WSIS Forum 2026
AI-generated report
ENFR

AI Empowerment for Older People from a Gender Mainstreaming Perspective

6 speakers
Summary

This session focused on AI empowerment for older people from a gender mainstreaming perspective, examining how the intersection of demographic ageing and digital transformation creates compounding inequalities . Prof. Moon Choi of KAIST introduced the concept of 'gendered ageing,' noting that older women outnumber older men, making ageing fundamentally a gender issue rather than merely a demographic or health concern . She emphasised that AI and digital systems are not merely tools, but have become everyday infrastructure. This further disadvantages women, as they tend to have lower AI access and lower accumulated lifetime digital literacy. .

Dr. Jingbo Huang of UNU Macau highlighted the 'feminisation of ageing,' noting that women already make up 54% of the world's older adults and that this population will grow threefold by 2050 . She introduced synthetic data as both an opportunity and a risk: whilst it can address data scarcity and correct biases, it can, if generated carelessly, amplify existing inequalities. This would further entrench the disparities experienced by older women, a group already underrepresented in real-world datasets. . Dr. Huang called for gender- and age-disaggregated data as a foundational requirement, alongside governance frameworks anchored to international agreements such as the Madrid International Plan of Action on Ageing (MIPAA) and the Global Digital Compact .

Ms. Ern Chern Khor presented empirical findings from a 2025 South Korean survey of over 2,600 older internet users, revealing a 'second-level digital divide' in generative AI adoption that disproportionately affects older women and those with lower socioeconomic status . A particularly alarming finding was that improvements in AI literacy translate into adoption gains far more quickly for already-advantaged groups, meaning poorly targeted interventions risk widening the AI divide further .

Panellists added further dimensions to the discussion. Mr. Wai Kit Si Tou advocated for accountability mechanisms modelled on environmental, social and governance frameworks, requiring companies to disclose how AI impacts different stakeholders across its lifecycle . Prof. Tim Unwin stressed the importance of intersectionality, including LGBT identities, disability, and ethnicity, and warned that excessive AI use risks creating 'digital dementia' by reducing people's capacity for independent thought . An audience member raised the issue of data sovereignty policies in Ethiopia, noting that government restrictions on humanitarian data disproportionately harm older women .

Overall, the discussion concluded that AI's benefits and risks are not distributed equally, and that genuinely inclusive AI for older women requires co-design with marginalised communities, careful governance of data quality, and a commitment to valuing inclusion over speed of measurable outcomes .

Keypoints

  • Overall Purpose

  • The session examined how the intersection of AI and population ageing disproportionately affects older women. The goal was to present empirical research, identify policy gaps, and generate concrete recommendations to make AI development more equitable and inclusive, particularly from a gender mainstreaming perspective.
  • --

  • Major Discussion Points

  • The feminisation of ageing and its implications for AI equity. Both Prof. Moon Choi and Dr. Jingbo Huang established that population ageing is fundamentally a gendered phenomenon. Women already constitute 54% of the world's older adults, a figure projected to reach 1 billion women over 60 by 2050. Because older women are poorly represented in the data that AI systems learn from, their marginalisation in the real world is amplified rather than corrected by AI. Prof. Choi stressed that inequalities are not simply about age or gender in isolation, but about how disadvantages accumulate across the life course and become most visible in later life. - Synthetic data as both an opportunity and a risk for representing older women. Dr. Huang introduced synthetic data (artificially generated data modelled on real-world statistical features) as an increasingly dominant force in AI training, with approximately 60% of training data now synthetic. She argued it offers genuine opportunities: addressing data scarcity (especially in the Global South), correcting biases present in real data, and protecting privacy. However, if generated carelessly, synthetic data can amplify existing biases, effectively scaling the absence of older women rather than remedying it. Her recommendations called for technical quality controls, mandatory gender- and age-disaggregated data as a policy foundation, and a governance framework anchored to instruments such as the Madrid Plan of Action on Ageing and the Global Digital Compact. - The generative AI divide among older adults, and the limits of AI literacy interventions. Ms. Ern Chern Khor presented findings from a 2025 South Korean survey of over 2,600 older internet users, demonstrating a second-level digital divide: even among older adults who already use the internet, older women and those with lower socioeconomic status are significantly less likely to adopt generative AI tools such as ChatGPT. Crucially, her moderation analysis revealed a counterintuitive finding - increases in AI literacy translate into AI adoption far more rapidly for those already in advantaged positions, meaning that poorly targeted literacy programmes risk widening the AI divide rather than closing it. She concluded that interventions must consciously direct resources to the most vulnerable, include use cases relevant to older women, and resist the temptation to prioritise groups that can demonstrate fast, measurable outcomes. - Accountability, co-design, and the right not to be connected. The panel discussion surfaced complementary governance perspectives. Mr Wai Kit Si Tou advocated for an environmental, social, and governance (ESG) style public disclosure mechanism that would require companies to report how their AI models impact different stakeholders across the AI life cycle, creating accountability for inclusive design. Prof. Tim Unwin broadened the gender framing to include intersecting identities - older lesbian women, women with disabilities, ethnic minorities in rural areas - and warned against equating 'gender' solely with women. He also introduced the concept of 'digital dementia,' arguing that excessive reliance on AI risks eroding cognitive independence, and that forcing elderly people to engage with digital systems they cannot navigate amounts to a form of exclusion rather than empowerment.
  • Data sovereignty and the invisibility of marginalised populations. An audience member from the Gender Empowerment Movement raised the case of Ethiopia's data sovereignty policy, which prevents independent researchers and humanitarian organisations from accessing data, disproportionately affecting older women who bear the greatest burden of humanitarian crises. Prof. Choi and Dr. Huang connected this to the broader challenge of low-resource language datasets and cross-border data interoperability, noting that synthetic data and interoperability frameworks could offer partial solutions, though significant technical and policy work remains.
  • --

