Do monkeys see the way we do? Qualitative similarities and differences between monkey and human perception
Curation statements for this article:-
Curated by eLife
eLife Assessment
This important study presents a comprehensive comparison of human and macaque monkey behavior across a range of visual perceptual phenomena using a unified oddball visual search paradigm, which enables direct cross-species comparison while minimizing task-related confounds. It provides convincing evidence that visual perception is largely similar between these two species, but with some interesting exceptions. These insights into qualitative and quantitative differences between species are relevant for evaluating macaques as a model organism for understanding human vision.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Monkeys are widely used as models for human vision. While their anatomy and physiology are similar to humans, it is unclear to what extent their perception matches with ours. Previous studies on monkeys have evaluated specific perceptual phenomena after extensive training that could have even altered perception. To resolve these issues, we trained monkeys to perform a highly general oddball visual search task on natural images, and then tested them on carefully controlled image sets to detect a variety of perceptual phenomena and compared their performance with humans. This revealed a number of qualitative similarities and differences. Like humans, monkeys showed similar object relations, Weber’s law and amodal completion. However, unlike humans, monkeys did not show mirror confusion or a global advantage. Our results represent a first comprehensive comparison of visual perception in monkeys and humans, revealing both strengths and limitations of monkeys as a model organism for human visual perception.
Article activity feed
-
eLife Assessment
This important study presents a comprehensive comparison of human and macaque monkey behavior across a range of visual perceptual phenomena using a unified oddball visual search paradigm, which enables direct cross-species comparison while minimizing task-related confounds. It provides convincing evidence that visual perception is largely similar between these two species, but with some interesting exceptions. These insights into qualitative and quantitative differences between species are relevant for evaluating macaques as a model organism for understanding human vision.
-
Reviewer #1 (Public review):
[Editor's Note: this version has been assessed by the Reviewing Editor without further input from the original reviewers. The original reviews for this study were unanimously positive. Reviewers requested more extensive discussion of the study's limitations (including potential cross-species differences in familiarity with the object stimuli used), and clarification of the training procedure and some design choices (such as the choice of tasks). The revised manuscript comprehensively addresses these comments. Since reviewers did not question the accuracy or validity of the study's conclusions and mentioned that the paper can serve as an impactful example of cross-species comparisons in other fields (R2), the eLife assessment was adapted to state that this is an important study providing convincing results.]
S…
Reviewer #1 (Public review):
[Editor's Note: this version has been assessed by the Reviewing Editor without further input from the original reviewers. The original reviews for this study were unanimously positive. Reviewers requested more extensive discussion of the study's limitations (including potential cross-species differences in familiarity with the object stimuli used), and clarification of the training procedure and some design choices (such as the choice of tasks). The revised manuscript comprehensively addresses these comments. Since reviewers did not question the accuracy or validity of the study's conclusions and mentioned that the paper can serve as an impactful example of cross-species comparisons in other fields (R2), the eLife assessment was adapted to state that this is an important study providing convincing results.]
Summary:
The authors set out to conduct a behavioral comparison of macaque and human vision across a wide range of visual properties. Such a comparison is critical for evaluating the use of macaques as a model system for understanding human vision and the underlying neural mechanisms. This goal represents a unique endeavour since prior studies have typically focused on only highly specific tasks. While the authors found consistent coarse representational structure for objects, evidence for Weber's Law and amodal completion, there was divergence for mirror image confusion and the use of global or local image properties.
Strengths:
There are three major strengths of the study. First, the authors employed a behavioral paradigm (oddball search) that allowed them to test multiple perceptual phenomena without having to train the macaques on the specific type of stimuli tested. Second, humans and macaques could be tested in an identical manner. Third, the authors tested a range of different visual properties and phenomena, allowing a broader comparison between species.
There are also some weaknesses to the study (described below), but that doesn't change the fact that the authors have demonstrated and validated a novel approach for systematic and comprehensive comparisons of vision across species.
