When and How to Implement and Expand Artificial Intelligence in Government: Perspectives from an Expert Roundtable

(Tatiana Sokolova and Claude Yusti of IBM contributed to this post)

Last January, the IBM Center joined with the Partnership for Public Service to release The Future Has Begun:  Using Artificial Intelligence to Transform Government, which highlighted early public sector service innovators leveraging AI to improve service delivery and operational effectiveness.  The IBM Center then published Delivering Artificial Intelligence in Government: Challenges and Opportunities, which outlined a set of underlying conditions that could enable successful use of AI by government agencies.

In June, we announced the next phase of work by the Center and the Partnership in this emerging area:  two special reports that describe the impact and potential performance improvements that AI can bring to government in areas such as effective workplaces, skilled workforces, and mission-focused programs. The special reports will highlight how AI technology can assist agencies to deliver positive outcomes for their constituents, based on practical experiences and lessons learned with AI.

To gain insights that will inform these special reports, the Center and the Partnership are hosting a series of four roundtables, starting now and continuing into next year. These sessions seek to explore pressing issues surrounding AI, share best practices for addressing challenges, and work toward a roadmap for government to maximize the benefits of AI. They are being conducted in a non-attribution setting to promote candid and transparent dialogue among participants.

We held the first such Roundtable on July 17^th, which explored the potential of AI to help government and identified pressing and solvable AI challenges.  Attendees discussed steps an agency can take to go from detecting a problem to determining whether AI is the best solution to that problem, and finally to implementing the AI solution. 

Key questions included:

Foundational Considerations: AI’s current and future state

• What is the state of artificial intelligence in the federal government?
• What are the mission-related challenges AI is helping federal agencies address?
• How are agencies using AI to address their internal, organizational challenges?

How agencies can solve problems and share best practices with AI

• How do agencies assess and decide whether AI is the right solution to a problem?
• What should an agency know about AI before it starts using it?
• How can governments transfer learning about AI across topic areas and agencies?
• What should be the private and nonprofit sector roles to help government solve problems with AI?

A summary of key findings from the Roundtable, which benefitted from framing remarks by NIST and GAO, follows.

Foundational Considerations: AI’s current and future state

• AI involves algorithms that think and act rationally in the same manner as humans. AI capabilities include knowledge representation, perception, reasoning, prediction, planning, simulation, anomaly detection, control, and navigation.
• AI is not new, but does represent a significant advance in computing through algorithms that enable real-time analysis, as well as action- and decision-making across vast data stores – often in a mobile mode.
• The US Government has expanded AI activities, setting up a cross-agency committee led by the Office of Science and Technology Policy and addressing priorities that include funding for research and development, removing barriers to innovation by reforming policies and regulations, training the future workforce, and improving government services.
• GAO reports on a recent DARPA taxonomy regarding the evolution of AI in three ongoing phases.Agencies are making varying levels of progress in using AI for: 1) expert knowledge and logical reasoning (e.g. online tax preparation), 2) improving machine learning (e.g. advanced facial recognition), and 3) adapting data-based responses to different contexts (e.g. powering autonomous vehicles).
• A key advance in the use of AI will enable agencies to provide transparency for users on how AI improves their work – the concept of “explainable AI”, consistent with an .
• Some challenges to AI Implementation include:
  • The demanding levels of trust and safety required of AI systems;
  • The need for cybersecurity, using risk management to ensure robust system operations;
  • The acknowledged difficulties of machine learning architectures and training protocols in dealing with incomplete, sparse, and noisy data;
  • The difficulty in knowing what questions can be asked of an AI system and whether the answers make sense;
  • The need for explicit and standardized success metrics to avoid unintended consequences;
  • The risk that flawed AI implementation will raise opposition, which could lead to unnecessary legislative constraints on future AI use;
  • The need to ensure accountability, transparency, and ethical activity in AI use.

How Agencies Can Solve Problems and Share Best Practices with AI

• AI needs appropriate resource planning in a budget framework in a manner that allows for flexible adaptation with the technology advances – innovation sandboxes are a useful approach.
• AI is a powerful tool to improve testing in IT systems and monitor performance in programs – agencies can set up systems that do automated checks and reports for human decision-making, rather than require extensive repetitive data gathering.
• AI has helped the financial sector to address risks and reduce fraud, making it easy to apply advanced mathematics in accelerating analysis of anomalies and sorting through correlations to identify true causation.
• Social services agencies can use AI to improve speed and accuracy in claims processing and benefits delivery.
• In the national defense space, AI supports mission areas ranging from drone operations to air traffic control – the key in all of these areas is to ensure that data and algorithms are reliable, unbiased (or transparent about potential bias), and can be trusted as part of AI operations.
• The Intelligence Community uses AI to augment human expertise in areas ranging from counterterrorism to insider threats, moving from predictive to anticipatory modeling.
• State and local governments are making significant advances in the use of AI, in areas that include smart cities, customer service, cybersecurity, public safety, and geospatial information systems.
• AI is not always the solution to a problem. For example, AI will not fix poor data quality or problems with cars, when it is better to improve the data or fix the street.
• Agencies should think about implementation in an agile and iterative fashion – start small, learn from experience, and build incrementally, rather than attempt to construct large systems de novo.
• For AI to have sustained impact within and across agencies, leaders of all major functions must collaborate.Communities to engage include:
  • Chief Human Capital Officers to address workforce retraining and longer timer career impacts
  • Chief Financial Officers to review efficiencies gained from leveraging AI, and how resources could be channeled to address pressing agency challenges
  • Chief Acquisition Officers to develop improved procurement solutions that leverage emerging technologies.
  • Chief Information Officers to incorporate user experience and human centered design in developing IT solutions.
• Agencies should build capacity to share information and best practices, clearly and transparently, with one another and with the private, non-profit, and academic sectors. NIST and the Federal CIO Council have models to build on; the challenge will be to move from sharing information to collective action among experts in building systems and leveraging best practice.

AI Opportunities Going Forward

GAO recently reported numerous findings about the road ahead for AI:

• “Data are the new oil” – tremendous opportunities exist for all economic sectors to embrace key trends and shape them toward positive ends
• Rumors of the deaths of institutions and vocations are greatly exaggerated – jobs will be lost, but others gained (with possibly a net gain)
• Human capital development will require fundamental reconsideration and will need transformation to meet the future demands of most professions
• Vocations – including those in the public sector service – will be required to adapt to the probabilistic (vs. deterministic) paradigm
• The greatest challenges ahead are socio-economic and cultural, not technical

In the next Roundtable, the Partnership and the IBM Center will bring together experts for a discussion on how individual agencies and governments collectively could build a robust AI workforce, train employees (or hire AI experts), and identify skills these employees need to help their agencies succeed.  Future Roundtables will expand the discussion to an additional range of topics, such as how to leverage the economic and market activity around AI for government.

These Roundtables and the ensuing reports are intended to help scale integration of AI into the processes and work of government. Agencies have shown great interest in AI, though most initiatives are still exploratory and relatively small scale.  Early steps will likely continue until agencies have enough experience to understand the criteria and approaches for successful expansion of AI in a public sector context, rather than merely replicating commercial experience. In this next phase, agencies will build on lessons from organizations that have pioneered AI adoption – we hope that these Roundtables and reports will accelerate this progress, providing agencies asking: "Why AI?" and "When?", with actionable insights to move forward.

Image courtesy of thesomeday1234 at FreeDigitalPhotos.net