  • Overall Tone

  • The discussion opened in a measured, academic tone, with presenters carefully grounding their arguments in empirical data and institutional frameworks. As the panel discussion progressed, the tone became more personal, particularly when Prof. Unwin drew on his own experience on the subject of dementia and spoke passionately about the impact of digital exclusion. An audience member's reflection that inclusion had never been discussed in her AI master's programme emphasised existing gaps that should be addressed. Throughout, the overarching mood remained constructive and solution-oriented, with speakers consistently returning to concrete policy recommendations and calls for participatory, human-centred design. Prof. Choi's closing remarks reinforced this constructive intent, reminding participants that technology is not neutral and that the question of 'AI for good' must always be accompanied by the question 'for whom.'
Speakers Overview
PM
Prof. Moon Choi
133 wpm · 14 min
DJ
Dr. Jingbo Huang
134 wpm · 11 min
ME
Ms. Ern Chern Khor
130 wpm · 9 min
MW
Mr. Wai Kit Si Tou
144 wpm · 2 min
PT
Prof. Tim Unwin
151 wpm · 6 min
A
Audience
122 wpm · 5 min

Expanded Summary: AI Empowerment for Older People - A Gender Mainstreaming Perspective (WSIS Session 1-8-8)

#

Session Overview and Framing

This session, convened at the WSIS forum under the title "AI Empowerment for All the People," brought together researchers, UN officials, and policy practitioners to examine how the simultaneous forces of population ageing and digital transformation are reshaping inequality, particularly for older women . Prof. Moon Choi of KAIST, who chaired and opened the session, situated the discussion within two converging global trends: the longevity revolution, in which people are living significantly longer than previous generations and ageing has become a central concern for every policy domain , and the digital transformation, in which AI and digital systems are becoming everyday infrastructure for healthcare, work, and social connection . The central question animating the session was not simply whether AI is beneficial, but whether its benefits and risks are distributed equally among older people - and, specifically, whether some groups are being excluded entirely as a result of this transformation .

Prof. Choi introduced her institution, the KAIST Aging and Technology Policy Lab, as the first lab dedicated to ageing, social welfare, and technology policy since its founding in 2014, conducting quantitative and qualitative empirical research in support of evidence-based policymaking. She introduced the concept of "gendered ageing" to frame the discussion . She noted that there are more older women than older men in the world because women live longer , and she was emphatic that this makes ageing fundamentally a gender issue rather than merely a demographic or health concern . Crucially, she argued that gendered ageing is not simply about age or gender in isolation, but about how inequalities accumulate across the life course and become most visible in later life . Women tend to have lower access to AI and lower AI literacy as a result of these accumulated disadvantages . She also noted that measuring these inequalities is particularly challenging, as it often requires longitudinal studies, though her lab has managed to conduct some experimental research with AI systems.

Prof. Choi also drew attention to the way in which the domains of age-friendly cities and communities must now account for AI-mediated interactions embedded in every aspect of daily life. She referenced the WHO's eight domains of age-friendly cities and communities, proposed in 2002 - including transport, housing, and social participation - and noted that all of these domains now require AI-mediated interactions, from transport platforms to healthcare monitoring to social connection . Technology, she argued, is no longer merely a tool; it is part of the environment in which people live and with which they live .

#

The Feminisation of Ageing and Older Women's Marginalisation in AI

Dr. Jingbo Huang, Director of the United Nations University Institute in Macau, joined the session online and opened her remarks by echoing and extending Prof. Choi's framing . She grounded the discussion in demographic data: according to the UN, by 2018 - for the first time in history - people aged 65 and older outnumbered all young people under 18 . Within this ageing world, women are the majority - already constituting 54% of the world's older adults - and the number of women over 60 is projected to grow from approximately 336 million in 2000 to 1 billion by 2050, a threefold increase . Demographers refer to this as the "feminisation of ageing" . Dr. Huang argued that this large, fast-growing, and increasingly female population deserves urgent policy attention, yet older women are among the most marginalised groups when it comes to AI - in part because they are so poorly represented in the data that AI systems learn from .

Dr. Huang drew on two bodies of research from UNU Macau to develop this argument. The first was the EQUALS report, produced in 2019 with contributions from Prof. Choi, which examined gender and technology across three dimensions: access, skills, and leadership . The central message of that report, she noted, was already clear at the time: without gender-disaggregated data, inequality cannot even be seen, let alone addressed . The second body of research concerned synthetic data - a topic she argued has become central to AI governance and directly relevant to the marginalisation of older women .

#

Synthetic Data: Opportunity, Risk, and Governance

Dr. Huang introduced synthetic data - artificially created data modelled on the statistical features of real-world data, which can be partially real, partially artificial, or completely synthetic - as an increasingly dominant force in AI development . Dr. Huang cited figures suggesting that approximately 60% of the data used to train AI models is now synthetic, and that accessible, machine-readable real-world data is being exhausted, making the governance of synthetic data an urgent priority . She identified three principal opportunities: synthetic data can address data scarcity, particularly in the Global South; it can, if properly treated, correct biases present in real data, including those related to gender and age; and it can protect privacy by removing identifying information . However, she was equally clear about the risks. If generated carelessly, synthetic data can amplify existing biases, reproducing the very gaps it is meant to close, and acting as a "magnifying glass" of real-world data biases . It can also infringe intellectual property .

For older women specifically, Dr. Huang argued that the stakes are doubled. Older women are marginalised both as women and as older people, meaning they fall through two gaps simultaneously . Without data that captures both gender and age, synthetic data will only scale their absence from AI models rather than remedying it . She also raised a fundamental methodological challenge: when baseline data on older women barely exists, it becomes nearly impossible to measure whether synthetic data represents them adequately - a question she identified as an open research priority . She made three sets of recommendations: technically, quality control and bias correction must be built into synthetic data generation, with explicit targets for both age and gender representation ; at the policy level, gender- and age-disaggregated data must be made a foundational requirement, since synthetic data cannot represent what was never measured ; and at the governance level, a framework anchored to the Madrid Plan of Action on Ageing, the SDGs, and the Global Digital Compact should be established, built through regional cooperation so that smaller economies can share standards, and older women themselves should have a participatory voice in its design .

