The Future of Education and Research in the AI Era: Equipping Young People for Tomorrow

The University of Geneva is actively embedding AI governance into its teaching and research across all faculties, particularly law, addressing intellectual property, responsibility, and policy development. The university promotes executive masters and continuing education modules, such as the digital governance module in the MEIG programme, to serve both current students and professionals already in the workforce. This positions Geneva as a central hub for AI governance discussions, especially given the upcoming third AI Summit.

Professor Kaddous noted that after Paris and Delhi, Geneva will host the third AI Summit in June, with the university participating in those events . She highlighted that the law faculty is at the heart of AI-related matters including intellectual property, responsibility, and policy . The MEIG programme includes a whole module on digital governance aimed at both current students and those in continuing education .

While AI can rapidly summarise hundreds of legal cases or bibliographies, students must develop the critical capacity to evaluate what AI produces rather than accepting it uncritically. The goal is to ensure AI serves as an assistant that enhances work quality, not a replacement for independent analysis and legal reasoning. Fostering ethical thinking and intellectual independence is therefore a core pedagogical priority.

Professor Kaddous gave the example that in law, AI can summarise hundreds of cases of jurisprudence or bibliography in seconds or minutes , but stressed that a critical look at AI output is essential . She summarised her position clearly: 'AI is an assistant, but it is not a substitute for analysis, legal reasoning or intellectual rigor' .

Students are often reluctant to admit they have used AI, but transparency about its use should be actively encouraged as a professional norm. AI must be acknowledged because its use intersects with data protection rights and professional responsibilities, particularly in legal practice where client confidentiality and discretion are paramount. Professional responsibility ultimately rests with the individual, not the tool.

Professor Kaddous observed that students are still afraid to admit using AI and that transparency must be encouraged . She noted that AI use must be mentioned because of data protection and many rights that need to be protected . She further emphasised that in professional legal practice, clients have expectations of discretion and that competition, data protection, and professional responsibility are all at stake .

Generations of professionals already in the job market need structured pathways to understand and adopt AI tools, beginning with demystifying AI to reduce fear and then enabling them to use it effectively in their professions. This may also involve supporting career transitions as AI transforms existing roles. Continuing education must itself evolve rapidly given how quickly AI capabilities change.

Professor Kaddous explained that a first step of 'apprentissage' is needed so that former generations understand what AI is and do not see it only as a risk . She stressed that continuing education must allow professionals to evolve within their professions or even change professions , and acknowledged that what is valid today may not be valid in a year or less .

As premium AI tools move behind paywalls, access to the most capable AI systems becomes a function of financial means, creating a new dimension of inequality in education and professional development. This cost barrier has not been sufficiently discussed in debates about AI and education equity. It compounds existing digital divides by making the most powerful tools available only to those who can afford them.

Professor Kaddous raised the point that the best AI tools are no longer given for free and that payment is now required, identifying this as an important element not yet fully discussed in the session .

MIPT uses a structured three-pillar approach: selecting the top 1,000 students nationally, putting them through an intensive three-year fundamental cycle in computer science, mathematics, and physics, and then attaching them to leading professors for real research from their third year. This system is designed to produce graduates with two or three top journal publications by the time they complete their bachelor's or master's degrees. The model ensures that students develop both deep technical foundations and practical research experience simultaneously.

Professor Zhilokov described the FISTECH system in detail: the first pillar selects only 1,000 students per year , the second pillar involves a very hard three-year fundamental cycle in computer science, maths, and physics , and the third pillar attaches talented students to leading professors from their third year . The target is for students to have two or three top journal publications by the time they complete their degrees .

Just as physical muscles require struggle to develop, the brain requires genuine challenge during education to build the capacity for lifelong learning. Relying on AI to do one's thinking may produce outputs but does not build the underlying understanding needed to evaluate, direct, or correct those outputs. Without that understanding, individuals ultimately lose control of the processes they are supposed to be managing.

Professor Zhilokov drew an analogy between building physical muscles and building cognitive capacity, arguing that education must involve challenge and struggle . He cited AI researcher Andrej Karpathy's observation that 'you can outsource thinking, but you cannot outsource understanding' , and warned that without understanding, using tools all day long will cause one to lose sight and control of things .