-
Reviewer #2 (Public review):
Summary:
The macaque monkey is often considered as the animal model of choice to study the neural correlates of visual perception, due to the close similarities to humans in terms of anatomy, physiology and behaviour (Van Essen and Dierker, 2007; DiCarlo et al., 2012; Roelfsema and Treue, 2014; Picaud et al., 2019; Van Essen et al., 2019; Hesse and Tsao, 2020). Quite some studies have been performed to compare the behaviour of macaque monkeys and humans on visual perception tasks. However, it remains difficult to compare the results of these studies as the methods that are used differ significantly between these studies. Furthermore, behavioural studies of macaque monkeys often involve extensive training as the tasks were relatively hard, making it difficult to compare the results with humans, who generally …
Reviewer #2 (Public review):
Summary:
The macaque monkey is often considered as the animal model of choice to study the neural correlates of visual perception, due to the close similarities to humans in terms of anatomy, physiology and behaviour (Van Essen and Dierker, 2007; DiCarlo et al., 2012; Roelfsema and Treue, 2014; Picaud et al., 2019; Van Essen et al., 2019; Hesse and Tsao, 2020). Quite some studies have been performed to compare the behaviour of macaque monkeys and humans on visual perception tasks. However, it remains difficult to compare the results of these studies as the methods that are used differ significantly between these studies. Furthermore, behavioural studies of macaque monkeys often involve extensive training as the tasks were relatively hard, making it difficult to compare the results with humans, who generally require very little training. The authors present a set of experiments to compare visual perception between macaque monkeys and humans, using the exact same behavioral task that is easy to learn and therefore requires very little training. As expected, they overall find that the two species behave similarly. However, they find a number of interesting exceptions.
Strengths:
A major strength of the current study is the relatively large number of tasks that were tested in the same subjects. This is made possible by using the oddball visual search task, which macaque monkeys can learn very quickly. This means that few trials are sufficient to obtain a significant difference between conditions, minimizing learning effects. Although this type of task has been used in previous studies (Sablé-Meyer et al., 2021), the current manuscript makes better use of the advantages and explains them more explicitly.
In addition, the study finds a number of interesting differences between macaque monkeys and humans. In particular, while humans can dissociate horizontally mirrored images better than vertically mirrored images, monkeys show no difference between these two conditions (Experiment 4). Also, while humans dissociate images better based on the global shape of a stimulus, monkeys dissociate images better based on local elements of a stimulus (Experiments 5 and 6). Although these findings are largely a replication of previous results, they have not yet been studied together with other tasks within the same individual subjects, and the low number of trials avoids any learning effects.
-
Reviewer #3 (Public review):
Summary:
In this study, Cherian and colleagues compare visual perception between humans and monkeys using a common oddball visual search task across a battery of perceptual phenomena. By keeping the task constant and varying only the stimulus sets, the authors aim to isolate perceptual similarities and differences between species. Across six experiments, they report that monkeys and humans share similarities in coarse object representations, Weber's law, and amodal completion, but differ in mirror confusion and global/local processing.
Strengths:
A major strength of the study is the unified experimental framework. The authors designed the experiments such that the task procedures are identical across conditions and species, differing only in the images shown. This is a significant methodological advantage, …
Reviewer #3 (Public review):
Summary:
In this study, Cherian and colleagues compare visual perception between humans and monkeys using a common oddball visual search task across a battery of perceptual phenomena. By keeping the task constant and varying only the stimulus sets, the authors aim to isolate perceptual similarities and differences between species. Across six experiments, they report that monkeys and humans share similarities in coarse object representations, Weber's law, and amodal completion, but differ in mirror confusion and global/local processing.
Strengths:
A major strength of the study is the unified experimental framework. The authors designed the experiments such that the task procedures are identical across conditions and species, differing only in the images shown. This is a significant methodological advantage, as it minimizes task-related confounds that often complicate cross-species and cross-experiment comparisons. As a result, observed similarities and differences can be more directly attributed to perceptual processes rather than differences in training or task demands. This allows for a more comprehensive evaluation of visual perception than is typical in the literature, where individual studies often focus on a single effect with specialized training. The data are carefully collected, and the analyses are systematic and appropriate for the questions posed.
-
Author response:
The following is the authors’ response to the original reviews.
eLife Assessment:
This valuable study presents a comprehensive comparison of human and macaque monkey behavior across a range of visual perceptual phenomena. The use of a unified oddball visual search paradigm enables direct cross-species comparison while minimizing task-related confounds. It provides solid evidence that visual perception is largely similar between these two species, with some interesting exceptions. These insights into qualitative and quantitative differences between species are relevant for evaluating macaques as a model organism for understanding human vision.
Thank you, we are glad to note the positive overall assessment. We hope that our revision will lead to an even stronger assessment.
Reviewer #1 (Public review):
Summary:
Th…
Author response:
The following is the authors’ response to the original reviews.
eLife Assessment:
This valuable study presents a comprehensive comparison of human and macaque monkey behavior across a range of visual perceptual phenomena. The use of a unified oddball visual search paradigm enables direct cross-species comparison while minimizing task-related confounds. It provides solid evidence that visual perception is largely similar between these two species, with some interesting exceptions. These insights into qualitative and quantitative differences between species are relevant for evaluating macaques as a model organism for understanding human vision.
