normal range. Entries within 20 sec. were recorded at 9 out of the 39 trials (23%), about as often as in the control series (26%) and the average time spent there on completed detours was, if anything, rather less (15 sec., cf. 22 sec. by controls). During the successful runs the orientation of these operated animals was essentially normal if somewhat unsteady (Fig. 5 shows tracings from a film of a run by K 99). Successful runs were made when the animals remained closely applied to the floor or sides of the maze, and the increasing proportion of completed runs by the operated animals as tests continued (see Fig. 6) appear to be attributable in the main to a reduction in the tendency to move swiftly, using the arms as levers, rather than gripping and pulling the animal along. Levering and the use of the jet in locomotion tend to produce violent lurches and disorientation in animals without statocysts.

An examination of the 3 errors made by K 99 shows that on two of the three occasions the animal left the corridor heading straight towards the back wall of the choice compartment, and spent a minute or more wandering there before finally going into the wrong feeding compartment. On the third occasion K 99 spent 25 sec. in the choice compartment, but what it did in this time was not recorded. On successful runs the delay in the choice compartment was rarely as long as 10 sec. Such errors as happened, in short, occurred under the same circumstances as errors by control animals.

It is concluded that removal of the statocysts does not seriously affect the capacity to make successful detours under these conditions.

Unlike removal of the statocysts, section of the optic nerves seriously damages performance. For the experiments four octopuses were each blinded in one eye by section of the optic nerves. All four had previously been trained in runs of 20 or more trials and had performed consistently well, making only 3 errors in a total of 80 pre-operational trials (36 R, 44 L).

After the operations the animals continued to perform well when making a detour in one direction (to the right for animals blinded in the right eye, left for the single animal blinded on the LHS), but they made many errors when required to make a detour towards the feeding compartment on the other side. Thus, in a total of 63 trials towards the ‘unblinded’ side, the four octopuses together erred 16 times and failed to make a detour 28 times. In 63 trials towards the ‘blinded’ side there were only 4 errors and 9 failures (Table 2).

The usual cause of the errors is shown in Fig. 7, taken from tracings of two filmed sequences of detours by the same octopus, K 33. This animal was blinded by section of the RHS optic nerves. Attacks were therefore always made using the left eye, so that the animal was always headed towards the RHS of the tank. On going into the corridor on RH detours, this heading could be maintained and the left eye used to guide progress down the corridor and round into the feeding compartment. Detours in the other direction, however, produced difficulties.

An octopus detouring left after using the left eye to direct its initial attack would normally turn through 180°, switch from the left to the right eye and lead down the LHS wall using that. Having no effective right eye K 33 tended to continue heading in the same direction, to the right, and on LHS detours commonly crossed to the RH wall, leading with the left eye to round the exit doorway into the (wrong) RHS feeding compartment (Fig. 7). In 20 LHS trials K 33 made 9 errors, 8 of them of this type, the ninth following a period spent wandering in the choice compartment.

An alternative cause of failure to respond correctly was seen in octopus K 65, blinded in the right eye. This octopus made only 2 errors, but regularly failed to complete LHS detours. In 17 attempts to get K 65 to make a detour in this direction, only 7 correct responses were made; 13 out of 17 attempted RHS detours were successful. The cause of LHS failures generally appeared to be that the animal, having observed the crab with the right eye, swung round the entrance doorway onto the (normally ‘correct ‘) LH wall. This left the seeing eye facing along the wall homewards, instead of onwards down the tank; the octopus led with it, and returned to the home compartment (Fig. 8). Details of abortive entries into the corridor were not at first kept with this octopus, but in the last 7 LHS trials, it failed to make a detour 6 times and on every one of these occasions made one or more abortive entries into the corridor. In the corresponding 7 RHS trials it made rapid and complete detours on five occasions, made one return home from the corridor before completing the detour at a sixth, and failed to detour once.

The other two unilaterally blinded octopuses (K 10 and K 77) made errors by failing to detour and by turning into the wrong feeding compartment. The mistakes made by the four animals are summarized in Table 2; they failed to make a correct detour at 70% of the trials towards the intact side.

The performance of these animals in their less common successful runs towards the intact side also throws light on the means by which octopuses orient themselves in detours. Correct, completed detours to the unblinded side were made at 19 out of 63 trials, and one of these correct runs was filmed. The animal’s posture during this run is shown in Fig. 9. It was clearly abnormal; the octopus was heading towards the ‘wrong’ side of the tank during the greater part of its run down the corridor. But it will be noted that though the animal was heading in the wrong direction, it led throughout with the good eye either close to, or facing directly towards the correct wall. This, it seems, is the key to the making of successful detours under these conditions. At 10 out of the 19 successful runs made by unilaterally blinded animals detouring towards the unblinded side, the animals led with the good eye clearly facing mainly towards the ‘correct ‘wall ; presumably the octopus was watching the wall, and working along it, seeking a break in the screen separating it from the goal. At most of the remaining trials the correct answer seems to have been achieved rather by chance, following a period spent wandering in the choice compartment.

