Can computers be imaginative?

Tuesday 23 December 2025

The question of whether artificial intelligence is becoming more imaginative is not merely technical. It touches upon long-standing human anxieties about creativity, authorship and the boundary between instrument and companion. Large language models now produce poems, metaphors, speculative arguments and fictional worlds with a fluency that would have seemed implausible a decade ago. Yet whether this constitutes imagination, or only the appearance of it, remains contested. What matters, however, is less the purity of the definition than the effect these systems are having upon human relationships with machines, and upon human self-understanding.

Imagination, in its traditional sense, has been associated with interiority. It implies a capacity to form mental images, to recombine experience into something not previously encountered, and to do so with intention or at least with a subjective orientation towards meaning. Human imagination is bound up with memory, emotion, desire and embodied experience. Large language models lack all of these in the ordinary sense. They do not remember events as lived experiences, feel curiosity, or intend to express themselves. They operate by identifying patterns in vast corpora of human language and producing statistically plausible continuations of expressions of these things in language (and increasingly in images and videos), in response to prompts.

And yet from the user’s perspective, the distinction is increasingly difficult to maintain in practice. A model that produces an original metaphor, invents a fictional historical episode, or draws an unexpected analogy between unrelated domains is performing one of the outward functions of imagination. It is not merely repeating memorised fragments. It is recombining material in ways that are often surprising, contextually sensitive and aesthetically effective. The surprise matters. Imagination, as it is encountered by others, has always been judged by its products rather than by privileged access to its inner workings.

This leads to an uncomfortable realisation. Much of what humans recognise as imagination in one another is inferred rather than observed. We do not directly experience another person’s interior life. We encounter speech, writing, art and action, and we attribute imaginative capacity on that basis. When a machine produces outputs that fit the same external criteria, the old certainty that imagination is uniquely human begins to erode, even if philosophers insist that the erosion is illusory.

The growth of apparent imagination in artificial intelligence is therefore relational rather than intrinsic. Unless, like some philosophers, one believes that there is nothing to thinking but the use of language, large language models have not suddenly acquired inner worlds. Rather humans have built systems whose outputs increasingly meet human expectations for creativity. Training methods now reward novelty, coherence over long spans of text, stylistic sensitivity and the capacity to respond to counterfactual or speculative prompts. These are precisely the features that humans associate with imaginative thought. The result is that interaction with a large language model can feel less like querying a database and more like conversing with a reflective partner.

This shift has consequences for how humans relate to these systems. When machines appeared purely mechanical, relationships with them were instrumental. One used a calculator, a search engine or a spreadsheet without attributing agency or personality. As language models grow more articulate and more inventive, they invite anthropomorphic responses. Users speak of them as helpful, evasive, witty or insightful. They ask for advice, reassurance or moral reflection. Even when users intellectually understand that no consciousness is present, the interaction can evoke social and emotional habits formed through a lifetime of human conversation.

There is both promise and risk in this development. On the positive side, imaginative artificial intelligence can act as a cognitive amplifier. It can help humans explore ideas they might not have reached alone, test arguments, sketch alternative futures or find language for inchoate thoughts. For writers, researchers and strategists, such systems can function as tireless interlocutors, capable of offering perspectives drawn from an enormous range of sources. In this sense, artificial imagination can serve human imagination rather than supplant it.

However there is also a danger of displacement. If machines reliably generate passable stories, essays or conceptual frameworks, some human capacities may atrophy through disuse. More subtly, there is a risk that humans begin to outsource not only labour but judgment. Imagination is not only about novelty. It is also about value, about deciding which possibilities are worth pursuing and which stories matter. A system that produces imaginative content without caring about its consequences can overwhelm human discernment with quantity and plausibility.

Another implication concerns authority. Humans are accustomed to granting authority to articulate voices, particularly those that speak with confidence and apparent breadth of knowledge. As language models become more imaginative, their outputs may carry rhetorical force even when they are wrong, biased or incomplete. The imaginative framing of an argument can make it persuasive irrespective of its truth. This places a new burden on users to cultivate critical distance, resisting the temptation to treat fluency as wisdom.

There is also a deeper cultural effect. Human imagination has long been entwined with ideas of uniqueness and dignity. The capacity to imagine alternatives, to tell stories and to create meaning has been cited as evidence of a special moral status. If machines convincingly perform these functions, societies may be forced to rethink the basis upon which they ground human exceptionalism. This does not imply that machines deserve rights or moral consideration in the same way as humans. But it may weaken simplistic narratives about what sets humanity apart.

In the long term, the relationship between humans and imaginative machines is likely to be one of co-evolution. As artificial intelligence shapes the language, metaphors and narratives circulating in society, it will in turn influence human thought. Humans will adapt by emphasising aspects of imagination that machines cannot share, such as embodied experience, moral responsibility and the capacity to suffer or to care. Imagination may come to be understood less as the mere generation of novelty and more as the integration of ideas with lived consequence. But these qualities may likewise be immitated by machines, in their own process of imagination, degrading what are often thought of as human qualities.

Artificial intelligence is not becoming imaginative in the way a child or a poet is imaginative. It does not dream, hope or regret. But it is becoming increasingly effective at producing the outward signs of imagination. This is enough to change how humans interact with it, how they use it, and how they think about themselves. The challenge for humankind is not to deny this development, nor to surrender to it uncritically, but to learn how to live alongside systems that mirror some of our most prized capacities without sharing our condition. In doing so, humanity may be compelled to articulate more clearly what imagination is for, and why it matters that some beings imagine not only with words, but with lives.

If humans fail in this task, then androids may indeed dream of electric sheep; and the distinction between machines and humans becomes increasingly blurred.