The DOI matches its Crossref record; 83 of the 95 references it cites were found in a registry or an index; 11 of the 31 sources its claims cite were checked for support.
Design and evaluation of a global workspace agent embodied in a realistic multimodal environment
Open accessAbstract
As the apparent intelligence of artificial neural networks (ANNs) advances, they are increasingly likened to the functional networks and information processing capabilities of the human brain. Such comparisons have typically focused on particular modalities, such as vision or language. The next frontier is to use the latest advances in ANNs to design and investigate scalable models of higher-level cognitive processes, such as conscious information access, which have historically lacked concrete and specific hypotheses for scientific evaluation. In this work, we propose and then empirically assess an embodied agent with a structure based on global workspace theory (GWT) as specified in the recently proposed “indicator properties” of consciousness. In contrast to prior works on GWT which utilized single modalities, our agent is trained to navigate 3D environments based on realistic audiovisual inputs. We find that the global workspace architecture performs better and more robustly at smaller working memory sizes, as compared to a standard recurrent architecture. Beyond performance, we perform a series of analyses on the learned representations of our architecture and share findings that point to task complexity and regularization being essential for feature learning and the development of meaningful attentional patterns within the workspace.
Overview
CoChat Summary
TL;DR
The authors design an embodied global-workspace agent satisfying GWT indicator properties and train it on realistic audiovisual 3D navigation. It outperforms a GRU baseline mainly at smaller working-memory sizes, while larger workspaces yield little benefit and attention patterns become less dynamic.
Key findings
- The global workspace agent achieved ~80% success across all working-memory sizes apart from size 256, where some runs performed poorly and were considered possible outliers; the baseline GRU achieved only over 60% across all memory sizes and became competitive only at higher sizes. For reference, a random policy achieved 0% and the original SAVi RL agent reached 25% on average.
- Ablating working-memory size from 32 to 512 showed negligible performance differences for the global workspace agent, suggesting the task’s short-term memory demand is small and that SAVi is not an adequate benchmark for working-memory capacity; larger agents did not perform significantly better.
- Attention analyses showed all agents primarily used cross-attention across input modalities; smaller, more bottlenecked agents developed more mixed attention patterns integrating modalities over time, whereas larger workspaces produced more binary attention that saturated early, with working memory and null keys largely unattended.
- Probing found acoustic features encoded target object category and visual features encoded scene; target category from visual features ranged 14–37% versus chance 4%, scene from audio ranged 6–14% versus chance 1.79%, and after acoustic cue cutoff target-category probing accuracy dropped abruptly for both variants, though for memory sizes ≤128 global workspace agents had higher average probing accuracy than GRU counterparts. Broadcast weight analysis indicated previous working memory was prioritized over current sensory inputs in both audio and vision encoders.
Methods
- Architecture: an embodied agent with independent visual and acoustic CNN+GRU encoders, a recurrent working memory/global workspace, cross-attention bottleneck with a null input, and global broadcast from previous working memory to encoders; the baseline GRU ablates global broadcast and cross-attention while retaining independent encoders and recurrence.
- Task and data: SoundSpaces/SAVi semantic audiovisual navigation in 3D scanned real environments, with 21 target object categories and 56 training scenes; observations were 128×128×3 RGB images and 65×25×2 spectrograms, with four discrete actions (forward, turn left, turn right, stop).
- Training and analysis: behavioral cloning on 500,000 oracle tuples, batch size 10 contiguous trajectories of length 150, entropy coefficient 0.2, Bayesian hyperparameter optimization, final five seeds for 50,000 iterations across memory sizes 32, 64, 128, 256, and 512; each run took ~5 days on an NVIDIA RTX 3090, requiring more than 8,616 GPU hours total. Evaluation used interquartile mean ±95% CI over the last 50 evaluations, probing MLPs, attention weights, and normalized input-weight magnitudes.
Limitations
- Stated: SAVi is not an adequate benchmark for testing working-memory capacity; a more sophisticated task or environment is needed; developing realistic simulator environments is labor-intensive and beyond the scope of this work. Performance sometimes declined after the halfway point, which may be an artifact of hyperparameter optimization runs stopping early to reduce computational requirements by ~4,000 GPU hours.
- Stated: neither the conventional GRU nor the proposed attention mechanism preserved high enough target-category probing accuracy after sound cutoff, contrasting with the hypothesized ability of the global workspace to sustain a representation over time; no consistent attention patterns appeared across runs, and all agents achieved similar task performance despite different attention.
- Evident: findings come from one audiovisual navigation task using imitation learning from an oracle rather than online RL; the work evaluates indicator properties, not consciousness itself; and the baseline retains some GWT/RPT properties (GWT-1, RPT-1/2), so the ablation is partial.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures









Verification
Claims the paper makes
7 Valid · 4 Not valid · 12 Not checked · 63 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“However, prior work has been restricted to unimodal input data, namely either visual or textual information (Goyal et al., 2021; Juliani et al., 2022b).”
