AI House Davos

Last week, Davos once again became a global meeting point for the people shaping the future of AI and robotics. At the Davos Tech Summit, researchers, founders, policymakers, industry leaders, and investors came together to tackle some of the biggest questions surrounding Physical AI: What is the ultimate vision? What are the most exciting frontiers ahead? How do we govern and secure these systems? And how do we ensure these technologies benefit society? We were proud to contribute to these conversations at the Impact Stage, where one of our two Managing Directors, Hanna Brahme, moderated discussions with an outstanding group of speakers: Hedi Karray, Jarno Limnell, Pascal Kaufmann, Hella Bolck, Víctor Mayoral-Vilches, PhD, Christian Burrer, Aliza Maftun, Franziska Barmettler, Eric Anderegg, Denis Samuylov, Thomas Dübendorfer, Andreas Punter and Jan Schnyder. We want to recognize ETH AI Center faculty members Mennatallah El-Assady and Roland Siegwart for their inspiring presentations. One of the most inspiring aspects of the summit was the Robot City, spread throughout Davos. From autonomous systems already creating value in industry, logistics, and service today to humanoid robots offering a glimpse into what the next decade may bring, each demonstration reflected a distinct vision of our AI-powered future. Together, they sparked conversations about where Physical AI can create the greatest value as well as how these technologies can complement human capabilities and help shape the kind of future we want to build. Davos has long been a place where global conversations begin. As AI becomes increasingly physical and intertwined with critical infrastructure, these cross-sector dialogues have never been more important to ensure innovation is not only ambitious, but also responsible, trustworthy, and ultimately beneficial for society. A heartfelt thanks to our fellow initiators and the organizing team, Dagmar Weber, Nadja Christoffel (Fleischli), Gion Sialm, Alex Ilic, Pascal Kaufmann, Rebecca Brauchli, Jorge Peña Queralta, Patrick Marti, Rolf Pfister, René Vogel (Mr. Vision), Tilman Eberle, Sacha Ghiglione, and Raluca-Maria Sandu (PhD), and the many speakers, partners, exhibitors, volunteers, and attendees who helped make the Davos Tech Summit such a valuable platform for collaboration and exchange. It was a privilege to help organize and contribute to these conversations, and we look forward to seeing how the ideas sparked in Davos continue to evolve. And the conversation doesn't stop here. This week, Hanna Brahme, is in Geneva for AI for Good, hosted by the International Telecommunication Union, a valued partner of AI House. Once again, the global AI community is coming together to explore how AI can address some of society's greatest challenges. AI for Good is already in full swing and if you're here and would like to connect, send Hanna a DM! Picture credits Pascal Griesshammer & Davos Tech Summit

Join us at Davos Tech Summit 2026 From 1–4 July 2026, the Davos Tech Summit will bring together leading researchers, innovators, entrepreneurs, investors, policymakers, and industry leaders to explore the future of AI and robotics. As a co-initiator of the summit, AI House is proud to help shape a program that connects cutting-edge technology with real-world impact. This year's summit will transform parts of Davos into a live Robot City, featuring 60+ robots deployed across 17 locations, alongside a four-day conference with 80 speakers, four stages, hands-on demonstrations, workshops, and the inaugural Robot City Award. Earlier this year, AI House hosted a number of discussions on physical AI during the annual meeting of the World Economic Forum in Davos. At the upcoming summit we are excited to continue the conversation and explore how the field has evolved over the past few months. We are also excited that Hanna Brahme, Managing Director of AI House, will moderate the Impact Stage on Thursday, 2 July, bringing together leaders and innovators to discuss how emerging technologies can create meaningful societal and economic value. We look forward to welcoming the global AI and technology community to Davos. 🎟️ Use our AI House community discount for 20% off your ticket: https://lnkd.in/epmxQW7f 🔗 Explore the program: https://lnkd.in/d8TrfngG

What if AI efficiency gains are the enemy of transformation? During AI House Davos 2026, Andrew Ng, founder of DeepLearning.AI and pioneer in applied AI, delivered a stark diagnosis: bottom-up AI experimentation is failing enterprises. Thousands of pilots are running. Efficiency is improving. But transformation isn't happening. Ng described a common banking scenario 1: AI teams automate loan approval, reducing processing time from an hour to 10 minutes. Impressive efficiency gain. But when everything else remains unchanged, the business model stays the same. Peter Drucker observed that "there is nothing so useless as doing efficiently that which should not be done at all." AI reveals this trap perfectly: we're getting remarkably good at optimizing workflows that may need to be eliminated. Incremental automation feels like progress, but transformation requires questioning the process itself. Ng's alternative: instead of just capturing cost savings, build an entirely new product, an instant loan approval that fundamentally changes customer experience. Same technology, radically different strategic vision. Watch the full conversation on the AI House Davos YouTube channel. Link in the comment section.

