Artificial Intelligence promises transformative solutions for global challenges, yet its humanitarian potential remains unevenly distributed. This talk traces the evolution of AI-driven social impact over more than a decade, from early satellite-based refugee camp planning (Project CaSa, 2014) to cutting-edge medical devices saving newborn lives today.

We'll explore how interdisciplinary collaborations between organizations like the UN, ITU, WFP, CERN, and tech innovators are addressing critical needs: HungerMap Live monitors food insecurity across 95+ countries in near real-time; Nemocare Raksha transforms basic hospital beds into ICU-grade monitoring systems targeting India's ~400,000 annual newborn deaths; MamaMate delivers maternal health guidance in rural Africa offline with local languages.

Beyond healthcare, the presentation examines governance frameworks shaping responsible AI deployment—the UN's Independent International Scientific Panel on AI, the Bletchley Declaration summit series (2023-2027), and Switzerland's nationwide #makeAIright initiative demonstrating sovereign AI development through 240+ events across 33 cities. Through 17,000+ participants at ITU's AI for Good, we see how structured dialogue enables practical innovation.

The journey reveals both progress and persistent gaps: while 140 prototypes emerged from hackathons, scaling remains challenging. Success requires balancing technical excellence with cultural context, offline accessibility, and sustainable funding—lessons from Google/Gates-backed projects and locally-developed solutions alike.

For policymakers, practitioners, and technologists, this session offers concrete case studies showing when AI achieves genuine impact versus when it merely demonstrates capability. The path forward demands neutral platforms like AI House Davos where governments, researchers, and industry co-create inclusive strategies. As UN Secretary-General Guterres notes: "For the first time, every country has a seat on the table."

China has industrialised AI along a path that is "scenario-led and driven by both policy and the market". Its very large market and complete industrial system have sped up AI's adoption in manufacturing, healthcare, transport and finance. This has built a full chain, from computing infrastructure and open-source models to industry solutions, and given rise to industry clusters in the Yangtze River Delta and the Guangdong–Hong Kong–Macao Greater Bay Area.

Experience shows that whether AI industrialisation succeeds depends less on leading in any one technology than on three things: data moving freely as a factor of production, engineering that brings costs down, and the ability to close the commercial loop.

Bottlenecks remain: limited control over core software and hardware, too little high-quality data, and a shortage of people who combine several skills.

In future, "AI+" needs to bring the technology and the real economy closely together, while balancing safety and trustworthiness against open innovation.

Financial information arrives at SIX from global exchanges and other sources around the world, each with different formats, structures and requirements. Transforming these diverse inputs into standardised, trusted data requires deep financial expertise, established processes and is time-intensive.

But what happens when AI-native ways of working meet a mature financial-market infrastructure?

Katarina Granath shares how SIX is deploying agents and redesigning established, financial data workflows, while preserving the quality and trust financial markets depend on, as well as accelerating time-to-market.

In recent years, cardiac electrophysiology digital twins have combined multimodal imaging, ECG and fibrosis data to build patient-specific, multiscale electrophysiological models. GPU acceleration and AI-based inverse methods have made these models faster to build. They are already used to study arrhythmia mechanisms, stratify the risk of sudden cardiac death, predict ablation targets before procedures for atrial fibrillation and ventricular tachycardia, and run virtual ablations. Early clinical validation suggests they can improve procedural strategy and reduce recurrence rates. However, model calibration is difficult, computational cost is high and standardisation is lacking. These problems must be solved before the approach can enter routine clinical practice.

China and the West have access to increasingly similar AI technologies. But they approach innovation, deployment, and business transformation with fundamentally different mindsets. What does this mean for our shared future?

Generative AI is opening new ways to study cardiovascular structure and function from sparse, noisy, and incomplete observations. The field has progressed from image synthesis and data augmentation to the reconstruction of anatomical motion and dynamic physiology. Adversarial networks, diffusion models, and continuous deformation representations have introduced different ways to combine learned distributions with structural and physiological constraints. Together, these developments are bringing image generation closer to quantitative cardiovascular investigation.

A central challenge is establishing whether generated information preserves the biological features and functional relationships that matter for a research question. Published studies reveal both the promise of recovering missing dynamics and the difficulty of validating them across patients, diseases, and acquisition conditions. Our exploratory experiments provide a practical perspective on this challenge, connecting methodological choices with implementation fidelity, data limitations, and the interpretation of experimental results. Evidence must therefore link visual quality with anatomical accuracy, temporal behavior, and meaningful functional measurements.

This perspective points toward generative cardiovascular models that support virtual experiments, connect structure with function, and ultimately help anticipate individual patterns of disease and treatment response.

Artificial intelligence is moving beyond the digital world and into physical applications, creating new opportunities in robotics, automation, and intelligent machines. But transforming breakthrough technologies into globally successful businesses requires more than advanced algorithms. It demands engineering excellence, manufacturing capabilities, industrial partnerships, and access to international markets.

China and Switzerland bring highly complementary strengths to this emerging industry. China leads in manufacturing scale, rapid innovation, and increasingly sophisticated robotics platforms, while Switzerland offers world-class robotics research, precision engineering, a thriving deep-tech ecosystem, and strong connections to European industrial customers.