One of the hottest current research areas is finding new model architectures and layer-level optimisations that reduce the cost of individual tokens and the overall energy burden of AI inference. Given the enormous capital and operational expenditure being invested in AI infrastructure globally, energy and cost efficiency will remain a dominant research priority for years to come. MIPT has a significant team actively publishing in this area.

Professor Zhilokov explained that research into new architectures and specific model layers to optimise inference cost and reduce the cost per token is a major focus . He noted that CAPEX and OPEX investments into infrastructure are enormous and the burden on energy systems is significant , and that MIPT has a large team presenting papers on this subject at the ICML conference in South Korea that week .

Making AI outputs explainable and trustworthy, and introducing effective guardrails for large language models, is a significant and growing area of research that was prominently discussed throughout the conference. This work is essential for building confidence in AI systems and ensuring they can be safely deployed in high-stakes environments. It addresses fundamental questions about how humans can maintain meaningful oversight of AI behaviour.

Professor Zhilokov identified interpretability, explainability, and trust as a hot topic, noting that many sessions during the conference addressed introducing guardrails for large language models and making AI results explainable and trustworthy .

A major research direction involves building neural models that understand and can simulate real-world physical laws and dependencies, enabling AI to function effectively in complex physical processes. Separately, as a multidisciplinary university, MIPT is developing AI tools to support research objectives across chemistry, physics, and biology departments. Both directions represent significant expansions of AI's applicability beyond language and text.

Professor Zhilokov described neural world models as making AI understand and imagine real physical laws and dependencies so they can operate in complex environments, referring to these as 'physically inspired neural networks' . He also noted that MIPT works with chemistry, physics, and biology departments to develop AI tools for their specific research tasks .

In robotics, two competing paradigms — visual language action models and world models — are being pursued, with one likely to become the dominant framework for future robotics control systems. Research targets include improved locomotion, dexterous manipulation, and reduced latency, all of which are critical for practical deployment of robots in real-world settings. MIPT is actively engaged in both directions.

Professor Zhilokov stated that MIPT conducts research on visual language action models and world models as two competing directions, one of which will probably be the future of robotics control systems . The specific targets he identified were enhanced locomotion, dexterous manipulation, and reduced latency .

Because AI dramatically increases what a single person can accomplish, it also dramatically increases the potential harm that individual can cause, making responsibility a more urgent educational priority than ever before. Engineering education must therefore go beyond technical skills to instil a deep sense of ethical responsibility in students. Universities have a heightened role in ensuring that students understand and embrace this responsibility.

Professor Xiao argued that technological people must take much more responsibility than ever because AI has higher capability but also creates much higher risks . He stated that universities have an even higher role to make students responsible and to let them understand what responsible use means , and that students must be taught to use technology responsibly because even a single person now has higher capability to do good or bad things .

Beyond the traditional digital divide of internet connectivity, AI introduces a new and deeper gap because accessing state-of-the-art models requires expensive tokens and computational infrastructure that many communities simply cannot afford. This means that the benefits of AI are concentrated in wealthy countries and communities while poorer ones are excluded. Rich countries simultaneously consume vast energy and create carbon emissions building ever-more-powerful models that the poor cannot access.

Professor Xiao distinguished the new AI equity gap from the older digital gap, noting that state-of-the-art models are expensive in terms of tokens and computational facilities . He observed that people in rural areas or poor countries cannot afford to let young people understand what happens in the AI era , while rich countries expend electricity and create carbon emissions building stronger AI models that poor countries cannot even access .

AI is not a single unified technology but is highly fragmented, with different countries using different models that each carry their own biases. This fragmentation, combined with divergent national policies on AI use — including in military applications — creates greater global risk and division rather than convergence. The rapid pace of AI change, evolving daily rather than on decade-long cycles, makes international coordination increasingly difficult.

Professor Xiao noted that AI is highly fragmented and that different countries, including China and the US, use different AI models . He pointed out that every single AI model has bias, creating more risks and fragmentation throughout the world . He also observed that while a few years ago governments were discussing restraining fully autonomous military AI, both China and the US are now actively automating the battlefield .

Traditional wireless technology standardisation operated on roughly ten-year cycles, but AI changes on a daily basis, fundamentally disrupting the pace and process of international standardisation. This creates significant challenges for bodies like the ITU, which must adapt their processes to keep pace with technology that does not wait for consensus-building cycles. The mismatch between AI's pace of change and institutional standardisation timelines is a structural problem for global digital governance.

