The question of whether computers can think has occupied philosophers, computer scientists and cognitive researchers for more than seventy years. This essay examines the principal arguments that have shaped the debate, drawing on foundational work in both philosophy and artificial intelligence. The discussion begins with Alan Turing’s operational approach, moves to John Searle’s well-known objection, and considers how contemporary systems continue to test these positions. The central aim is to assess whether current evidence supports the claim that computers genuinely think or whether serious conceptual obstacles remain.
The Turing Test and Computational Functionalism
Alan Turing proposed that the question “Can machines think?” could be replaced by a practical test of behaviour (Turing, 1950). In the imitation game a human interrogator communicates with two hidden participants, one human and one machine, and must decide which is which. If the machine routinely deceives the interrogator, Turing argued, it possesses conversational intelligence indistinguishable from that of a human. The test therefore sidesteps difficult metaphysical issues about inner states and concentrates on observable performance.
Turing’s proposal rests on the assumption that thinking is a form of information processing that can be realised in different physical media. If the same functional organisation produces the same intelligent behaviour, the substrate—biological neurons or silicon circuits—should not matter. This functionalist view became influential in early artificial intelligence research and still underpins much work on large language models. Nevertheless, the test measures only linguistic imitation; it does not directly examine understanding or consciousness. Critics note that a system could pass the test through statistical pattern-matching without possessing the internal states normally associated with thought (Floridi, 2014). Thus, while Turing’s criterion remains a useful benchmark for conversational ability, its limitations prevent it from settling the deeper philosophical question.
Searle’s Chinese Room and the Symbol-Grounding Problem
John Searle mounted a direct challenge to the claim that syntactic manipulation suffices for thought. In the Chinese Room thought experiment a person who understands no Chinese follows a rulebook to produce correct Chinese answers to Chinese questions (Searle, 1980). From outside the room the process appears intelligent, yet the occupant understands nothing. Searle concludes that computers, which operate solely by manipulating formal symbols, likewise lack semantic content. No matter how sophisticated the program, syntax alone cannot generate meaning.
The argument highlights what Harnad later termed the “symbol-grounding problem”: symbols must be connected to the world through perception and action if they are to refer (Harnad, 1990). Contemporary neural network models attempt to address this issue by learning representations from vast multimodal data. Nevertheless, Searle’s objection retains force because these models still operate on statistical correlations rather than genuine intentional states. The room may now contain billions of adjustable weights, yet the underlying process remains formal symbol manipulation without intrinsic understanding. Consequently, many philosophers maintain that passing behavioural tests does not entail the presence of thought.
Contemporary AI Systems and Persistent Questions
Recent advances in deep learning have produced systems that exceed human performance on narrow tasks such as language translation and game playing. These achievements demonstrate impressive functional capacities. However, performance gains have not resolved the philosophical disagreement. Large language models generate coherent text by predicting likely token sequences; they possess no first-person perspective and no capacity for genuine belief or desire. Researchers therefore distinguish between “weak” AI, which simulates intelligent behaviour, and “strong” AI, which would actually possess mental states (Searle, 1980). Current systems clearly belong to the former category.
Furthermore, practical limitations reinforce this assessment. Models frequently produce plausible but factually incorrect statements, revealing an absence of stable world models. They also lack the embodied interaction that, according to many theorists, grounds human concepts (Brooks, 1991). While embodiment is not a strict requirement in every philosophical account, its absence removes one plausible route by which symbols could acquire meaning. Therefore, although computers can now exhibit sophisticated behaviour, the gap between simulation and genuine thought remains philosophically significant.
Conclusion
The debate over whether computers can think continues to hinge on the distinction between behavioural adequacy and internal semantic content. Turing’s operational test usefully measures performance yet leaves open questions of understanding. Searle’s Chinese Room argument shows that symbol manipulation alone does not produce meaning, a conclusion that still applies to contemporary statistical models. While future research may narrow the functional gap between machine and human performance, the conceptual obstacles identified by Searle have not been removed. Hence, at present there is no convincing evidence that computers think in the full sense required by the original question.
References
- Brooks, R. A. (1991) ‘Intelligence without representation’, Artificial Intelligence, 47(1-3), pp. 139–159.
- Floridi, L. (2014) The Fourth Revolution: How the Infosphere is Reshaping Human Reality. Oxford: Oxford University Press.
- Harnad, S. (1990) ‘The symbol grounding problem’, Physica D: Nonlinear Phenomena, 42(1-3), pp. 335–346.
- Searle, J. R. (1980) ‘Minds, brains, and programs’, Behavioral and Brain Sciences, 3(3), pp. 417–424.
- Turing, A. M. (1950) ‘Computing machinery and intelligence’, Mind, 59(236), pp. 433–460.