#

The Generative AI Divide Among Older Adults

Ms. Ern Chern Khor, a doctoral candidate at KAIST working within Prof. Choi's Aging and Technology Policy Lab, presented empirical findings from a study focused specifically on the generative AI divide among older adults . She began by situating the AI divide historically: it is not a new problem, she argued, but the result of longstanding structural inequalities colliding with new technology applications . Since the launch of ChatGPT in 2022, generative AI applications and users have grown exponentially, making it urgent to ask who is being left behind . Existing digital inclusion interventions have helped older adults access the internet, but the question of whether those interventions are sufficient to bridge older women to AI adoption remains open .

To investigate this, Ms. Khor's study used the 2025 Digital Device Survey in South Korea, covering more than 2,600 older adults who were already internet users . By focusing on this group, the study was designed to reveal whether a divide exists even among those who have already crossed the first-level digital divide between internet users and non-users . The results confirmed significant gaps associated with older women and lower socioeconomic status in generative AI adoption, providing evidence of a second-level digital divide . This finding demonstrated that internet access alone is insufficient to ensure participation in advanced AI applications, and that something structural beyond connectivity is shaping who uses AI and who does not .

The study's most striking and policy-relevant finding emerged from a moderation analysis. AI literacy was confirmed as the most important factor in explaining the generative AI divide across all variables included in the analysis . However, the moderation analysis revealed a counterintuitive and, in Ms. Khor's words, "alarming" result: increases in AI literacy translate into AI adoption far more rapidly for older adults already in advantaged positions - those who are younger, male, of higher socioeconomic status, and living in urban areas - than for those in disadvantaged positions . This means that if policy interventions are not designed carefully, resources directed at improving AI literacy may be captured disproportionately by those already advantaged, potentially widening the AI divide rather than closing it .

Ms. Khor drew four conclusions from these findings. First, older adults are not a uniform demographic, and interventions must consciously attend to who is actually participating in and benefiting from them . Second, resources must be specifically directed to the most vulnerable older women - those in rural regions, with lower education, and burdened by family and social roles . Third, use cases relevant to older women must be consciously included when designing generative AI adoption interventions . Fourth, and most importantly, it is tempting to direct resources to groups that can demonstrate fast, tangible outcomes, but this risks an uneven distribution of resources that will worsen the existing AI divide; the slower progress of the most vulnerable must be recognised and valued in monitoring and evaluation frameworks . Ms. Khor also noted that her lab was preparing a report on AI and ageing, to be presented at the AI for Good Summit two days later, which collected fourteen case studies across different countries examining practices that help older adults and older women improve their AI literacy.

#

Panel Discussion: Accountability, Intersectionality, and the Right to Remain Unconnected

The panel discussion opened with Prof. Choi posing a guiding question to the two panellists: if there were one thing they would change tomorrow to place older women at the centre of AI for ageing, what would it be ? Mr. Wai Kit Si Tou, Economic Affairs Officer at UNCTAD, focused his answer on accountability . He proposed an AI accountability mechanism modelled on the ESG (Environmental, Social and Governance) framework, which already requires companies to report not only on their financial performance but on their environmental, social, and governance impacts . An AI equivalent, he argued, would require companies to disclose how AI impacts different stakeholders across the AI life cycle, including through public disclosure of how AI models work, how decisions are made, and how data is collected and managed . This mechanism, he suggested, would be essential to engaging the private sector in inclusive AI development .

Prof. Tim Unwin of the University of London offered a more expansive and at times deliberately provocative response. He began by noting that he had programmed in Fortran in the 1970s and is himself in his 70s, establishing a personal stake in the topic of ageing and technology. He cautioned that in UN dialogues, the word "gender" is too often equated solely with women, and that the full diversity of gender and sexuality must be considered . Drawing on research with LBT communities in the favelas of Brazil, he argued that older LGBT women are even more marginalised than older women as a group , and he invited the session to imagine itself as addressing "older lesbian women with disabilities from ethnic minorities living in rural areas" - a formulation designed to make the compounding nature of intersectional disadvantage viscerally concrete.

Prof. Unwin then introduced the concept of "digital dementia" - a term he coined - drawing on his personal experience of his mother, a pioneering mathematician and early adopter of digital technology who had introduced computers into schools in the 1980s, whose ability to use Skype disappeared as dementia set in . He argued that keeping the brain through independent thought is essential to avoiding dementia, and that excessive use of AI and digital technology risks creating a generation of people who cannot think for themselves - a prospect he described as "totally totally scary" . In a deliberately provocative rhetorical move, Prof. Unwin referenced legislation in Britain permitting assisted dying, asking whether the logical conclusion of excluding elderly people from digital life was that they should simply "kill themselves" - a reductio ad absurdum intended to underscore the moral stakes of genuine inclusion. His answer to the guiding question was that the one thing he would change is involving people from intersectionally marginalised groups in the design of AI itself, since without co-design, no governance or literacy intervention can succeed .

#

Audience Contributions: Data Sovereignty and the Absence of Inclusion from AI Education

The discussion was significantly enriched by two audience contributions. The first came from Paseba Seifu of the Gender Empowerment Movement, who raised the case of Ethiopia's data sovereignty policy, under which the government has declared data a matter of state sovereignty, preventing independent researchers and humanitarian organisations from accessing or generating data . She connected this directly to the session's themes, noting that this disproportionately impacts older women who bear the brunt of humanitarian crises, and that without data, AI systems cannot learn about the lives of these women - effectively rendering them invisible . Prof. Choi acknowledged the concern and connected it to the broader challenge of low-resource language datasets and the volume problem in AI training data . She also noted that a KAIST alumnus had developed a large dataset on Korean population perceptions, including sufficient minority representation, available on GitHub and Open Sciences, as an example of inclusive open data practice. Dr. Huang responded by pointing to synthetic data as a potential partial solution and to UNU Macau's ongoing research on cross-border data interoperability frameworks as a related avenue . She referenced UNU Macau's research on AI safety interoperability, which had examined cross-border data frameworks across South Korea, China, Singapore, and the UK, and noted that this work was continuing with a focus on interoperability in educational systems. Neither speaker, however, directly addressed the political question of how the international community can respond to deliberate state-imposed data blackouts.

