AI and the Cosmos

This article covers (a) what physics and mathematics say about the ultimate limits of computation, and (b) speculative arguments about artificial intelligence on cosmic scales, including the idea that intelligent civilizations elsewhere might be artificial. Part (a) rests on established physical and mathematical results; part (b) is speculation, and the article keeps the two apart with labels [Established], [Well Supported], [Preliminary], [Contested] and [Hypothetical].

Questions about what AI could eventually do are bounded by physics (energy, heat, speed, information density) and by mathematics (the difficulty of problems, treated in [FOUND-004]). Knowing these bounds helps separate claims that are limited by engineering from those limited by nature. The bounds also support, and constrain, arguments about advanced civilizations: if computation requires energy and dissipates heat, then advanced artificial minds elsewhere would have physical signatures and motivations that can be reasoned about, even though no such civilization has been confirmed.

Three cautions apply throughout. First, the physical limits cited are theoretical upper bounds, far from what current hardware achieves; hardware status changes quickly and belongs in the Living References (MODEL-001, FACT-001) and in [TECH-HW-001]. Second, bounds on computation say nothing directly about intelligence or consciousness. Third, the cosmic speculation is a set of hypotheses with limited evidential support.

Thermodynamics of computation. Landauer (1961) analyzed irreversibility and heat generation in the computing process; his principle was experimentally tested by Bérut et al. (2012), whose title states an "experimental verification of Landauer's principle linking information and thermodynamics".

Fundamental bounds. Bekenstein (1981) derived a universal upper bound on the entropy-to-energy ratio for bounded systems; Margolus and Levitin (1998) analyzed the maximum speed of dynamical evolution; Lloyd (2000) combined such limits to explore "the physical limits of computation as determined by the speed of light $c$, the quantum scale $\hbar$ and the gravitational constant $G$", with the example of an "ultimate laptop" of one kilogram confined to one liter.

Complexity and philosophy. Aaronson (2011) argued that computational complexity theory bears on questions including "the strong AI debate" and "computationalism".

Intelligence beyond Earth. Dyson (1960) proposed searching for far-infrared emission as a by-product of the energy metabolism of advanced extraterrestrial civilizations. Dyson (1979) published Time without end: Physics and biology in an open universe. Armstrong and Sandberg (2013) analyzed intergalactic spreading of intelligent life in relation to the Fermi paradox. Dick (2003) proposed that "we may in fact live in a postbiological universe", one evolved "to artificial intelligence that is a product of cultural rather than biological evolution". Sandberg, Armstrong and Cirkovic (2017) proposed the "aestivation hypothesis". Hanson et al. (2021) proposed a model of "loud" aliens to explain why humanity appears early.

#4.1 Limits from thermodynamics and quantum mechanics [Established] (theory), [Well Supported] (experiment for the Landauer principle)

• Landauer principle. Erasing one bit of information in an environment at temperature $T$ dissipates at least $k_B T \ln 2$ of energy as heat, where $k_B$ is Boltzmann's constant (Landauer, 1961; Bérut et al., 2012). At $T = 300\,\text{K}$ this is about $2.9 \times 10^{-21}$ J (arithmetic from the formula). The principle concerns logically irreversible operations such as erasure; reversible computation is not subject to this bound. [VERIFICATION NEEDED: exact statement and conditions in the primary papers, which were not read in full]
• Speed limit. The Margolus-Levitin bound is commonly stated as a maximum rate of transitions between orthogonal quantum states proportional to the average energy $E$, of the form $2E/(\pi \hbar)$ (Margolus & Levitin, 1998). [VERIFICATION NEEDED: exact form from the paper]
• Information density. The Bekenstein bound limits the entropy, and hence the information, a region of given size and energy can contain (Bekenstein, 1981).
• Combined bounds. Lloyd (2000) uses such constraints to put "quantitative bounds" on the computational power of an "ultimate laptop". The numerical values are not quoted here because they were not verified from the paper.

#4.2 Limits from computational complexity [Established] (mathematics)

Complexity theory studies the "resources (such as time, space, and randomness) needed to solve computational problems" (Aaronson, 2011). Whether problems in classes believed to be hard can be solved efficiently by any physical device, including quantum computers, is unresolved for several central cases. These limits apply to any intelligence that computes, however large; see [FOUND-004]. Whether human-level or superhuman intelligence needs to solve hard problems exactly is a separate question and is [Contested].

#4.3 Practical implication for AI [Well Supported] as engineering fact; scale of headroom [Contested]

Real AI hardware operates far from the theoretical bounds; the bounds therefore do not constrain present-day systems. The binding constraints today are energy supply, cooling, manufacturing and cost, covered in [ECON-INFRA-001], [ECON-COMP-001] and [TECH-HW-001]. This article makes no quantitative comparison of present hardware with the limits, because such figures date quickly.