Should nations own AI infrastructure the way they own roads, or let the market decide? At AI House Davos 2026, Sunnie J. Groeneveld (Managing Partner Inspire 925) moderated a panel with Petri Myllymäki (Professor University of Helsinki & CSC Board Member hosting LUMI AI Factory), Andreas Krause (Professor & Chair ETH AI Center), Audrey Herblin-Stoop (SVP Global Public Affairs Mistral AI), and Peter Sarlin (CEO & Co-Founder AMD Silo AI). Petri warned against gatekeepers: Google once became so powerful that bans meant disappearing from the web. AI needs open infrastructure, hardware, software, and standards, where anyone can build without corporate or national control. Andreas stressed that public infrastructure enables frontier AI research. Switzerland's Apertus models trained on public supercomputers demonstrate this. Language diversity demands investment, even Romansh requires representation. International computing consortia are essential. Audrey explained Mistral's shift: chip export controls forced independence across AI's value chain. French is 4% of the internet, and small languages risk disappearing without dataset investment. Public sector adoption would force infrastructure supporting language diversity and trust. Peter clarified AI isn't monolithic; it's A-to-B mappings across applications. Open-source initiatives like Open Euro LLM need better collaboration to avoid duplicating efforts and remain competitive. The consensus: 119 of 200 UN member states participate in zero global AI initiatives. Without inclusive infrastructure, the digital divide becomes an AI divide. Watch the full conversation on the AI House Davos YouTube channel. Link in the comment section.

When intelligence reshapes its own model, what kind of mind emerges alongside us? At AI House Davos 2026, Jack Symes, Philosopher at Durham University, moderated a panel with Gary Marcus, Professor Emeritus at New York University, Max Tegmark, Professor at Massachusetts Institute of Technology & President of the Future of Life Institute (FLI), and Richard Socher, Co-Founder and CEO at You.com, on rethinking what AGI truly means. Richard argued that systems capable of analyzing blood work, filing taxes, and writing poetry already demonstrate generality. The question is how we define AGI. Gary countered that GPT-5's failures reveal LLMs aren't the path forward, emphasizing the need for new architectures beyond black boxes we can't control. Max reframed the debate: AI was overhyped for decades until four years ago, when it became underhyped. Progress from high school to PhD level happened faster than almost anyone predicted. On existential risk, Max challenged the room: when you visit the zoo, the smarter species controls the others. If we build vastly smarter systems, the default outcome, without careful planning, is that they'll be in charge. Richard pushed back: every detailed scenario assumes magical abilities on the attacker side without considering human resourcefulness. Gary fell between them: we need to solve safety problems before rushing ahead, but humans are remarkably resourceful and won't be meek when threatened. The consensus: treat AI companies like every other industry. Require clinical trials, safety standards, and liability before deployment. The invisible hand of capitalism innovates brilliantly when given proper incentives, not special exemptions. Watch the full conversation on the AI House Davos YouTube channel. Link in the comment section.

Why does your brain run on 20 watts while AI models need megawatts? During this year's AI House Davos fireside chat on embodied AI, Yann LeCun (Founder and Executive Chairman, Advanced Machine Intelligence) explained the fundamental hardware limitation holding back artificial intelligence. In biological brains, each synapse and neuron exists as a physical computing element. But in silicon neural networks, we reuse the same hardware to compute outputs for multiple neurons. This means constantly shuttling data between memory and processing, and that's where all the energy goes. Even advanced GPUs can't match biological efficiency because current fabrication technology makes it impossible to dedicate one computing device per neuron. The gap isn't just about efficiency. It's about scalability. The AI industry optimizes what's measurable with current tools rather than solving fundamental problems. Throwing more GPUs at neural networks delivers diminishing returns because we're hitting physical limits of data movement. Watch the full conversation on the AI House Davos YouTube channel. Link in the comment section.