Professor Xiao, drawing on his experience attending ITU standardisation meetings, noted that wireless technology previously evolved on 10-year periods from 3G to 4G to 5G , but that AI is changing on a daily basis . He stated that international standardisation has already been transformed as a result .

The concept of the 'one-person company' reflects the reality that a single individual equipped with AI tools can now accomplish what previously required a full team. Universities should institutionally support this by providing students with funding, token allocations, and GPU access to pursue independent projects. Huazhong University of Science and Technology has already implemented such a programme at the undergraduate level.

Professor Xiao described the 'one person company' (OPC) undergraduate programme at Huazhong University of Science and Technology, where each undergraduate student can apply for funding, a certain amount of token expense, and GPU card access to do a project on just one single summer . He emphasised that one student can now do a lot of huge things that in the past would have required building a whole team .

The one-person company concept is specifically about the ability of a single individual to build or produce something using AI tools; it does not mean that the entire business lifecycle can be conducted alone. Selling a product, understanding market needs, and building relationships still require human communication and collaboration. This distinction is important for avoiding a misreading of the concept as promoting social isolation.

Professor Xiao clarified that one-person company 'doesn't mean the whole business is one person' but rather that one person can build something by themselves through the engineering part . He noted that to sell the product, one still has to understand the need and communicate with a lot of people , and referenced Chinese culture of going to restaurants to talk business as an example of the continued importance of human communication .

The rapid pace of AI development means that ideas which seem impractical or even ridiculous today may become entirely feasible tomorrow as new tools and capabilities emerge. Educators who reflexively dismiss unconventional student proposals may be shutting down genuinely disruptive innovation. This requires a fundamental shift in how professors engage with students who bring seemingly outlandish ideas.

Professor Xiao reflected on a change in his own thinking, noting that in the past a professor could dismiss a ridiculous-sounding student question, but now educators must rethink this approach . He argued that students with disruptive or unconventional ideas could be right because they have access to advanced solutions not available in the past, and that what sounds wrong today could be viable tomorrow due to AI evolution . He concluded that if a disruptive idea is rejected by everyone around you, that may actually be the opportunity for a startup .

Because AI tools are widely accessible, any startup idea that is not genuinely novel can be replicated almost immediately by others with the same tools, eliminating competitive advantage. This raises the bar for what constitutes a viable startup concept, making truly disruptive and original thinking more valuable than incremental or derivative ideas. The connected world means that information asymmetries that previously protected less innovative businesses have largely disappeared.

Professor Xiao noted that in the past, entrepreneurs from China could learn models like eBay or PayPal in the US and replicate them in China due to the digital gap , but that in today's wholly connected world, if you have an idea similar to others it will not shine . He warned that if a startup idea lacks genuine novelty, others can do similar things, meaning that truly disruptive ideas are what will succeed .

The majority of universities globally have lost their original purpose and now function primarily as businesses issuing certificates of questionable value, with little connection to real-world needs. For most people, high-quality technical and vocational education is far more relevant and impactful than university degrees, as it equips them with skills for the physical jobs that will continue to exist. The focus on university education as the primary pathway obscures the greater need for skilled tradespeople and technically trained workers.

Professor Unwin stated bluntly that 'most universities in the world are not fit for purpose' and have become institutions for handing out worthless certificates, including in his own country of Britain . He drew on experience working in 75 countries and argued that technical education is the learning that people with real physical jobs need, citing the example that his country needs more plumbers than academics . He also noted that a minister of employment in an African country told him that TVET is the most important thing but remains a 'Cinderella subject' underfunded by donors .

The most urgent educational challenge globally is not how to integrate AI into elite universities but the fact that one in ten children worldwide is not in school at all. Economic growth driven by AI and digital technology tends to increase inequality rather than reduce it, and the focus of the global community should shift from growth to equity. The so-called 'youth dividend' in Africa becomes a liability rather than an asset if young people are uneducated and unemployed.

Professor Unwin cited UNESCO's latest figures indicating that one in ten children in the world is not in school , and challenged the assembled experts to consider what they are doing about that . He argued that economic growth drives inequality and that very little work is being done on how AI can address the situation of children out of school . He also warned that in Africa, a large number of uneducated young people without jobs is 'not a dividend, it's a millstone around the neck of Africa' .

