Challenges and strategies to manage artificial intelligence in the public sector
Most of the debate about artificial intelligence (AI) in the public sector revolves around its potential benefits, application areas, and risks involving privacy, security, and ethical concerns. But little in-depth research has been conducted about the organisational challenges of developing AI – a critical aspect considering the potential costs.
In the United States alone, federal government expenditure on non-defence AI research and development was expected to be over one billion dollars in 2020. Other governments are devoting far more. The Chinese government, for instance, is expected to invest more than 30 billion dollars in AI and related technologies, and the city of Beijing has already committed 2 billion dollars.
What are the organisational challenges that public organisations face when trying to develop artificial intelligence projects? Marc Esteve (Esade), Averill Campion (Esade), Mila-Gasco Hernandez (State University of New York), and Slava Jankin Mikhaylov (Hertie School) provide answers to this question in the journal Computer.
Developing artificial intelligence in the public sector involves several challenges that may hinder its success. “Our study shows that most challenges arise during implementation and relate to skills, culture, and resistance to sharing information driven by data challenges,” says Esade professor Marc Esteve.
3 challenges of artificial intelligence implementation
1. Lack of skills
One of the common challenges when adopting artificial intelligence in the public sector is a lack of data literacy and skills. There is a general lack of awareness about how to apply AI techniques to solve problems and, therefore, about the ability to ask data-oriented questions.
Addressing this skills gap is crucial for AI projects: “Developing AI projects in some cases requires that people in organisations have a basic understanding of what a data-oriented question is and how it can be solved using an AI technique,” says Marc Esteve.
One of the common challenges when adopting artificial intelligence in the public sector is a lack of data literacy and skills
Another important challenge is training high-level executives in public organisations so they can also ask the right sets of questions and be involved in pushing forward AI-driven projects.
2. Lack of a collaborative culture
AI projects are often collaborative in nature, which means that obtaining the data needed may require data sharing among units within the same or different organisations.
For AI projects to flourish, organisations in the public sector must be willing to share their data. “Our findings show that the lack of a collaborative culture, based on data sharing, becomes a key challenge during the first phase of implementations,” warn the researchers.
Their findings also show that the history of past collaborations, along with a lack of trust, play a major role in shaping the lack of a collaborative culture.
3. Resistance to share data driven by data challenges
The authors explain why there is resistance to sharing data: “Most of the participants in our study recognised that they needed to share data with other organisations for AI to be successful, but that doing so is not always easy because of some important features of the data.
“For example, although most interviewees referred to the importance of data integration, they also agreed that AI projects require a lot of data, which may result in availability and quality issues. Data quality, which includes issues such as consistency, integrity, accuracy, and completeness, is key when data must be shared and integrated."
Another issue which increases resistance to sharing data, say the authors, is a legitimate fear about privacy and ensuring that private data is not compromised.
5 strategies to manage artificial intelligence in the public sector
In their findings, the authors identify the following five strategies to overcome the major challenges of artificial intelligence implementation. These strategies cover data standardisation, training, data-sharing agreements, political will, executive support, and stakeholder management.
1. Data standardisation
To address resistance to sharing data, the researchers recommend tackling data-quality issues, which means developing a standardised set of guidelines and procedures for quality. This could result in a formal process of moving away from fragmented quality standards held by various individuals and organisations and towards more systematic rules and frameworks.
2. Training
Although training and skills development are an important strategy for successfully implementing AI, the researchers identify the need to develop awareness capabilities and a strategic understanding of data. In other words, training public sector employees how to turn a challenge into the appropriate data-based question and helping them understand how to better leverage their data.
For AI projects to flourish, organisations in the public sector must be willing to share their data
3. Data-sharing agreements
Data-sharing agreements should be made that help build trust among the units and organisations involved in AI projects and so increase the willingness to share data. This type of agreement becomes a tool to build trust among organisations.
4. Political will and executive support
During the implementation process, political and executive sponsorship must provide the necessary legitimacy to keep pushing AI projects. The authors explain: “If you have executive sponsorship and a clear directive from the executive, then doors open pretty quickly. But if that is not the case, you can be knocking on doors for a long time and get nowhere.”
5. Stakeholder management
"Our findings show that AI development often involves collaborating with other organisations because these projects may cut across many fields, such as domestic abuse and child welfare. Identifying the stakeholders of an AI project and planning and implementing actions designed to formally and informally engage with those stakeholders becomes an important management strategy." Stakeholders involved in AI projects may include the AI team, the organisations that hold the needed data, the beneficiaries (usually the general public), and any other partners.
In addition to these specific actions, building long-term cooperation also requires increasing trust at the individual level in public organisations. “Balancing long-term and short-term strategies is key in addressing implementation challenges, and leadership plays a key role in doing so”.
The authors conclude: “Short-term actions contribute to addressing specific problems that hinder daily operations – while long-term strategies provide legitimacy, stability, and the sustainability of AI development”.