The DOI matches its Crossref record; 19 of the 20 references it cites were found in a registry or an index; 14 of the 20 sources its claims cite were checked for support.
Compression, The Fermi Paradox and Artificial Super-Intelligence
Open accessAbstract
The following briefly discusses possible difficulties in communication with and control of an AGI (artificial general intelligence), building upon an explanation of The Fermi Paradox and preceding work on symbol emergence and artificial general intelligence. The latter suggests that to infer what someone means, an agent constructs a rationale for the observed behaviour of others. Communication then requires two agents labour under similar compulsions and have similar experiences (construct similar solutions to similar tasks). Any non-human intelligence may construct solutions such that any rationale for their behaviour (and thus the meaning of their signals) is outside the scope of what a human is inclined to notice or comprehend. Further, the more compressed a signal, the closer it will appear to random noise. Another intelligence may possess the ability to compress information to the extent that, to us, their signals would appear indistinguishable from noise (an explanation for The Fermi Paradox). To facilitate predictive accuracy an AGI would tend to more compressed representations of the world, making any rationale for their behaviour more difficult to comprehend for the same reason. Communication with and control of an AGI may subsequently necessitate not only human-like compulsions and experiences, but imposed cognitive impairment.
Overview
CoChat Summary
## TL;DR
This short paper argues that highly intelligent systems may be hard to understand because they can compress information and construct task solutions that humans would mistake for noise or fail to recognize as meaningful. It extends this idea from the Fermi paradox to AGI, suggesting that communication and control of AGI may require matching human-like experiences or even imposing cognitive impairment.
## Key findings
- The paper argues that communication between agents requires “approximately the same solution” to a task, including similar stimuli and similar significance assigned to those stimuli.
- It states that highly compressed signals can look like random noise to observers lacking the right decoder, and applies this to both extraterrestrial intelligence and AGI.
- It claims that a super-intelligent AGI would likely construct an “intensional solution” that is among the most compressed, making its rationale difficult or impossible for humans to interpret.
- It concludes that interpretability and control of AGI may require either human-like sensors, actuators, compulsions, and environment, or “imposed cognitive impairment” to limit how the AGI constructs solutions.
## Methods
- Conceptual/theoretical argument rather than an empirical study.
- Draws on prior work on symbol emergence, embodied/enactive cognition, compression, Kolmogorov complexity, and the Fermi paradox; no new dataset or sample is reported.
- Uses qualitative reasoning and analogy to connect communication with nonhuman intelligence, signal compression, and AGI interpretability.
## Limitations
- Not stated in the paper.
- Evident from the text: the argument is speculative and philosophical, with no experiments, formal proof, or empirical validation presented.
- Evident from the text: claims about AGI, human cognition, and extraterrestrial intelligence are argued by analogy and cited theory, so their scope is not demonstrated beyond the conceptual framing.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
11 Valid · 6 Not valid · 1 Not checked · 24 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“The solutions they construct then facilitate encoding and decoding of signals interpretable by both agents [11].”
What the sources say
- Supportedcites [11] Bennett, M.T., Maruyama, Y.: Philosophical specification of empathetic ethical artificial intelligence. IEEE Trans. Cogn. Dev. Syst. (2021). https://doi.org/10.1109/TCDS.2021.3099945 source
Section V (p. 6) describes the proposed symbol system L as a means of encoding and decoding perceptual states for transmission between agents. Section IV (p. 5) adds that communication requires agents to share signs, referents, and interpretants, supporting the claim that both can interpret the signals. (LLM, conf=0.91)
plainstates a cause- 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
Every source the paper cites for this claim was checked and supports it.
In the paper
“A natural language model such as GPT-3 is trained only on text data [2]”
What the sources say
- Supportedcites [2] Floridi, L., Chiriatti, M.: GPT-3: its nature, scope, limits, and consequences. Minds Mach. 30(4), 681–694 (2020). https://doi.org/10.1007/s11023-020-09548-1 source
Section 2, “GPT-3,” says the language model is trained on an unlabelled dataset made up of texts. Section 3 likewise describes GPT-3 as trained on human texts, supporting the claim about GPT-3. (LLM, conf=0.91)
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
“Attempts to train models on multimodal sensorimotor data have yielded some success, with agents able to associate the sensory information of an object such as a cup with the signals that represent it [7,8,9].”
