The effect of observing trained conspecifics on the performance and motivation of goldfish, Carassius auratus , in a spatial task

Spatial and social cognition are two aspects of fish behaviour that have been subject to an increasing amount of research in recent years, but few have investigated potential behaviour overlaps. Testing the ability for an individual to socially learn a spatial task would bridge this gap in understanding. We provided naïve goldfish, Carassius auratus , the opportunity to observe a trained conspecific navigate a T-shaped maze, and then recorded how many trials it took for them to learn the maze, time taken per trial, motivation, and acceptance of the food reward. We also recorded how many trials it took a control group to learn the maze without the opportunity to observe a demonstrator. The observer group took significantly longer to learn the maze than the control group. Although the observer group were significantly less motivated (trials without a choice made), they were significantly more likely to accept the food reward. The social learning of reward acceptance was taking place, but the process of the demonstration disrupted the training of the spatial task, with possible explanations as the passenger effect and trade-off mechanism being discussed. Future studies are needed to determine whether goldfish can acquire spatial information socially; however, this study contributes to the feasibility of studying social learning of environmentally information in goldfish.


Introduction
Spatial memory is paramount to navigation, foraging and predator avoidance in animals.
Spatial memory is how animals retrieve, encode, store and present information about the external environment (Bshary & Brown, 2014).Animals learn locations in space by exploration of their environment and there are two ways this spatial information is represented; egocentrically, where space is represented by the relationships between the individual and the landmarks, and allocentrically, where space is represented by the objective relationships between different landmarks (Broadbent et al., 2014).An allocentric representation of space suggests that more complex behaviours are taking place, such as the use of a cognitive map, which is a mental analogue of a topographic map (Wehner & Menzel, 1990).The use of allocentric navigation, also known as place learning, is often considered evidence for the presence of cognitive maps, as it demonstrates an understanding of the spatial relationships of landmarks, without the individual's starting position affecting their understanding (Rodriguez et al., 2021).Animals retrieve this information in a variety of methods.Animals use different senses to retrieve spatial information, using sight during exploration or sounds for echolocation to discover landmarks, or detecting the sun or magnetic field to orientate to important places.
Another possible method that individuals use to learn about space is through social learning, when animals live in groups or observe competitors to gather spatial information (Brown & Laland, 2003).Social learning is a distinctive behaviour pattern that observing practitioners acquire in part through the behaviour of a demonstrator (Fragaszy & Perry, 2003).Social learning allows animals to gain information from a safer position, such as learning to recognise a predator without having to personally encounter the predator (Ferrari et al., 2007), or learning J o u r n a l P r e -p r o o f a novel area is safe to explore or food is abundant by observing a conspecific take the dangerous venture first (Coolen et al., 2003).There is evidence that animals can socially learn some spatial information.Captive-reared North American Whooping Cranes socially learned their migratory routes, making fewer deviations after migrating with experienced cranes (Mueller et al., 2013).Bem et al. (2018) found that rats can acquire spatial information from observing a demonstrator complete a spatial task from a fixed point.French grunts exhibit social traditions; individuals placed in new schooling sites were able to use the new migration routes and return to their new schooling sites (Helfman & Schultz, 1984).Schooling and shoaling behaviours greatly benefit from social learning, often with efficient social transmission taking place.Naïve fish have been found to learn foraging behaviour (Reebs, 2000), food patch profitability (Pitcher & House, 1987), routes (Laland & Williams, 1997;Swaney et al., 2001) and escape responses (Brown & Warburton, 1999) socially from trained fish within the shoal.Guppies prefer socially learned routes over alternative self-discovered routes, even when they are more energetically costly (Laland & Williams, 1998).
The presence of trained conspecifics does not necessarily facilitate learning for observer individuals.Roy and Bhat (2017) investigated whether zebrafish can socially learn a maze.All fish gradually improved in performance over the course of the experiment; however, the naïve group of observer fish that were each paired with a demonstrator showed no significant difference in their ability to learn, compared to the learning ability of the control group that paired two naïve fish together during observation.Performance did not improve for the observers, and the number of mistakes did not change.Likewise, Burt de Perera and Guilford (1999) found that pigeons performed a spatial task more effectively alone than they did after having the presence of a demonstrator pigeon.This is sometimes referred to as the passenger effect, and has been found in homing pigeons (Banks & Guilford, 2000), crab-eating macaques (Stammbach, 1988) and zebrafinches (Beauchamp & Kacelnik, 1991).Burt de Perera and J o u r n a l P r e -p r o o f Guilford hypothesised that this was either due to the passenger not learning anything in the observation period, or was learning something different to the control subjects in their respective spatial tasks.Therefore, the possibility that a trained conspecific might inhibit the learning of a naïve individual should be equally as anticipated as the possibility the learning is facilitated.
Social learning differs between species depending on the lifestyle of the species.For example, social shoaling species show social facilitation in predator recognition, which benefits the group as a whole (Hall & Suboski, 1995), whereas other species can use socially gained information to capitalise on food patch profitability for an intraspecific competitive advantage (Coolen et al., 2003).As a shoaling species, shubunkin goldfish are expected to behave in a way to strengthen the group, while dealing with factors such as intraspecific competition.
Goldfish, Carassius auratus, is a social fish (Blanco et al., 2018).Goldfish have demonstrated the ability to navigate mazes and the use of cognitive maps (Rodriguez et al., 1994).Shubunkin goldfish have varied and colourful blemishes which make them easy to identify by eye, and habitually eat regardless of the level of hunger.For these qualities we selected shubunkin goldfish.
The main aim of this investigation is to develop our understanding of the role of social learning in the formation of spatial memory.Particularly, to see whether social learning has any impact on the rate of learning.Other aims include understanding whether the presence of a trained individual affects other facets of the experiment, such as reward acceptance, motivation, and time taken to complete the trial.We hypothesise that unconditioned fish, which observe a conditioned fish undergo a maze, will learn the maze faster than fish that did not observe a conditioned fish.We hypothesise this because of the findings from Laland and Williams (1997) and Pouca et al., (2017) indicate similar behaviour.
J o u r n a l P r e -p r o o f