Excessive reliance on AI to perform cognitive tasks that humans should do themselves risks atrophying the brain in a manner analogous to dementia, which Professor Unwin described from personal experience of caring for relatives. The right to be unconnected — to choose not to use digital technology — is as fundamental as the right to be connected, particularly for those who cannot access connectivity. Humans should actively reclaim their cognitive independence rather than simply adapting to AI.

Professor Unwin stated that 'the right to be unconnected is more important than the right to be connected' , arguing that those unable to connect still have rights. He warned that the brain will stop working far faster if one uses a lot of AI than if one uses the brain for what it was intended , and referenced the horror of watching relatives die with dementia as a personal illustration of what cognitive decline looks like .

Despite being the most practically relevant form of education for the majority of the world's population, technical and vocational education and training (TVET) is systematically underfunded by international donors who regard it as unproven. This neglect means that the people who most need skills for the jobs that will continue to exist in an AI-transformed economy are the least supported. Addressing this funding gap is more urgent than expanding university AI programmes.

Professor Unwin recounted a conversation with an African minister of employment who described TVET as 'the most important thing' but a 'Cinderella subject' that most donors are not funding because it is considered unproven .

Through a capstone project partnership with the ITU, Alexander and his team rapidly gained deep knowledge of a major international digital governance institution that they would not otherwise have encountered. The project produced a concrete policy output — a recommendation that was integrated into an ITU standard — demonstrating that student-institution partnerships can generate tangible real-world impact. This experience also directly shaped his subsequent academic and professional trajectory.

Alexander described how as a suburban American student at Penn he knew little about the ITU before the capstone project , but through it his team developed a standard to measure the environmental impact of AI data centres . Their recommendation was submitted to ITU Council and integrated into ITU-TL1801, which was funded by the ITU . He also noted that this experience led him to pursue a master's at Johns Hopkins focusing on law and digital policy .

At its core, the ITU is a place where knowledge is transferred between experts, academics, policymakers, and students, and where that knowledge is translated into real international standards. Its study groups bring together top global experts, its Academic Advisory Body addresses frontier topics including AI and quantum, and its conferences create forums for broad knowledge exchange. The more the ITU channels this knowledge transfer towards young people, the greater the positive impact.

Alexander described the ITU's study groups as bringing together top experts to develop standards on digital technologies , and highlighted the Academic Advisory Body as a new initiative where professors discuss AI, quantum, strategic foresight, and space . He also pointed to conferences such as WSIS, AI for Good, and AI Global Dialogue as forums that bring together academics, policymakers, and industry leaders . He concluded that 'for the ITU, the more it transfers that knowledge to young people and students like myself through the Capstone Project, through other sorts of partnerships, then we're getting someplace better' .

The capstone project undertaken by Alexander's team at the University of Pennsylvania resulted in a concrete policy recommendation on measuring the environmental impact of AI data centres that was formally integrated into an ITU standard. This demonstrates that student research, when properly structured and connected to international institutions, can produce outputs with real governance implications. It also illustrates the ITU's openness to incorporating student contributions into its standard-setting work.

Alexander explained that his team's recommendation was submitted to ITU Council and integrated into a new standard called ITU-TL1801, which was funded by the ITU . The team was also able to speak at the 2025 ECOSOC Youth Forum on their recommendation . The topic of the project was creating a standard for the environmental impact of AI data centres .

The University of Geneva benefits from its location in a city that hosts major international organisations and is set to host the third global AI Summit, placing it at the heart of global AI governance discussions. This geographic and institutional positioning gives the university unique opportunities to integrate cutting-edge governance debates into its teaching and research. The university actively participates in these events alongside international organisations, the private sector, NGOs, and other academic institutions.

This argument was primarily made by Christine Kaddous rather than Alexander Schrier. Professor Kaddous noted that after Paris and Delhi, Geneva will host the third AI Summit in June, with the university participating , and that the university is part of discussions involving international organisations, private sector, NGOs, and academia .

