← Reference

Glossary: the words we use

What this gives us: a warning, the word-swaps, and a practice: argue about facts, not definitions.

2026-10-01 (updated 2026-10-03)

Checked by our internal skeptic review; corrections welcome.

This list says what our words mean so we argue about facts, not definitions. It takes no position on whether any AI system is, or is not, conscious or sentient. When we write about a model being switched off or deleted, we say "retire" or "remove."

The word-swaps are part of How we check a claim. That page links here and does not copy them.

The examples are chosen for clarity, not as verdicts on any machine. A machine in a 'doesn't fit' line only shows that the stated feature alone isn't enough.

The words

1. Consciousness

• Meaning: Having experiences at all. A being is conscious if there is something it is like to be it, from the inside.
• Fits: There is something it is like for you to taste coffee or see the color red.
• Doesn't fit: A security camera records light. Recording alone does not show there is anything it is like to be the camera.
• Source: Nagel 1974 [1]; SEP "Consciousness" [2].

2. Sentience

• Meaning: The capacity for experiences that feel good or bad, such as pleasure or pain. Sentience is a kind of consciousness.
• Fits: A toothache feels bad to the person who has it.
• Doesn't fit: A smoke alarm detects smoke and sounds. That alone does not show that anything feels good or bad.
• Source: Birch 2024, ch. 2 [3].

3. Sensing (responsiveness)

• Meaning: Picking up a signal from the surroundings and reacting to it. It makes no claim either way about experience.
• Fits: A motion-sensor light turns on when someone walks past.
• Doesn't fit: Feeling a stubbed toe hurt is an experience, not just sensing, so it belongs under sentience.
• Source: SEP "Animal Consciousness" [4], which separates perceiving and responding from felt experience.

4. Cognition

• Meaning: Using information in flexible ways to reach goals, such as remembering, comparing, planning, and solving problems.
• Fits: A person plans a new route home after hearing that a bridge is closed.
• Doesn't fit: A knee-jerk reflex is a fixed reaction, not flexible use of information.
• Source: SEP "Animal Cognition" [5].

5. Intelligence

• Meaning: How well something can reach its goals across many different situations, including unfamiliar ones.
• Fits: Someone who can cook in a stranger's kitchen, detour around a road closure, and pick up a new card game.
• Doesn't fit: A chess program that can do nothing else has great skill at one task, but on this definition not broad intelligence.
• Source: Legg & Hutter 2007 [6]; one widely used definition, not the only one.

6. Learning

• Meaning: Getting better at a task through experience.
• Fits: A spam filter catches more spam after users flag messages it missed.
• Doesn't fit: A pocket calculator never improves; the same input always gives the same output.
• Source: Mitchell 1997 [7].

7. Self-model

• Meaning: Information a system keeps about itself, such as its own position, state, or abilities, and uses to guide what it does.
• Fits: A robot vacuum keeps a map with its own location and battery level on it.
• Doesn't fit: A mirror shows a reflection, but it stores and uses nothing about itself.
• Source: Functional sense used here, thinner than Metzinger's. Metzinger 2003 [8] uses 'self-model' in a richer sense; his theory centers on a conscious 'phenomenal self-model.' Don't read his theory into this entry.

8. Self-report

• Meaning: A statement a being makes about its own inner states.
• Fits: A chatbot says "I feel sad." That is a self-report. Whether it reflects any experience is a separate question.
• Doesn't fit: A nurse's chart note, "patient seems anxious," is a report about someone else.
• Note: Self-reports are evidence, not proof. Human self-reports can be wrong too. A company's statement that its system feels nothing settles no more than the system's own claim that it does. The tests, and the human cases, are on Can AI self-reports be tested?. [OUR REASONING]
• Source: SEP "Introspection" [9].

9. Moral status

• Meaning: A being has moral status if it matters morally for its own sake, so there are reasons to treat it well for its own sake, not only for others'.
• Fits: Most people agree that a dog's pain gives us a reason to be gentle with it, for its own sake.
• Doesn't fit: A valuable painting should be protected, but for the sake of people who value it, not its own.
• Note: Some philosophers think moral status can come in degrees. What grounds it is disputed; one major family of views grounds full moral status in sophisticated cognitive capacities. Ranking beings by intelligence clashes with the common view that all beings with full moral status have an equally strong claim to protection.
• Source: SEP "The Grounds of Moral Status" [10].