Subjects
Shubunkin goldfish were bought from an online aquarium trader and maintained for a month before the experiment started.The goldfish lived in 40 cm x 80 cm x 45 cm glass aquaria with aerated filtered water at approximately 22°C.The experiment room used a light-dark cycle of 12 hours, and the goldfish were fed flakes once a day, every evening after the testing.100 goldfish were separated evenly and randomly between five tanks that were situated in the experiment room.The tanks were identical in contents, containing 20 goldfish, fake plants, a water pump and the water filtration system, and were assigned an experimental group each at random.Each goldfish had profile photos taken of them, which were printed onto laminated paper, and were assigned a number for identification.Their unique patterns and blemishes made the breed ideal for eye-identification (Figure 1).Due to the highly inbred nature of the breeding of shubunkin goldfish, there are no non-invasive methods to determine sex, and therefore sex was not accounted for in this experiment.Each tank was assigned a group (demonstrators, observers, spare observers, control, spare control), which prevented any goldfish being familiar with any goldfish from any other group before any trials.Each tank had 20 goldfish.If any goldfish from the observer or control group needed to be replaced, they were replaced with goldfish from their respective spare tank.Throughout the experiment, 10 of the original observer fish were replaced, and 12 of the original control fish were replaced.

J o u r n a l P r e -p r o o f
The circumstances when fish were switched out were due to a highly-unmotivated subject, a subject that did not accept the food reward, or a subject suffered health problems during the course of the experiment.As only two demonstrators were required, the other fish from the demonstrator group were used to improve the design of the experiment.We used only two demonstrators for the core experiment to limit the variance in demonstrator performance.One demonstrator was used to demonstrate the left route, and the other for the right route.They only demonstrated once per day.Stanley (left) had a success rate of 91.6% throughout demonstrating, and Hobnob (right) had a success rate of 92.3%.

