Most of what business leaders hear about AI and creativity is either breathlessly optimistic or quietly dismissive. Dr Todd Lubart is neither.
As Professor of Psychology at Université Paris Cité and one of the world’s leading researchers in the science of creativity, Todd has spent his career building rigorous frameworks for understanding, measuring, and developing creative potential. His recent work on Cyber-Creativity maps how AI is reshaping every dimension of the creative process, and what that actually means for the organisations trying to use it. In a wide-ranging conversation on the Arkaro Insights podcast, we explored where AI genuinely adds value in innovation, where it falls short, and what business leaders should do differently starting today.
Executive Summary
Key insights from Todd Lubart’s conversation with Mark Blackwell:
- AI is strong at divergent thinking. This is both its biggest contribution and its most misunderstood limitation.
- The most important skill in an AI-augmented innovation process is not generating ideas. It is asking the right question.
- Pick and Mix prompting across multiple AI systems consistently outperforms asking a single system to do better.
- Problem definition remains a distinctly human advantage, and the research behind it is unambiguous.
- Plagiarism 3.0 is a real and growing risk, and most organisations have no policy to address it.
- AI can be trained to evaluate ideas reliably, but only when humans first design the measurement framework.
- Slop is the hidden cost of careless AI use: output that looks reasonable on the surface but requires as much time to fix as if you had done it yourself.
- How you integrate AI into your innovation process will become part of your organisational culture, whether you design it that way or not.
Mapping the Territory: The Seven Cs
Todd’s starting point is a framework he developed some years ago to map the whole territory of creativity research. He called it the Seven Cs: Creators, Creating, Collaboration, Context, Creations, Consumption, and Curricula. Each C represents a different dimension of the creative process, from who is doing the creating to how their output is taken up and used.
The framework was built around human creativity. What Todd’s recent work has done is take each C and ask a new question: how does AI change this?
The answers are not uniform. Some Cs are barely affected. Others are being fundamentally reshaped. The creating process itself, the way ideas are actually generated, is one of the most significantly changed. Humans tend to search associatively, drawing on experience, emotion, and context. AI systems search statistically, finding patterns across vast datasets and recombining them. It is a genuinely different process, and it produces genuinely different results.
That difference is both the opportunity and the risk.
AI Is Strong at Divergence — and That Changes Everything
The single biggest contribution AI makes to the creative process right now is in divergent thinking, the phase where the goal is to generate as many ideas as possible before narrowing down.
Humans find this phase genuinely difficult. Our brains pull us back towards familiar territory. Creativity techniques were developed specifically to help people break out of those ruts. AI, by contrast, can generate large volumes of ideas very quickly, drawing on a breadth of sources no individual could match.
Todd’s research team, including Florent Vinchon and Dimitris Grammenos, has tested this systematically, asking systems such as ChatGPT the same task a hundred times and mapping the distribution of responses. The results show something instructive: AI ideas cluster around the statistically average, the things most likely to appear on the internet. The genuinely unusual ideas, the kind that might represent a real breakthrough, are rare in AI output and common in the tails of human creative performance.
The implication for business leaders is clear. AI is a powerful tool for getting a large volume of ideas onto the table quickly. It is not a reliable source of the ideas that will change your industry. The human role in divergent thinking has not disappeared. It has shifted, from generating everything to recognising what is genuinely novel among a much larger set of options.
The Most Important Skill Is Asking the Right Question
One of the most consistent findings in Todd’s research is that the quality of AI output is determined largely by the quality of the input. This is not a surprising observation in isolation. What makes it significant is the specific skill it points to: problem definition.
The ability to frame a problem well before attempting to solve it has long been recognised in creativity research as a key predictor of creative performance. John Dewey’s formulation, that a problem well posed is half solved, has been in the literature for a century. What AI has done is make this skill newly visible and newly urgent.
Todd and his colleague Sudapa Chompunuch have studied specialised GPT systems designed specifically to help with problem formulation. Tools such as AhaApple include functions that prompt users to generate multiple ways of framing the same problem before generating solutions. The research on teams that invest time in this way, explored in a recent Arkaro Insights conversation with Roni Reiter-Palmon, consistently shows better outcomes.
This is not a soft observation. It is one of the most practically actionable insights in the field. Before your team opens a chat window, ask: have we actually defined the problem we are trying to solve?
.
AddPick and Mix Beats Try to BeatYour Heading Text Here
One of the more counterintuitive findings from Todd’s recent research concerns what happens when you use multiple AI systems together. The instinct most people have when they get a mediocre output from an AI system is to push back: ask it to do better, be more creative, try harder. Todd’s research, conducted with Dimitris Grammenos, shows that this instinct leads to worse results.
The concept of better is not well defined for an AI system. When asked to improve on its own output, the system tends to produce something similar, or worse. What works significantly better is what Todd calls “Pick and Mix”: presenting multiple systems with a composite of the best outputs from several different AI tools and asking each to select and combine the strongest elements.
The practical implication is straightforward. Rather than pushing one system harder, use Gemini, ChatGPT, Claude, and others in parallel. Pool the outputs. Then ask each system to build on the composite. The results are consistently stronger, and the approach plays to what these systems actually do well, which is recombination rather than origination.
The Four Scenarios: Where This Could Go
Todd and his colleagues, including Florent Vinchon and many others, published what they called the Manifesto for Human-AI Creativity. It sets out four possible scenarios for how human-AI co-creation might develop.
The first, and the one they argue should be favoured, is what they call co-creaition, their intentional spelling, used to signal that AI is genuinely part of the creative act rather than merely a tool alongside it. In this scenario, humans and AI systems work together across the creative process in ways that produce outcomes neither could reach alone.
