Maya Ackerman: “Hallucinations are the underlying mechanism at work in creativity, whether human or AI”

There is a question that has been circulating in academic circles for a long time and has also reached the general public in recent years:  can machines be creative? Since the widespread adoption of generative artificial intelligence following the launch of ChatGPT in November 2022, questions about its reliability, its potential to distort reality through so-called “hallucinations,” and its effects on human creativity have become central topics in public debate.

To address these topics, researcher Maya Ackerman spoke at BBVA’s FinAI Summit,, a four-day event held across Mexico, Colombia, Türkiye and Spain that explored the potential of artificial intelligence in various fields and brought together more than 6,000 participants from 30 countries.  Ackerman has dedicated over a decade to studying the relationship between AI and creativity, first as a researcher and now as a professor of artificial intelligence at Santa Clara University in California, as well as co-founder of WaveAI, a startup focused on developing AI models for music creation.

Before addressing whether machines can be creative, Ackerman considers what creativity actually is and whether it is a unique human trait:  “When we talk about intelligence and creativity, we believe that these concepts should begin and end with us, and we tend to judge others based on our own.”

One of the defining elements of human creativity is intention.  “When humans create music, art, or a business idea, we do so with an intention,” she points out, adding:  “Even when an artist paints a sketch and is not sure why they chose a particular color, we say that they at least have an unconscious intention, something we do not attribute to machines.”  For her, machines follow a different process when it comes to creating something, but the fact that they do not have intention should not invalidate it.

The second key element that links creativity to human nature is emotion:  “Our brain depends on emotions.”  Once again, Ackerman emphasizes that machines are fundamentally different from humans and their processes do not need to mirror our own.  She points out that “the brain of artificial intelligence does not have emotional centers, yet this does not appear to diminish the quality of the outcomes they generate.”  For this reason, Ackerman suggests that we should evaluate the outputs produced by machines, rather than focusing on the methods they use to arrive at those results.

Creativity judged by outcomes

Ackerman’s view that creativity should be assessed based on outcomes rather than process echoes the work of David Cope, a music professor at the University of California, Santa Cruz, and a pioneer in the study of artificial intelligence and music. In the early 1980s, David Cope was commissioned to compose an opera.  When he encountered writer’s block that stalled his progress, he decided to develop a computer program to assist with the creative process.  The resulting software, called EMI (Experiment in Musical Intelligence), not only enabled him to complete the commissioned opera but also paved the way for analyzing and recomposing works by composers such as Bach, Vivaldi and Chopin.

However, Cope observed an intriguing phenomenon.  When he played the compositions for his audience, they were impressed by the quality. Yet, as soon as he revealed that the pieces had been created by a machine, the audience dismissed them.  “There is incredible discrimination against the creativity of machines,” Ackerman notes.

To further investigate this bias, Cope designed an experiment: he selected an original composition by Bach, a piece generated by EMI imitating Bach’s style, and a work by a human composer also emulating the German master. He then played all three for a panel of experts.  Most of the experts thought the EMI composition was the authentic Bach piece. “So, we should evaluate the quality of music produced by machines based on its outcome, not the process behind it,” Cope contends.

Hallucinations as the basis of creativity

Since the launch of ChatGPT in 2022 and its rapid adoption by the public, concerns have emerged about the accuracy of the large language models that power generative AI.  The most well-known of these concerns is the phenomenon called “hallucinations:” instances where AI generates text, video, or audio that deviates from reality or produces incoherent results.  However, Ackerman offers a different perspective on this issue:  “I believe that the central mechanism underlying creativity, both in machine intelligence and in humans, is hallucinations.”

Hallucinations, Ackerman argues, aren’t limited to what psychedelics trigger.  “We experience controlled hallucinations all the time,” she says.  Our brains act as prediction machines, constantly guessing what will happen next. When our guesses miss the mark, we revise our own algorithm.”  Machines, she notes, create in much the same way: they try to anticipate the next step, sometimes accurately, sometimes not.

Ackerman emphasizes that both people and algorithms need room to make mistakes if they are to be creative.  “For a machine to be creative, we have to let it get things wrong. We need to stop fearing that machines will lie to us,” she says.

Tips for sparking creativity with AI

Large language models have quickly become popular tools for streamlining tasks like writing emails or drafting reports —a familiar way, Ackerman says, that people tend to view machines.  But she believes these new systems offer far more than just speed and convenience.  “Instead of simply doing more, faster, we can create things that are truly better,” she explains.  To unlock this potential, Ackerman suggests we stop seeing artificial intelligence as a mere replacement and begin treating it as a creative partner. “We need to bring our best effort if we want the results to be outstanding,” Ackerman says. She offers several practical suggestions for working with AI:

• Stay in control: AI can supply expertise that users may lack, such as fluency in another language or advanced knowledge of mathematics and algorithms. However, Ackerman stresses the importance of remaining in charge of the creative process. “Maintain control and always be skeptical,” she advises.
• Break down problems into parts: When facing a complex challenge, one of the most effective strategies is to divide it into manageable pieces.  Generative AI often stumbles with large, complicated tasks.  But if you break the problem down and give it to the AI in smaller steps, it typically performs much better.
• Inspiration: When you hit a creative block, AI can suggest new directions or highlight possibilities you might not have considered.  “Sometimes, just seeing the options helps you create something different,” Ackerman says, reflecting on how she uses AI to choose the right words when writing.  Yet she cautions against unconsciously adopting the language patterns of AI models.  “I’ve noticed that even in my own thinking, I’m using phrases typical of ChatGPT.”
• Get feedback: More artists now use tools like ChatGPT to review and compare their ideas or works, treating the AI as a sounding board for feedback. “You shouldn’t accept all of AI’s feedback at face value, but if you find something useful, it can make a real difference,” Ackerman says.
• Adopt new perspectives: Prompting AI to take on different roles or answer from various viewpoints can yield fresh insights you might not reach on your own.
• Stay curious: Ackerman also highlights the value of curiosity and exploration when working with AI. “Real progress happens by showing up, experimenting, and investing time and effort,” she notes.

She leaves one final thought:  “It’s fine if we prefer human-made art.  The real mistake is believing machines can’t be just as creative as we are.” So even if we do not value machine creativity as highly as our own, Ackerman argues, “we should consider how it can help improve society, and how that talent can lift us all.”