Since successful trials by unilaterally blinded animals often involved headings 90–180° away from the goal, it seems that the orientation of the head and the position of the body relative to the goal cannot be a critical factor in determining correct detour responses. The view from the intact eye, on the other hand, would appear to be vital. If the octopus fixes on the wrong wall as it goes into the corridor, it will end up in the wrong feeding compartment more often than not.

From this, and from the lack of effect of statocysts removal, it is concluded that the correct orientation of octopuses in these detours is achieved visually, and that the animals do not use information about their own bodily position in learning to make appropriate responses to crabs seen in the feeding compartments.

In addition to the above trials, all of which were carried out in Naples, experiments were made with two animals brought from Naples and kept in Cambridge. These octopuses were trained to make detours similar to those used in the rest of the experiments, except that the length of the corridor was variable and generally rather longer. Detours through passages up to 90 cm. long were made without difficulty. Each of the two Cambridge octopuses was trained by trials to one side only (one R, one L) at a rate of one trial per day, until 30 successful runs had been completed. They were then tested for their capacity to make a detour in the other direction. In both cases the first detour in the unaccustomed direction was accomplished without hesitation within 40 sec. of entering the corridor, and only one of the octopuses erred (once) in the following 5 trials.

It seems reasonable to suppose that if the animals were learning to orient themselves relative to cues from inside the maze, or from elsewhere in their external environment, the switch from one side to the other would have upset performance of the detour. Since the change did not have this effect, these experiments confirm (1) that the octopus does not learn to carry out a specific set of movements in response to the detour situation, and (2) that orientation is in relation to a remembered location of the crab and not due to learning to recognize appropriate environmental cues for each side of detour.

In the past, conflicting results have been obtained when trying to teach octopuses to run mazes. Thus Boycott (1954) reports two experiments in which octopuses failed to learn to make detours round partitions in order to get crabs. One of these experiments was with the maze shown in Fig. 10b. Five octopuses were tested. Each was given 3 trial runs at which it was led through the maze and fed on a crab (at X in Fig. 10b). Only one of the five subsequently proceeded to X when shown a crab at the entrance to the maze. In a second experiment three octopuses were required to make a detour round a glass plate to get a crab seen on the other side. Only one of them learned to do so; in 44 trials with this octopus the time spent pushing against the glass before making the detour fell from about 20 to about 10 sec. Bierens de Haan (1926) reports on a similar experiment in which octopuses failed to go round the edge of a wire-netting screen to get crabs. In contrast to most of these results Pieron (1911) found that octopuses learned to make a detour successfully in the situation shown in Fig. 10c, moving round the glass wall of a large bottle to get in at the neck and obtain a crab.

Schiller’s (1949) experiments with the maze shown in Fig. 10a appeared to settle the matter, and to explain some of the failures. If, as he supposed, successful making of detours required maintained contact between the lower surface of the octopus and the intervening wall, then Boycott’s (1954) results (Boycott was unaware of Schiller’s report when he did the experiments) are not surprising; his maze would have been unusually difficult for an octopus.

The present series of results confirm Schiller’s in showing that octopuses can readily be taught to perform with a high degree of reliability in at least one sort of maze experiment. But it appears that correct performance does not depend, as he believed, on the animal maintaining contact with the intervening wall, ‘Although in a few cases it was noted that the body posture was radically changed while traversing the passage, yet it was obvious that the octopus mostly crawled along the wall to which it was oriented while making contact with the bait at the window ‘, and (in the summary), ‘Performance was correlated with maintaining bodily orientation, the beak and the majority of suction cups being turned in the direction of the bait, and with persistence of crawling along the continuous wall. If this postural factor were disorganized by requiring the octopus to change it, chance choices were evinced’ (Schiller, 1949).

The present series show that the making of correct detours depends on visually directed orientation, and that tactile contact between the lower surface of the octopus and the intervening wall is not essential, although it commonly follows.

These findings link the present work with previous research on the visual system of octopuses which has shown that the orientation of the eyes is normally dependent on the correct functioning of the statocysts. With the statocysts removed, constant retinal orientation is lost, and with this the capacity to make visual orientation discriminations. Octopuses previously trained to distinguish between horizontal and vertical rectangles (Wells, i960), or between sources of light polarized in two planes at right angles (Rowell & Wells, 1961), make many errors after the operation because they continue to respond as if the eyes were still correctly oriented. Since eye position depends, post-operationally, simply on how the animal happens to be sitting, the animals continue to make errors for as long as training is continued or until they cease to respond altogether. The statocysts, in short, set the eyes to an ‘artificial horizon’ and the animal’s visual learning system is constructed on the assumption that retinal orientation remains constant. The two eyes are locked so that removal of a single statocyst is without effect.

