Recently, I was a panellist in a discussion about women in AI, held in March as part of the lead-up to International Women’s Day, alongside researchers and industry practitioners. The conversation covered a wide range of topics — from technical progress in modern AI systems to questions of governance, infrastructure, and societal impact.
In this piece, I want to separate myself slightly from the live discussion and instead reflect on some of the questions that stayed with me personally. These are my own interpretations of the issues raised, shaped by both the panel and my perspective as a PhD student working in AI-adjacent research.
AI systems as invisible infrastructure
One example that often comes to mind when thinking about AI in practice is weather and climate prediction systems. These models increasingly inform decisions in agriculture, energy management, and disaster preparedness. While most people never interact directly with them, their outputs influence when crops are planted, how cities prepare for extreme weather, and how risk is managed at a national level.
Alongside this, AI and data-driven systems are also increasingly embedded in governance and digitised bureaucracy — from welfare systems and tax processing to administrative decision-making and eligibility assessments. These systems often operate quietly in the background, shaping how institutions function and how individuals experience public services, even when the underlying models are not visible to end users.
This is one illustration of a broader shift: AI systems are becoming embedded in infrastructure that shapes everyday life, even when they remain invisible to end users.
AI and the scaling of expertise
One of the questions I reflected on most was the potential of AI to scale expertise.
AI systems can, in principle, expand access to high-quality support in areas such as personalised education, translation, and medical assistance in contexts where specialists are scarce. This makes them particularly relevant in discussions around global inequality and access to knowledge.
However, in practice, many of these tools are still developed and deployed in environments where strong infrastructure, data availability, and institutional resources already exist. This creates a structural imbalance between where capabilities are built and where they are most needed.
If AI is to meaningfully reduce inequality, distribution matters as much as capability.
Technology reduces inequality only when access spreads faster than advantage.
From this perspective, the central question is not only what AI systems can do, but where they are deployed and who they ultimately serve.
Data, privacy, and constrained choice
Another question I found myself thinking about was data and privacy in large-scale AI systems.
In principle, individuals should be able to understand when their data is being used and have meaningful ways to opt out of certain uses. In practice, however, the complexity of modern digital ecosystems makes it unrealistic to expect individuals to actively manage continuous consent decisions across platforms.
This shifts the responsibility towards system design, governance, and collective protections rather than individual decision-making alone.
Often, the choice presented is not really between participation and non-participation, but between participation and exclusion from digital life.
I also do not think it is accurate to say that younger people care less about privacy. Rather, many have grown up in systems where these trade-offs are already embedded into everyday digital participation.
Representation as pathways and signals
The question of representation in AI also stayed with me, particularly how it is often framed.
While representation is frequently discussed in terms of pipeline size, I find it more useful to think in terms of pathways and signals — how people come to see a field as accessible and open to them.
Many women may not see themselves reflected in AI early on, whether in school environments, technical spaces, or public narratives about what AI work looks like. Over time, these small gaps in visibility accumulate and shape who enters and remains in the field.
Representation, in this sense, is not only about recruitment at later stages, but about earlier signals of belonging.
One important step is making these pathways visible early — in schools, universities, and public conversations about AI. Equally important is creating environments where people can see long-term belonging, not just entry points.
Representation usually changes when the signals about who belongs in a field start to change.
Concluding reflections
Overall, my reflections from this discussion reinforce a view of AI less as a purely technical system and more as a form of societal infrastructure.
It shapes how expertise is distributed, how participation is structured, and how trust is formed between individuals and systems. It also raises questions that extend beyond capability — into access, governance, and the design of participation itself.
These are the questions I continue to think about: not only how AI systems are built, but how they are deployed, who benefits from them, and what kinds of futures they make more or less likely.