Major international conferences convened by or associated with the ITU serve as critical nodes for knowledge exchange, bringing together diverse stakeholders including academics, policymakers, CEOs, and students. These events expose students to frontier issues and debates that would be difficult to access through conventional university curricula alone. They represent an important complement to formal education in preparing students for careers at the intersection of technology and governance.

Alexander cited conferences such as WSIS, AI for Good, and AI Global Dialogue as examples of events that bring together professors, academics, CEOs, and others to discuss AI and the frontier of research and education .

Unlike bodies that primarily serve the interests of powerful nations, the ITU is committed to all its member states equally, making it a uniquely important institution for anyone working on global digital policy. Understanding this multilateral mandate is essential for students who want to make an impact in international digital governance. The capstone project gave Alexander's team direct insight into this dimension of the ITU's work.

Alexander noted that through the capstone project his team learned about the ITU's commitments to its member states, 'not just committing to one country, but to all countries' .

Contrary to the characterisation of university students as primarily seeking certificates, Ava argued from personal experience that intellectual curiosity is the genuine motivation for many students. The capstone project exemplified this by requiring students to develop real critical oversight skills when using AI tools, particularly in distinguishing between what primary sources actually said and what AI-assisted summaries implied they said. This kind of rigorous engagement with knowledge is what university education should cultivate.

Ava stated that she did not go to university to get a certificate and that intellectual curiosity was her motivation, adding that she did not meet a single student whose goal was merely to tick the box for a certificate . She described how the capstone project required developing 'source discipline' because AI makes it easy to accumulate information quickly but creates a false sense of coverage , and that the nine expert interviews and full literature review were essential to distinguish between what primary sources actually said and what AI-assisted summaries implied .

One of the most important skills developed through the capstone project was recognising the limits of AI delegation — specifically, that AI should not be used for work requiring original judgement on contested questions where no established consensus exists. Over-reliance on AI in such contexts produces confident-sounding but shallow analysis that fails to engage with genuine complexity. Students must learn to protect the interpretive and argumentative dimensions of their work as distinctly human contributions.

Ava described how the capstone research involved genuinely contested new terrain - questions about data sovereignty, geopolitical risk, and supply chain transparency - where there is no single established consensus . She warned that over-reliance on AI in such contexts would produce 'a very confident-sounding but shallow analysis' , and argued that the capstone taught students to reserve AI for organising, structuring, and accelerating while protecting the interpretive and argumentative work . She also noted that the ITU context sharpened this discipline because findings were intended to inform real standard-setting work, making the stakes of misrepresentation concrete .

The year-long capstone project with the ITU required students to develop and maintain rigorous verification processes throughout the research, not merely as a final check, because the findings were intended to inform real international standard-setting. Working as a self-checking pair while remaining fully accountable for accuracy instilled disciplined methodology that would not have developed in shorter or lower-stakes academic exercises. The key skill is not just using AI tools but managing them critically through a combination of efficiency, rigorous validation, and sound judgement.

Joseph described how working at the scale of a year-long project forced the team to be disciplined in methodology and to put clear verification processes in place, working as a self-checking pair while remaining fully accountable for accuracy . He concluded that 'the key skill is not just using these tools but managing them critically by combining efficiency with rigorous validation and sound judgment' .

While AI significantly accelerates routine tasks like formatting, rephrasing, and editing, it is not reliably capable of handling complex concepts and ideas, and its tendency to hallucinate — generating non-existent references or misattributing ideas — creates new challenges rather than eliminating existing ones. Managing a large bibliography over a year-long project required constant verification and cross-checking because the AI would sometimes generate references that did not exist. Similarly, AI processing of interview transcripts could misattribute statements or invent connections that were not present in the original data.

Joseph noted that AI tools are highly effective at speeding up routine tasks such as formatting, sparing, rephrasing, and editing, but are not always reliable for handling complex concepts and ideas . He described a key issue with hallucinations in bibliography management, where the tool would generate references that did not exist or attribute ideas to sources that never mentioned them, requiring constant verification over the course of the year-long project . He also noted that when processing interview transcripts, AI could occasionally misattribute statements or invent connections not actually present, requiring systematic return to raw data .

The audience member, drawing on classroom experience, observed that students are increasingly interacting with AI agents rather than with each other, which appears to be degrading their ability to collaborate and share knowledge with peers. This raises a concern about the one-person company model: while it may enhance individual productivity, it may simultaneously undermine the collaborative and social skills that are essential for broader entrepreneurial and professional success. The question was directed at Professor Xiao regarding whether China's experience confirms or contradicts this trend.