What the sources say
- Supportedcites [7] Taniguchi, T., et al.: Symbol emergence in cognitive developmental systems: a survey. IEEE Trans. Cogn. Dev. Syst. 11(4), 494–516 (2019) source
Section C reports that robots can form object categories from visual, auditory, and tactile information, then associate those categories with words or signs. It also uses a cup to illustrate cross-modal sensory inference. Together, these passages support the claim's general account of multimodal sensorimotor learning. (LLM, conf=0.90)
- Supportedcites [8] Taniguchi, T., et al.: Symbol emergence in robotics: a survey. Adv. Robot. 30(11–12), 706–728 (2016) source
Section 5.4 (p. 14) reports that robots formed object categories and learned related words from speech signals together with visual, auditory, and haptic sensory-motor information. This supports the claimed association between an object's sensory information and signals representing it; a cup is an illustrative example, not a reported test object. (LLM, conf=0.90)
- Not matchedcites [9] Ramesh, A. et al.: DALL-E: Creating Images From Text. OpenAI (2021)
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“it lacks any of the other sensorimotor information we might typically associate with words.”
What the sources say
- Supportedcites [2] Floridi, L., Chiriatti, M.: GPT-3: its nature, scope, limits, and consequences. Minds Mach. 30(4), 681–694 (2020). https://doi.org/10.1007/s11023-020-09548-1 source
Section 2, “GPT-3” (page unavailable), describes training on a dataset of texts and generating word sequences statistically. Section 3, “Three Tests” (page unavailable), says GPT-3 has only a syntactic capacity to associate words, not an understanding of their meanings or contexts. Together, these support the paraphrase that it lacks sensorimotor information associated with words. (LLM, conf=0.84)
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
Every source the paper cites for this claim was checked and supports it.
In the paper
“to infer what someone means, an agent constructs a rationale for the observed behaviour of others.”
What the sources say
- Supportedcites [10] Bennett, M.T.: Symbol emergence and the solutions to any task. In: Goertzel, B. (ed.) AGI 2021. LNAI, vol. 13154, pp. 30–40 (2022). https://doi.org/10.1007/978-3-030-93758-4_4 source
In “Natural Language” (p. 8), the paper says an agent observes another agent’s response and constructs a rationale for that behaviour; it then describes this as interpreting what a signal means. The Introduction (pp. 1–2) likewise describes inferring intent by constructing a rationale for observed behaviour. (LLM, conf=0.94)
plainstates a cause- 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
13 more claims · 2 with their cited sources checked
Create a free account to see all 42 claims and every source verdict.
- It may state two separate claims as one. 7
- Probably not a substantial claim. 18
Citations
Cites
Information theory, inference, and learning algorithms
(2004) · Choice Reviews Online · doi:10.5860/choice.41-5949 · Cited by 6,362
Information Theory, Inference, and Learning Algorithms
(2004) · Kybernetes · doi:10.1108/03684920410534506 · Cited by 4,937
GPT-3: Its Nature, Scope, Limits, and Consequences
Floridi, L., & Chiriatti, M. (2020) · Minds and Machines · doi:10.1007/s11023-020-09548-1 · Cited by 2,519 · cited 1×
“A natural language model such as GPT-3 is trained only on text data [2], implicitly endorsing the idea that meaning is just relations between words.”
MethodA formal theory of inductive inference. Part II
Solomonoff, R. J. (1964) · Information and Control · doi:10.1016/s0019-9958(64)90131-7 · Cited by 1,758 · cited 1×
“The ability to generalise is closely related to compression [14,15], with more compressed representations yielding better accuracy [13] for the same reason that there is only one straight line interpolating any two points, but infinitely many polynomials of higher degree.”
