AI Creativity Unpacked: Marcus de Sotoy on AI's role in art, music and human imagination

In this New Scientist feature, Marcus de Sotoy explores whether artificial intelligence can be truly creative, how creativity can be defined, and how AI systems learn to surprise and add value. The discussion covers novel ideas such as creative adversarial networks, the dependency on data and training, and examples spanning visual art, jazz improvisation, Go and beyond. The conversation emphasizes three kinds of creativity exploratory, combinational and transformational, and asks whether AI can express intention or a form of inner drive. It looks at how AI might push human creativity forward rather than merely imitate the past, and what this means for artists, audiences and the future of culture.

Overview and Context

The transcript centers on a two-part discussion featured by New Scientist that probes the evolving relationship between artificial intelligence and human creativity. Marcus de Sotoy, a prominent science communicator and author of The Creativity Code, frames the conversation around existential questions about AI’s growing capability to write, compose, paint and create as convincingly as humans. The discussion makes a deliberate effort to define creativity, explore how AI learns, and present concrete examples across music, visual art and performance to illustrate both the opportunities and the challenges of AI-assisted creativity.

Defining Creativity: Novelty, Surprise and Value

De Sotoy references a useful definition attributed to Margaret Boden: a creative act is novel, surprising, and valuable. He notes that novelty can be assessed more objectively, while surprise and value are highly subjective. This framework becomes a lens for evaluating AI’s outputs, asking whether AI-generated works meet human expectations of novelty and whether they carry meaningful value beyond replication of past patterns.

From Static Rules to Learning Systems: The Rise of Machine Learning

The talk contrasts earlier AI with modern approaches that learn from data. Instead of fixed code, contemporary AI uses machine learning to adapt by interacting with data, enabling more sophisticated interpretation of sensory input such as vision. This shift is crucial for enabling AI to participate in creative domains traditionally dominated by human intuition.

Three Kinds of Creativity: Exploratory, Combinational and Transformational

De Sotoy delineates three pathways for creativity. Exploratory creativity involves exploring the edges of established rules, for example, a jazz context where an AI pushes riffs to the extreme. Combinational creativity merges seemingly unrelated styles to produce novel hybrids. Transformational creativity involves breaking the rules to reveal new ways of thinking, a category where human artists such as Picasso and groundbreaking AI moves may converge to redefine what is possible in art and practice.

Intention, Soul and Autonomy in AI

The conversation tackles the philosophical question of whether AI can have intention, essence, or soul. While AI lacks consciousness in the human sense, its evolving algorithms can develop decision-making processes that feel autonomous. The discussion highlights that intention in AI is often human-provided yet can become self-directing as models learn to navigate data landscapes in unexpected ways. This raises questions about authorship, responsibility, and the nature of creativity as a shared human-machine endeavour.

Examples and Case Studies

The transcript references several concrete examples. Creative adversarial networks pit two AI systems against one another to generate novel art while maintaining recognizability, producing works that elicit emotional reactions when audiences are unaware of their digital origin. The Jazz Continuator demonstrates how AI can study jazz riffs and then push beyond them to reveal new expressive possibilities for human performers. AlphaGo and AlphaZero illustrate how AI discovers non-intuitive strategies that expand human perception of what is possible in strategic games, with Move 37 in particular cited as a landmark moment that shifted human thinking about the game.

Data, Bias and Ethical Considerations

The discussion turns to data bias as a practical and ethical concern. An MIT example shows a robot failing to recognize a Black woman and only recognizing her when wearing a white mask, underscoring how training data shapes AI perception and performance. The Algorithmic Justice League is mentioned as a response to these biases, highlighting the importance of fairness and accountability as AI systems become embedded in artistic and cultural contexts. The broader point is that AI, while powerful, is not value-free and requires careful curation and oversight to avoid reinforcing social biases.

AI in Music, Dance and Visual Arts: Human-AI Collaboration

Beyond theory, the transcript explores practical applications in music and dance. Projects involve AI-generated sheet music, deep fake musical textures, and raw audio generation trained on broadcasts to create evolving live experiences. Dancers interact with predictive AI to generate new choreographic language, enabling performers to respond creatively in real time. The collaboration between AI and live performance is framed as a means to expand the expressive palette rather than supplant human creativity.

Algorithmic Curation and Cultural Impacts

Another thread examines algorithmic curation in platforms like Instagram, comparing machine-led selections with human curators. The case study demonstrates how algorithms tend to favor recognizable, commercially viable imagery and may suppress novelty, raising questions about how our cultural reality is shaped by data-driven filters. This has implications for artists, audiences, and the public understanding of culture in the age of algorithmic influence.

Future Prospects and Human Significance

The dialogue concludes with a nuanced view: AI can push human creativity into new domains, revealing better ways to think and create, but there remains a unique space for human artistry, intention, and emotional resonance. The availability of data limits AI capabilities to replicate iconic masterpieces, which provides a form of protective boundary for human creativity. The overarching message is one of cautious optimism: AI is a powerful collaborator that can expand the horizons of art and science, not merely imitate the past.