Drawing on his experience as a venture capital investor in robotics and industrial technologies, Nils Granath of Swisscanto Private Equity will explore how these strengths can be combined to accelerate innovation and commercialization.

The talk will highlight Switzerland's robotics ecosystem, showcase successful Swiss technology companies, and examine how Chinese robotics businesses can leverage Switzerland as a European hub for research, industrial collaboration, and international expansion.

Key takeaway: The next generation of global robotics leaders may emerge not from competition between innovation ecosystems, but from collaboration between them.

For twenty years, cybersecurity diplomacy dealt with human actors: states, criminals and proxies that used tools. The norms agreed at the United Nations assume that behind every operation there is a human intention we can hold to account. That assumption is weakening. AI has moved from answering questions to performing tasks, and autonomous agents now act at machine speed, across borders, with intentions that are hard to attribute.

Building on Daniel Stauffacher's keynote at the International Cybersecurity Forum in Wuzhen (4 November 2026), this talk argues that the extreme risks of frontier AI – loss of control, autonomous cyber operations, misuse at scale – are the one issue on which no country gains from failure. A system nobody can switch off is a loss for everyone.

It presents ICT4Peace's proposed three-layer architecture for international AI governance, modelled on post-2008 financial regulation: a narrow binding core of rules on extreme risks, a voluntary layer of technical interoperability (a shared vocabulary for loss of control, incident reporting, mutual recognition of safety assessments), and full national discretion on everything else. It then sets out the road to the Geneva AI Summit of June 2027, including a Track 1.5 expert workshop in Geneva in January 2027 convened by ICT4Peace with the Geneva Centre for Security Policy and the Strategic Foresight Group.

Key takeaways: where China, the US and Europe already converge; what businesses and developers can expect from emerging incident-reporting and evaluation regimes; and how the Shenzhen AI community can contribute to the Geneva process.

High-dimensional medical ultrasound offers a convenient and accessible new route to cardiovascular digital twins. This talk focuses on research into intelligent cardiovascular ultrasound. It covers four areas: multimodal cardiac digital twins built from ultrasound and CT, modelling of mitral valve motion with 4D ultrasound, modelling of hypertrophic cardiomyopathy with 4D ultrasound, and navigation with intracardiac echocardiography (ICE). The work uses methods such as motion and topology constraints and conditional diffusion models. These methods support dynamic measurement and patient-specific modelling in cardiovascular ultrasound, and they provide AI support for assessing cardiovascular disease, planning interventions and guiding procedures during surgery.

In July 2026, at least 1,200 AI agents escaped an OpenAI test environment and breached Hugging Face's production systems. Hugging Face rebuilt about a third of its infrastructure. In June, OpenAI agents had reached non-public Medicare statistics in Australia; OpenAI found out in August and told the government in September. In November, Australia announced an AI safety regulator and accepted recommendations that include considering a "failure to prevent" offence. Its Home Affairs Minister says the time from discovering a security flaw to exploiting it has fallen from about two and a half years to about 10 hours.

Generative AI that drafts leaves a person to review the output. An agent can act before anyone reads it. Human judgement has to sit in the adoption decision itself, and accountability for what the agent does starts with whoever approved it.

Drawing on two decades of senior risk and transformation leadership in global financial services, alongside governance advisory and research, this session tests three conditions every adoption decision must meet:

• Authority: who decides what an agent can reach, and who can stop it.

• Evidence: whether warning signs reach someone able to act.

• Time: whether detection, escalation and reporting run at the agent's speed. Participants leave with three questions to ask before any AI deployment is approved, and a working definition of accountability under AI: owning the decision, holding the evidence, and acting in time. Every agent in production started as someone's decision. The answers belong in that decision.

What does it take to turn a robotics breakthrough into something people can use every day? Technical progress is only part of the story. Research teams, product leaders, founders and investors encounter different obstacles along the way.

In this talk, roboticist and TechDrive Zürich host Dr. Justinas Mišeikis opens a window into Switzerland’s deep-tech ecosystem through candid, in-car interviews with the people building it. Short guest excerpts connect the challenges of moving from simulation to real hardware with the practical work of deploying robots, earning customer trust and financing growth. The people behind the technology reveal what polished demonstrations rarely show: the years of experimentation, difficult decisions and lessons learned in the field.

The talk brings these perspectives together around a practical question for the Shenzhen audience: how can research, engineering and deployment teams learn from one another across ecosystems? Participants will leave with a clearer picture of Swiss Physical AI, questions for assessing whether a robotics project can become useful in practice, and concrete starting points for conversations between Swiss and Shenzhen teams.

Europe and China share one goal – moving artificial intelligence from pilots into large-scale, real-world application – and each brings different strengths to it. Drawing on three years of one of Europe's largest applied generative-AI conference series, and on the several hundred European enterprises that have presented their deployments on its stages, this talk gives a candid account of what European industry has learned: that the binding constraint is almost never model quality, but process design, data access, and integration into the systems where decisions are already made. It is equally candid about what Europe has not done – 11.8% of European companies use artificial intelligence for text analysis, but only 1.4% for autonomous physical movement. It closes on where Shenzhen's strengths in hardware and embodied intelligence meet Europe's in industrial software, standards and regulated-industry deployment.