§ Design and evaluation of a global workspace agent embodied in a realistic multimodal environment > 2 Materials and methods > 2.1 Functional theories of consciousness
What the sources say
- Disputedcites Goyal et al., 2021 sourceExperimental — accuracy not yet measured
The Experiments section (p. 8) describes a shared-workspace task whose input combines an image with a question. It says the model concatenates image patches with a representation of the question, contrary to the claim that this prior work was restricted to visual-only or textual-only input. (LLM, conf=0.90)
Checked by a reasoning model — may be wrong.
- Not matchedcites Juliani et al., 2022b
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“While Transformer-based architectures have meteorically risen in popularity and become widely used across numerous academic and industrial applications, they lack the overall structure of a system with a global workspace.”
§ Design and evaluation of a global workspace agent embodied in a realistic multimodal environment > 2 Materials and methods > 2.2 Implementations of global workspace theory
What the sources say
- Could not verifycites Vaswani et al., 2017 sourceExperimental — accuracy not yet measured
The Model Architecture section (p. 2) describes the Transformer's encoder–decoder structure, and §3.2.3 (p. 5) describes how its attention layers operate. Neither assesses that structure against a global workspace; a search for “global workspace” found no such discussion in our copy. (LLM, conf=0.76)
- Not checked — no full text of the sourcecites Butlin et al., 2023 doi:10.48550/arxiv:2308.08708
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Despite these additions, the agent was only able to achieve an average success rate of 25%.”
§ Design and evaluation of a global workspace agent embodied in a realistic multimodal environment > 2 Materials and methods > 2.7 Agent training
What the sources say
- Partly supportedcites Chen et al., 2021 sourceExperimental — accuracy not yet measured
Our copy reports a success rate of ~25% for SAVi, but only on the unheard-sounds split: Section 7.2 (Implementation Details and Analysis of the Baselines) states 'our SAVi model 24.8% success rate and 17.2% SPL' on the unheard sound test split, and Table 1 (Sec. 4) lists 24.8 unheard / 33.9 heard. The claim's characterization as an 'average success rate of 25%' is not supported — the paper nowhere averages across settings, and averaging its two reported success rates would give ~29%, not 25%. So the magnitude traces to the source but the 'average' framing and the implied overall performance figure do not. (LLM, conf=0.72)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“each of these previous studies have lacked evaluation of the model while embodied within a multimodal environment.”
§ Design and evaluation of a global workspace agent embodied in a realistic multimodal environment > 1 Introduction
What the sources say
- Supportedcites Goyal et al., 2021 source
The Experiments section (pp. 6, 8–9) describes evaluations on benchmark tasks, including image-and-question reasoning, video object tracking, physical prediction, and Atari games. Although some tasks involve more than one input type, the paper does not describe evaluating its model while embodied in a multimodal environment. (LLM, conf=0.84)
- Not matchedcites Juliani et al., 2022b
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not checked — no full text of the sourcecites Butlin et al., 2023 doi:10.48550/arxiv:2308.08708
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The SAVi task (Chen et al., 2021) features two semantic concepts that can be queried from the features learned by the agent.”
§ Design and evaluation of a global workspace agent embodied in a realistic multimodal environment > 2 Materials and methods > 2.8 Agent analysis
What the sources say
- Partly supportedcites Chen et al., 2021 sourceExperimental — accuracy not yet measured
Section 4.2 (page unavailable) says SAVi’s learned goal descriptor predicts two properties from acoustic observations: the target’s location and object category. It explicitly calls location spatial and object category semantic, so the paper does not support describing both as semantic concepts. (LLM, conf=0.86)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
37 more claims · 1 with their cited sources checked
Create a free account to see all 86 claims and to verify the 3 references still open.
- It may state two separate claims as one. 26
- Probably not a substantial claim. 23
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“Researchers aiming to i mprove AI capabilities have proposed -and in some cases built -systems that intentionally reproduce c omputational features associated with human consciousness [ 4,5].”
From Identifying indicators of consciousness in AI systems (2026) · cites this paper as [5]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Design and evaluation of a global workspace agent embodied in a realistic multimodal environment > 2 Materials and methods > 2.2 Implementations of global workspace theory
“the Perceiver architecture notably lacks global broadcasting to the independent, pre-processing modules, thus falling short of satisfying GWT-2.”