In the present series of experiments removal of the statocysts did not prevent octopuses from making detours, even though it upset retinal orientation, because visual contact with the intervening wall could be maintained despite some movement of the image on the retina. Removal of the statocysts did not deprive the octopuses of a sensory input that they would normally have taken into account in learning to make the detour; their behaviour, apart from some unsteadiness in movement, was essentially normal and the proportion of errors that they made was no larger than that of unoperated controls.

Failure to take internal sensory inputs into account is also the probable explanation of why unilateral blinding upsets performance so badly. Octopuses normally watch prey with one eye at a time and in moving from place to place regularly switch from one eye to the other. There is no indication that individuals have a ‘preferred’ eye. Interocular transfer is complete in simple situations and very nearly so in all but the most complex learned discrimination experiments (Muntz, 1961). In the present experiments the change from one eye to the other sometimes failed because the second eye was blind. On these occasions the animals naturally reverted to the first eye. But they were not, it appears, able to detect the fact that in order to do this and continue down the corridor they had to rotate the head and alter the orientation of the whole body. The implication is that the necessary information from the statocysts or other internal receptors never reaches the highest parts of the cephalopod brain. As a result errors of predicatable types follow attempts to detour by animals blinded on one side; in one direction they perform correctly because the detour does not normally involve a change of eye; in the other they proceed into the corridor and either return home or continue along the wrong wall and into the wrong feeding compartment.

The failure to use information from statocysts or other indicators of bodily position in detour experiments confirms a generality that has been made elsewhere on a basis of tactile discrimination experiments. Octopuses, it seems, never learn to take information from internal receptors into account when they learn to recognize objects that they pick up and handle. The position of the arms and suckers is irrelevant in touch discrimination of shape and texture (Wells, 1964) as is the muscle tension that the animals exert in attempted weight discriminations (Wells, 1961). Proprioceptive inputs would appear to be the basis of a number of reflex acts and must surely be concerned in the regulation of movement. The relevant sense organs have been shown to exist (Graziadei, 1964), but the information that these provide is, it seems, always used more or less locally in the nervous system and never penetrates to those lobes in the supra-oesophageal part of the brain that experiment has shown to be concerned in learning (Wells, 1962, 1963a; Young, 1961, 1963).

That vertebrate animals are not in a comparable condition, able to order movements but unaware of their progress other than indirectly through exteroceptors, would seem to be a consequence of their particular type of skeleton. A hydrostatic skeleton, as in most invertebrate animals, permits great flexibility in movement. Many such animals can and do undergo gross changes in shape as they move through their environment, and presumably derive advantages from this. But a penalty of flexibility is the loss of the possibility of a centralized detailed control of movement. Detailed control of the movements of an animal as flexible as an octopus would necessitate central monitoring of the relative positions of a series of extremities, each of which is free to extend and contract, bend and twist independently of the others—an operation of an altogether different order of complexity from the integrative task to be carried out by the brains of a vertebrate or arthropod. With brains no larger than normal, movement control in animals with hydrostatic skeletons must of necessity be heirarchic.

In a vertebrate or arthropod, by contrast, movement is severely restricted by joints. But such movements as are possible can potentially be integrated and regulated with very great delicacy precisely because the restrictions imposed by their particular skeletal systems mean that central monitoring of effector positions is, relatively speaking, an easy task. Learning to make skilled movements becomes a possibility. It is worth noting that even in ourselves conscious control of movement is done on a basis of information from deep pressure and joint receptors and not on information supplied by muscular stretch receptors. The latter, as in the octopus, provide information that is used locally for postural adjustment. Like the octopus, we cannot use this proprioceptive input in learning to regulate what we do (Merton, 1964); we do what we can because we are able to learn to recognize inputs from joint, pressure, and perhaps tendon receptors, possibilities that are apparently absent in a hydrostatic animal. It is interesting to speculate about the development of this capacity throughout the evolution of vertebrates. Fish swim, but can they detect the progress of an ordered movement, other than through skin receptors, should the movement be halted? What is the condition in amphibia and reptiles—can these animals learn to manipulate? There seems to be surprisingly little data. Yet this is the capacity above all others to which we owe our ascendancy; why a mammal can learn a skilled movement while a cephalopod cannot; why—from our viewpoint—the behaviour of in-vertebrates so often appears to be so very limited.

This work was carried out at the Stazione Zoologica di Napoli, and I should like to thank the Director and his staff for providing the facilities that made it possible. I should also like to thank Prof. J. Z. Young, both for bringing two of the experimental animals to Cambridge from Naples and for his comments on the work in manuscript. The expenses of this research were paid by a grant the from Rockefeller Foundation.