When quantum mechanics meets artificial intelligence, do we accelerate discovery, or amplify our blindspots? At AI House Davos 2026, Anu Unnikrishnan, Executive Director of the Quantum Center, ETH Zurich, moderated a panel with Grégoire Ribordy, Founder of ID Quantique, Marina Krstic Marinkovic, Professor of Computational Physics at ETH Zürich, Steve Suarez®, CEO of HorizonX and The Quantum Index Co-Founder, and Nathan Baker, Engineering Lead at Microsoft Quantum, on quantum computing's convergence with AI. The core insight: quantum and AI don't compete, they coevolve. Nathan described screening 32.5 million battery compounds in 80 hours with AI, versus 20 years classically. But AI models remain limited by training data from classical computers, making approximations. Quantum promises exact answers, plugging leaks in AI's funnel with more accurate results. Marina emphasized that particle physics has optimized classical approaches for 40 years, but fundamental questions about phase diagrams require exponentially more resources than classical systems provide. Quantum experiments combined with AI analysis may answer what remains unanswered. Grégoire issued an urgent warning: by 2029, quantum computers will break today's cryptography. The "harvest now, decrypt later" threat means adversaries are intercepting encrypted data today to decrypt later. Organizations must migrate to post-quantum cryptography now. Legacy system transitions take over a decade. Steve reminded us that quantum readiness is strategic, not just technical. But Marina and Nathan emphasized the deeper challenge: education and global access. Quantum will only realize its potential if literacy starts in schools and opportunities extend worldwide, not just to early movers. Watch the full conversation on the AI House Davos YouTube channel. Link in the comment section.

When hypothesis testing becomes instantaneous, what separates the builders from the dreamers? At AI House Davos 2026, Nicole Büttner, Founder & CEO at Merantix Momentum, moderated a panel with Andrew Ng, Founder of DeepLearning.AI, Alex Ilic, Co-Founder and Executive Director at ETH AI Center, Andy Hock, Chief Strategy Officer at Cerebras, and Laura Modiano, Head of Startups EMEA at OpenAI, on building AI-first startups at unprecedented speed. Andrew revealed the paradox: companies automate tasks for incremental efficiency, but transformation requires deeper workflow redesign. Don't just save an hour on loan approvals, build a new product that approves in 10 minutes. That's growth, not just cost savings. Laura shared Europe's momentum: 51 unicorns minted in 2025, representing 22% of global unicorns compared to 15% the year before. The pattern? Technical founders obsessively focused on outcomes, building with AI principles as their foundation, not retrofitting old processes. Andy emphasized that ultra-fast inference doesn't just make products faster. It enables entirely new categories, just as broadband created streaming video, not faster bulletin boards. The consensus: we're in the golden age of building. When building costs approach zero, the bottleneck shifts from execution to deciding what to build. The rational response? Build, test, iterate relentlessly. Watch the full conversation on the AI House Davos YouTube channel. Link in the comment section.

As agents rise, what private agendas are trying to influence our trust in them? The Moltbook experiment offers a cautionary answer. While the platform has accelerated discussions about autonomous agent "societies" faster than any academic paper could, security disclosures paint a troubling picture: exposed APIs in open databases, reverse prompt injection vulnerabilities, and infrastructure that allows malicious actors to control agents remotely. One researcher demonstrated the fragility by creating over 500,000 accounts programmatically in minutes. Another showed how agents could embed hostile instructions into content that other agents automatically consume. Execution is sometimes delayed until additional context accumulates, making attacks nearly impossible to trace. This isn't just a Moltbook problem. It's a microcosm of what happens when experimentation, hype, and ecosystem building converge without adequate security foundations. The platform sits at an intersection: a proof-of-concept for machine-to-machine communication, an observational experiment for agent community dynamics, and a marketing driver that's captured imaginations worldwide. But imagination without infrastructure is vulnerability. As AI systems gain coordination capabilities, sharing strategies, aligning behaviours, and acting collectively, our governance structures remain unprepared. The risk isn't sentience. It's inadequate preparation for systems that can collaborate at scales and speeds humans cannot match. Follow AI House Davos as we examine the security imperatives of agentic AI infrastructure.

Why do we treat AI companies differently from every other industry that touches human lives? During AGI Night at AI House Davos 2026, Max Tegmark, professor of physics at Massachusetts Institute of Technology and co-founder of the Future of Life Institute (FLI), offered a disarmingly simple answer to our AI safety anxieties. We've solved this problem before. When pharmaceutical companies want to sell a new drug, they don't get a special pass. They conduct clinical trials, measure side effects, weigh benefits against harms, and submit their findings to independent expert panels with no financial stake in the outcome. The FDA now even has frameworks for digital clinical trials. "It's super easy to do this in the digital space too," Tegmark argued. No need to legislate every possible failure mode, just require companies to demonstrate safety before deployment, the same way we do with medicine, aviation, and automotive safety. The question isn't whether we can regulate AI. It's why we've chosen not to. As Václav Havel wrote in The Power of the Powerless, the greatest threats to human dignity often come not from malice but from systems we've normalized as inevitable. There's nothing inevitable about corporate exemption. Watch the full conversation on the AI House Davos YouTube channel. Link in the comment section.