Thank you, we are glad to note the positive overall assessment. We hope that our revision will lead to an even stronger assessment.
Reviewer #1 (Public review):
Summary:
The authors set out to conduct a behavioral comparison of macaque and human vision across a wide range of visual properties. Such a comparison is critical for evaluating the use of macaques as a model system for understanding human vision and the underlying neural mechanisms. This goal represents a unique endeavour since prior studies have typically focused on only highly specific tasks. While the authors found consistent coarse representational structure for objects, evidence for Weber's Law and amodal completion, there was divergence for mirror image confusion and the use of global or local image properties.
Thank you for your accurate assessment.
Strengths:
There are three major strengths of the study. First, the authors employed a behavioral paradigm (oddball search) that allowed them to test multiple perceptual phenomena without having to train the macaques on the specific type of stimuli tested. Second, humans and macaques could be tested in an identical manner. Third, the authors tested a range of different visual properties and phenomena, allowing a broader comparison between species.
Thank you, we are glad that you found our study to be interestsing.
There are also some weaknesses to the study (described below), but that doesn't change the fact that the authors have demonstrated and validated a novel approach for systematic and comprehensive comparisons of vision across species.
We have now done our best to respond below to these comments and acknowledge limitations explicitly in the revision.
Weaknesses:
The weaknesses of the study arise in part because of the breadth of the work. In cases where there are divergences between the two species, it would be helpful to know what might account for such divergence, to have more depth. Is it really a species difference, or could there be a different account? For example, does the difference in mirror image confusion arise because the stimuli were objects that would have been highly familiar to the humans but not the macaques? Further, the authors often used small sets of stimuli (e.g. 8 objects only in the test of object similarity; a small set of highly specific occlusion stimuli), and how well the findings will generalize beyond those stimuli is unclear.
Thank you for raising these important points. We have now acknowledged these issues in the Discussion (p.37) and the relevant paragraphs are reproduced below.
“Although we have done our best to perform a comprehensive comparison of several perceptual phenomena between monkeys and humans, our study has several limitations.
First, we were only able to test these phenomena on three monkeys, because training monkeys on these tasks is extremely challenging, and requires months of focused training, because of which it was simply not feasible for us to test larger numbers of animals. However, we consider it unlikely that our results would not generalize, because many of the tested perceptual phenomena were robust across our monkeys, and are highly robust across humans. There was also striking similarity between the three monkeys, which would not have been observed had there been large inter-individual differences. Even if our results did not generalize, the fact remains that a subpopulation of monkeys do show similarities and differences as observed here. Testing these phenomena on larger populations of animals will require installing automated training stations in group enclosures where many animals can be tested simultaneously.
Second, the perceptual phenomena tested in this study are based on small numbers of images and conditions that again might not generalize when repeated on a larger set of images/conditions. While this is possible in principle, it was simply not possible for us to test a wide range of phenomena with a wide range of images, in the interest of having compact experiments that could be completed within one testing session. Testing a larger set of conditions would require multiple sessions for each animal, which would require counterbalancing sessions and introduce potential practice effects. However, we consider it unlikely that our results would not generalize across a wider set of conditions/images, because these effects are highly robust when tested in humans with relatively little inter-individual variability.
Third, some of the perceptual phenomena tested in this study are based on testing natural objects, that are likely to be familiar to humans but not necessarily to our monkeys. As a result, the results observed could potentially be explained by familiarity to these images in humans but not in monkeys. We consider this unlikely, for several reasons. Our monkeys were housed in natural arenas where they encountered many natural objects, although not necessarily the images used in our experiments. While our human participants might have indeed been familiar with some objects, their responses during visual search may not have been driven by familiarity, particularly since their task involved searching for oddball targets without specifying their features. Indeed, oddball search shows only subtle effects of familiarity, as evidenced by highly correlated performance between readers and non-readers on letter searches (Agrawal et al., 2019, 2022). Understanding how familiarity shapes object representations in humans and monkeys would require careful testing with increasingly familiar objects and potentially with arbitrary shapes that are unfamiliar to both species. These are important considerations for future studies of cross-species comparisons.”
