Artificial intelligence and equity: A conversation with Professor Parashkev Nachev

Parashkev Nachev is Professor of Neurology in the UCL Queen Square Institute of Neurology and was the lead for the roundtable on Equity. He sat down with roundtable coordinator, Audrey Tan, to share his thoughts on the most salient points of the discussion. Read more about the series and the Insights for Public Policy on the AI for People and Planet website.

What sort of policy work have you been involved with since the roundtable?

PN: I’ve been working with a team to develop a method of calibrating machine learning models in terms of equitability. Currently, models rarely assess a constellation of features beyond a patient’s health indicators so our model will incorporate performance as a function of other variables, such as age, ethnicity, and location. The power of machine learning is its ability to support precision medicine and to ensure that each individual has the best possible treatment and outcome. The idea of equitable measurement in performance reorients us much more towards the experience and the outcome of the individual patient, rather than the population being treated as a homogeneous group.

Are there other pieces of work that you’ve been working on or has it mainly been this equity tool?

PN: My research program involves the modelling of data so that we can achieve better predictive outcomes and better descriptions of treatments in patients. It’s divided into a number of domains and a suite of applications. I’m working with colleagues to explore other conditions where the complexity of the underlying factors cannot be captured in current machine learning models. We’ve been working with Professor Olga Ciccarelli (Department of Neuroinflammation) whose research is using machine learning to predict individual treatment responses for people with Multiple Sclerosis. Crucially, their tool will also account for elements other than just biomedical factors, such as demographics, diet, comorbidities and lifestyle. We have seen that there is a breadth of applications that these technologies can have for bringing greater equity to healthcare. These machine learning methods are broadly applicable across sectors, such as education, and the roundtable highlighted how our task is to create a set of governance and principles on which to transform the way we do things, while ensuring it is equitable.

Was there anything from the roundtable discussion that surprised you?

PN: What I thought was most interesting was there was a remarkable receptivity to the idea that machine learning and artificial intelligence can be part of the solution to challenges in healthcare. I’m very encouraged because often the discourse on the subject is relentlessly one-sided and very negative and without acknowledgement of the possibilities. I think there was broadly a consensus that the way forward is to have the right type of governance and identifying precisely which problem we want the AI to solve. This is something that needs to be worked on from a multiplicity of angles, with focus needed on how to change perceptions on the use of AI in health care so that patients feel reassured that this is being done for their benefit. Ultimately, this element of wanting to achieve the best possible outcome for patients is what’s driving work in this area.

Was there anything new that you learned or anything that you hadn’t been aware of before?

PN: Again, it was that the participants in the roundtable were so open to these ideas and being given the current political discourse is so one-sided.

What are the next steps in terms of, sort of public policy around this area, and the equity lens around AI?

PN: I think what’s crucial is that we foster more trust amongst people in the capabilities of AI. Solutions also need to be intelligible and work to address the problems that need to be resolved. The next step is then to make and deploy these new tools and technologies. That’s the purpose of the proposal that I described earlier — to create a platform that will enable anyone, regardless of the model or the dataset — to model performance against international standards. We want to think about the process of certification, just as you would with a new diagnostic test or therapeutic. We need to know the sensitivity and specificity of a model given characteristics that might be of ethical concern. So it’s a natural extension of the current framework, it’s a way to quantify what currently is only discussed in the theoretical and will add the ethical and equitable dimension to treatments.

Where can we build engagement following on from the roundtable?

PN: There are two crucial elements and the first is with Government. As AI technologies are increasingly being adopted across the Government, there has to be a demonstrable and traceable link between why these technologies are being introduced and how it will benefit people’s lives. While there has been a move to use more evidence-based policies, we currently do not have policies that are in the necessary conceptual realm and a lot of AI is conceptual. First we need to invest in forming the right kind of logical framework and then this framework needs to be integrated with the empirical. Again, this comes back to the research proposal I’m involved with — trying to find a way to actually evidence whether something is equitable.

The second area for engagement is with industry and working with the people from the roundtable who are based with industry to integrate equity frameworks within their work. Currently, there is a risk when companies are cleaning datasets that they will be de-anonymised and render people identifiable. This, unfortunately, does happen, but it shouldn’t be allowed. It’s crucial that companies working with data ensure nonspecific traces.

Did you have any other final thoughts in terms of the roundtable and things that were discussed?

PN: I think the most important to me was that we should try to move as fast as possible to something practical. It’s not enough to just discuss ideas, we need to implement tools that can have an impact on society.

_

More about the Roundtables

Jointly supported by UCL Public Policy, UCL Grand Challenges, the Business and Innovations Partnerships team and UCLB from UCL Innovation and Enterprise, UCL hosted a series of roundtable discussions on the topic of artificial intelligence (AI) throughout 2020–2021. This series brought together leading voices in policy, industry, third sector and academia with the aim of stimulating dialogue and forging consensus on how to deliver ‘AI for People and Planet.’