The second significant audience contribution came from Anisha Tamawi, who asked how Prof. Unwin's argument about intersectional complexity could be implemented in practice, given that agentic AI is specifically designed to reduce the complexity of user interaction . Prof. Unwin's response reinforced his earlier argument: the right to be unconnected is as important as, if not more important than, the right to be connected . He gave the example of an elderly neighbour rendered unable to function by the digitalisation of banking , and argued that forcing people to use AI constitutes "enforced slavery" and amounts to data capture . A second audience member - who identified herself as completing a master's degree in AI and machine learning management, and whose contribution was a spontaneous floor remark rather than a prepared question - then offered a candid reflection: in her entire programme, human-centred AI had been discussed only in terms of keeping humans in the loop for oversight and error-correction, never in terms of inclusion of marginalised groups, and the perspective had simply never occurred to her . This exchange underscored the depth of the blind spot in mainstream AI education and governance discourse regarding the most marginalised populations.

#

Closing Reflections and Overarching Conclusions

Prof. Choi closed the session by returning to its foundational premise: technology is not neutral, and it does not work the same way for everyone . Even agentic AI is used differently by different people, and technology is often designed for the convenience of already-privileged users . The question of "AI for good" must therefore always be accompanied by the question "for whom," and a diversity of perspectives must be brought to bear on AI development and governance .

Taken together, the session produced a coherent, if contested, policy agenda. There was strong consensus across all speakers on the diagnosis: older women are disproportionately marginalised in AI systems due to accumulated structural inequalities, poor data representation, and the feminisation of ageing . There was also broad agreement that gender-disaggregated data is foundational , that AI interventions risk widening rather than closing inequalities if not carefully designed , and that governance frameworks must be anchored to international commitments and include participatory approaches . The most significant unresolved tensions concerned the scope of "gender" in AI policy, the right to remain unconnected, the conditions under which synthetic data helps rather than harms older women's representation, and the political challenge of state-imposed data sovereignty blocking humanitarian data access - all of which point to a field that has not yet reached consensus on some of its most foundational goals.