The authors discuss the implications of training macaques to perform specific tasks on specific stimuli in comparing across species. While I agree that extensive training in monkeys could change perception, it is important to also consider that humans have been extensively trained through the types of visual tasks we conduct throughout our lives, so I'm not sure it is universally true that the best comparison is between humans and untrained monkeys. But this just consideration just highlights the general problem of comparing across species.
We agree this is an important point. We now acknowledge this in the Discussion (p. 37) – see the paragraphs above.
Reviewer #2 (Public review):
Summary:
The macaque monkey is often considered as the animal model of choice to study the neural correlates of visual perception, due to the close similarities to humans in terms of anatomy, physiology and behaviour (Van Essen and Dierker, 2007; DiCarlo et al., 2012; Roelfsema and Treue, 2014; Picaud et al., 2019; Van Essen et al., 2019; Hesse and Tsao, 2020). Quite some studies have been performed to compare the behaviour of macaque monkeys and humans on visual perception tasks. However, it remains difficult to compare the results of these studies as the methods that are used differ significantly between these studies. Furthermore, behavioural studies of macaque monkeys often involve extensive training as the tasks were relatively hard, making it difficult to compare the results with humans, who generally require very little training. The authors present a set of experiments to compare visual perception between macaque monkeys and humans, using the exact same behavioral task that is easy to learn and therefore requires very little training. As expected, they overall find that the two species behave similarly. However, they find a number of interesting exceptions.
Thank you, we are glad that you found our study interesting.
Strengths:
A major strength of the current study is the relatively large number of tasks that were tested in the same subjects. This is made possible by using the oddball visual search task, which macaque monkeys can learn very quickly. This means that few trials are sufficient to obtain a significant difference between conditions, minimizing learning effects. Although this type of task has been used in previous studies (Sablé-Meyer et al., 2021), the current manuscript makes better use of the advantages and explains them more explicitly.
In addition, the study finds a number of interesting differences between macaque monkeys and humans. In particular, while humans can dissociate horizontally mirrored images better than vertically mirrored images, monkeys show no difference between these two conditions (Experiment 4). Also, while humans dissociate images better based on the global shape of a stimulus, monkeys dissociate images better based on local elements of a stimulus (Experiments 5 and 6). Although these findings are largely a replication of previous results, they have not yet been studied together with other tasks within the same individual subjects, and the low number of trials avoids any learning effects.
We thank the reviewer for highlighting the strengths of this study.
Weaknesses:
A weakness of the study is that while the objects that were used can be considered to be familiar to humans, they are not familiar to macaque monkeys
In Experiment 4, humans can be expected to have 3D representations of familiar objects such as a Roman helmet or an office chair. Humans can therefore be expected to have view-invariant representations of these objects, predominantly for rotations around the vertical axis of the object (as movements are most common in the horizontal plane). This can explain why only humans confuse objects more often when mirrored vertically than when mirrored horizontally.
Similarly, in Experiment 5, humans can be expected to be familiar with abstract geometric shapes such as squares and circles, while monkeys likely are not. This could explain why monkeys find it hard to recognize these geometric shapes in the global shape of the stimuli, even when thin grey lines are drawn to connect the local elements that constitute the global shape (Experiment 6). Instead, the combination of local shapes can be expected to form a texture that might be more easily recognized by the monkeys
More generally, as proposed by Fagot et al, it might well be that monkeys tend to conceive stimuli as a combination of low-level visual features, instead of as references to objects in the outside world, as humans have learned to do (Fagot et al., 2010). This line of critique would be relevant to take into account.
Thank you for raising these important points, which are similar to the concerns raised by Reviewer 1. We now acknowledge these limitations in the Discussion (p. 37), please see our responses above.
Another weakness could be that only three monkeys are tested, while 24 human subjects are tested. According to some theoretical work, a finding in 3 animals is not sufficient to make a claim about an animal species (Fries and Maris, 2022). However, it seems that the results are largely consistent between the different monkeys. Moreover, the results generally agree with the results from previous literature. The conclusions by the authors are therefore largely supported by the results. Some results could be strengthened by additional experiments, or at least a more extensive discussion of the potential weaknesses.
Thank you for this suggestion. We have now included a detailed section in the Discussion on limitations (p. 37).
The potential impact of the paper is significant, as a start of a comprehensive comparison of visual perception between humans and macaque monkeys, which can inspire other labs to contribute to. This comparison can also be extended to other animal species (e.g. crows and rodents), as well as to different types of artificial neural networks (Leibo et al., 2018).
Thank you, we now acknowledge this point in the Discussion (p. 40).
