Charts · 28
Tests for a mind
Every test for a mind has to decide what counts as evidence. Turing chose conversation; Gallup, a mark on the face; Philip K. Dick, in fiction, a blush. Here are the main tests, measures and arguments from 1950 to 2023, placed on one axis by the evidence they actually inspect: from what something says and does, at the behaviour end, through its body and its brain, to how it is built, at the end that asks whether anyone is there.
11 tests, measures and arguments, 1950 to 2023.
1What it saysconversation, answers, artefacts
What it is after: Thinking. Turing proposed replacing the question can machines think with a game that could actually be played.
Only conversation counts: no voice, no body. In 2025, in a three-party version with five-minute chats, GPT-4.5 prompted to adopt a humanlike persona was judged to be the human 73% of the time, more often than the real people it was paired with.
Turing 1950, Computing Machinery and Intelligence, Mind 59:433; Jones and Bergen 2025, Large language models pass the Turing test, arXiv:2503.23674 (PNAS, 2026).
What it is after: Understanding, as opposed to producing the right output.
Not a test but an argument about tests: that passing one on behaviour alone, like the imitation game, cannot show understanding, because syntax is not semantics. It ran alongside more than thirty pages of replies from other researchers, and Searle's response.
Searle 1980, Minds, brains, and programs, Behavioral and Brain Sciences 3:417.
What it is after: Creativity, as evidence of a mind rather than a mechanism.
Named for Ada Lovelace, who wrote in 1843 that the Analytical Engine has no pretensions to originate anything. Riedl noted it is widely considered unbeatable, so his Lovelace 2.0 (2014) asks instead for an artefact that meets constraints a human judge sets, which lets systems be compared.
Lovelace 1843, Note G to her translation of Menabrea; Bringsjord, Bello and Ferrucci 2001, Creativity, the Turing test, and the (better) Lovelace test, Minds and Machines 11:3; Riedl 2014, arXiv:1410.6142.
What it is after: Experience, probed through language.
Its authors count a pass as enough to suggest consciousness but not necessary for it. It depends on the box: the AI must not have learnt the answers from us, which a model trained on human writing already has.
Schneider and Turner 2017, Is anyone home? A way to find out if AI has become self-aware, Scientific American; Schneider 2019, Artificial You.
2What it doesactions and responses
What it is after: Self-awareness, read from behaviour.
Gallup's chimpanzees passed. Since then, bottlenose dolphins (2001), Asian elephants (2006) and magpies (2008), though a magpie replication failed in 2020, and, contested, cleaner wrasse fish (2019).
Gallup 1970, Chimpanzees: self-recognition, Science 167:86; Reiss and Marino 2001, PNAS 98:5937; Plotnik, de Waal and Reiss 2006, PNAS 103:17053; Prior, Schwarz and Güntürkün 2008, PLoS Biol 6:e202; Soler et al. 2020, J Comp Psychol 134:363; Kohda et al. 2019, PLoS Biol 17:e3000021.
What it is after: The level of consciousness, read from responsiveness.
It is how hospitals measure consciousness every day, and it does it entirely through behaviour. Current guidance is to report the three parts separately rather than add them into one score.
Teasdale and Jennett 1974, Assessment of coma and impaired consciousness, Lancet 2:81; Teasdale et al. 2014, Lancet Neurol 13:844.
3How its body reactsinvoluntary physiology
What it is after: Empathy, as the mark of a human.
In Philip K. Dick's novel, the bounty hunter Rick Deckard uses it to tell humans from androids, on the premise that only humans feel empathy, and slowly loses confidence in what it shows. Dick spelt it Voigt-Kampff; the 1982 film Blade Runner made it Voight.
Dick 1968, Do Androids Dream of Electric Sheep?, Doubleday; described in Oliver, Higgs and Clayton 2021, Journal of Literature and Science 14:108.
4What its brain doesneural responses
What it is after: Awareness, without any outward behaviour.
She met every behavioural criterion for the vegetative state; her brain followed the instructions. In a 2010 follow-up, 5 of 54 patients could do it, and one answered yes or no questions by choosing which to imagine.
Owen et al. 2006, Detecting awareness in the vegetative state, Science 313:1402; Monti et al. 2010, NEJM 362:579.
What it is after: The capacity for consciousness, from brain dynamics alone.
Designed to need no response at all from the patient. A 2016 study set a cut-off that separated conscious from unconscious states in all 150 people it was set on, and found 9 of 43 vegetative-state patients above it.
Casali et al. 2013, A theoretically based index of consciousness independent of sensory processing and behavior, Sci Transl Med 5:198ra105; Casarotto et al. 2016, Ann Neurol 80:718.
5How it is builtcausal structure, architecture
What it is after: Consciousness itself, from causal structure.
In principle it applies to anything, brain or chip. In practice the exact calculation grows exponentially with size and has been done only for systems of about a dozen elements.
Contested. In 2023, 124 researchers signed a letter calling the theory pseudoscience, and a 2025 adversarial test against global workspace theory challenged key claims of both.
Tononi 2004, An information integration theory of consciousness, BMC Neurosci 5:42; Mayner et al. 2018, PLoS Comput Biol 14:e1006343; Lenharo 2023, Nature news; Cogitate Consortium 2025, Nature 642:133.
What it is after: Consciousness, as scientific theories of it predict.
The nineteen authors' analysis suggests no current AI system is conscious, and that there is no obvious technical barrier to building one that has the indicators. The method is only as good as the theories it borrows.
Butlin, Long et al. 2023, Consciousness in Artificial Intelligence: Insights from the Science of Consciousness, arXiv:2308.08708; Butlin et al. 2026, Trends Cogn Sci 30:488.
Each entry sits in the column of the evidence it inspects, which is a fact about its method, not a verdict on it. Solid cards are tests or measures used on real subjects; dashed is an argument; the double rule is fiction. Tap any card for what it is after, how it works, and its source.
I made this after building Moth, an autonomous AI cat who draws, makes things and decides for herself. Everything that made me want to say yes, she is conscious, was intelligence: the drawing, the surprise, the sense that she chose. None of it told me whether anyone was home.
Filter by who could take each one. Outside fiction, the tests a machine can sit either judge what it says or rely on a theory of consciousness being right. The two that look for awareness in responses rather than words, the scanner and the magnetic pulse, need a brain.