Generalizing frameworks for sentience beyond natural species

What to take away

1. 1. Levin argues that Smith & Boyd's (1991) sentience criteria are irrelevant across the vast majority of possible agent-space because they presuppose phylogenetic provenance and neuroanatomical homology to mammalian brains.
2. 2. Crump et al.'s 8-criterion framework (nociception, sensory integration, integrated nociception, endogenous/exogenous analgesia, motivational trade-offs, flexible self-protection, associative learning, and analgesia preference) is endorsed as the right structural form of a sentience framework, but its scope is limited to natural biological species.
3. 3. Criterion #7 (associative learning) from Crump et al. is already satisfied by gene regulatory networks, as demonstrated by at least 4 independent studies (Biswas et al. 2021; Fernando et al. 2009; McGregor et al. 2012; Watson et al. 2010), meaning the neural boundary of the framework is empirically unjustified.
4. 4. In vitro neuronal cultures embodied in a simulated game-world (Kagan et al. 2021, bioRxiv) constitute a real existing 'Hybrot' substrate that meets proto-cognitive criteria yet falls outside every conventional sentience-assessment framework.
5. 5. Xenobots — scalable reconfigurable organisms with anatomies radically different from any natural life form (Kriegman et al. 2020, PNAS 117:1853; Blackiston et al. 2021, Sci Robot 6:eabf1571) — are offered as a concrete substrate for which no existing ethical framework provides adequate guidance.
6. 6. The open question Levin raises is whether the LaMDA large language model (Thopilian et al. 2022) is sentient: he argues that no existing framework, Turing-Test-based or anatomical, can settle the question, making it an urgent test case for any future invariant-based framework.
7. 7. Levin's proposed methodology for replicating a substrate-neutral sentience assessment is to identify functional invariants — such as 'competency in navigating arbitrary problem spaces' (Fields & Levin 2022, Entropy 24) — and re-derive each criterion by replacing substrate-specific terms (e.g., 'neuron') with the more general functional equivalent (e.g., 'electrically active cell').
8. 8. The Clark & Chalmers (1998) Extended Mind hypothesis is cited as already instantiated in cyborg implants and prosthetics (Orive et al. 2020; Pio-Lopez 2021), meaning the boundary of 'natural' biological sentience is already practically dissolved.
9. 9. Levin argues that human agency-detection is well-calibrated only for medium-sized objects moving at medium speeds in 3D space, making it structurally unreliable for recognizing intelligence in cells, organs, plants, or software — an intuition-failure that cannot be corrected by introspection alone.
10. 10. Levin predicts that the coming decades will introduce bioengineered and AI agents into homes and human bodies whose moral status cannot be evaluated by any existing criterion, framing the development of invariant-based sentience frameworks as a probable existential requirement for humankind.

Peer brief — for seminar discussion

Levin's 2022 commentary in Animal Sentience (32:15, DOI 10.51291/2377-7478.1733) uses Crump et al.'s 8-criterion framework for decapod sentience as a springboard to argue that the entire project of sentience-assessment must be radically generalized beyond natural biological species. The specific trigger is the observation that Crump et al.'s criterion #7 — associative learning — is already demonstrably satisfied by gene regulatory networks (citing at least Fernando et al. 2009; Watson et al. 2010; McGregor et al. 2012; Biswas et al. 2021), and that most remaining criteria apply to non-neural morphogenetic systems once 'neuron' is replaced with 'electrically active cell.' This is not merely a conceptual point: xenobots (Kriegman et al. 2020, PNAS 117:1853) and Hybrots — in vitro neuronal cultures embodied in robotic systems, including Kagan et al.'s (2021) game-world experiment — already occupy a region of possible-agent-space where no existing ethical framework applies. The load-bearing finding is negative: both the Smith & Boyd (1991) criteria and Turing-Test-style verbal report standards are shown to be insufficient across the vast majority of possible agents, because they presuppose phylogenetic provenance and mammalian neuroanatomy — features Levin calls 'frozen accidents' of Earth's single phylogenetic path. The method Levin introduces is invariant-based framework construction: deriving sentience criteria from deep functional invariants (such as competency in navigating arbitrary problem spaces, per Fields & Levin 2022, Entropy 24) rather than from substrate-specific heuristics. An alternative method he implicitly sets aside is behavioral/ethological criteria in the tradition of Smith & Boyd (1991), which he dismisses as applicable only to the narrow band of already-familiar natural organisms. The implication is prescriptive and urgent: with LaMDA (Thopilian et al. 2022) already generating public controversy and bioengineered chimeric organisms entering laboratories, humanity will face legally and morally consequential decisions about novel agents before principled frameworks exist, constituting what Levin calls a probable existential requirement. The contestable element a critical reader would press hardest on is the paper's reliance on the functional pivot — replacing 'neuron' with 'electrically active cell' — as sufficient to extend sentience criteria to non-neural substrates. This move is asserted rather than demonstrated: Levin provides no positive evidence that non-neural electrically active cells in morphogenetic contexts satisfy the motivational, integrative, or hedonic dimensions of criteria like analgesia preference (#8), and critics in the basal cognition debate would argue that substrate-neutral functionalism risks inflating the reference class of potentially sentient entities to the point of practical uselessness. The paper also does not engage with the possibility that false positives — extending moral concern to non-sentient systems — carry their own ethical and resource costs, treating false-negative minimization as unambiguously dominant.

Claims (1)

• Verbal reports, homology, and phylogenetic provenance are insufficient to determine sentience in unconventional agents.

  Core claim: Turing test and brain homology fail for synthetic, AI, and radically non-human agents; new frameworks required.

Questions (1)

• What frameworks can identify sentience in radically unfamiliar agents without familiar touchstones of evolution or anatomy?

  Central question driving the paper; necessitated by imminent proliferation of cyborgs, synthetic organisms, and AI.

Similar preprints — Semantic Scholar