On the 'user' in User-Centric AI

Written on October 2, 2021

AI and ML have touched numerous aspects of our lives in today's world. Generally, AI-powered products or services are put out into the world by developers and innovators and user consumption and approval take a natural route. Depending on perceived utility value, the service may catch fire and become popular.

Off late, there has been a deliberate shift in how AI applications are developed and deployed. Academia and industry have started to shift their focus to user-centric AI research. I think this stems from the recognition that when valuable time and resources are invested into the research and development of new AI-based products, it might as well be to directly achieve end-user goals and needs. The question is still, what is the role the 'user' has in user-centric AI? And how do we facilitate it?

My research over the last 6-7 years has been in the space of assistive robotics. By definition, assistive robotics as a field exists to 'assist' humans. My work primarily focuses on building AI/ML-based technologies to assist people with motor impairments in activities of daily living. My research is worthwhile if and only if it brings value to the end-user, particularly, in terms of perceived improvement in quality of life. As a researcher working in an academic research lab, I grapple with the question of how to streamline my research and identify useful end-user objectives that I can direct my focus and attention to.

One of the critical components to the solution to this question is the need to have an established iterative framework that would bring the end-user into the design and development phase of these assistive technologies right at the outset. Empower and emphasize the user! There are few aspects that I would like to highlight:

Firstly, there has to be a greater emphasis on developing common terminologies and communication protocols that would effectively bridge the gap between the two primary stakeholders in this endeavor namely, 1) the researchers and engineers who are responsible for building the technologies and 2) the end-users who are the consumers. This might require some outreach efforts through which we can enlighten the end-user about the technical and engineering aspects of these technologies. On the other hand, researchers and engineers need to develop skills beyond technical writing communication and become skilled in approaches to community engagement, and participatory design.

Secondly, the process has to be iterative and continual and should target long-term engagement. This is important because it is unlikely that both parties are fully aware of what is needed or what is possible. The initial designs and prototypes can become a catalyst for collaborative innovation and can inspire both parties to think of ideas that didn't exist before. Anchoring new technologies in something that the end-user is already familiar with can act as a great hook to start the conversation. By being 'on the edge', we can always feel a sense of security and yet push the horizons of what is possible.

Lastly, and this is particularly important for academic researchers working in this domain, the right incentives should be in place for researchers. Academia, particularly in AI and machine learning space, aims to achieve high churn rate when it comes to publications. This can be detrimental for more long-term and slower research cycles. This can disincentivize researchers engaged in longitudinal studies, as they run the risk of producing less number of publications in top-tiered journals and conferences, thereby diminishing their future prospects.

It is indisputable that AI and ML will become a dominant aspects of lives whether we like it or not. The challenge is to focus the energies on the right priorities (and how to implement them) in a way that the ultimate beneficiaries of these technologies are not left out in the process.