Reviewer #2 (Recommendations for the authors):
If additional experiments with the same monkeys are feasible, it would be valuable to test objects with which the monkeys are already familiar. In particular, there is evidence that monkeys interpret images of faces as referring to real-world objects (Fagot et al., 2010), which may be relevant for Experiments 4-6. For Experiments 5 and 6, for example, it may be useful to use face-like or emoticon-like stimuli rather than abstract geometric objects.
Unfortunately, our study is now concluded and performing additional experiments is not possible at least for this study.
In additionn, monkeys appear to discriminate geometric objects relatively well when they are presented as solid shapes (Sablé-Meyer et al., 2021). Rather than presenting only outlines, it may therefore be helpful to use filled shapes composed of local elements, in order to direct attention to the global shape (for example, a disk rather than a circle).
Thank you for this interesting suggestion. We acknowledge this in the Discussion (p. 36).
In case additional experiments are not feasible, it may still be worthwhile to examine the current dataset in more detail. In particular, there may be variation across stimuli in the degree of familiarity monkeys have with the objects used in Experiment 4, or in how easily they can distinguish the global shapes used in Experiments 5 and 6. It could therefore be informative to analyse performance separately for different pairs of objects, as in Figures 2B, C. For example, the contrast between an "X" and a circle may be easier for monkeys to distinguish at the global level than some other shape pairings.
While these are interesting possibilities, we did not observe major differences in the monkeys’ performance on these shapes, and therefore did not include them here.
Minor comments:
(1) The fact that the reciprocal of reaction times seems to make it easier to compare between species could be due to the fact that the distribution of latencies then becomes symmetric instead of skewed, see the work on the LATER model (Noorani and Carpenter, 2016).
Yes, thank you for noting this. We now acknowledge this in the Introduction (p. 5).
(2) The study by Hopkins & Washburn 2002 is taken as evidence that macaque monkeys have a global advantage, while the conclusion of the paper itself is the opposite, and reaction times are indeed shorter for local differences than global differences (their Figure 3), as in the current manuscript.
Thank you for pointing this out, we have corrected this copyediting error.
(3) The discussion section about behavioral tasks that can be used to quickly probe perception is relevant, relating it to the free-choice paradigm. In addition, it may be useful to discuss the reward-association paradigm developed in the lab of Barry Richmond to quickly probe perceptual categorization after cortical lesions in macaque monkeys (Minamimoto et al., 2010). Although behavior in this paradigm is somewhat noisier, several dozen trials are generally sufficient to pick up differences in reaction times, which might provide an even more sensitive measure of the animals' ability to distinguish stimulus categories.
We thank the reviewer for sharing this observation. While this is an interesting paradigm, the interaction with various reward volumes can make it extremely hard to interpret the results. We are not sure how this could be used for the kind of cross-species comparison we are attempting here.
Reviewer #3 (Public review):
Summary:
In this study, Cherian and colleagues compare visual perception between humans and monkeys using a common oddball visual search task across a battery of perceptual phenomena. By keeping the task constant and varying only the stimulus sets, the authors aim to isolate perceptual similarities and differences between species. Across six experiments, they report that monkeys and humans share similarities in coarse object representations, Weber's law, and amodal completion, but differ in mirror confusion and global/local processing.
We thank the reviewer for their careful evaluation of our manuscript and constructive feedback. We hope that our responses satisfactorily address the reviewers concerns.
Strengths:
A major strength of the study is the unified experimental framework. The authors designed the experiments such that the task procedures are identical across conditions and species, differing only in the images shown. This is a significant methodological advantage, as it minimizes task-related confounds that often complicate cross-species and cross-experiment comparisons. As a result, observed similarities and differences can be more directly attributed to perceptual processes rather than differences in training or task demands. This allows for a more comprehensive evaluation of visual perception than is typical in the literature, where individual studies often focus on a single effect with specialized training. The data are carefully collected, and the analyses are systematic and appropriate for the questions posed.
Thank you for your accurate summary of the strengths of our study.
Weaknesses:
Despite its strengths, the study is largely descriptive and provides limited mechanistic or theoretical explanation for the observed similarities and differences. While the authors document several convergences and divergences between humans and monkeys, there is relatively little discussion of why these patterns arise or how they relate to existing theories of visual processing. As a result, it is difficult to assess the broader implications or generalizability of the findings beyond the specific task and stimuli used.