Apparatus
The experimental apparatus consisted of a four-armed maze (Figure 2) constructed with transparent glass.The maze was on a 76 cm tall table with a 1 cm thick white polystyrene sheet between the tank and the table.Each arm entrance had a slot to slide a transparent sheet of plastic to act as a door (30 cm tall).Another of these doors was used to separate the inner and outer halves of the starting arm.Three arms were used for one given trial, requiring an opaque light blue plastic door to block the one arm not in use for the respective trial.These doors were moved manually.
Timing was taken using a stopwatch.Each experiment day started with putting fresh water into the maze up to a height of 22 cm.The water was approximately 20°C and was circulated around the maze between each trial.The layout of the procedure room is shown in Figure 3.The maze was situated in the same room as the goldfish were housed in, within line of sight of the tank.
However, the distance was too far for the goldfish to gain information of the maze from the tank.The experimenter positioned themselves randomly in one of five marked spots around J o u r n a l P r e -p r o o f the maze to observe the fish.Resources used during the trials, such as the doors and the bloodworms, were kept closeby on some shelving.There are no windows to the outdoors in the room, and the adjacent room is a large aquarium room with translucent windows.
Each fish was assigned an arm (left/right) that was considered the 'correct' arm which resulted in a reward.The target arm would remain the same for the entirety of the respective fish's training.The control and observer groups each had ten that were randomly assigned the left arm, and ten that were randomly assigned the right arm.

Procedure
Training: To train an individual, they were placed into the inner segment of the starter arm and given two minutes to acclimatise to the maze environment.To begin the trial, the door for the starter arm was lifted and the timer started.Once the subject had completely passed across the threshold of the choice arm entrance (excluding the starting arm) (Figure 4), a door was slotted into the arm entrance and the timer was stopped.If the fish got the choice correct, two bloodworms were immediately dropped into the further end of the now inhabited arm.If incorrect, the individual was given two minutes of isolation in the incorrect arm.Each trial was recorded for success and time taken.After the trial, the individual was netted back to the starting arm, and water was circulated around the maze to remove the influence of olfactory cues post-successful trial by saturating any remaining cues.This was repeated until the goldfish had completed ten trials -this formed a block of trials.The goldfish was then returned to the tank it came from, and the next subject was brought to the maze.Each goldfish was trained for one block of trials every weekday until it reached the learning criterion of 80% trial success rate for three consecutive trial blocks.If the subject had not made a decision by five minutes, J o u r n a l P r e -p r o o f it was considered an unmotivated trial.Unmotivated trials were not re-ran.Three unmotivated trials in a single block resulted in the omission of that block.
Demonstrators: Fish to become demonstrators (N = 2) were trained first.One individual was trained to turn left, the other was trained to turn right.Once trained, they were tested for another two weeks to ensure that they could maintain the high success rate post-training, as their consistency while demonstrating was important (90% success rate for both demonstrators).If this was the case, demonstrating started the day after the two week post-training period.
Observers: Observer fish (N = 20) were put into the observer segment of the maze (a portion of the start arm, separated by a see-through barrier, as seen in Figure 3) for their observation period of three blocks of trials.During this, a demonstrator, that was trained to turn the same direction that the observers were going to be trained to take, completed three blocks of trials as normal, including positive and negative reinforcement.Half of the observer subjects (N = 10) observed the left-turning demonstrator, and the other half (N = 10) observed the right-turning demonstrator.The data about these trials were recorded.After three blocks of observations, the observer underwent training as usual.
Control: To eliminate the influence of familiarity with the maze on training, control goldfish (N = 20) spent 30 minutes in the observer segment for three days in a row, with no other fish in the maze.This provided control fish with as much time as observers in the maze before training, and then they too underwent training as usual.

