Dr Melanie McGrath heads to Europe for AI conference tour
Image: Steven Waller/Unsplash.
In her recent trip to Europe, Collaborative Intelligence FSP and Responsible Innovation FSP researcher Dr Melanie McGrath engaged with the emerging community of interest around how people interact with AI.
Her current research is focused on building a framework for trust in collaborative intelligence; in other words, how trust works in teams where humans and AI agents collaborate to achieve shared goals. “I’m interested in understanding what role trust plays: how we can build it and make sure that it’s appropriate for the capabilities of the system,” said Melanie.
“We’ve been studying what makes it possible for teams to work together and coordinate outputs for decades,” Melanie explained. Trust in these teams is often influenced by factors like effective communication, and clarity around team members’ progress towards shared goals. “My question is whether we can adapt these processes from teams of people, to teams with people and AI?”
An important step toward answering that question has been identifying relevant processes from existing research, and building them into a framework to see how interaction with an AI agent affects patterns of trust development.
Exciting applications
An example of where collaborative intelligence can add value in CSIRO includes ongoing research with our Agriculture and Food Business Unit, on how to bring together AI and human expertise to annotate plant genomes with greater accuracy and speed.
“Currently, genomes are annotated by human experts, which can take significant time. But when you apply AI to complex biological data, it can be very inaccurate. We’re looking at combining researchers’ expertise with AI efficiency, to streamline the process,” Melanie explained.
Another example builds on the work being done with radio telescopes that CSIRO manages. Substantial amounts of cross-referencing is required to identify new celestial objects. By finding ways for researchers to engage AI systems in the task, it could help identify anomalies in hours, rather than days.
“The key thing we’re discovering so far is that trust in AI is going to be highly dependent on context: on the task you’re doing, the environment, time pressure, and risk factors. This means we’re unlikely to develop one model of trust to suit all contexts: instead, we’ll need to find models that apply to particular systems or applications, and test them.”
From Canterbury to Kraków
Melanie’s first stop on her conference tour was Canterbury, UK, for the Moral Psychology of AI conference. Discussions focussed on what we can learn about ourselves from how we interact with technology, with presentations from philosophers, economists, computer scientists, and psychologists addressing the question from diverse perspectives.
They also examined the moral questions arising from the development of more complex and capable AI, as we start to assign it tasks and responsibilities that impact others. “It was a conference of big ideas, as we think through the consequences and implications of AI’s growing role in our societies,” said Melanie.
Next up was Munich, for the Hybrid Human AI Conference (HHAI 2023) – an interdisciplinary conference aimed at understanding human-AI teaming. Bringing together expertise from computer science to psychology, the focus here was on the “how”: how to approach building human-AI teams, and enable them to work together effectively on tasks.
Melanie’s final stop was Kraków, Poland, for the 19th General Meeting of the European Association of Social Psychology (EASP). The largest social psychology conference held in Europe, EASP 2023 attracted more than 2,000 submissions and 1,700 attendees. With emergent research on the social psychology of AI set to take a bigger and bigger role in coming years, Melanie’s presentation attracted a keen audience.
“Across the board, there was so much curiosity about what we do at CSIRO,” Melanie said. “It’s widely accepted that we need to bring together different perspectives, skillsets, and disciplines, and it was great to connect with communities that are working on the same things we are, but who may not have the capacity to address the full range of questions within a single organisation. People were excited to hear about how we have truly interdisciplinary teams at CSIRO, applying all our sciences together under one roof to deliver impact.”