Thank you for raising this point. Our goal was to characterize similarities and differences in perception between monkeys and humans. As previous comparative studies have typically focused on individual perceptual effects using different behavioural paradigms, it has remained difficult to obtain an integrated view of similarities and differences across multiple aspects of perception. Therefore, we believe that the establishment of this comparative benchmark is an important first step. We have deliberately chosen to avoid a deeper causative analysis of any observed differences in perception as that would necessitate further, and specific experiments to test out any competing hypothesis. This is an important aspect that can be studied in future experiments.
Relatedly, the rationale for selecting the particular set of perceptual phenomena is not fully developed. Some tasks appear motivated by prior work comparing humans and deep neural networks, but it is unclear whether this set constitutes a representative or theoretically grounded sampling of visual perception. Without a clearer justification, it is difficult to interpret the absence or presence of specific effects (e.g., mirror confusion or global advantage) as reflecting fundamental species similarities/differences.
We appreciate this insightful comment. The perceptual phenomena included in this study collectively span multiple aspects of visual processing while remaining measurable within the same oddball visual search paradigm. Our intention was not to provide an exhaustive survey of visual perception, but rather to demonstrate that a diverse range of perceptual signatures can be evaluated under identical behavioural conditions. These signatures are well-established and can be measured using the oddball visual search task. The primary aim of our study was to establish a general comparative framework rather than proposing any set of phenomena as the only or definitive dimensions of visual perception.
Reviewer #3 (Recommendations for the authors):
A key strength of this study is the use of a general oddball search paradigm that avoids task-specific training for each perceptual phenomenon. However, the manuscript provides very limited detail about how monkeys were trained to perform the task, with only a brief statement that they learned it in less than two weeks. Given that one of the central claims of the study is that it avoids training-induced biases, a more detailed description of training procedures is essential. In particular, it would be important to clarify what images were used during training, and whether these images were distinct from those used in the six experiments. If the monkeys were trained on a separate set of images and then generalized to novel stimuli, this would represent a significant strength that could be emphasized more clearly in the manuscript. If not, and if training involved the same or similar stimuli, this raises potential concerns about experience-dependent effects that should be addressed explicitly. Additionally, it would be helpful to clarify whether monkeys required retraining or familiarization when transitioning between experiments with different stimulus sets.
Thank you for this important suggestion. We have now included details of the training procedures in the Methods (p. 42).
The manuscript would benefit from a clearer and more principled justification of the selected perceptual phenomena. In the current version (e.g., L97-102), the choice appears to be motivated in part by the authors' prior work comparing humans with deep neural networks. While this is a reasonable starting point, it does not fully justify the selection in the context of cross-species comparison. The authors could strengthen the paper by explaining whether these phenomena were chosen to span distinct levels of visual processing, to test specific theoretical predictions, or to provide a representative sampling of perceptual functions. Making this rationale explicit would help readers interpret both positive and negative findings. For example, in its current form, there might be cross-species differences only in mirror-confusion and global advantage, and whatever is driving the differences does not affect any other visual processing.
As we have detailed above, we attempted to employ a broad range of perceptual phenomena and unfortunately these do not have any common theoretical justifications, other than that many of them cannot be explained by simple pixel or low-level differences. We now acknowledge this point in the Introduction (p. 6).
The results for coarse object representations and for global/local processing may benefit from further clarification. In Experiment 1, animate objects appear to be predominantly lateral views and therefore asymmetric, whereas inanimate objects are more often symmetric. As a result, within-category similarity may partly reflect shared global shape properties. It would be useful for the authors to clarify how this observation relates to the finding of a local (rather than global) advantage in monkeys in Experiments 5 and 6.
Thank you for this interesting observation. While it is true that some of our inanimate objects are asymmetric, some are symmetric as well. Since the objects in this experiment were different in both their global and local features, it is not clear which features were being used by humans and monkeys, particularly since their performance was correlated. We now acknowledge this point in the Discussion (p. 35).
-
eLife Assessment
This valuable study presents a comprehensive comparison of human and macaque monkey behavior across a range of visual perceptual phenomena. The use of a unified oddball visual search paradigm enables direct cross-species comparison while minimizing task-related confounds. It provides solid evidence that visual perception is largely similar between these two species, with some interesting exceptions. These insights into qualitative and quantitative differences between species are relevant for evaluating macaques as a model organism for understanding human vision.
-
Reviewer #1 (Public review):
Summary:
The authors set out to conduct a behavioral comparison of macaque and human vision across a wide range of visual properties. Such a comparison is critical for evaluating the use of macaques as a model system for understanding human vision and the underlying neural mechanisms. This goal represents a unique endeavour since prior studies have typically focused on only highly specific tasks. While the authors found consistent coarse representational structure for objects, evidence for Weber's Law and amodal completion, there was divergence for mirror image confusion and the use of global or local image properties.