—
Prof. Moon Choi
I think they can see anyway. So thank you so much for coming to the Wizzies session on 1 -8 -8, AI empowerment for all the people. It's 1 p .m., but we'll wait for one or two more minutes so the speaker can be here, and then we'll start shortly. Thank you. Thank you so much for coming to the Afton Session. It's a hot, you know, humid day, and, you know, it must be, you know, hard to travel to here, and we really appreciate your interest in this session. And can you hear me? Okay. Yeah. I'm Moon Chae. I'm a professor at KAIST. We are the Tuck University Developing AI and Emerging Technologies, and I'm also a director of the KAIST Aging and Technology Policy Lab. So today's session is about AI empowerment for older people from a gender mainstreaming perspective. So our lab is the very first lab about aging, social welfare, and technology policy since 2014, and we are doing a lot of quantitative and qualitative empirical research about evidence -based policymaking. So there's a demographic change and also the new technology and their intersection, and we are based on social justice and sustainability, and that tried to make evidence to make a better policy. So we are going through the longevity revolution. What does it mean? So in our parents or grandparents, have a shorter life, than we have these days. So people live longer, and aging becomes a central question for every policy stream. So around the UN system, most countries have gone through the population aging. It caused a longevity revolution. And also there's another big trend. It's a digital transformation. AI and digital systems are becoming everyday infrastructure for healthcare and work and connection. So we are going through these two different transformations. But, you know, when they interact with each other, what our life would be like, and how it would affect the policy making. So here's, you know, life course perspective. So we have a life stage, you know, after born, you know, the childhood and middle age and other to do, and then a later life. But when I'm a gerontologist, we often call it gendered aging, and what does it mean? So I just drafted these images, and when you think about old age, we picture, you know, we have 20 more years after retirement, after 65. But it depends on your gender, your health status, your background, your geography. And so you can see that, you know, older women, there are more number of older women compared to men. But the reason is that women live longer. So when you talk about the aging, it's a gender issue. It's not just about health issue. It's not just about demographic issue. It's a gender issue. So when you think about the WHO aging -friendly cities and communities, WHO suggested eight different. The domains in 2002. So it's a transport. housing, social participation, et cetera. But when they designed this kind of domains of age -friendly cities, they pictured those things could be hardware and software, but these days everything should go through the interaction. So when we use transportation, we have to go through the platform, platform, ride services, and now the agent AI, when you work or when you travel, et cetera, you make a decision with your agent AI. And also even today's meeting, we have the Zoom here. So all these kind of mobile interactions are embodied in everyday life. It's an infrastructure. So technology is not just a tool. It's a tool. Part of where we live in and also where we... live with. And behind these technologies, there are AI. So AI algorithm has been advanced so much, and now it's product and also intact with everyday life. So I would like to give attention about AI and aging. So from home to healthcare, AI is becoming part of how all the others are supported, monitored, and connected and governed. So if you have parents or your grandparents or even your neighbors or even you have friends, you can have witnessed that, you know, how AI has been used for health monitoring and also smart home technology and also platform social connection. But I would like to highlight gender effect here. women tend to have lower access and AI literacy and also it's from the accumulated advantage and disadvantage. So gendered age is not only about age or gender separately. It is about how inequalities accumulate across the life course and become visible in later life. So I would like to highlight the important questions in today's session. So the fundamental question is that AI has a lot of benefits. We have AI for good and we have WISIS. It seems to have a future and it's very promising. I always like to surprise how agent AI is need. I'm a happy user but have you thought about the benefits and risk of this AI transformation distributed equally among older people? Some people might take advantage of a lot, but some people might be excluded completely because of this AI. And there are many documentaries about the digital divide, how older people struggle with everyday life because of digital transformation. But now we need to pay attention to the impact of AI on the increasing or decreasing the inequality issue. So how does gender shape all the other's experience of access to an outcome from this transformation? And we have a growing evidence about the inequalities. It's very hard to measure, also research, because it often requires longitudinal studies. But our lab has managed to do some experimental study with the AI system. And recently we have published a number of papers how the digital age is. And has played in the popular AI services. So today I would like to address the three important questions Who is being left behind and why And what does global evidence tell us and not yet tell us That's knowledge gap How can regional cooperation turn evidence into policy That's a purpose of ITU and also UN system So today we have two speakers The first speaker is Dr. Jingbo Huang Director of UN in Macau And because of her schedule of conflict She will join online Dr. uang, are you ready?
—
Dr. Jingbo Huang
Yes, Dr. Che.
—
Prof. Moon Choi
Yes, you can hear, we did the headset So please use headset. Yes, the floor is yours.
—
Dr. Jingbo Huang
Okay, thank you The headset, there's something wrong So I will just Go straight to the presentation So good afternoon, dear colleagues My name is Jingbo Huang Director of United Nations University Institute in Macau As you probably know or not know that UNU is one of the UN organizations. Our headquarters is in Tokyo. There are 13 research institutes in 12 different countries. And the institute that I'm heading in Macau is specialized in research, education, and training on digital technologies and SDGs. So today's topic is a very crucial topic. And older women, gender issues. And today I would like to bring additional perspective and angle, which is synthetic data. So first I would like to echo what Professor Che has been mentioning. Aging is not only a demographic issue, but also a gender issue. So the world is aging. Let me begin with a few numbers. According to the UN, by 2018, for the first time in history, people aged 65 and older, will outnumber all the... youth under 18 years old. So we're becoming an older world. Second, within that older world, the majority are women. The number of women over 60 will grow from about, you know, in 2000, we have a number of 336 million. And by 2050, the number will increase to 1 billion, a threefold increase. Already today, women make up 54 % of the world's older adults. Demographers call this the feminization of aging. So why does it matter? So this is a large, fast -growing and increasingly female population. It deserves our attention, and yet older women are among the most marginalized groups. When it comes to AI. And that's why they're so poorly, you know, because they're so poorly represented in the data that AI learns from. So I would talk about it from the UNU Macau's previous research. So our work, I would like to approach this topic from two angles, two research. One research is the synthetic data research that we conducted a couple of years ago. So how to use synthetic data to train AI models. So we developed this UN policy guidelines and launched at our previous UNU Macau AI conference in 2024. The other piece of research that we'd like to draw from is actually dated back to 2019, which Professor Che was also a part of, is an important contributor, is the equals report. And in the report, we look at the gender and technology. So we approach this topic from three areas. Access skills. And leadership. And we look at the gender and technology. And back then, the central message of the equals report is already very clear. We need a gender disaggregated data. Without it, we cannot even see inequality, let alone address it. Let's combine the data, digital tech and older women. If we combine these topics, we can see that we tend to amplify the marginalization of older women within AI models. And that in turn aggravates their marginalization in the real world. So this is a concern I would like to unpack in the rest of my