For a discussion that takes a more nuanced and practical view of how AI is set to impact work, Stanford Online recently brought together Stanford d.school instructor Perry Klebahn and Sebastian Krakowski, an assistant professor at the Stockholm School of Economics (House of Innovation) and digital fellow at the Stanford Digital Economy Lab whose work focuses on human-algorithm interaction, for an interactive webinar discussion.
While acknowledging risks and uncertainties, Klebahn and Krakowski aimed to reframe the conversation from one of centering on either the dangers or opportunities of full automation to one discussing the possibilities of AI augmentation of the creative process. They also offered practical advice for how creative professionals can start using AI tools to enrich and accelerate their work right now.
An option engine, not an answer engine
After a brief but essential overview of how generative AI tools like ChatGPT and Stable Diffusion work as well as the major risks users should be aware of (if you’d like more information on these topics, a more in-depth primer is available here), Klebahn and Krakowski dive into the heart of their talk, arguing that generative AI can be a powerful tool for ideation in the creative industries.
“It's really to get a head start in your thinking when you're looking at that blank page. So concept generation, visualizing different concepts, ideas, and new ways that maybe you wouldn't have thought of, or at least getting options, alternatives to consider,” Krakowski says, describing this as one of the best current use cases for new AI tools for creative professionals including writers, marketers, product teams, designers, and artists.
“Think about it as an option engine, not an answer engine,” suggests Klebahn. With its well known tendency to “hallucinate” and its lack of human discernment, ChatGPT cannot be relied on to create final products, but it is a fantastic way to push yourself to explore new creative avenues.
Putting this advice into practice
Maximizing these technologies' potential as brainstorming partners requires new ways of thinking and knowledge of best practices in eliciting the highest quality responses from the tools. The essential first step is accepting that, as Krakowski puts it quoting chess grandmaster Gary Kasparov, “if you can't beat them, then join them.”
Together with colleagues in Switzerland and Denmark, Krakowski conducted a study of chess players that showed those who performed best were those who learned to lean on chess computers, incorporating the technology into their game play. “We found that the old kind of skills of playing chess no longer determine who wins. Instead, it's who can leverage the computer, the algorithms, best,” Krakowski insists.
Chess is one narrow domain, but Klebahn and Krakowki argue the general principle applies more broadly. In a world where these AI tools are available, those who thrive and succeed are those who neither fully reject them nor expect them to replace human judgment altogether. The highest performers take a blended approach.
“AI is not able to be curious. It's not able to be surprised. It's not able to infer meaning,” says Krakowski, “but it is relentlessly able to generate material and evaluate material if you direct it.” And when it comes to creativity, the best final output is strongly correlated with the volume of initial ideas generated. AI helps you generate more and more varied ideas, while humans remain responsible for evaluating and editing and selecting among them.
How exactly can you push a tool like ChatGPT to generate the most and the most useful ideas for consideration if you’re a creative? Klebahn and Krakowki offered several practical tips drawn from their own work:
Cross domain prompting. This is a technique Klebahn has been experimenting with in his Stanford Launchpad class, which helps engineering students develop startup ideas. The idea is to ask generative AI tools to take models or frameworks from one area or industry and apply them to another area or industry. This might look something like “How would you apply the business model of eBay to short-term leases of office space?” but doing this many times through using different domains. Many responses will be bizarre and useless, but some may generate a creative spark.
The process works by supercharging one of the chief mechanisms of human creativity — combining ideas from different domains. “What happens in our head is we pull up all of this information related to whatever experiences we've had, and we put them together with our idea and we generate new ideas. The only limitation is it's limited to the knowledge I have. The computer is relentless. It can keep going,” explains Klebahn.
Ask for different story structures. Generative AI can create mashups of different domains of knowledge, but it can also apply different storytelling structures and conventions to the same basic information. We’ve all seen funny examples where someone asks ChatGPT to write a letter to their landlord in the form of a Shakespearean sonnet, for example.
But this approach can also be useful for business brainstorming. “This is helpful with writing copy for a website. This is helpful to communicate your idea to an investor,” says Klebahn who gives examples to try such as instructing the AI to, write about your company as if it's a haiku; write about your company as if it's a review on Amazon; write about your company as if it's a New York Times obituary about you as the founder. Again, most of all this material will not end up getting used. The aim is to spur creatives to think in new directions. Human taste and judgment remain essential.
Noticing patterns with AI. For creatives that utilize a lot of customer interviews and who do a lot of sentiment analysis on those interactions, AI can also be useful for sifting through transcripts and interviews and pulling out instances where customers are expressing particular emotions, such as anxiety, tension, or engagement that might be worth a human examining more closely. This saves creative teams from a lot of tiresome review. “It was an insane amount of effort for them to review all of it in the past as opposed to have something go through and have them sort of go back through their reasoning,” explains Klebahn.
Low effort in, low quality out
In each of these use cases, good outcomes depend on good “prompt engineering.” The more detailed information you give AI tools and the more you push them for more creative, detailed, or diverse responses, the more useful you will find them, both experts insist.
“Low effort prompts means low quality output,” says Krakowski. “So we actually need to help it along a little bit and, and give it an indication of what you're looking for. Then you can get much, much more useful output.”
But assuming you’re willing to familiarize yourself with best practices for using generative AI to improve your creative practice and then put in that effort to optimize your prompts, AI is already a profoundly useful tool for creative professionals to expand the palette of options and ideas they consider. Rather than thinking of these technologies as a silver bullet or a terrifying threat, it is more useful to see them as a help tool to augment ideation and spur creativity.
If you are interested in learning more about how to generate ideas and increase your creative output, check out Stanford Online’s Creativity and Design Thinking Program for much more information and opportunities for further study.
