Reframing AI: Becoming Centaur
Almost everyone is confused about AI, because of the many conflicting claims about it and the speed at which it keeps changing. Karim clearly explains the scientific basis underlying the current models, how they work, what they can and cannot do, and what that means for your workflow and the value proposition of your team and sector. Participants leave with a working understanding of AI models, two fundamental paradigm shifts in the way we will work in the future, and two design questions to design their own customised interaction with AI models for their workflow. The session works on several levels, from beginners who get introduced to the key points to advanced users challenged to expand their ambition.
Key takeaways
Understanding the models. What the different types of AI can and cannot do, and what you can build with them.
Rethinking your workflow and value proposition. Design how you interact with AI models, and how to build your own team of assistants and support programmes.
Applying a beginner’s mind. The encouragement to experiment as much as possible with the models available today.
Interaction
This is an interactive session in which Karim puts several questions to the audience. Participants map their current AI use, design the AI collaborator they would want, and decide what they could build to support their workflow. Throughout, Karim demonstrates interactive AI tools he has built himself. Q&A at the end.
Session overview
In 1997 the world chess champion Garry Kasparov lost to a computer. A year later he set up a new kind of tournament, featuring a combined team of a human and a computer, and found that this combination beat both the best humans and the best machines. He called this a centaur: a mythical half-horse, half-human figure. We are all becoming AI-assisted centaurs now, and the question for every team is how to reframe its current workflow and value proposition into an AI-assisted one.
To do that, we need to understand what the models actually do. Karim explains three kinds of AI in plain terms. Traditional AI is symbolic and deterministic: rules, if-then, current software programming. Generative AI is probabilistic: it matches patterns in enormous datasets and produces new text, images, and code, which is why it is creative and makes mistakes. Reinforcement learning, which produced mind-blowing breakthroughs in gaming and biology, is used with both. Understanding this is not academic: it tells us how to use the models in our work. Karim shows interactive AI tools he has built to illustrate this, using his own work and examples from different sectors. Participants see what can be built with AI and how it reframes what a job consists of.
The second part is about reframing your own workflow, at three levels: automation, augmentation, and reframing. Two paradigm shifts run through it. The first is from monologue to dialogue, from working alone to working with collaborators you define yourself: this is what your interaction with a GPT or an agent really is, and it fundamentally changes how you work. Participants reflect on how they would make that shift in their current role and see how to set up a system to build their own team. The second shift is from standard to customised. Our learning, our software, and our tools are all being built just for us. The consequence is simple: stop doing things yourself and instead ask a carefully designed agent to do them for you. We examine how to provide maximum context and training to design such agents.
This session works across all levels: from beginners still struggling to understand AI to people who already build their own AI agents. Beginners are told which key points really matter and how to get started; advanced participants are encouraged to become more ambitious with their technical knowledge. Everyone leaves with a clearer picture of how these models work, a sharper sense of how their own work is going to change, and the encouragement to start building.