§ Design and evaluation of a global workspace agent embodied in a realistic multimodal environment > 2 Materials and methods > 2.1 Functional theories of consciousness
“| GWT-3 | Global broadcast makes information in the workspace available to all modules |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Design and evaluation of a global workspace agent embodied in a realistic multimodal environment > 2 Materials and methods > 2.7 Agent training
“We augment this with an entropy regularization term, scaled by a coefficient η = 0.2, to mitigate overfitting of the agent's policy”
§ Design and evaluation of a global workspace agent embodied in a realistic multimodal environment > 2 Materials and methods > 2.7 Agent training
“| Entropy coefficient | 0.497 | 0.066 | 0.221 | 0.055 | 0.363 | 0.420 | 0.223 | 0.396 | 0.200 | 0.331 |”
Citations
Cites
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318 · cited 3×
“Similarly, biological research on other aspects of cognition, such as memorization, has benefited from comparative studies between populations of neurons and their artificial analogs (Bedia et al., 2007; Li and Fan, 2019), such as recurrent neural networks (RNNs), which are a prominent class of architecture used to proc…”
BackgroundAdam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946
Exploiting Generative AI to Scale up Intelligent Tutoring Systems
Jakubův, J., Chvalovský, K., Goertzel, Z., Kaliszyk, C., Olšák, M., Piotrowski, B., Schulz, S., Suda, M., & Urban, J. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.itp.2023.19 · Cited by 76,658
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380 · cited 1×
“…(CNNs; LeCun and Bengio, 1995; Krizhevsky et al., 2012; Simonyan and Zisserman, 2014) in the ImageNet visual object recognition competition (Russakovsky et al., 2015) spurred comparisons between trained CNNs and areas of the brain related to visual processing (Afraz et al., 2014; Seijdel…”
MethodNeural Machine Translation by Jointly Learning to Align and Translate
Bahdanau, D., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1409.0473 · Cited by 16,213
Parallel Distributed Processing: Explorations in the Microstructure of Cognition: Foundations
Rumelhart, D. E., McClelland, J. L., & AU (1986) · doi:10.7551/mitpress/5236.001.0001 · Cited by 15,686
Proceedings of the 24th international conference on Machine learning
(2007) · doi:10.1145/1273496 · Cited by 11,630
Research works printed but not linked: 3
- Baker, 2019, Emergent Tool use from Multi-Agent Interaction
- Goodfellow, 2016, Deep Learning
- 2018, OpenAI Five
Cited by
Identifying indicators of consciousness in AI systems
Butlin, P., Long, R., Bayne, T., Bengio, Y., Birch, J., Chalmers, D., Constant, A., Deane, G., Elmoznino, E., Fleming, S. M., et al. (2026) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2025.10.011 · Cited by 56
Self-citationNakanishi, J., Baba, J., Yoshikawa, Y., Kamide, H., & Ishiguro, H. (2025) · Frontiers in Robotics and AI · doi:10.3389/frobt.2025.1607190 · Cited by 3
“…Attention: It is thought that broadcasting allows each specialized module to concentrate its resources on information that is deemed to be extremely important according to the current goals and environmental conditions, thereby improving the efficiency and accuracy of task execution [16], [20].”
A Review of the Current State of Multimodal Data Annotation in Traditional Chinese Medicine
Yutong, D., Yuan, Y., & Hongwei, Z. (2025) · American Journal of Health Research · doi:10.11648/j.ajhr.20251306.17
Goyal, K., Dixit, R. S., Choudhary, S. L., & Goyal, P. (2026) · Journal of Business Analytics · doi:10.1080/2573234x.2026.2617555
Metadata
- Authors
- Dossa, R. F. J., Arulkumaran, K., Juliani, A., Sasai, S., & Kanai, R.
- Affiliations
- Araya Inc. (Japan) (JP)
- Microsoft Research New York City (United States) (US)
- Corresponding author
- Dossa, Rousslan Fernand Julien
- Venue
- Frontiers in Computational Neuroscience
- Publisher
- Frontiers Media SA
- Published
- 2024-06-14
- Volume
- 18
- Pages
- 1352685-1352685
- ISSN
- 1662-5188 (ISSN-L)
- PMC
- PMC11211627
- OpenAlex
- W4399687862
- Cited by
- 5
- References
- 95
- Citations by year
- 2026: 2 · 2025: 3
- FWCI
- 0.73
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Frontiers in Computational Neuroscience (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Psychiatry, Mental Health, Neuroscience
- Social Sciences › Arts and Humanities › Philosophy › Philosophy and Theoretical Science
- Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
- Keywords
- global workspace theory, consciousness, artificial neural networks, working memory, multimodal environments, recurrent neural networks, attention, representation learning, regularization, task complexity
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-13, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-13 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Design and evaluation of a global workspace agent embodied in a realistic multimodal environment. https://app.cochat.ai/papers/10.3389/fncom.2024.1352685
- The paper
- Dossa, R. F. J., Arulkumaran, K., Juliani, A., Sasai, S., & Kanai, R. (2024). Design and evaluation of a global workspace agent embodied in a realistic multimodal environment. Frontiers in Computational Neuroscience. https://doi.org/10.3389/fncom.2024.1352685
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.