This article is written by André Corrêa d'Almeida, Adjunct Associate Professor of International and Public Affairs at Columbia University, and Bernardo Rivera Muñozcano, specialist in urban affairs, data and technology governance.

As governments adopt and scale Artificial Intelligence (AI) solutions, AI tools are becoming more complex, and increasingly deal with citizens’ sensitive and private information.

They have a growing — sometimes unsupervised — influence in public decision making processes.

Algorithms have of course made the provision of public services more efficient. However, theoretical and empirical research shows that these tools can contain and even magnify human biases, enabling the violation of basic civil rights and liberties by producing unintended consequences that can negatively impact hundreds or thousands of lives.

The critical conversation these days is not whether to use AI and data-driven methods to aid policymaking, but how to build and use them ethically. Fairness, bias, accountability, and transparency are increasingly frequent concepts in discussions between AI technical experts, scholars, industry professionals and legislators.

There are at least two critical aspects public entrepreneurs need to consider when venturing into the design and implementation of AI tools and systems for the common good.

First, a community-led identification and prioritisation of concerns that members of their communities have regarding the use of these tools. Second, the ability to work with technical experts in order to encode these context-specific values and priorities into the algorithms and their regulations. This article focuses on the first critical issue: community engagement in AI management.

Encoding People’s Voices in Localised Algorithms

Societies have different perspectives around issues of privacy and fairness based on their specific needs, attitudes, culture and institutions. The specific disparities between these priorities are critical, since there are always trade-offs when encoding ethics into computing models. As Michael Kearns and Aaron Roth argue in their book “The Ethical Algorithm”, there are unavoidable trade-offs when ‘tuning’ algorithms: accuracy versus fairness in classification, privacy versus analytical relevance of the data, among others.

Future studies and foresight methodologies may open a path towards community engagement in a complex public issue as this one

Instead of trying to come up with a one-size-fits-all framework for ethical algorithmic management, as it seems to be the mainstream discourse, policymakers should focus on their local desired outcomes (e.g. tenant’s privacy above efficiency in accessing their building) and reverse engineer their solutions and regulations accordingly. By doing so, the prioritisation scheme will vary based on these social preferences, historical circumstances and prevailing structural inequalities of local contexts. Promoters of AI localism, a growing field of research, highlight how by calibrating algorithms for local conditions, policymakers have a better chance of establishing feedback loops that will result in greater effectiveness and accountability.

This argument follows the logic of the immediacy and proximity of local governments to their communities, when compared to other levels of government, such as regional and international regulatory frameworks. An example of these disparities may be drawn around racial biases in algorithms. Even when this issue is paramount in one community, it may be difficult to use ethical guidelines drawn from this perspective and implement them elsewhere. The difference might easily show up in a community which takes in no clear legal categorisation of race, or where discrimination and social narratives are drawn around different racial identities that derive from other social constructs.

Implementing any community engagement strategy is challenging enough, and successfully engaging a non-specialised public in a highly technical conversation may seem impossible. However, future studies and foresight methodologies may open a path towards community engagement in a complex public issue as this one. In fact, some cities around the world have developed experimental exercises on AI governance with these future studies methodologies. Their uses could be explored beyond the experimental realm and into streamlined governance mechanisms.

Causal Layered Analysis (CLA), for example, provides an appropriate framework in scenarios where risk aversion and resistance to change are present, such as in the public sector, and most importantly for situations where inclusion of different world views, expressions and layers of participation are required.

By providing a space where conflicting worldviews intersect, this methodology also enables the creation of conversations where a broad audience can engage in the dialogue on new technologies, enhancing public trust in new approaches to policy making. By being able to incorporate this valuable participatory input, organisations can align the policy outcomes that can be informed by alternative layers of analysis (technical, communitary, legal, among others).

Anticipatory Governance offers perhaps one of the most interesting methods for community engagement in AI management.This methodology allows decision makers to choose and design approaches that incorporate prospective capabilities in their decision making processes, incorporating long-term social challenges, but remaining flexible enough to respond to unforeseen circumstances.

According to the Institute for the Future, anticipatory governance could mean incorporating forecasting, visioning, and participatory processes when setting public goals, engaging government institutions in committing to those goals, and measuring progress against them. The methodology allows governments to set up immediate feedback loops between technological tools and their social impacts.

Technology Roadmapping is a technique that can be used to explore technology developments and create frameworks to integrate them with larger organisational planning, and most importantly, with predefined social values and norms. This methodology is also useful to assess the impact of new technologies and market developments, and to capture their environmental and social landscape, threats and opportunities for a particular group of stakeholders in a technology or application area.

In the context of governments and societies facing external technological shocks, this technique can help policymakers prepare for solutions that account for organisational culture, new (unforeseen) technologies, and how the organisation’s internal capabilities can deal with these factors.

Given the open-ended nature of these methodologies, their flexibility to external shocks, and the (arguably) early stage of current adoption of AI in governments, their implementation is appropriate for contexts where the local ecosystem is yet to define the long-term AI objectives.

CLA and Anticipatory Governance can be used simultaneously to create a comprehensive roadmap that allows governments to adapt decision making processes on they go. The immediacy of community engagement and feedback loops at a local level has the potential to enhance the benefits of these tools in government services provision, while limiting the risks and harms to the communities where they are deployed. These local societal roadmaps could even influence future decisions in both academia, industrial research and development.

As we enter the third decade of the 21th Century, public entrepreneurs and innovators must continue to build tools, techniques and methodologies for the design of ethical algorithms in public decision making. As Kearns and Roth argue, this is critical to expand the principles on which our technology is based. If we are to demand that they incorporate many of the values we care about as individuals and as societies, we must do so hand in hand with the communities these technologies will be serving and impacting. — André Corrêa d'Almeida and Bernardo Rivera Muñozcano

We want to hear what you think about this article. Pitch an article to us or send your comments to hello@apolitical.co

(Photo credit: UnSplash)