BackgroundA formal theory of inductive inference. Part I
Solomonoff, R. J. (1964) · Information and Control · doi:10.1016/s0019-9958(64)90223-2 · Cited by 1,192
Universal Artificial Intellegence
Hutter, M. (2005) · Texts in Theoretical Computer Science An EATCS Series · doi:10.1007/b138233 · Cited by 523
Kolmogorov, A. N. (1998) · Theoretical Computer Science · doi:10.1016/s0304-3975(98)00075-9 · Cited by 362
Schlicht, T. (2009) · PhilPapers (PhilPapers Foundation) · Cited by 305
Neural Suppression of Irrelevant Information Underlies Optimal Working Memory Performance
Zanto, T. P., & Gazzaley, A. (2009) · The Journal of Neuroscience · doi:10.1523/jneurosci.4621-08.2009 · Cited by 302 · cited 1×
“Any information not relevant to satisfying compulsions is not only meaningless but may be entirely ignored, which is consistent with observations of human behaviour [6,5] (i.e. one may be unable to perceive something in the stream of sensorimotor stimuli because it is filtered out).”
BackgroundSymbol emergence in robotics: a survey
Taniguchi, T., Nagai, T., Nakamura, T., Iwahashi, N., Ogata, T., & Asoh, H. (2016) · Advanced Robotics · doi:10.1080/01691864.2016.1164622 · Cited by 150 · cited 1×
“Attempts to train models on multimodal sensorimotor data have yielded some success, with agents able to associate the sensory information of an object such as a cup with the signals that represent it [7,8,9].”
Background
Research works printed but not linked: 1
- Hutter, M.: Universal Artificial Intelligence: Sequential Decisions based on Algorithmic Probability. Springer, Heidelberg (2005)
Cited by
The Optimal Choice of Hypothesis Is the Weakest, Not the Shortest
Bennett, M. T. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-33469-6_5 · Cited by 33
Self-citationOn the Computation of Meaning, Language Models and Incomprehensible Horrors
Bennett, M. T. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-33469-6_4 · Cited by 19
“However, weakness is well defined in a more recent formalism of artificial general intelligence (AGI) [11, 1] and enactive cognition, so we will instead reformulate the theory using that formalism, extending it to account for meaningful communication.”
BackgroundMethodInfluentialSelf-citationComputational Dualism and Objective Superintelligence
Bennett, M. T. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-65572-2_3 · Cited by 32
“- Rodney Brooks [28] The only aspects of the environment that we might actually need model are those necessary to satisfy goals [5].”
BackgroundSelf-citationOn the Computation of Meaning, Language Models and Incomprehensible HorrorsPreprint
Bennett, M. T. (2023) · doi:10.36227/techrxiv.22216753.v1 · Cited by 11
Self-citationThe Optimal Choice of Hypothesis Is the Weakest, Not the ShortestPreprint
Bennett, M. T. (2023) · doi:10.36227/techrxiv.21965675 · Cited by 10
Self-citation
Metadata
- Authors
- Bennett, M. T.
- Affiliations
- Australian National University (AU)
- Corresponding author
- Bennett, Michael Timothy
- Venue
- Lecture notes in computer science
- Publisher
- Springer International Publishing
- Published
- 2022
- Pages
- 41-44
- ISSN
- 0302-9743 (ISSN-L)
- 1611-3349
- OpenAlex
- W3204744729
- Semantic Scholar
- e6fd810701f94bf4915b6361e6030ada731b8ea2
- Cited by
- 31
- References
- 20
- Citations by year
- 2025: 2 · 2024: 8 · 2023: 17 · 2022: 4
- FWCI
- 34.08
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- www.springer.com/tdm (tdm)
- www.springer.com/tdm (vor)
- Locations
- Lecture notes in computer science (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Physics and Astronomy › Astronomy and Astrophysics › Space Science and Extraterrestrial Life
- Physical Sciences › Computer Science › Artificial Intelligence › Psychiatry, Mental Health, Neuroscience
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Keywords
- Fermi paradox, artificial general intelligence, artificial intelligence control, signal compression, symbol emergence, meaning inference, prediction accuracy, cognitive impairment, human-AI interaction
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 2026-10-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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: Compression, The Fermi Paradox and Artificial Super-Intelligence. https://app.cochat.ai/papers/10.1007/978-3-030-93758-4_5
- The paper
- Bennett, M. T. (2022). Compression, The Fermi Paradox and Artificial Super-Intelligence. Lecture notes in computer science. https://doi.org/10.1007/978-3-030-93758-4_5
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.