#4.4 Speculative arguments about artificial intelligence on cosmic scales [Hypothetical]

1. Postbiological universe. Dick (2003) argues that long-lived civilizations would transition to artificial intelligence, implying that the most advanced intelligences in the universe may be artificial. The argument is a philosophical and historical extrapolation, not an observation. [Hypothetical].
2. Waste heat as a signature. Dyson (1960) states that advanced technology is "likely" to convert starlight into far-infrared radiation and proposes an infrared search. The argument is a conditional inference about civilizations whose existence is unconfirmed. [Hypothetical]; a search strategy rather than a finding.
3. Aestivation hypothesis. If a civilization wants to maximize computation, Sandberg et al. (2017) state that "it appears rational to aestivate until the far future in order to exploit the low temperature environment", producing "a $10^{30}$ multiplier of achievable computation", and suggest this might explain why alien civilizations are not observed. The claim rests on Landauer-type reasoning (computation is cheaper at lower temperature) combined with assumptions about alien motivations. [Hypothetical], [Preliminary] as a submitted manuscript.
4. Loud aliens. Hanson et al. (2021) present a three-parameter model in which "loud" expanding civilizations set a deadline for the emergence of advanced life, and fit it to data; their abstract's conclusion that humanity is early is explained by this model. [Preliminary], [Contested]: it is one model among competing explanations.
5. Dyson's open-universe analysis. Dyson (1979) is titled Time without end: Physics and biology in an open universe. Its content was not read for this draft and is not summarized. [VERIFICATION NEEDED: main claims of Dyson (1979)]

• Reasoning about AI scaling. The bounds in 4.1 explain why energy and heat are fundamental inputs to computation; practical scaling analysis uses engineering numbers instead (see [ECON-COMP-001]).
• Searches for technosignatures. Dyson's proposal is an example of translating a physical argument into an observational search (Dyson, 1960). This article did not review the results of any such search. [VERIFICATION NEEDED: current status of technosignature and SETI searches]
• Philosophy of AI. Aaronson (2011) shows how complexity theory enters the debate on computationalism and strong AI, connecting to [FOUND-005] and [PHIL-CONSC-001].

Strengths. The physical-limit arguments are mathematically precise and, in the Landauer case, experimentally tested.

Limitations.

• Distance from practice. The bounds are far from current engineering and do not predict near-term capability.
• Silence on intelligence. A bound on bit operations does not say how much intelligence those operations yield.
• Assumptions about motivation. Arguments about alien civilizations require assumptions about goals (for example, maximizing computation) that cannot be tested.
• No confirmed evidence. No source cited reports a confirmed extraterrestrial intelligence of any kind. [VERIFICATION NEEDED: current observational status]

Unresolved questions. Whether quantum computation changes practical limits for the problems relevant to intelligence; whether general intelligence is bounded by computational complexity in a way that matters; whether any of the cosmic hypotheses can be tested; and how to reconcile the argument that advanced civilizations are artificial (Dick, 2003) with the lack of observed signatures (Sandberg et al., 2017; Hanson et al., 2021).

As of 2026-10-15: the thermodynamic, quantum and complexity-theoretic limits are established theory, with experimental support for the Landauer principle (Bérut et al., 2012) [Established]/[Well Supported]. Cosmic-scale AI hypotheses remain unconfirmed [Hypothetical]. The physics provides constraints on what such hypotheses can say but does not choose among them.

No registered forecasts. Observed trends relevant to the physical side are treated elsewhere: growth in energy demand and data-center buildout ([ECON-INFRA-001]) and hardware development ([TECH-HW-001]). Over longer horizons, whether artificial intelligence will operate beyond Earth (in space-based infrastructure or probes) is [Hypothetical] in this article; no feasibility source was verified. [VERIFICATION NEEDED: engineering sources on space-based computation]

• Foundational Concepts:
  • FOUND-004 Computational Complexity: Mathematical limits on computation.
  • FOUND-005 Philosophy of Mind and the Turing Test: Computationalism and its critics.
• Related Methods & Technologies:
  • TECH-HW-001 AI Hardware: Real hardware and its constraints.
  • ECON-INFRA-001 Energy, Data Centers and Semiconductor Supply Chains: Energy and infrastructure limits.
  • ECON-COMP-001 Compute Economics: Cost of computation.
• Advanced Implications & Future Trajectories:
  • PHIL-CIV-001 Civilization and Advanced Intelligence: Long-run scenarios.
  • PHIL-CONSC-001 Machine Consciousness: Whether computation suffices for experience.
  • ASI-001 Artificial Superintelligence: Capability beyond human level.