‹ Back

 

Role of product managers in AI ethics

by James Kingston

 

AI ethics is product ethics

We live in a world of products. As a human in social space, every item you touch - your hat, your laptop, your cup of tea - is the product of choices made by those who built it. The hat producer has made choices about the quality and the sourcing of its fabric; the laptop producer has made multiple choices about the sourcing of its components, the design of the finished product, and the marketing and distribution by which it meets the market; the maker of my tea mug has considered its shape, aesthetic, and pricing. But it goes further, for the the makers of these products are also making choices about the behaviour of their users, actively reshaping them. My purchase of a hat alters my daily routine; I now can’t stand a rainy day without it. My laptop socialises me to a day spent working or playing at my desk, somewhere not too far from a power supply; it may also lead to me buying painkillers for neck pain. My mug trains me to expect upright liquid holders rather than bowls, to signal my aesthetic tastes via its design, and the fact of it sitting next to me as I type makes me want another bag of Twinings. 

 

Product choices are nothing new, and nor is it new that these choices have an impact on human behaviour. What is new is the sheer ever-expanding variety of products we interact with in the digital world, and the increasing power that products have to remould the behaviour of individuals and society at large. Twenty years ago, there were no phone addictions, no social media influencers, no recommendation algorithms reshaping the content preferences of visitors to Youtube or any of a hundred different platforms, no viral video challenges. As AI capabilities advance at dizzying speed, and with AI-enabled tools, services, and platforms already altering societies and economies throughout the world, the way we choose to design products - whether physical tools, or tech platforms, or service systems - is of huge ethical, social, and economic importance. 

 

Cut away the hype - the grand Google statements on ‘AI ethics’, the letters about ‘Killer Robots’,  Trolley Problems, or the risk of the observable universe being turned into paper clips - and what you have, even as regards many of the more exotic problems around superintelligence, is a debate around the practices and principles of product development. Most of ‘AI ethics’, just like the wider space of ‘tech ethics’ is product ethics, and product ethics is determined by those who make decisions on the development and design of products - product managers. 

 

The changing role of product managers

The role of product management is in rapid evolution. As technology companies have moved away from on-premise systems and into the cloud, the role of the product manager has expanded. Described as a ‘mini-CEO’, and as a ‘conductor’ of product development, product managers must orchestrate and prioritise multiple technical, commercial, and operational workstreams in order to build effective and scalable products and product features. It is a central strategic role in modern tech business. Sitting as it does across multiple workstreams, it is also the one best-placed to have impact in designing and implementing ‘ethical AI’. 

 

Many product choices are potentially ‘ethical’. There is, for instance, the choice of data - what training data do you use, what do you seek to do with it, and where have you got it from? There are choices around what 3rd party services one chooses to use - what APIs do you use, and where are they getting their own data services from? There is the question of what behaviour you want to promote in your users, or of what interactions you allow them to have with one another. Consider the success of Bumble vs Tinder - very similar apps with highly dissimilar user experiences made possible through a simple choice of product feature, the decision that conversations can only occur when a woman messages first. 

 

Both investors and product managers increasingly recognise the importance of responsible product management. Of three dimensions of responsible product management - privacy sustainability, inclusion -  McKinsey found that product managers rated privacy and inclusion above ‘performance’. In some ways this is unsurprising - customers, investors, and other stakeholders are awakening to the importance of responsible design. 

 

Many product designers are already attempting to build responsible products. Privacy is a central concern - years of data collection scandals, and regulatory requirements such as those imposed by GDPR, have socialised PMs and companies into the need to ensure their products have some level of privacy. Indeed, McKinsey found that some PMs went so far as to call it an ‘existential reputational risk’. A desire for privacy drives choices around user interfaces - how often should users be asked, for instance to give permission for data use - as well as more fundamental features of data collection and storage, such as decisions to use differential privacy frameworks, or federated learning systems. 

 

Sustainability and inclusion are also both increasing salience. Historic datasets may contain patterns that reflect racist and gendered presumptions or behaviours that we would not want to see reflected in present-day systems; no product manager wants to create a racist or sexist AI system. Thinking through use of data, as well as how a system will be used in the wild is key.

 

Uniting each of these, however, is a common thread of difficulty. In order to be acted upon, the various issues and challenges we ascribe to ‘privacy, sustainability’, and ‘inclusion’ must first become legible to product managers and the organisations in which they operate. In order for this to happen, such concepts - often highly subjective and variable - need to be measurable and quantifiable, constructed in such a way that metrics can be created and outputs assessed against them.

 

A practical agenda

As ‘mini-CEOs’, product managers are among the most busy members of any tech organisation. In a world of rapidly evolving priorities, not every priority is equal. Investing in training, the development and articulation of internal metrics around responsible and ethical development practices, and a degree of flexibility are important. There are various solutions.  

 

Training: The first is investing in the training of product managers themselves, helping them via workshops and other coaching methods to understand the ethical dimensions of their work. 

 

Assessments: Responsible product management should be a component of personnel assessments for product managers, and have an impact upon promotion and compensation.

 

End-to-end ethics: Responsible product management must be implemented from the very beginning of the product design and development process. PMs should work with stakeholders including policy teams and customer advocates to ensure that development occurs in a responsible way, one alert to the impact of product choices upon users and to the wider policy positioning of the organisation as a while

 

Professionalisation: Another solution is that product management develops itself as a new professional discipline, imposing standards of ethical training, review, and sanction. Here, an ‘Institute of Product Managers’ could give accredited training on principles of design thinking, risk management, and design accountability

 

To summarise, ‘ethical’ or ‘responsible' product management for AI should become an element of risk management - one baked in from the beginning, and explicitly and implicitly saturating the overall product development perspective.