Strengths:
There are three major strengths of the study. First, the authors employed a behavioral paradigm (oddball search) that allowed them to test multiple perceptual phenomena without …
Reviewer #1 (Public review):
Summary:
The authors set out to conduct a behavioral comparison of macaque and human vision across a wide range of visual properties. Such a comparison is critical for evaluating the use of macaques as a model system for understanding human vision and the underlying neural mechanisms. This goal represents a unique endeavour since prior studies have typically focused on only highly specific tasks. While the authors found consistent coarse representational structure for objects, evidence for Weber's Law and amodal completion, there was divergence for mirror image confusion and the use of global or local image properties.
Strengths:
There are three major strengths of the study. First, the authors employed a behavioral paradigm (oddball search) that allowed them to test multiple perceptual phenomena without having to train the macaques on the specific type of stimuli tested. Second, humans and macaques could be tested in an identical manner. Third, the authors tested a range of different visual properties and phenomena, allowing a broader comparison between species.
There are also some weaknesses to the study (described below), but that doesn't change the fact that the authors have demonstrated and validated a novel approach for systematic and comprehensive comparisons of vision across species.
Weaknesses:
The weaknesses of the study arise in part because of the breadth of the work. In cases where there are divergences between the two species, it would be helpful to know what might account for such divergence, to have more depth. Is it really a species difference, or could there be a different account? For example, does the difference in mirror image confusion arise because the stimuli were objects that would have been highly familiar to the humans but not the macaques? Further, the authors often used small sets of stimuli (e.g. 8 objects only in the test of object similarity; a small set of highly specific occlusion stimuli), and how well the findings will generalize beyond those stimuli is unclear.
The authors discuss the implications of training macaques to perform specific tasks on specific stimuli in comparing across species. While I agree that extensive training in monkeys could change perception, it is important to also consider that humans have been extensively trained through the types of visual tasks we conduct throughout our lives, so I'm not sure it is universally true that the best comparison is between humans and untrained monkeys. But this just consideration just highlights the general problem of comparing across species.
-
Reviewer #2 (Public review):
Summary:
The macaque monkey is often considered as the animal model of choice to study the neural correlates of visual perception, due to the close similarities to humans in terms of anatomy, physiology and behaviour (Van Essen and Dierker, 2007; DiCarlo et al., 2012; Roelfsema and Treue, 2014; Picaud et al., 2019; Van Essen et al., 2019; Hesse and Tsao, 2020). Quite some studies have been performed to compare the behaviour of macaque monkeys and humans on visual perception tasks. However, it remains difficult to compare the results of these studies as the methods that are used differ significantly between these studies. Furthermore, behavioural studies of macaque monkeys often involve extensive training as the tasks were relatively hard, making it difficult to compare the results with humans, who generally …
Reviewer #2 (Public review):
Summary:
The macaque monkey is often considered as the animal model of choice to study the neural correlates of visual perception, due to the close similarities to humans in terms of anatomy, physiology and behaviour (Van Essen and Dierker, 2007; DiCarlo et al., 2012; Roelfsema and Treue, 2014; Picaud et al., 2019; Van Essen et al., 2019; Hesse and Tsao, 2020). Quite some studies have been performed to compare the behaviour of macaque monkeys and humans on visual perception tasks. However, it remains difficult to compare the results of these studies as the methods that are used differ significantly between these studies. Furthermore, behavioural studies of macaque monkeys often involve extensive training as the tasks were relatively hard, making it difficult to compare the results with humans, who generally require very little training. The authors present a set of experiments to compare visual perception between macaque monkeys and humans, using the exact same behavioral task that is easy to learn and therefore requires very little training. As expected, they overall find that the two species behave similarly. However, they find a number of interesting exceptions.
Strengths:
A major strength of the current study is the relatively large number of tasks that were tested in the same subjects. This is made possible by using the oddball visual search task, which macaque monkeys can learn very quickly. This means that few trials are sufficient to obtain a significant difference between conditions, minimizing learning effects. Although this type of task has been used in previous studies (Sablé-Meyer et al., 2021), the current manuscript makes better use of the advantages and explains them more explicitly.
