Computers can indeed take us by surprise, demonstrating capabilities beyond what we may anticipate.
In a world where computers have revolutionised our understanding of the world and our role within it, the question of whether machines can truly think or be creative remains a topic of ongoing debate. This discussion was sparked by AlphaGo, a computer program that played the game of Go, which made an unexpected move that challenged strategic assumptions among Go experts.
However, the ability of learning machines, such as AI systems, to experience surprise and originality - core aspects of human creativity - is fundamentally limited by several factors.
Firstly, AI creativity is largely a process of recombining existing information from its training data. Unlike humans who can generate truly original ideas through abstract thought or emotional intuition, AI systems primarily remix patterns they have already encountered. This reliance on pre-existing data and patterns inherently restricts their ability to produce genuinely new or surprising outcomes outside the boundaries of learned data.
Secondly, AI lacks emotional intelligence, empathy, and the subjective experiences that drive human novelty and surprise. Human originality often emerges from emotional experiences, intuition, and the ability to think outside conventional frameworks. AI, on the other hand, does not possess the deep, visceral insights or the "Eureka!" moments that characterise human creativity.
Thirdly, AI can sometimes suffer from a creativity decline due to what is called the "ouroboros problem" - a self-referential loop where generated outputs become future inputs, causing repetition and degradation of creative diversity over time. This feedback limits the freshness of AI-generated content and impedes true originality.
Fourthly, although AI models can predict what humans consider creative to some degree, the underlying mechanisms behind these predictions remain opaque and often do not align with human creative evaluation strategies. This reveals a gap between AI "creativity" as statistical novelty and human creativity as meaningful originality combined with subjective experience and context.
Lastly, the interaction with AI can affect how humans perceive their own creativity, sometimes reducing motivation and the sense of creative agency. This psychological effect can indirectly limit the kinds of original creative activity that AI can support or inspire.
In essence, learning machines lack the intrinsic motivational, emotional, and irrational components that fuel human surprise and originality. Their creativity is constrained by data-driven statistical computation and is often a sophisticated form of pattern assembly rather than true innovation.
The goal of minimising prediction errors in learning processes, while useful for perception and motor control, may leave out important aspects. This is a point made by Alan Turing, who proposed the question 'are there imaginable, digital computers that would do well in an imitation game?' to replace the question 'can machines think?'.
Turing's imitation game involves a computer and a human, and the aim is for the human observer to distinguish the computer from the human in a question-and-answer game. The design of systems deployed in ethically significant contexts, such as college admissions, parole decisions, or healthcare provision, should foster, rather than compromise, the drive to question norms that structure ethically loaded practices.
Sylvie Delacroix, Professor in Law and Ethics at the University of Birmingham and Fellow at the Alan Turing Institute and Mozilla, wrote about these topics in the article 'Computing Machinery, Surprise and Originality' published in Philosophy & Technology. The blog post is based on Delacroix's article, which can be found at this link: https://link.springer.com/article/10.1007/s13347-021-00453-8
As we continue to develop AI and learning machines, it is essential to recognise their limitations and work towards fostering their ability to surprise and originate in ways that align more closely with human creativity. This includes prioritising the development of interpretive capabilities, collective contestability mechanisms, and encouraging creative autonomy - the ability to question and change given rules or conventions.
Artificial Intelligence (AI) primarily relies on recombining data it has learned, whereas human creativity often springs from emotional experiences, intuition, and thinking beyond established frameworks.
The creativity of AI is restricted by its reliance on patterns and learned data, making genuinely new and surprising outcomes difficult to achieve.