Data recording and analysis
To evaluate a possible link between social learning and spatial memory, the following measures were recorded; success of trials, time taken per trial, whether the reward was accepted within J o u r n a l P r e -p r o o f the first three blocks (subjects that consistently refuse the reward within the this period are replaced after the third block), and the motivation of the subject (unmotivated subjects do not make a decision within 300 seconds).Time taken per trial started when the starting arm's door was lifted, and stopped when the door had been placed in the decided arm.The rate of learning was recorded as the number of trials that were undertaken before reaching the trained threshold.
For each individual, single measures of performance (e.g.number of blocks to learn, mean trial time, % success) were used in the following statistical analyses.Therefore, there were no repeated performance values per individual.Trials were not recorded with video technology.
RStudio (version 2023.09.0+463), with R version 4.3.1, was used to process the statistical tests.
Group means for every measure required nonparametric tests for comparison, due to the not normally distributed nature of the data.Mann-Whitney U tests were utilised to compare group rates of first trial block performance, group performance, group food acceptance likelihood, the effect of food acceptance on performance and group motivation.A Pearson's Chi-squared test was used to compare the success rate of the first trial between the two groups.An Independent Samples t-test was used to test for a difference on the observer performance, depending on which demonstrator was used.Spearman's rank correlation tests were used to test the correlations of time taken per trial and trial success rate, acceptance of reward likelihood, subject motivation and trial success rate, the trial success rates of demonstrations and observers, and the effect of demonstrator motivation on both observer motivation and trial success rate.The R packages used for data visualisation were ggplot2, ggpubr and ggthemes.

Statement of ethical review
Ethical approval for this study was granted by Bangor University Animal Welfare and Ethics Review Board.The protocol number for the ethical review: CNS2018JCB01A.

Control VS Observer
There was a significant difference in the number of trials it took for controls and observers to complete the training.According to a Mann-Whitney U test, the control group (Md = 70, SD = 46.521)trained, on average, in fewer trials than the observer group (Md = 115, SD = 81.323),W = 120, p < 0.05 (Figure 5).
According to a Mann-Whitney U test, there was no significant difference in the mean trial success rate of the groups in the first block (control Md = 50%, observer Md = 50%), W = 171, p = .430. Figure 6 shows the progression of mean trial success rate over the first five blocks of trials.A chi-squared test of independence revealed that there was no significant difference between the success rate of the first trial of the two groups, X 2 (1) = 0.110, p = .740. Figure 7 shows the overall progression of the training of all the subjects over the course of the experiment.

Motivation
The final factor investigated was the motivation of the subjects.The proportion of motivated trials of a fish was calculated with the following equation; J o u r n a l P r e -p r o o f The motivation of goldfish was analysed against their group, and how quickly the subject trained.A Mann-Whitney U test showed a significant difference between the motivation of observers (Md percentage of unmotivated trials = 3.125%) and control (Md percentage of unmotivated trials = 0.000%), W = 131, p < 0.05.Furthermore, as seen in Figure 8, there was a significant positive correlation between the percentage of unmotivated trials a goldfish had, and the number of trials it took for them to be considered trained, r(6969.100)= 0.346, p < .05.

Demonstrator performance
The rate of learning did not significantly differ depending on which demonstrator was being observed, t(38) = 0.403, p > 0.05, 95% CI [7.34,22.96].Furthermore, the mean success rate of the demonstrators over the course of the demonstrations was 92%, and any variation of success rate observed by a particular observer did not affect their rate of training, r( 1571