The second scenario is Plagiarism 3.0. AI makes it easier than ever to pass off generated work as one’s own. Detection tools exist but are imperfect. The temptation is real and, as Todd puts it, humans are a rather lazy species. Most organisations have no policy that adequately addresses this risk.
The third scenario is the human purist position: I did it entirely by myself, the old way. Some creative professionals will make this a point of value, and there is a market for it. But as a default position for an organisation, it is increasingly hard to sustain.
The fourth is shutdown: the belief that since AI can do it, there is no need to worry about developing creative capability at all. Todd is direct about this: the AI systems do not actually do that well. They can produce a genuinely good idea occasionally. But not every day.
AI Can Evaluate Ideas — With the Right Framework
One of the most promising, and most misunderstood, applications of AI in innovation is idea evaluation. Most organisations either over-rely on senior judgement, with all the boss’s pet project biases that entails, or have no systematic evaluation process at all.
Todd’s research shows that AI can perform reasonably well as a judge of ideas, but only when it is given a proper framework to work with. An open-ended request to evaluate a set of ideas produces poor results. Give the system a scoring rubric, with clear criteria and calibrated examples from one to seven, and the picture changes significantly. A panel of AI judges, averaged together, correlates with a panel of human experts at around 0.8, which is a meaningful level of agreement, though not perfect.
The more powerful approach is to train a system on a large database of human-produced and human-scored work in a specific domain. A system trained this way becomes a genuinely useful evaluator for that task. But the critical point is that humans must design the measurement system. Defining what a one looks like and what a seven looks like is itself a creative and judgmental act. It cannot be outsourced.
This matters enormously for sectors like agriculture, food, and chemicals, where the cost of backing the wrong idea through a development pipeline is very high. Rigorous idea evaluation is not bureaucracy. It is risk management.
The Hidden Cost of Slop
Todd’s most practically useful warning for business leaders is about what he calls slop. It is output that looks reasonable on the surface, reads adequately, and passes a cursory glance, but when you actually know the subject, reveals itself to be riddled with errors, gaps, and imprecision that require as much time to fix as if you had done it from scratch.
The organisations most at risk are those where AI adoption has outpaced AI literacy, where people are generating and submitting AI output without the domain expertise to know when it is wrong. The antidote is not to use AI less. It is to ensure that the people using it have enough depth in the subject to evaluate what comes back.
This is a direct argument for investing in expertise rather than assuming AI can substitute for it.
What to Do on Monday Morning
Todd’s practical advice to business leaders is to resist the temptation to standardise too early. The organisations most likely to get genuine value from AI in innovation are those that approach it as an exploratory process: trying different tools at different stages, building a team of AI systems rather than relying on one, and staying genuinely curious about where the value actually is rather than where it is supposed to be.
Before your next innovation workshop, define the problem more carefully than you normally would. Ask the question in five different ways before you start generating solutions. When you do use AI for divergence, use more than one system and combine the outputs using “Pick and Mix” rather than asking any single system to do better. And when it comes to evaluation, invest in building a proper framework. The time it takes will be recovered many times over.
The goal is not to use AI more. It is to use it in the places where it actually adds something you would not otherwise have.
Listen to the full conversation
📺 YouTube: www.youtube.com/@arkaro
🎧 Audio: https://arkaroinsights.buzzsprout.com/
About Dr Todd Lubart
Todd Lubart is Professor of Psychology at Université Paris Cité, where he has spent his career researching the science of creativity. He is the originator of the Investment Theory of Creativity and the developer of the Creative Profiler. His recent work on Cyber-Creativity explores how AI is reshaping every dimension of the creative process. He can be contacted at Todd.Lubart@gmail.com.
Research mentioned in this episode:
- The Manifesto for Human-AI Creativity (with Florent Vinchon and colleagues)
- AI as a Helper: Leveraging Generative AI Tools Across Common Parts of the Creative Process (with Sudapa Chompunuch, PhD)
- Are Eight Chatbots Better Than One? (with Dimitris Grammenos)
Related Reading from Arkaro Insights
On AI and the Workplace
How to Use AI Without It Going Wrong — Ray Eitel-Porter on the governance frameworks organisations need to put in place before AI adoption outpaces their ability to manage the risk. A natural companion to Todd Lubart’s argument about slop and the hidden costs of careless AI use.
Why AI Transformation Fails — and What Leaders Must Do in 90 Days — Charlene Li on why most AI transformations stall not on technology but on leadership. Directly supports Todd’s point that how you integrate AI will become part of your organisational culture.
On Creativity and Innovation
Creative Problem-Solving Starts Before the Brainstorm — Roni Reiter-Palmon on why teams that spend the majority of their time defining the problem consistently outperform those that rush to solutions. The scientific foundation for Todd’s argument about problem framing as the most important AI skill.
Possibility Thinking and the Art of Creative Leadership — Vlad Glaveanu on Slow AI, possibility thinking, and why the most productive relationship with AI may be one in which it asks us questions rather than answers them.
Is Your Organisation Ready to Innovate With AI?
The conversation with Todd Lubart is a reminder that the biggest obstacle to getting value from AI in innovation is rarely the technology. It is the absence of a clear process for using it well.
At Arkaro, we work with leaders in agriculture, food, and chemicals to design innovation processes that are both rigorous and genuinely collaborative. Our do-it-with-you approach connects the technical and the human, ensuring that AI tools are integrated in ways that add real value rather than generating impressive-looking slop.
We don’t just coach — we get on the pitch with you.
Contact Mark Blackwell: mark@arkaro.com
Connect on LinkedIn: Mark Blackwell
Subscribe to the Arkaro Insights newsletter: Arkaro Insights on LinkedIn
💬 Where in your innovation process do you think AI adds the most value right now?
