Antoine Bordes (Helsing) & Gaël Varoquaux (INRIA): Where Public Research Meets Innovation

Event Context & Growth:

• The panel "AI Futures: Public Research Meets Private Innovation" took place at RAISE, an event that has grown from approximately eight exhibiting companies last year to a significantly larger scale this year.
• The growth mirrors parallel trends in AI development, with public research bodies achieving breakthroughs, private innovation accelerating, and capital deployment reaching record levels.
• A primary structural challenge identified is the "handoff" between public and private sectors, which often suffers from friction and inefficiency.

Panelist Introductions & Expertise:

• Isabelle Riel (PSL University/PRISME AI): Head of the Paris School of AI research institution, focusing on public research and knowledge creation.
• Kaël Varroco (INRIA): Research director in machine learning/AI and co-founder of startups including "Probable."
• Joëlle Barral (Google DeepMind): Director of research and engineering leading foundational research teams across Europe and the US, sponsoring all healthcare-related work.
• Antoine Borde (Helsing): Chief Scientist of a European defense AI company founded four years ago, previously at Meta Fair.

10-Year Forecasts (2035) & Vision:

• Antoine Borde: Hopes for the freedom to use diverse AI models and infrastructure without monopolistic constraints; emphasizes the need for business-creating flexibility.
• Joëlle Barral: Predicts generative AI will become a universal healthcare assistant within 10 years, comparable to calling a family physician for triage, democratizing access previously limited to those with medical connections.
• Isabelle Riel: Foresees "rebound effects" where technology adoption increases rather than decreases administrative burdens (e.g., bookkeeping); believes societal change will depend on the equilibrium between technology and social adaptation.
• Kaël Varroco: Argues public research cannot predict specific outcomes without stifling creativity; predicts the emergence of technologies currently unimaginable, citing the 2015-to-today shift from military construction concepts to drones as an example of "dreaming behind the frontier."

Healthcare & Medical Innovation Trajectory:

• Surgical Robotics (2015 vs. Now): Early projects lacked clarity on data connectivity; 10 years later, operations now leverage data from prior similar procedures to inform new surgeries.
• Large Language Models (LLMs): Unexpectedly accelerated adoption in healthcare; MedPaLM was fine-tuned on medical corpora to generate radiologist-level reports from X-rays, achieving research-grade results faster than anticipated.
• Commercialization Gap: While research prototypes exist, transitioning from 0% to 99% reliability required for broad clinical adoption takes significantly less time than previously envisioned.
• Education & Skills: Medical training timelines (10 years) mean the workforce entering practice today will operate in the AI-integrated future being described.

Defense Innovation & Helsing's Methodology:

• Investment Strategy: Utilizes self-funding to test high-impact defense technologies before procurement contracts exist, aligning technological vision with stakeholder gaps (military/government).
• Product Lifecycle: Shifts focus from purchasing static hardware to "drone programs" with update cycles every 3–6 months to maintain relevance against rapidly evolving tech.
• Data & Quality Assurance: Emphasizes a "data flywheel" where collected flight data and simulated data are used to train and statistically validate models against strict military quality standards.
• Testing Loops: Requires rigorous validation using both real-world data and AI-generated simulations to guarantee model performance boundaries before deployment.

Public Research vs. Private Innovation Dynamics:

• Public Sector Role: Serves as the "beginning of the pipeline" by creating fundamental knowledge and educating students/postdocs who act as bridges between academia and industry.
• Knowledge Distillation: Acadia consolidates and "distills" vast amounts of research (e.g., New York's papers), preventing talent from being overwhelmed by unstructured data.
• Long-Term Stability: Universities provide a necessary "balancing force" against the short-term, product-focused pressure of private markets, focusing on longevity and deep scientific inquiry.
• Interdisciplinary Impact: Generative AI is expected to revolutionize not just CS, but chemistry, physics, and humanities (e.g., redefining historical research through vast knowledge searches).
• Open Source Ecosystem: Google DeepMind released "Gemma" (open weights) to allow academia and broader society to leverage and build upon foundational models.

Funding & Administrative Challenges:

• Funding Volume vs. Velocity: While France and Europe have committed significant capital, funding is often restricted to high-profile topics, leaving other domains (e.g., biology) under-resourced.
• Administrative Bottlenecks: Excessive bureaucratic rules prevent institutions from spending available funds efficiently; the speed of private competition outpaces public administrative agility.
• Sovereign AI: The concept of "Sovereign AI" highlights the tension between the need for independent data sovereignty and the difficulty of managing complex power structures within regulations like the AI Act.

Talent, Boundaries, & Ecosystem Friction:

• Boundaries as Missed Opportunities: Progress often occurs at the intersections of fields, economic actors, and research/innovation silos; current boundaries cause energy and information loss.
• The "One-Way Door": Moving from academia to private sector in Europe often results in academic exile, whereas the US culture supports frequent back-and-forth movement (e.g., Fei-Fei Li).
• Talent Retention: Europe risks losing top professors and researchers to countries offering better deals and more innovation-friendly environments.
• Inspiration Gap: Young people lack the connection between foundational skills (math, physics, biology) and tangible impact (healthcare, climate), hindering motivation for long-term academic tracks.
• Power Dynamics: Unaddressed power struggles (e.g., medical doctors fearing AI loss of control) and bureaucratic power structures are significant, unspoken blockers to adoption.

Strategic Outlook for Europe:

• Product vs. Technology: Europe often prioritizes "technology" over "product"; success requires building scalable products that address global markets rather than just the domestic French market.
• Market Scaling: Scaling from California to the US is easier than scaling from a single European country to the entire continent due to fragmentation; products must be designed for global scale from inception.
• Success Cases: Mistral and others demonstrate that fundraising and success are possible in France, but the focus must shift from funding narratives to product-market fit and customer understanding.
• Circular Innovation Model: The ideal ecosystem involves public research creating knowledge, private sector commercializing it, and results being returned to the public domain for further iteration and education.