AI can accelerate delivery – but overreliance can dull human expertise. Claire Mason, Principal Research Scientist at CSIRO’s Data61, has built her career studying that tension: how organisations can embrace emerging technologies without losing the human judgment, skills, and creativity that make them thrive.

As a leading authority on the social and workforce impacts of AI, Claire advises governments and industry on how automation and digital transformation are reshaping work itself. 

We sat down with Claire to discuss how organisations can navigate this balance — unlocking AI’s potential without losing the expertise that underpins Australia’s critical infrastructure.

FuturePlace: How should infrastructure and engineering organisations prepare their people for the impact of generative AI on roles, workflows, and project delivery

Claire: It is interesting that you ask how organisations can prepare their people for generative AI tools. When it comes to generative AI tools, it seems that people are jumping on board while the organisations are playing catch-up. Of course it takes time for organisations to translate AI guidelines and governance into practice. But survey data suggests that people are using these tools regardless of whether they are officially endorsed by their employer. Rather than attempting to prescribe or proscribe generative AI use, perhaps organisations should focus on ensuring that their organisational values, objectives and culture are well-defined and their people are able to access relevant training and information? I say this because there is lots of evidence that individuals are more productive when they use these tools but far less evidence that enterprise-level AI initiatives deliver value.

Your question reminds me of conversations I’ve been hearing lately about sovereign AI. These are conversations about whether Australia needs to build its own, ‘safe’ generative AI that we can rely upon. The truth is that relatively few organisations in the world have the computational resources, expertise and R&D budget that is needed to build their own, state-of-the-art generative AI tools. Fortunately, generative AI tools are highly flexible and can be adapted for a range of uses and there are open source generative AI models that we can fine-tune with local data. However, no generative AI tool is ‘safe’ in the sense that we can trust it to be consistently accurate and unbiased. The point I’m trying to make is that achieving efficiency gains and retaining agency and autonomy in a generative AI augmented workforce is not so much about making big investments in technology as it is about having well-informed generative AI users.

FuturePlace: What kinds of training and reskilling strategies are most effective in building AI-ready talent pipelines

Claire: One of the great things about the latest AI tools is that we can direct them in natural language – we no longer need to be a software developer or familiar with coding languages in order to adapt generative AI tools for our own use. On the other hand, human intelligence and context-specific expertise are critical to using it well. Formal training is part of the answer but with new and improved tools emerging all the time, we need more than a one-shot AI literacy course. One answer is to keep researching, experimenting and evaluating to stay abreast of the changing capabilities of these tools. If you can’t find the time to do that, I’d recommend following other people (ideally people working in your own field) who do, so that you can learn from them. Responsible AI frameworks are helpful as a source of guidance but these need to be complemented with human intelligence. Research suggests that human metacognition (the ability to think about our own thinking) becomes increasingly important when working with generative AI. AI is a powerful tool but it isn’t self-aware and it is through exercising our uniquely human self-awareness and metacognition that we optimise the use of generative AI.

FuturePlace: Are AI tools an answer for industries facing skills shortages?

Claire: The latest AI tools are best at supporting ‘knowledge work’ or cognitive tasks but they are more limited in their usefulness when it comes to the physical world of people and objects. It is interesting to note that many of the occupations where we’re experiencing skills shortages (e.g., truck-driving, welding, nursing and teaching) are roles that demand an in-person, human presence, involve inherently challenging work (dealing with people with complex needs, managing expensive and dangerous machinery) and do not attract high wages. In the long-term we may have autonomous trucks, remote carers and virtual tutors but in the short-term, AI can only reduce some of demands experienced by these workers, for example by optimising routes for truck drivers, suggesting engaging content for teachers to use with their students or assisting with information management in health care. But I suspect that other issues underlie the skills shortages that are seen in these occupations. Perhaps what we’re calling ‘skills shortages’ are really failures to incentivise workers to do inherently challenging work such as dealing with people with complex needs, being responsible for managing and potentially dangerous machinery or working outside in physically demanding roles.

FuturePlace: How should organisations balance short-term productivity gains from AI with the long-term need to preserve human expertise and critical decision-making capacity?

Claire: This is a great question and ties in well with emerging research which points to the mixed outcomes that can derive from AI use. For example, randomised control studies find that workers with access to generative AI generally achieve higher productivity and improved quality of work. However, in some instances, generative AI use can actually have a negative effect on human performance. Similarly, generative AI tools have been found to support learning in some contexts and inhibit learning in other contexts. These mixed findings serve to illustrate that the outcomes derived from AI use are not determined by the technology.

What we get from generative AI use depends on the choices we make. For example, we can use generative AI to get answers or to gain understanding. In the latter case, we are still cognitively engaged and we are likely to learn. However, if we offload the cognitive effort to the generative AI and rely on it for a solution or answer, we are less likely to learn. In contexts where the goal is to get something straightforward done quickly, cognitive offloading might be appropriate. It is difficult to prescribe how generative AI should be used but valuing and preserving human expertise and decision-making capacity so that it can be applied to making the best use of generative AI for a specific task or context, is definitely part of the answer. To this end, I should mention that the CSIRO is delivering generative AI training for organisations which focuses on building the knowledge, skills and experience to support productive and appropriate use of generative AI tools.

If you are interested in this topic, then check out DBW26 which is a national summit dedicated to Integrating AI-driven engineering and digital delivery in the construction and infrastructure sectors. Claire is leading a roundtable on AI and the workforce and is delivering a presentation on AI strategies for ensuring climate-resilient critical infrastructure.

At the 5th annual Digital Built World Summit, Claire will be bringing a crucial human perspective to the conversation on AI in infrastructure. She will lead a roundtable on “AI and the Workforce – Building Human-Centred Digital Talent for Australia’s AI Future,” exploring how organisations can prepare for generative AI’s impact on attracting, developing, and retaining digital talent. The discussion will highlight inclusion, human-centred design, and how to embed diverse perspectives in AI-driven teams.

Claire will also deliver an R&D Insight session on “AI Strategies for Ensuring Climate-Resilient Critical Infrastructure,” sharing how data-driven modelling and predictive AI are being used to anticipate vulnerabilities, inform maintenance, and support more resilient, sustainable infrastructure systems across Australia.

You can view the full agenda here.

Tags: ,