Teaching about the Future of Creativity

Generative artificial intelligence (GenAI) can produce art, videos, stories and other outputs normally associated with creativity. But whether that makes it creative is debatable. Two new Common Core courses, one led by Kathryn Goldstein, Lecturer in the Faculty of Arts, and the other by Professor Bo Dai of the HKU Musketeers Foundation Institute of Data Science, are pushing students to think more deeply about a technology that is reshaping their future. Tellingly, the professors themselves are far from in accord on the topic.

Ms Goldstein, in her course ‘Who Made This? Authorship and Art in the Age of AI’, takes a guarded view. She presents creativity as both a series of choices and a social experience, and she explores whether past technological innovations contain insights into whether AI can meet this definition.

“We look at photography, which in the beginning was seen more as a form of non-creative documentation of history or scientific findings. Gradually, it came to be seen as a real art form, so I ask my class whether AI can be viewed similarly – whether it expands creative choices or steers us towards the path of least resistance,” she said.

They soon find AI is not really designed to promote choice. “Most people want AI to answer questions and complete tasks, not help them make creative choices. Even if you ask AI to, say, have a conversation with you about a paper you are writing, it will try everything it can to write the paper for you. This is the way the models are designed and marketed, to give people what they want,” she said.

To demonstrate the choice limitations, she asked students to participate in an alternative uses test – commonly used by computer scientists to benchmark AI creativity – by suggesting different uses for an HKU hoodie, then seeking ideas from AI. The students’ own ideas were more varied and unusual, such as using it as a slingshot, while they found different AI models all gave similar answers, such as using the hoodie as a belt or for expressing school spirit. Most of the students also found AI frustrating for creative work, even when it could improve individual performance.

“On average, AI may generally outperform the average person on creative tasks, but it’s a technological feature of these models that they arrive at similar, average understanding of different concepts and words, whereas people come up with strange and unique responses. If we rely on them as a society, we are going to be led towards less creativity,” Ms Goldstein said.

The social experience was also undermined by the uncanny feeling evoked upon learning that content, such as music or videos, had been created by AI, not people, she added.

Best and worst of times

But this cautious perspective is not the only view on AI and creativity. Professor Dai takes a more optimistic view in his course ‘GenAI and the Future of Creativity’, because he thinks AI offers opportunities for more people to be creative.

“Someone without much expertise in AI technology or creative skills can find unique uses for AI, for instance, turning their holiday photos into a Disney-style animation,” he said. “In that sense, it is the best of times because everyone can shine.”

He also thinks AI has some features that qualify as creative – such as the ability to find connections and generate new ideas from that. Take hallucinations, where AI generates different (and sometimes wrong) answers from identical prompts.

“People who see AI as a tool are very annoyed by this result and want to diminish hallucinations. But those who see GenAI as an assistant, or a source of inspiration, like the hallucinations and want to brainstorm together with AI,” he said. “You can hardly describe GenAI as just a tool if it gives you something you did not think of yourself.”

His students have designed new applications for GenAI and their results – human-led but AI-enabled – show AI’s potential in collaboration. They turned a story into a video with music, created a variety of quizzes based on teaching materials and lecture notes, and built a digital avatar from audio recordings and images of a deceased relative to ‘converse’ with them again.

Nonetheless, Professor Dai says the current state of AI is also the worst of times because of the lack of governance and clear boundaries for AI use and the proliferation of fake content. Technical limitations also persist. “Current AI just learns in a passive way – it does not know what it does not know. It also never asks questions,” he said, echoing Professor Goldstein’s concerns.

“I think it is impossible that AI will replace people,” he added, “but the people who know how to use it will have the advantage. Since there is no standard answer on how to use it yet, everybody can be creative.”