In addition, the study finds a number of interesting differences between macaque monkeys and humans. In particular, while humans can dissociate horizontally mirrored images better than vertically mirrored images, monkeys show no difference between these two conditions (Experiment 4). Also, while humans dissociate images better based on the global shape of a stimulus, monkeys dissociate images better based on local elements of a stimulus (Experiments 5 and 6). Although these findings are largely a replication of previous results, they have not yet been studied together with other tasks within the same individual subjects, and the low number of trials avoids any learning effects.
Weaknesses:
A weakness of the study is that while the objects that were used can be considered to be familiar to humans, they are not familiar to macaque monkeys.
In Experiment 4, humans can be expected to have 3D representations of familiar objects such as a Roman helmet or an office chair. Humans can therefore be expected to have view-invariant representations of these objects, predominantly for rotations around the vertical axis of the object (as movements are most common in the horizontal plane). This can explain why only humans confuse objects more often when mirrored vertically than when mirrored horizontally.
Similarly, in Experiment 5, humans can be expected to be familiar with abstract geometric shapes such as squares and circles, while monkeys likely are not. This could explain why monkeys find it hard to recognize these geometric shapes in the global shape of the stimuli, even when thin grey lines are drawn to connect the local elements that constitute the global shape (Experiment 6). Instead, the combination of local shapes can be expected to form a texture that might be more easily recognized by the monkeys.
More generally, as proposed by Fagot et al, it might well be that monkeys tend to conceive stimuli as a combination of low-level visual features, instead of as references to objects in the outside world, as humans have learned to do (Fagot et al., 2010). This line of critique would be relevant to take into account.
Another weakness could be that only three monkeys are tested, while 24 human subjects are tested. According to some theoretical work, a finding in 3 animals is not sufficient to make a claim about an animal species (Fries and Maris, 2022). However, it seems that the results are largely consistent between the different monkeys. Moreover, the results generally agree with the results from previous literature.
The conclusions by the authors are therefore largely supported by the results. Some results could be strengthened by additional experiments, or at least a more extensive discussion of the potential weaknesses.
The potential impact of the paper is significant, as a start of a comprehensive comparison of visual perception between humans and macaque monkeys, which can inspire other labs to contribute to. This comparison can also be extended to other animal species (e.g. crows and rodents), as well as to different types of artificial neural networks (Leibo et al., 2018). -
Reviewer #3 (Public review):
Summary:
In this study, Cherian and colleagues compare visual perception between humans and monkeys using a common oddball visual search task across a battery of perceptual phenomena. By keeping the task constant and varying only the stimulus sets, the authors aim to isolate perceptual similarities and differences between species. Across six experiments, they report that monkeys and humans share similarities in coarse object representations, Weber's law, and amodal completion, but differ in mirror confusion and global/local processing.
Strengths:
A major strength of the study is the unified experimental framework. The authors designed the experiments such that the task procedures are identical across conditions and species, differing only in the images shown. This is a significant methodological advantage, …
Reviewer #3 (Public review):
Summary:
In this study, Cherian and colleagues compare visual perception between humans and monkeys using a common oddball visual search task across a battery of perceptual phenomena. By keeping the task constant and varying only the stimulus sets, the authors aim to isolate perceptual similarities and differences between species. Across six experiments, they report that monkeys and humans share similarities in coarse object representations, Weber's law, and amodal completion, but differ in mirror confusion and global/local processing.
Strengths:
A major strength of the study is the unified experimental framework. The authors designed the experiments such that the task procedures are identical across conditions and species, differing only in the images shown. This is a significant methodological advantage, as it minimizes task-related confounds that often complicate cross-species and cross-experiment comparisons. As a result, observed similarities and differences can be more directly attributed to perceptual processes rather than differences in training or task demands. This allows for a more comprehensive evaluation of visual perception than is typical in the literature, where individual studies often focus on a single effect with specialized training. The data are carefully collected, and the analyses are systematic and appropriate for the questions posed.
Weaknesses:
Despite its strengths, the study is largely descriptive and provides limited mechanistic or theoretical explanation for the observed similarities and differences. While the authors document several convergences and divergences between humans and monkeys, there is relatively little discussion of why these patterns arise or how they relate to existing theories of visual processing. As a result, it is difficult to assess the broader implications or generalizability of the findings beyond the specific task and stimuli used.
Relatedly, the rationale for selecting the particular set of perceptual phenomena is not fully developed. Some tasks appear motivated by prior work comparing humans and deep neural networks, but it is unclear whether this set constitutes a representative or theoretically grounded sampling of visual perception. Without a clearer justification, it is difficult to interpret the absence or presence of specific effects (e.g., mirror confusion or global advantage) as reflecting fundamental species similarities/differences.