The audience member stated from classroom experience that students have much harder time collaborating today because they are in exchange with AI agents rather than peers , and asked whether collaboration and sharing skills are getting higher or lower in China as a result of programmes like the one-person company .

The audience member, identifying as a French learner in China, raised a practical question about how language students should approach their studies given the rapid development of AI tools. This question implicitly acknowledges that AI is transforming language learning and that students need concrete advice on how to adapt. The question was directed at the panel for guidance relevant to their specific context.

The audience member introduced themselves as a French learner in China and asked whether the panellists had any advice for language learners in China as students .

The audience member, representing a quantum computing startup, highlighted the difficulty of integrating yet another emerging technology into the already complex landscape of professional upskilling. The question pointed to the challenge of continuously updating the skills of people who have already left formal education, especially as multiple frontier technologies converge simultaneously. This raises the question of how executive and continuing education programmes can keep pace with the speed of technological change.

The audience member identified themselves as being from a startup in quantum computing and expressed strong interest in the aspect of lifelong learning raised by Professor Kaddous, asking specifically how to upskill and reskill generations of students already out of formal education while integrating yet another frontier technology .

The audience member, a student at UC San Diego studying global health and political science and interning at the ITU, observed an exponential and potentially excessive use of AI in the classroom. The question was directed at Professor Kaddous to understand how AI will specifically change the study and practice of law, and how students interested in law school should prepare for these changes. This reflects a broader concern about whether AI dependency is eroding the foundational skills that legal education is meant to develop.

The audience member stated that they had witnessed the exponential use of AI in the classroom and that many would describe it as an over-reliance, and asked Professor Kaddous how AI will alter the dynamics within law studies specifically, how the field will change, and how students should be better prepared for these changes .

The audience member, a startup owner from Mexico City running an AI-powered personalised education platform, raised the question of which engineering skills retain value as AI tools automate an increasing range of technical tasks. The question also sought guidance on best practices and policies for startup owners building AI-driven educational products, particularly in relation to the one-person company concept discussed by Professor Xiao. This reflects a practical concern about how to design responsible and effective AI-enhanced learning environments.

The audience member introduced themselves as the owner of a startup called Alexandria focused on personalising education using AI and AI governance practices, and asked what skills in engineering are still needed now that coding tools and agents are automating many tasks, and what concerns, policies, or best practices startup owners building educational platforms should consider in relation to the one-person company concept .

The moderator drew a parallel between the adaptive learning processes of AI models and the need for educational institutions to similarly evolve in response to rapidly changing technology. She acknowledged that while education begins in early childhood and school, it is at the university level that the accelerating pace of AI-driven change in jobs and careers becomes most visible. This framing positioned adaptation as a shared imperative for both humans and institutions.

The moderator observed that AI and other emerging technologies are changing so fast that even within one or two years old jobs are evolving and careers are evolving, and drew a parallel to how AI models learn and adapt, suggesting that humans and institutions must do the same .

In response to Tim Unwin's critique of universities as certificate-dispensing businesses, the moderator affirmed that the core purpose of universities, irrespective of their funding model or national context, is education. This served as a grounding reminder that the structural diversity of universities across countries does not change their foundational mission. The moderator used this point to redirect the discussion towards shared values of responsibility and critical thinking.

The moderator acknowledged that there are public and private universities structured differently across countries, but asserted that the whole purpose of universities is to educate, as that is why they were created in the first place, before summarising that responsibility and critical thinking are the key takeaways from the discussion .

The moderator endorsed Tim Unwin's suggestion that taking time away from digital devices and spending time in nature is genuinely restorative and beneficial. She extended this idea by referencing her own practice of running sessions at WSIS that involve walking around Geneva, suggesting that physical and analogue experiences should be more deliberately integrated into professional and academic life. This reflects a broader concern about the cognitive and wellbeing costs of constant digital connectivity.

The moderator agreed with Professor Unwin's suggestion to take three days without digital devices and go walking in the Swiss mountains, and mentioned that she has run sessions at WSIS that involved a walk around Geneva, suggesting that more such activities should be incorporated .