Scaling AI in government

This article is written by André Corrêa d'Almeida* and Bernardo Rivera Muñozcano**.

A recent survey sent to more than 1500 decision-makers in multiple industries and regions found that 40% of respondents are currently implementing at least one artificial intelligence project or plan to do so in the short term.

But when Cognylitica, the firm behind the survey, asked those same respondents if they were already working on a project or were planning to launch one soon that had elements of AI in it (defined by Cognylitica as one of seven “patterns”), almost 90% said yes.

• Want to write for us? Take a look at Apolitical's guide for contributors

In other words, organizations around the world are already using AI tools without even knowing it, and this lack of awareness prevents government leaders and decision-makers from fully adopting different AI applications and machine learning (ML) techniques, keeping them away from the transformative potential these tools can bring to their organizations.

Most importantly, public institutions have not yet developed the appropriate regulatory and organizational frameworks to embrace technological change. Just like the early 1990s when government began adopting mainframe computing while the private sector was already phasing out of that technology, we believe that failing to adopt and scale up AI tools today will create greater risks and losses for the government agencies in the medium- to long-term.

Creating experimental sandboxes safe spaces where procedural or regulatory constraints are reduced by providing more leeway for public innovators — could allow organizations to adopt and scale AI tools. Also known as strategic innovation niches, these mechanisms open up possibilities for innovation by relaxing the restraining conditions of public sector organizations and reducing different risks associated with public experimentation. These spaces can materialize in different ways, like the creation of subsidiary government agencies — serving as consultants for other teams without invading their legal attributions — or through a legal instrument providing a greater manoeuvre margin for government officials.

Organizations around the world are already using AI tools without even knowing it

Sandboxes allow the frontrunners to create an open-ended organizational culture, where public servants develop new approaches on the basis of their own practices and insights, and not by following closed-ended step-by-step rules or procedures which by definition cannot integrate unforeseen technological developments. We present a roadmap towards the development of these kinds of institutional spaces, while learning from and integrating their public servants in the process.

Experiment, adapt, repeat, within a community space

There is a fair amount of literature on how governments have traditionally favoured a small-scale, iterative approach towards innovation in policymaking. When dealing with the integration of AI tools, this piloting logic can achieve a systemic or operational streamlined adoption of AI in an organization if leaders work together with their communities — whether it is their employees/civil servants or clients/citizens — and enable them to experiment and share their ideas and concerns with other members within the same organization (read more in Smarter New York City: How City Agencies Innovate).

This approach improves the efficiency of the feedback loops, while creating a space for community engagement: the space where top-down directives and objectives intersect with on the ground domain expertise and learning. More importantly, this community engagement may serve as a segue for scaling the experiment into full-scale systemic changes.

An experimental approach to AI adoption will create the conditions for both leadership and organizational culture to serve as enablers for AI adoption

Breaking siloed teams is not only widely recognized as a way to improve an organization’s efficiency, but it also makes it more resilient and able to respond to crises and adapt to external shocks, such as the development of new technological tools. Under this collaborative approach, organizations should be able to cocreate a holistic AI strategy that is unique to their mission, ethics, and values. Anticipatory Governance, Causal Layered Analysis, and similar future studies methodologies could help institutions navigate the inherent ethical dilemmas derived from the use of biased technology and data, an aspect that has become a requirement for any government that uses automated tools that consume and learn from data in public decision making processes.

When creating this common ground, a common language needs to be established for experts and non-experts to guarantee cross-sectional cooperation between different divisions in an organization. For this marketplace of ideas to work, practitioners who possess both domain knowledge and technical expertise are needed. These translators, or bilinguals, are becoming a required asset in any modern organization that is willing to adopt and scale AI tools and systems. The presence of these mediators is necessary for these early stages of technological adoption. However, with the development of more sophisticated tech tools, and the growing presence in every aspect of our lives, access to knowledge and basic digital literacy skills for everyone — not only experts or bilinguals — will become even more essential.

Find your AI-champions

The creation of such a sandbox or marketplace where AI-champions can work closely with domain and operational experts to create AI pilots, is a good first step towards the required cultural change that any organization needs to scale its use of AI.

Taking advantage of the existing in-house creativity and resources also prevents organizations from establishing long-term dependencies to external vendors that inhibit resiliency and adaptive capabilities, and that may turn out to be more expensive in the long-term.

Table 1. Common barriers to AI adoption and examples of open-ended strategies to tackle them.

There are a number of systemic changes that need to happen for a viable prototype to scale into a full organizational mechanism, namely restructuring information systems and architectures that sustain any algorithm. However, an experimental approach to AI adoption will create the conditions for both leadership and organizational culture to serve as enablers for AI adoption, and will ease the integration of out-of-date processes and systems to these new tools. Some even argue that Proof of Concept approaches towards the scaling of AI tools are doomed to fail, favouring instead a piloting, iterative approach towards the integration of minimum viable products and their subsequent versions.

The game has changed

The transition from experiment to full-scale initiatives can take place after a clear operational advantage has been identified, where public value is being created, and once there is a strong, cross-siloed working culture. Organizations that have an experimental sandbox as a transition buffer zone, will be better prepared as some of these technologies begin to enter the trough of disillusionment of the hype cycle.

If technological developments keep growing exponentially, the impacts that AI is having in every aspect of our society and its institutions should not be considered as a once in a lifetime event. As policy and decision-makers, we must equip organizations with open-ended mechanisms to enable them to embrace the next big technological “game-changer”. — André Corrêa d'Almeida and Bernardo Rivera Muñozcano

*André Corrêa d'Almeida, Ph.D., is an Adjunct Associate Professor of International and Public Affairs at Columbia University, Senior Advisor to the New York Academy of Sciences on AI, and the author of Smarter New York City: How City Agencies Innovate.

**Bernardo Rivera Muñozcano, MPA, is a policy consultant specialized in urban affairs, data, and tech governance. He’s a LabCDMX and NYC’s MODA alum.

(Picture credit: Unsplash)