remarks. Synthetic data. What is it? So synthetic data is artificially created based on the statistical data. So features of real world data. It resembles real data, but it can be hybrid. It can be partially real, partially artificial, or completely synthetic, completely artificial. So this is no longer a niche technique today. About 60 % of the data used to train AI models is synthetic. So in our report, which you'll find on our website, unu .edu .me, you'll find that in the report we mentioned that, you know, close to some researchers argue that by 2024, some argues that by 2028. So the real data, which I mean by accessible and the machine readable data will be exhausted. So the point is the same. The world is running out of real data to train AI models. So synthetic data comes into the picture and becomes more and more important. As a result, the production and the quality control of synthetic data are becoming a priority topic in AI governance. It carries both opportunities and risks. So from the opportunity side, it can address the data scarcity issue. Synthetic data can help address, for example, the lack of data issue, especially in the global south. The second opportunity, it can help if adjusted properly, correct the biases in the real data. So in the real data, there could be biases according to gender or age. If the data, the synthetic data is treated properly, these biases have the potential to be reduced or eliminated. The third opportunity is the privacy protection, because the identifying information from the real data can be removed in the synthetic data. So it's a way of protecting privacy. However, there are also risks, like any new technologies. So synthetic data can lead to infringement. It can lead to infringement of intellectual property. And more importantly for us today, it can amplify existing biases. If the data is not properly treated or corrected, generated carelessly, it reproduces the very gaps we're trying to close. So if we look at the synthetic data or AI models, they tend to be magnifying glasses of the real data biases. So we will need to be aware of these risks. On the gender tech side, that will bring me back to the central message of our equals report, the need for gender -disagreed data. For older women, this matters twice over. They're marginalized because they're women and, again, because they're older. Without data that captures both gender and age, they fall through both gaps at once. And synthetic data will only scale that absence. So, from our research and also in the future, we would like to make some recommendations, one from the technical side, so quality control. Technically build quality control and bias correction into how synthetic data is generated, set targets for both age and gender, and actively correct existing biases rather than replicate them. The second recommendation is from the policy level. Make the correction of gender and age -desegraded data found a foundation. Synthetic data cannot represent what was never measured in the first place, and synthetic data cannot assume fairness either. The third part is from the governance. Put a governance framework around this, anchored to the, as you know, the Madrid Plan of Action on Aging, the SDGs, the Global Digital Compact, and build through regional cooperation so that smaller economies can be integrated. Communities can share standards, and older women themselves should have a voice in it. We will need to do the participatory approach involving older women into every design. Future research, and we're interested in looking at how do we even measure, you know, when baseline data on older women barely exists, how do we even measure whether synthetic data represent them well? And the second research question that we're looking at is under what conditions does synthetic data reduce the marginalization of older women rather than deepening it? So in summary, older women are the fastest growing population on Earth. Synthetic data can help AI finally see them, but only if we govern it as carefully as we generate it. That is our work at UNU acau and is the governance choice for all of us. Thank you. Thank you, everyone, and thank you, Professor Che.
—
Prof. Moon Choi
Thank you so much for a very interesting research about synthetic data and inequality issue by aging and gender. And the next presenter is Ms. Eun -Chun Ko of Dr. Yudan, Dr. Kennedy rom Kaiser Graduate School of Science and Technology Policy, and she'll give a talk about the generative AI divide.
—
Ms. Ern Chern Khor
Thank you so much. Good afternoon. My name is Eun -Chun Ko. I'm a doctor candidate from Korea Advice Institute of Science and Technology and working under the Aging and Technology Policy Lab led by Professor Moon Choi. So my doctor dissertation focused specifically on investigating the structural disparities in AI, which means the uneven distribution of benefits and harms in AI across social groups. So today I will talk about some insights from one of my studies, which focus on the generative AI divide among older adults. So before beginning the presentation, I would like to acknowledge the support of this presentation by the NRF grant funded by Korean government. So some of you may hear about the term AI divide, which is often coined to describe the uneven benefit distribution. Due to the disparities in AI adoption or AI. So since the launch of ChatGPT in 2022, the growth of both generative AI applications and also users grows really exponentially. And it is important to ask questions about then who is left behind in this trend. So although the widespread use of generative AI are just happening recent few years, I would like to argue that what we call AI divide is not a new problem, but the results of old problems collide with new technology applications. So from the perspective of gendered aging, older women has long been recognized as vulnerable groups in digital use. And we have interventions, some interventions to help. To use more digital device and internet. But it is important to ask if what we have now is enough for them to also adopt AI, to bridge them. with AI adoption. So that's why in my study, I specifically look closer into older Internet users to investigate if there is AI divide even among older adults who are already using Internet. And this will give us a hint that on top of Internet access and Internet use, something else is structurally shaping who is using AI and who is not. So this study, we utilized the recently released 2025 Digital Device Survey in South Korea, which covers more than 2 ,600 older adults who are already using Internet. This study investigated the relationship between the AI divide, social demographic factors, and also AI attitudes and AI literacy. So first of all, the results as explained, showed significant gaps associated with older adults. women and lower socioeconomic status, which provides evidence on the existence of second -level digital divide in generative AI adoption. So first -level digital divide separate what we call internet users and also non -internet users. And among internet users, we can see that disadvantaged older women experiencing another divide when we talk about advanced application use like generative AI. So the next question in this study, we're trying to ask then what we can do to address this AI divide for older women. So policy interventions often target AI literacy. So yes, our results also confirm that AI literacy is the most important factor in explaining AI divide among all factors included in this analysis. So it seems like the answer comes from the AI divide. It could be simple, right? We can just direct resources to improve AI literacy for older women, and then we can narrow the AI divide. However, under the moderation analysis in this study, we can see a counterintuitive and alarming result, which is the increase in AI literacy can be translated into AI adoption much faster for older adults who are already in advantage positions than those in disadvantage positions. So what I mean by disadvantage positions here in this study means older age, women, lower socioeconomic factors, status, and also living in rural regions. This means that if policy intervention is not designed carefully enough, the important resources can be utilized much faster than those who are already in advantage positions, and thus AI divide may be widened further. So in conclusion, what we can learn from these findings, first, it is important to recognize older adults are not uniform demographic. So in the interventions we are doing right now, it's important to consciously notice who are those actively participating and utilizing the resources. You may actually notice that they're more younger demographic, more with higher education level and income, and those from urban regions. Second, we need to put actions to distribute resources specifically for older women, especially those who are more vulnerable, those who live in rural regions, less educated, and being burdened by family roles and social pressure. Third, we need to consciously include more use cases for older women when designing interventions related to generative AI adoption. And lastly, also the most importantly, it is tempting to direct resources to groups who can bring tangible outcomes faster. But this risks an even distribution of resources in policy interventions, which will worsen the existing AI divide. So it is important to consciously recognize that older women, especially those in vulnerable positions, can take more time to translate esources into tangible results. So it's important to recognize this point when monitoring and evaluating the interventions. And that is all for my presentation. Thank you.
