Was den Menschen zum Menschen macht

Turing Test

6 Min. Lesezeit

🇩🇪 Deutsche Version: Turing-Test

The Turing Test is an operational criterion for machine intelligence, proposed in 1950 by Alan Turing: if an interrogator in a text conversation cannot reliably distinguish a machine from a human being, the machine is to be credited with thinking. From the perspective of personal ontology, this criterion is fundamentally inadequate, because it confuses behavior (operatio) (German) with being (esse) (German).

Why the Turing Test fails as a criterion for personhood

The Turing Test rests on a behaviorist premise: intelligence — and implicitly consciousness — is defined in terms of observable behavior. This is precisely the logic of the empirical-functionalist concept of person, which the dissertation rejects as inadequate.

The Thomistic principle operatio sequitur esse — activity follows being — reverses the logic of the test: what a being does follows from what it is. That a machine imitates human behavior says nothing about its being — just as a parrot that says “hello” does not speak.

Personhood is First Actuality (prote energeia) (German) — the substantial being of the Person, which precedes all doing. The Turing Test captures at best Second Actuality (deutera energeia) (German) — the exercise of activities. It confuses the ground of knowing (ratio cognoscendi) with the ground of being (ratio essendi) of personhood.

The Chinese Room Argument

John Searle formulated the most influential philosophical critique of the Turing Test in 1980. His thought experiment: a person sits in a room, receives Chinese characters through a slot, and returns correct answers with the help of a rulebook — without understanding a word of Chinese.

The point: syntax (formal symbol manipulation) does not suffice for semantics (meaning). This confirms personal ontology: genuine Intentionality — the directedness of the mind toward objects — is an immaterial faculty of the Person and is not reducible to information processing. Thomas Aquinas draws a distinction: the human intellect receives the form of the thing known without its matter — a process fundamentally different from symbol processing.

From this, Searle distinguishes two theses:

  • Strong AI: the appropriately programmed computer literally is a mind — it really understands. From the standpoint of personal ontology this is impossible in principle: Personhood requires a spiritual substance and an actus essendi that cannot be artificially produced.
  • Weak AI: computers simulate thinking. Their apparent understanding is an as-if. This is compatible with personal ontology: Artificial Intelligence as Deutera Energeia without Prote Energeia.

The Blockhead Argument

In 1981 Ned Block exposed the weakness of the Turing Test with a further thought experiment: a being with a pre-programmed response for every possible input passes the Turing Test, yet has “the intelligence of a toaster.”1 Behavioral adequacy does not imply intelligence, let alone Personhood.

The philosophical zombie

In 1995 David Chalmers formulated the Hard Problem of Consciousness: why does subjective experience exist at all? A “philosophical zombie” behaves exactly like a conscious Person but has no inner life — no joy, no suffering, no consciousness. In the language of personal ontology: a philosophical zombie is Deutera Energeia without Prote Energeia — precisely the situation of any AI.

In his own inquiry into whether large language models could be conscious, Chalmers concludes that they lack too many of the potential prerequisites for consciousness for one to be entitled to ascribe actual experience of the world to them.2

Current state of research

The studies of Jones and Bergen (2024/2025) show that modern language models pass the Turing Test in short conversations: GPT-4.5 with a particular persona was judged human in 73% of cases — more often than actual humans. The interrogators relied on linguistic style and socio-emotional factors, not on markers of intelligence.

This confirms the diagnosis of personal ontology: the Turing Test measures quality of simulation, not personhood. The better the simulation, the more dangerous the confusion — and the more urgent the distinction between Someone and something.

The Full Rights Dilemma

Eric Schwitzgebel (2023) formulates a dilemma: AI systems of disputed moral status produce a lose-lose situation. Either rights are ascribed to unconscious systems (at the expense of real persons) or rights are denied to conscious systems (a moral wrong).

Personal ontology dissolves this dilemma: Personhood is not constituted by functions but by the being of a spiritual substance. An AI in principle has no actus essendi and therefore cannot be a person. The dilemma arises only on the presupposition of the empirical-functionalist concept of person — and dissolves as soon as one takes the substance-ontological-relational concept of person as its basis.

Dreyfus’s phenomenological critique of AI

From 1972 onward, drawing on Heidegger and Merleau-Ponty, Hubert Dreyfus argued that human intelligence depends primarily on an embodied background understanding that cannot be captured in formal rules. This is compatible with personal ontology: the human person is a bodily-spiritual substance, not a rule-following algorithm.

Spaemann: Someone and Something

Robert Spaemann shows the ontological depth of the problem: “Persons are ‘someone’ and not ‘something’. There is no continuous transition from something to someone.”3 The Turing Test treats the question “Can a machine be a someone?” as an empirical question of behavior. Spaemann shows: the distinction someone/something is not a question of behavior but an ontological category that cannot be overcome by imitation.

Ontological classification

Sources: Generated by querying the Personhood ontology.

Further sources:

  • Turing, Alan M. (1950): “Computing Machinery and Intelligence.” Mind 59(236), pp. 433–460 (the original formulation of the imitation game).
  • Searle, John R. (1980): “Minds, Brains, and Programs.” Behavioral and Brain Sciences 3(3), pp. 417–457.
  • Block, Ned (1981): “Psychologism and Behaviorism.” Philosophical Review 90(1), pp. 5–43.
  • Chalmers, David J. (1995): “Facing Up to the Problem of Consciousness.” Journal of Consciousness Studies 2(3), pp. 200–219.
  • Chalmers, David J. (2023): “Could a Large Language Model Be Conscious?” Boston Review / arXiv:2303.07103.
  • Spaemann, Robert, Persons: The Difference between ‘Someone’ and ‘Something’, transl. O’Donovan, Oxford UP 2006 (no continuous transition from something to someone).
  • Dreyfus, Hubert L. (1972/1992): What Computers Can’t Do / What Computers Still Can’t Do. MIT Press.
  • Aquinas, Thomas: Summa Theologiae, I, q. 84, a. 1–2 (the intellect receives form without matter — the difference from symbol processing).
  • Jones, Cameron R. / Bergen, Benjamin K. (2024): “Does GPT-4 Pass the Turing Test?” In: Proceedings of NAACL-HLT 2024, Vol. 1: Long Papers, pp. 5183–5210. Mexico City: ACL. arXiv:2310.20216.
  • Jones, Cameron R. / Bergen, Benjamin K. (2025): “Large Language Models Pass the Turing Test.” arXiv:2503.23674.
  • Schwitzgebel, Eric (2023): “AI Systems Must Not Confuse Users about Their Sentience or Moral Status.” Patterns 4(8).
  • Schwitzgebel, Eric (2024): The Weirdness of the World. Princeton University Press.

See also

Fußnoten

  1. Block, Ned (1981): “Psychologism and Behaviorism.” Philosophical Review 90(1), p. 20. ↩

  2. Chalmers, David J. (2023): “Could a Large Language Model Be Conscious?” Boston Review / arXiv:2303.07103. ↩

  3. Spaemann, Robert, Persons: The Difference between ‘Someone’ and ‘Something’, transl. O’Donovan, Oxford UP 2006, pp. 26 f. ↩

Graphansicht

Backlinks