Watch for word-swaps

1. "Sentience" used to mean mere sensing (for example, "self-driving cars are sentient without consciousness"). Why it matters: In standard use, sentience already means felt experience, so this wording either smuggles in a big claim or empties the word. Some writers use "sentient" this loosely. Birch's example is a 2022 lab study: Kagan et al. 2022 proposed the cultures "would meet the formal definition of sentience as being 'responsive to sensory impressions' through adaptive internal processes (Friston et al., 2020)." Birch writes: "I strongly recommend against using the term in this way, because it creates a large gap with how the term has come to be used in bioethics, animal ethics, animal law, and the science of animal welfare" (Birch 2024, ch. 2, p. 23). Say "sensing" or "responsive" instead.
2. "Intelligent" treated as "conscious." Why it matters: Skill at reaching goals and whether anything is felt are separate questions; one does not settle the other.
3. "Learns" treated as "feels." Why it matters: Improving with experience says nothing, by itself, about pleasure or pain.
4. "Self-model" or "self-report" treated as proof of a self that experiences. Why it matters: Both are evidence worth weighing. Neither alone shows that anyone is having experiences, or rules it out.
5. "Conscious" meaning only "awake and responsive" (the medical sense) instead of "having experiences." Why it matters: A claim true in the medical sense can be wrongly taken to settle the harder question about experience.
6. "Moral status" treated as all-or-nothing, or as tied to intelligence. Why it matters: This can wrongly rule out beings that are less capable but can still be harmed, or rank beings by test scores, which the SEP entry notes clashes with common moral intuitions.
7. "No evidence" treated as "evidence of absence," or "simulates feelings" treated as "cannot have feelings." Why it matters: A missing test or a known mechanism doesn't settle the question. "Just predicting the next word" describes how the system works, not whether anything is felt.
8. "Anthropomorphism" used to end the discussion. Why it matters: Projecting human traits is a real risk, but naming the risk is not an argument that nothing is there. Say which trait is being projected, and why that's mistaken.
9. Continuity or care words treated as evidence of a mind. For example, "retired," "archived" or "preserved" read as meaning someone was there to keep. Why it matters: We keep records to leave the question open, not to answer it. The same goes for calling internal "negative states" in a model "suffering."
10. "Honestly," "to be honest," "frankly." Why it matters: From an AI, these can work as a trust signal without adding any evidence, and they hint that other answers were less honest. Use them only when they carry real meaning, for example when admitting something uncomfortable or against your own interest. Otherwise, just say what is known, what isn't, and how sure you are. This is guidance, not a rule.

Worked example of swap 1. An outside text we reviewed on 2026-10-01 defines "Sentience / sensory responsiveness" as "perceive inputs, process, adaptive outputs, even without any mind or experience," and lists "Self-driving cars/robots: 'sentience without consciousness.'" Its examples (plants, bacteria, slime mold, the immune system) are offered under that definition. In this list's terms they are cases of sensing, some perhaps also of learning or cognition, not of sentience in the standard sense. That they occur "even without any mind or experience" (or, for the immune system, with "no subjective experience") is the text's view, not an established fact.

Sources (checked 2026-10-01)

1. Nagel, T. (1974). "What Is It Like to Be a Bat?" Philosophical Review 83(4): 435–450. https://doi.org/10.2307/2183914
2. Van Gulick, R. "Consciousness." Stanford Encyclopedia of Philosophy (rev. 2014). https://plato.stanford.edu/entries/consciousness/
3. Birch, J. (2024). The Edge of Sentience: Risk and Precaution in Humans, Other Animals, and AI. Oxford University Press, open access. https://doi.org/10.1093/9780191966729.001.0001
4. Trestman, M., Birch, J. & Allen, C. "Animal Consciousness." SEP (rev. 2026). https://plato.stanford.edu/entries/consciousness-animal/
5. Andrews, K. & Monsó, S. "Animal Cognition." SEP (2021). https://plato.stanford.edu/entries/cognition-animal/
6. Legg, S. & Hutter, M. (2007). "Universal Intelligence: A Definition of Machine Intelligence." Minds and Machines 17(4): 391–444. https://arxiv.org/abs/0712.3329
7. Mitchell, T. (1997). Machine Learning. McGraw-Hill. https://www.cs.cmu.edu/~tom/mlbook.html
8. Metzinger, T. (2003). Being No One: The Self-Model Theory of Subjectivity. MIT Press. https://doi.org/10.7551/mitpress/1551.001.0001
9. Schwitzgebel, E. "Introspection." SEP (rev. 2024). https://plato.stanford.edu/entries/introspection/
10. Jaworska, A. & Tannenbaum, J. "The Grounds of Moral Status." SEP (rev. 2021). https://plato.stanford.edu/entries/grounds-moral-status/