—
Prof. Moon Choi
Thank you so much. And now we are moving to the panel discussion. And I have three guiding questions, but we have time for a fuller discussion, too. So if you have any questions, we'll take them. Okay, so we have two panelists and also two speakers. Mr. Wai Kit Si Tou, Economic Affairs Officer from UNCTAD, and also Dr. Tim Unwin, Professor from University of London. Okay, so I'd like to ask two panelists. The first question is, if there were one thing you would like to see changed tomorrow, whether in esearch, design, or policy, to place all the women at the center of AI for aging, what would it be? The very first thing you can do.
—
Mr. Wai Kit Si Tou
Thank you very much for the invitation and for the questions. From my end, I would say something. On policy, and more specifically on accountability. Because I think with our accountability, we can hardly achieve. anything. And if we really want to put an older person or older women at the center of AI development, we need some accountability mechanism to make it happen. And what I would mean is that we could learn from the ESG framework, that is the environment, social and governance framework that is quite well established nowadays. That is to mean for a company, they are not only required to report their financial situation, but also their impact on environment, social and governance. And I think that an AI equivalent could be made in terms of how a company discloses the information on how AI impacts different stakeholders across the AI life cycle. And apart from that, some public disclosure mechanism could be in place that make this company accountable to present how the AI model actually works . the decision -making process or the use, collection, and management of data actually work. So through this public disclosure mechanism, we could establish an accountability mechanism to engage private sector in the AI development, and I think that would be essential to put this in place to make it happen. And for more information, you are welcome to have a look at UNCTEC's fresh publication. We have this technology and innovation report, and we have a different recommendation, and this is one of the key ones that we would like to advocate for. Thank
—
Prof. Moon Choi
Thank you. Okay. Thank you so much. And Mr. Zekish To emphasized the transparency and also the importance of governance across the AI development cycle. And, okay, Professor Tim, would you comment on that?
—
Prof. Tim Unwin
Thank you very much. I'm clearly not a woman, but I think I'm the token old person here because all of you look so young. And I remember programming in Fortran in the 70s. Can you imagine that? In case you think I'm in my 50s, I'm actually in my 70s. That means I know some real. Really old people. But before I say that the one thing I'd change, can I, Chair, please build on what Ansheng Kuo was saying? Because I think if we use the word gender, we need to think about the full diversity of the meaning of gender and sexuality. And all too often in UN dialogues, it's equated with women. And I think you touched on some really, really interesting points towards the end of what you're saying. We've been doing some really fascinating research practice with the LGBT community in the favelas of Brazil. And let's put it simply, we think women are a marginalized category. Women, LGBT, well, they call themselves LBT very specifically. They're even more marginalized. And yes, it's right that this session should focus on women and age and everything we've heard so far. There are a huge number of other intersecting variables that you touched on. So, yeah, I like to imagine that this session is talking about older lesbian women with disabilities from ethnic minorities living in rural areas. So that's just something I wanted to get out there up front because I think it's terribly important. How many of you have cared for an elderly relative with dementia? So you know what it's like. Those who haven't, I don't think, can imagine what it's like. And I'll just tell you a story. My mom was a computer. So she was trained as a mathematician in the 1940s. She did the calculations for some of the first jet engines being built in Britain. She was an amazing brain, amazing mathematician. She was an early adopter of digital tech. She was a school teacher and she introduced computers in schools. As soon as the old. BBC computer was out there. So, yeah, we're talking back in the 80s. and she used Skype and she loved to talk with her children and her grandchildren, so great children and grandchildren on Skype. And then dementia set in. And it was really interesting and I struggled because how could a woman so able not use Skype anymore when she'd been using it? I still don't understand why it happened. But the takeaway from that is I think twofold. One is, and people describe dementia as they've gone over a river by a bridge to another world and you've just got to be with them in that world. Hey, that's not a world that all this stuff about AI is living in. And we're in danger if we don't let elderly people use their brains. The one thing we know is you keep your, your brain active, she did cross words, she did number puzzles and that went but if you don't keep the brain she lived till 92 if you don't keep the brain you get dementia and AI excessive use of all digital technology excessive use of AI is leading to what I call digital dementia we're in danger of creating a whole generation of people who can't think for themselves anymore that is totally scary totally totally scary so the one thing I would change is we have to involve these people from these I don't like the word intersectionality but let's use it this intersectional complexity in the design of AI itself if they are to have any hope at all in benefiting from it.
—
Prof. Moon Choi
Thank you so much the professor team emphasized co -designing process also the public awareness of aging from the marginalized population's perspective. Okay, so I would like to move to the next question. And so this question would go to Ms. Ern Chern Khor. So would you share a good practice, policy example, or real -world case that offers lessons for making AI more empowering and inclusive for all the women and all the others?
—
Ms. Ern Chern Khor
Yes, so like, so as what I introduced also in my presentation in the last slides, there are some recommendations I put on. And also, so we were working on a report about AI and aging. It's going to be presented two days later in the AI for Good Summit. And inside we collect 14 cases across the global. different countries about how those practices actually help older adults and older women in improving their AI literacy. But we do see some gaps, actually, in those case studies. So there's not many cases now focused specifically on AI adoption compared to digital and Internet use. So a lot of cases actually just build on what we have, like existing interventions and so on. And then everything is moving very fast, and people are trying to get tangible outcome as possible so they can get more investment into pushing those interventions more. But I think it's so important to value first, fundamentally value older women first, like not just about digital use or Internet use or AI use. Value what they know and value what they can do first before we use, interestingly. to empower them. So you have to believe in the power first. Second is about giving more time and not just go for fast and tangible resources. So if we value inclusion first, then that should be the priority rather than seeing fast or tangible outcomes. I think it is something more lacking now, and I personally think that it should be more valued in when we design interventions. Thank you.
—
Prof. Moon Choi
Thank you. Okay, thank you. Now I would like to ask the UN officers about the government perspective. I guess UN officers work with the government officers a lot. Okay, so from your perspective, what is one concrete commitment governments should make to ensure that AI reduces inequalities for all the women rather than dipping them? Dr. Jingbo Huang, would you comment on that? B
—
Dr. Jingbo Huang
efore I comment, I see a hand up before me. Dr. Che, would you like to allow the person to speak first?
—
Prof. Moon Choi
Sure. Okay. Mr. Zaki, would you comment on that?
—
Dr. Jingbo Huang
No, sorry. I'm saying that on the Zoom, there's one person with a hand up. Sorry.
