Dances with Algorithms
Algorithms outshine us in ideation. And that's a good thing.
Last week I came across a recently published working paper suggesting that Large Language Models like ChatGPT-4 might have an edge over humans in generating business ideas (a “working” paper is one that hasn’t been peer-reviewed yet). Is AI on the cusp of overshadowing human creativity?
The idea is not new, mind you. I am sure you’ve seen plenty of examples of Generative AI used creatively. But when academics run controlled experiments, and they confirm what, until then, was just social media hype, it starts to get a bit serious.
I ran dozens, if not hundreds, of business ideation sessions. So I think I have a rough sense of what to expect from an average executive or an MBA student. Out of curiosity, I quickly generated 100 ideas with a moderately sophisticated prompt (as sophisticated as I could get it while walking my dog—I did it quickly on my phone). And, I have to admit, these ideas look pretty darn good. The first time I used GPT-3 for business ideation, about three years ago, it came up with ideas like “The audience in a Sumo match is provided with gloves so it can join in.” These ideas were intriguing but not yet good enough. Three years later, GPT-4-generated ideas are quite impressive! Just look at the video below.
ChatGPT has overtaken humans in business ideation. 🤖🚀
According to the working paper, ChatGPT-4 not only rivals human ideation but often exceeds it, particularly regarding the speed and diversity of its generated ideas. The paper authors write that Generative AI tools are “already significantly better at generating new product ideas than motivated, trained engineering and business students at a highly selective university.” They further state the logical conclusion: "The order of magnitude advantage in productivity itself is nearly insurmountable, and the higher quality of the best ideas further adds to the advantage of the LLM." Yet, instead of viewing it as a march towards an AI-dominated dystopia, it's paramount to grasp the deeper nuances.
But humans can learn to be as good as an algorithm. 😏💡
The idea that more is better in ideation makes sense, not just in algorithmic ideation. At QUT Business School, my colleagues and I have introduced a set of 'Ideation Lenses' to help our students generate diverse ideas as quickly as possible. The students learn the basic principles of idea generation and then are handed a deck of cards, each with a unique perspective to approach a problem. For instance:
Derive: Steal a leaf from other industries' books. How would an airline price your products? How would Google run your operation?
Utilise: Unlock the potential within. Can your existing assets be repurposed for new offerings or improvements?
Enhance: Reflect on your current processes. Can they be consolidated, eliminated, or reimagined to increase efficiency?
Equipped with these cards or remembering the prompts, our students could suddenly generate dozens of ideas within minutes. Ideation lenses are more than a method—using them is a strategy to systematize and amplify innovation.
You can see why I am quite intrigued by the developments in algorithmic ideation. You could argue that my work so far focused on helping humans become more like an algorithm when ideating. And at the same time, generative AI has emerged with the flipside of it—algorithms that become more like humans when ideating.
It’s not a duel. It’s a dance. 🕺💫
Is the dawn of AI-driven ideation signalling the beginning of a decline in human creativity in business? I don’t think so. Why?
AI-driven ideation is good on paper, in academic experiments, and in short youtube videos. But when the rubber hits the road, in business, there are challenges that pure algorithmic ideation cannot overcome. The three that immediately come to mind are the buy-in dilemma, the art of prompting, and quality assurance.
Buy-in Dilemma: Innovation thrives on involvement. If an idea stems from an algorithm, will human teams be as passionate about its execution?
The Art of Prompting: Large language models shine with the right prompts, which, in its essence, remains an art mastered by humans. The quality of the prompt directly influences the brilliance of the idea.
Quality Assurance: Presently, algorithms can't assess aspects like product-market fit. Such intricate evaluations remain a human prerogative.
And this is where humans can shine. In tandem with a generative AI tool, they can make ideas more human, fine-tune the algorithms to ensure the right mix of ideas, and continually improve such “human-and-algorithm” systems.
Human Touch: Use algorithms as an ideation aid. Let humans vet, refine, and give a face to these AI-generated ideas before presenting them.
Maximizing LLMs: Train your teams to extract the finest from LLMs. Strive for a blend of ideas: 80% that resonate with human-like practicality and 20% that push the boundaries, reminiscent of Arthur C. Clarke's words: "The only way of discovering the limits of the possible is to venture a little way past them into the impossible."
Feedback Loops: Establish workflows that sift, refine, and validate AI-driven ideas, fostering a culture of continuous improvement.
The intersection of human creativity and algorithmic ideation is not a battlefield but rather a dance floor where both partners lead and follow in turns. While AI has made significant strides in ideation, suggesting a potential transformation in business innovation, its effectiveness is ultimately shaped by human insight, guidance, and empathy. And if we dance it well, we will benefit from having even more good ideas than we could have ever hoped for.
This is not the end. The dance is only just starting!