Discussion
The finding of this research is that the fish that observed a trained individual learned the maze slower than their non-observing conspecifics.Therefore, it could be inferred that the act of observing a trained individual undertake the maze inhibited the learning process for the observer.Other studies found that fish can learn routes socially, and in some cases preferred detrimentally-longer socially learned routes (Laland & Williams, 1998).The observation process also reduced motivation and increased food acceptance, but only motivation had a significant effect on the rate of learning, as a higher rate of motivation led to a faster rate of learning.
Our results show that there was no significant difference in the success rate in the first stages of learning, within the first five blocks (Figure 6).Furthermore, there was no significant difference in the first trial success rate between the two groups, which is a strong indicator that social learning of the spatial task is unlikely to have occurred.We can be confident that the observer subjects were able to observe the demonstrators get the reward, as observers were significantly more likely to accept the reward than the control group, noting that fish can socially learn food acceptance (Suboski & Templeton, 1989).The data support the idea that there has been some social transmission of information, as the only rational explanation for the difference in food acceptance was the observation of the acceptance of food by the demonstrators.However, it remains unlikely that any social transmission of spatial data positively impacts the rate of learning in the spatial task.The remainder of the results indicate that the social aspect of the experiment may be detrimental to the observer subjects in training to learn this maze.
J o u r n a l P r e -p r o o f performance of observer zebrafish and naïve-paired zebrafish, whereas the control goldfish in this study significantly outperformed observer goldfish.This differs from our results, as our study found that observation negatively affected the performance of the test subjects.One of the main differences between the two experiments is that in Roy and Bhat's (2017) experiment, the observer followed the demonstrator through the maze during the observation period; whereas, in our experiment, the observer was held within a fixed point for the observation period.The argument could be made that this is not the cause for the differences between results, as a study by Pouca et al., 2020, found that non-social, juvenile Port Jackson sharks, can socially learn a new foraging route through observation of a trained demonstrator, and that observers learned faster than individual learners and sham-observers (paired with a naïve demonstrator).Our results contrast from Pouca et al., as the observers showed no improvement to performance/learning. Pouca et al. suggested that the results occurred due to the demonstrators drawing attention to the route, increasing the observer's learning opportunities.
The experiment design of both the Pouca et al. experiment and the Roy and Bhat (2017) experiment differ from ours by having the observers following the demonstrators in the task, but their results differ.Therefore, it is possible that the differences between our experiment and Roy and Bhat's (2017) are not attributable to this element of experiment design.These unusual findings allow for unique opportunities to investigate the reasons for these behaviours.Goldfish, as well as the species they descended from, live in streams and rivers in the wild.The hypothesis of this experiment was that it would be beneficial for goldfish to gain spatial information whilst observing their conspecifics foraging, in order to increase own success of foraging and reduce the chance of predation.Guppies, another river species of fish, are capable of socially learning foraging sites and escape routes (Reader et al. 2003).As this is not the case with these results, alternative behaviours may be more beneficial ecologically.As netting could J o u r n a l P r e -p r o o f be interpreted as a predation event by the subjects, it is possible they were avoiding the correct arm where netting took place during the demonstration.As it is evident that the observers could see the demonstrators accept a food reward in a different arm, we can be even more confident that they witnessed the netting process to return the demonstrators to the beginning arm, leading the observers to avoid that arm to avoid the same stressful event.It may also be possible that the goldfish are avoiding foraging areas where food depletion was observed, gambling that it would be more beneficial to forage elsewhere (Brown, 2011).This phenomenon has been studied in mammals, but not in fish.The lack of available data on this topic, particularly about fish in their natural habitat, reflects the research opportunities available, as well as the difficulty in providing accurate ecological perspectives to laboratory studies.
An explanation for the behaviour found in this experiment is the possible occurrence of tradeoff decision making.The observer group witnessed the demonstrators accepting a food reward, or spending two minutes in isolation.After both of these outcomes take place, the demonstrator is netted back to the starting arm.Observers witnessed demonstrators completing the spatial task with an average trial success rate of 92%, meaning 92% of observed trials ended with a food reward closely followed by the demonstrator being netted, as opposed to 8% of observed trials that ended with two minutes isolation, then followed by the demonstrator being netted.
The observer fish's decision making may be affected by the observation of the stressful event of netting.Therefore, we should consider the possibility that the observers' hesitation with decision making, emphasised through the less motivation shown by observers, may be the result of the worry of being netted.This may then result in a delay in the learning process, as observers may be focusing on a trade-off decision between food or not being netted as quickly.
It is not possible to rule this out as a factor from the experiment.However, we could argue that the observation of the netting procedure may itself be an effect of social learning, as their behaviour is impacted by socially-gained information.Future studies should use a stress-free J o u r n a l P r e -p r o o f replacement for netting to remove this factor.Alternatively, future studies could have fish netted in an observational condition and in a control condition.One possible explanation for the results of this study may be the occurrence of the passenger effect.Burt de Perera and Guilford (1999) found that pigeons learned the position of the food goal more effectively when performing the task alone than after completing it with a knowledgeable companion.Similar findings have been found in pigeons (Robertson et al., 1985;Giraldeau & Lefebvre, 1986;Biederman & Vanayan, 1988;Lefebvre & Helder, 1997;Banks & Guilford, 2000), zebra finches (Beauchamp & Kacelnik, 1991) and crab-eating macaques (Stammbach, 1988).Banks and Guilford (2000) hypothesised that observers are learning something different about the task compared to a control conspecific.It is possible that the presence of the demonstrator itself is an indicator of food, therefore removing the demonstrator would remove the indication that there is a food reward.Banks and Guilford (2000) suggest that spatial cues are not picked up from the demonstration by the naïve pigeon, or that the naïve pigeon fails to transfer them into the single phase, as homing pigeons can utilise knowledge observed from a demonstrator during a paired phase of a homing experiment.