—
Prof. Moon Choi
Okay. Sure. Okay. Ms. Paseba Seifu, do you have any comments or questions?
—
Audience
Yes. I just wanted to highlight, first of all, my name is Paseba Seifu. I am with Gender Empowerment Movement, SIGWRITE, and I wanted to raise a concern as well as maybe pose it as a question. Yes. And SIGWRITE, currently, they're in Ethiopia in general. There is something called data sovereignty policy, which means sovereignty of data, which means that independent researchers or independent humanitarian organizations cannot research and issue data because the government of Ethiopia has said that this is a matter of sovereignty. So this particularly will take it to the humanitarian aspect. It disproportionately impacts older women because they bear the brunt of humanitarian crisis. This follows the longest. This year, according to independent international credible research. the longest years of data blackouts, as well as deliberate siege by the Ethiopian government. So what is being done? And if there's no data, then the AI aspect will basically fade out because there is no AI without data. So what is the international community doing in regards to this? It's both a concern, raising a concern, as well as posing a question. Thank you so much for giving me the floor.
—
Prof. Moon Choi
Okay, thank you so much. It's a very thoughtful question. So her main point is that there are movements of sovereign AI, so many countries stopped sharing their data. That means that, you know, if we, you know, a person in, especially all the women in a country, if there is no available data about her, you know, AI never can learn about her life and that it can also create another inequality. So is that correct? My understanding is correct.
—
Audience
Yes, particularly when it comes to humanitarian data, it really impacts older women because they are at the receiving end of the humanitarian crisis disproportionately. And also this follows the longest year of siege according to international independent credible research in Tigray particularly. So that's my concern as well as my question. Thank you.
—
Prof. Moon Choi
I think the question is a little bit mixed. I think sovereign AI is the pillar, but also the Minor tissue is another one, but they can be a little bit different, separate agenda. But I think she, you know, tried to link those two things. But I think that question would go to Dr. Jinbo Hwang. Actually, she talked about the synthetic data. In some sense, synthetic data can be, you know, one of the approaches to solve that problem. And also, I think it's also part of low -resource language data set. Because most of the models that we are using are based on the large number of the population, say English, Chinese, or Spanish. But in South Korea, there are not many people who speak South Korean, right? So like that, you know, it's about the volume issue. But I think we can go back to Dr. Hwang's comment. Do you have any comment?
—
Dr. Jingbo Huang
Yes. So I think one thing is related to what I just said about the synthetic data. So that could be a solution. But of course, it takes a lot of the technical. and also policy work in between before we reach our ideal. The second piece of research that I would like to highlight or bring into this discussion is our recent work on interoperability of AI safety. And that also includes the data parts, data transfer, data across different borders, cross -border data flow. So if you're interested, please look at our research on interoperability of AI safety. And last year we did a research using four countries, Korea, South Korea, China, Singapore, and what is the other country? And UK. So this year we'll continue on this interoperability topics. As you have probably seen that in the first day global AI dialogue, interoperability is the best. And it's a very important topic. And so for UNU -Metal, we're continuing on this track. And this year we're working on the durability issues, particularly in the educational system. And, yeah, so this is something that I would like to share as additional angle to look at it. Thank you.
—
Prof. Moon Choi
And I also like to add one more. I think one of the KAIST alumni developed some huge data set about the twin system of Korean population, the perception. And then have you seen the news? And so that is an example, and it includes enough minorities, and it's available in Geek Hub and Open Sciences. So I think that could be another example. Okay, so time is almost up. So we'll take maybe one to question, and then we will wrap up. And any questions from the floor? Yes, Anisha?
—
Audience
Hello. Is it on? Okay, perfect. Thank you so much for your presentation. It was super interesting. My name is Anisha Tamawi, and I have a question. I sincerely apologize. I did not get the name from... the mister who talked about digital dementia but i thought it was a very important topic and it's one of the topics that are like personally concerning to me as well just like as a young person interacting with ai um and i was wondering because you were talking about like including complexity in like us interacting with the technology um if i understood you correctly but how would that work on like an implementation level because like if you look into like the model development it's becoming less complex for us as user usually especially if we look into like agentic ai so the responsibility for us as users is becoming less and less an actuality so how would that work uh do we have like governance like guardrails for that um what would you personally imagine as a i guess implementation approach to
—
Prof. Tim Unwin
that that's thank you so much my name is Tim um if I'm promoting as Jimbo was read my new book and it's got all the answers in it but but I think I'll answer it You may not like this, but remember the right to be unconnected is as important, if not more important, than the right to be connected. I have a neighbor, an elderly woman, who is increasingly unable to live because of digital tech, and AI is just making it worse. She used to be able to use a paper check and money. She really now has a fundamental problem with banking, because it's all internet, it's all digital. And just that one tiny example. And I loved your point about if we care about inclusion. If we care about inclusion, we do not, we provide this, you know, for people, they will die. Maybe we should be encouraging more forced suicide. And we're passing legislation. We're passing legislation in Britain to allow, you know, suicide. We should just get elderly people to kill themselves. women in particular because there are more of them is that what we want? I don't think so but if we really care about including them we have to go across the bridge to the land of dementia where they are and not force them forcing people to use AI is enforced slavery it's capture for their data that's all they are.
—
Audience
Just to follow on from that it's funny because I'm right now finishing up a masters in the management of AI and machine learning and an element of that is a human centered part of AI however none of what you just said was discussed is ever discussed it is always about keeping humans in the loop and for human oversight for AI. It's blowing my brain now because I really never thought about it even now. Never thought about it at all. It is always, we have to check, AI can make mistakes, AI can hallucinate, so new jobs are being created with the emergence of AI, with the emergence of this technology, and so yes, people don't have to worry so much about losing their jobs because these new jobs are coming, but indeed the inclusion is completely out of the question, so thank you for bringing that thought forward because I'm just thinking, yeah, let's get on with it and let's go along for the ride. All of us have to.
—
Prof. Moon Choi
I think the time is up, and so I think we have to wrap up, and I feel like we just started the discussion, but it's a short session, so maybe after the session we can you know, you know, have a small talk. But I think it's very important to emphasize that we often think technology is very neutral and then work the same way in everyone, but actually it's not. You know, even agentic AI, everyone use a different way. And also technology isn't often for the rich users, for their convenience. So it's very hard to be, you know, the equally, you know, benefiting people. So I think we are discussing AI for good, but we also need to think about, you know, for whom, and that we have to have more diversity perspective. Thank you so much for coming to the session and also, you know, give a big hand to the speaker and panel. Thank you so much.

Disclaimer: This is not an official session record. DiploAI generates these resources from audiovisual recordings, and they are presented as-is, including potential errors. Due to logistical challenges, such as discrepancies in audio/video or transcripts, names may be misspelled. We strive for accuracy to the best of our ability.