Summary
AI is extraordinary at finding answers. But who finds the problem?
In the 4th of our Beyond Human series, Anastassia and Matthias Röder go after a question: can an AI formulate a genuinely new problem, or is it forever confined to solving the ones we hand it?
Dr. Matthias Röder is a music and technology strategist based in Salzburg, and one of the most consistently original thinkers on creativity and machines that we know.
He is a former Managing Director of the Eliette and Herbert von Karajan Institute and a board member of the Karajan Foundation, as well as a trustee of the Mozarteum Foundation. He is co-founder and managing partner of The Mindshift, a consultancy on creative leadership and innovation strategy.
Matthias directed the Beethoven X project, the AI-assisted completion of Beethoven's Tenth Symphony, work that earned an Effie Bronze. He founded the Karajan Music Tech Conference in 2017 and launched the Classical Music Hack Day series in 2013. He holds a PhD in music from Harvard University and is an alumnus of the Mozarteum University Salzburg. His work has been recognized with the "Game Changer" award of the Salzburg Chamber of Commerce.
Chapters
00:01 Welcome, and the question behind the episode: can AI formulate problems, not only answer them?
01:48 In music, the problem is the composition — why framing the question is the hardest creative act
03:17 35 million papers, 4,000 a day: PubMed, AlphaFold, and what no human can read
05:58 GNoME and 2.2 million crystals — when the hypothesis machine outruns the laboratory
09:09 World models, physical intuition, and the questions AI does not know to ask
11:36 A map of all unsolved problems: knowledge graphs as an orchestration engine
15:16 Progress without consensus — why agreement may be the most expensive thing we do
17:24 Theory of mind, Belle and Rock Experiment, and why your chatbot keeps telling you that you are a genius
24:20 Context contamination, hallucinations, and professional AI hygiene
30:50 Organizational elasticity, the 2×2 of known and unknown, and three papers worth your time
Papers mentioned in the episode:
Towards Scientific Discovery with Generative AI: Progress, Opportunities and Challenges — a survey of AI across the full scientific discovery cycle, from literature analysis and brainstorming through theory refinement, experimental design and data-driven discovery.
The Future of Fundamental Science Led by Generative Closed-Loop Artificial Intelligence — an ambitious argument that AI is approaching the ability to close the scientific loop autonomously, from hypothesis generation through experimental design to validation.
Machine Learning as a Tool for Hypothesis Generation — a procedure for generating hypotheses from high-dimensional behavioural data and testing them on held-out human subjects, grounded in economics and behavioural science.
Also referenced: DeepMind's AlphaFold and GNoME; AI Feynman and the rediscovery of fundamental physics equations from data alone; Yann LeCun on world models; Jorge Luis Borges, The Library of Babel.
Hyperlinks:
Anastassia Lauterbach - LinkedIn
First Public Reading, Romy, Roby and the Secrets of Sleep (1/3)
First Public Reading, Romy, Roby and the Secrets of Sleep (2/3)
First Public Reading, Romy, Roby and the Secrets of Sleep (3/3)