However, when completing the task alone, they failed to benefit from their initial exposure of knowledge from a trained conspecific.They also suggested that the naïve pigeons' focus was on the demonstrator in fear of exposure by desertion, therefore inhibiting encoding of information function.The perception of being transported via netting as a stressful predation event might provide a fear of desertion for the subjects in this study, which would be further reinforced by the partition between the demonstrators and observers.However, one problem with the integration of the passenger effect into our results is that these experiments involve the naïve individual travelling with the demonstrator during observation, rather than observing from a fixed point.Unlike other experiments that are investigating the passenger effect, this one has the naïve fish only able to observe and do nothing else, arguably making it not a J o u r n a l P r e -p r o o f passenger at all.Therefore, if the mechanism for the passenger effect is rooted in the action of following, then that would not apply to this study.
The motivation of a goldfish to undertake the maze, combined with the mean speed at which goldfish completed their trials, may indicate boldness, familiarity with the situation and the strength of the attraction of the reward.Observers had a significantly higher percentage of unmotivated trials compared to control goldfish, but there was no significant difference in time taken per trial between the two groups.Regardless of group, each subject spent the same amount of time in the maze before commencing training, so the only difference in familiarity between the groups at the beginning of their training is any information the observers gain socially.Furthermore, goldfish as a species were selected for this experiment due to their insatiable hunger for food, which is unlikely to be satisfied with the amount of food offered throughout a day of trials.Considering that they are unfed in the aquarium until after the day's trials were complete, hunger is unlikely significantly impacting their motivation.Similarly, the social learning of food acceptance shows that the observers were able and willing to watch the demonstrator undertake the task.Ruling familiarity, attention, and hunger out, the lack of motivation for observers may be attributed to either boldness or strength of attraction to the reward.However, Roy and Bhat (2017) suggested that observers may be more comfortable in the maze due to the boldness of the demonstrators.Although increased familiarity may have influenced performance of wild-population juvenile zebrafish, this did not lead to higher learning capabilities (Roy & Bhat, 2016).If the reward was not a strong enough incentive to prompt the subject to embark the maze, then it would have been a struggle to maintain the process of training for all fish.Therefore, with this reasoning, the social aspect of the observers training might affect the boldness of the subjects.The higher the proportion of unmotivated trials a goldfish exhibited, the higher the number of blocks of trials it took for them to train (Fig. 8, P < 0.05).This could either mean that the lack of motivation affects the rate of learning, J o u r n a l P r e -p r o o f or these are two independent factors of observers and it just so happens that they are both slower to learn and are more unmotivated.
The use of shubunkin goldfish may have had an effect on the experiment that led to variance.
Shubunkins were used for their characteristic red, orange and black blemishes, unique to each individual, which made for easier identification with the naked eye.This variance in morphology is a product of inbreeding within the breed.This may well have also resulted in variance in spatial and social abilities between individuals.To avoid this, a wild type of breed of goldfish would be needed.The breed choice used also made it difficult to determine sex, therefore any difference due to sex could not be analysed.Acquiring wild-caught individuals from a single population would also eliminate any differences and influences from the aquarium industry, such as food preference.It should also be mentioned that when a goldfish did not accept any rewards within the first five blocks of trials, the goldfish was discontinued from the experiment and the results were not included in analysis.Food avoidance is a common sign of stress in goldfish (Carr, 2002), so perhaps the difference between the groups was due to a difference in stress through isolation, rather than social learning (Gaikwad et al., 2011).This suggests that the rejection of food data analysed in this experiment could be due to stress, but does not accurately represent the acceptance/rejection of food throughout the entire experiment.
It is difficult to determine a particular social learning mechanism for the spatial task, as the poor performance compared to the controls implies that social distraction, not social learning, took place.Considering how the demonstrators had a significant effect on observer behaviour, asocial learning can be ruled out (Heyes, 1994).In regards to the difference in food acceptance, social facilitation may be responsible.Guerin (1993, page one) defines social facilitation as 'when there is an increase or decrease in behaviour by an animal in the presence of another J o u r n a l P r e -p r o o f animal that does not otherwise interact with the first animal'.The reduction in the rate of learning by observers could be due to cognitive distraction, where the presence of conspecifics inhibits complex task processing (Guerin, 1993;Aiello & Douthitt, 2001).Similarly, a phenomenon where being part of a group results in less motivation, called social loafing, could account for both the lower rate of spatial learning and lower motivation (Paulus, 1983).
However, this mechanism typically applies to situations where effort from both individuals are given in a joint task.Although these social learning mechanisms are descriptive, they are poorly studied in animals, particularly in fish.
One way to expand the questions investigated in this experiment would be to test how the demonstrator success rate affects rate of learning.If a group of observers was exposed to a demonstrator completing 50% of the trials correct, and another group watch demonstrators that were 20% correct, it can be interpolated that either observing failures, successes, or an even mix provides the best chance for social learning.Alternatively, a parallel study where the observer fish follows the demonstrator through the maze, as opposed to being restricted to the observation segment, would provide more insight into the optimal situation for goldfish to socially learn spatial tasks.A suggested improvement to this experiment would be to include the presence of a demonstrator fish in the maze during the control's acclimatisation period.In addition to the netting problem, a limitation of the experiment design was the slight difference in temperature between the home tanks and the maze tank.Although any impact of the temperature difference would have affected all fish equally, it is difficult to rule out the possibility that this could have impacted decision making.Furthermore, future studies should ensure video recording of the experiment takes place, as the lack of video recording is a limitation on the reliability of the results found in this experiment.

J o u r n a l P r e -p r o o f
This investigation has shed some light onto the relationship between social learning and spatial memory capabilities in goldfish, and the impact of observing on an individual's motivation and likelihood of food acceptance in an experimental setting.This information could guide experimental design in future studies into social learning.
J o u r n a l P r e -p r o o f

Conclusion
Overall, our main hypothesis was rejected, as the act of observing a trained individual resulted in a slower rate of learning.It is not clear whether the socially gained information is inhibiting the learning process, or if the observation experience alters the subject's behaviour.In addition, the act of observing resulted in differences in motivation between the groups, which reflects a lack of motivation or desire to complete the spatial task, or a reduction in boldness.J o u r n a l P r e -p r o o f

Project funding
This project was entirely self-funded through Student Finance England.
J o u r n a l P r e -p r o o f 3.1 Results Individuals within the observer group (Md percentage of rejected rewards in the first three blocks = 0.000%) were significantly more likely to accept the food reward than individuals from the control group (Md percentage of rejected rewards in the first three blocks = 17.333%), according to a Mann Whitney U test, W = 287, p = 0.05.Furthermore, a Mann-Whitney U test showed there was no significant difference in the mean time taken per trial between control (control Md = 23.971seconds, control SD = 23.641) and observers (Md = 51.844seconds, observer SD = 29.833),W = 129, p > 0.05.

Figures
Figures

Figure 1 J
Figure 1 Figure 2 J o u r n a l P r e -p r o o f

Figure 7 J
Figure 7 J o u r n a l P r e -p r o o f

Figure 1 .
Figure 8 J o u r n a l P r e -p r o o f

Figure 2 .
Figure 2. Scale to plan view of the maze arena.The arms of the maze were 30 cm long, 30 cm

Figure 3 .
Figure 3. Plan to scale of the maze and the procedure room.X = marked spots for the

Figure 4 .
Figure 4. Illustration to show the point at which the fish were considered to have officially

Figure 5 .Figure 6 .
Figure 5.The number of trials taken for goldfish (Carassius auratus) to be considered trained

Figure 7 .
Figure 7.The mean trial success rate of goldfish (Carassius auratus) in a spatial task over the

Figure 8 .
Figure 8.The proportion of unmotivated trials (300 seconds without a choice made) per