A study commissioned by the European Commission highlighting the significant potential of AI to improve public sector services was published a few days ago. The study focused on the EU, however there are valuable lessons for every other government to consider, too.
The study looked at various AI initiatives from the EU countries and while 72 initiatives were identified, which for a union of 27 countries is relatively impressive, 33 of those were strategies, 13 legislation, 4 reports and guidelines respectively, 2 frameworks and 1 declaration. There’s nothing wrong with creating a supportive ecosystem for AI implementation, however the overall tendency should be more on the practical initiatives that bring with them both some type of tangible outcomes and better understanding for what the ecosystem really needs. 10 of the remaining initiatives were action plans and 5 programmes. Strategy is important, but it needs to be turned into practical outcomes. Every government that considers AI implementation needs a vision, core principles and key activities, however being a viable and active stakeholder with your own AI projects in development is just as crucial.
Talking about those AI projects, the study states that the four public sector areas with the greatest number of AI projects are general public services (public administration), transportation, police services, and public health (medical equipment). The most common typologies are chatbots, computer vision, machine learning and predictive analytics. Interestingly, chatbots are not only the most popular AI projects (mainly found on ministerial websites and thus seen as improving those “general public services”) but also are by far the biggest spending article with the estimated 42 000 000 €. The next typology on that list is machine learning with just over 10 000 000 € and computer vision and identity recognition with 7 300 000 €. The numbers are very surprising and a bit alarming. It is difficult to reason yet easy to understand why chatbots are so popular. They offer, in theory, great wins by alleviating customer service agents’ workload and being available 24/7, but in real life need lots of high quality data to be able to answer questions correctly and adequately. Chatbots need to be integrated with existing business processes and civil servants need to be educated on them. Moreover, the general public needs to be accepting and willing to use this new channel of communication. With all this in mind, it’s interesting that chatbots are so much more popular than image recognition for example, that can be used in so many ways (traffic management, wildlife detection) and with a more specific goal in mind thus also being an AI project that is not going to be that difficult to integrate with existing processes and technologies.
To parallel this, the study states that there’s a lot of unknowns in AI and this leads to the “AI for the sake of AI” situation where something is procured more because of the hype and less because of actual necessity. This in turn leads to only surface-level integration and the transformative change that is promised and theoretically possible, is not achieved.
The study also focused on core AI skills, or more the lack of these skills in the public sector. This is, understandably so, seen as hindering the AI uptake. What is odd, is that the first perceived solution by respondents is hiring in-house AI experts that with their higher expectations for salary is the variable behind the high cost of AI adoption. The lack of in-house AI experts is also seen as one of the causes for poor procurements and biassed algorithms. It is proposed that creating guidelines (“standard templates, policies and legal frameworks, as well as ethical considerations”) on procuring AI solutions should relieve that.
The lack of core AI skills is definitely a problem, however, because the perceived solution is hiring “AI experts”, it is first important to define “core AI skills”. A civil servant should not be an advanced Python coder - a civil servant should be well versed in the domain they work at, so that they could be a valuable partner to external AI experts. Most often the problem with any IT (AI incl.) project is the lack of understanding the root problem which comes from the lack of domain knowledge. Same with guidelines, as anyone who has ever had to write procurement documents can tell you, any template is much appreciated (although like with many other things, templates need to be kept up to date regularly). The best solution, however, is combining the two. The study mentions the Estonian public procurement guidelines. If you were to ask what resulted in most AI projects in Estonia, it was not (just) these guidelines, but bringing together civil servants with their very specific domain problem and different data scientists from the private sector telling them about different ways AI can be employed. This is why we at Digital Nation believe in the practical service called “AI accelerator” where we bring teams of civil servants together and with the help of very experienced data scientists start defining the domain problem into AI projects.
The study itself brought up the unrealistic expectations set on the technologies. Bringing the two sides together enables the civil servants to get practical feedback and answers to questions left unanswered on AI.
Digital is a leader’s job, so when it comes to skills and knowledge, managers cannot be overlooked. The study emphasises the importance of top management and their interest and understanding of AI. This is seen as the key to increase “buy-in” when it comes to any new ways of doing things. The way to get the point across, much like with other civil servants, is showcasing practical AI use cases and their business benefits.
As expected, data, the lack thereof, and the quality and processes were also heavily featured in the study. Most prevalent problem the civil servants are facing is data ownership. There’s a lot of grey area with sharing data: the general sense seems to be “the less you share the safer you are against any wrongdoing” because of vagueness stemming from no clear guidelines and the lack of data strategies. As with any type of strategy, a comprehensive and practical data strategy is a must-have for every government and will be of big help to those civil servants that deal with data by setting a clear vision and core values.
Next to a data strategy, many governments, unsurprisingly, were struggling with data governance. As AI enjoys the hype, data governance is the key to unlocking the promised benefits of AI. This link is still not understood by most governments as the lack of investments and interest on the topic are very prevalent. The study also suggests moving away from the old “data is the new gold” rhetoric as this is not conveying the practical value behind data and AI. Similarly to the AI accelerator, Digital Nation believes in empowering senior leaders with core data governance principles in a practical way to enable the previously mentioned transformative change within the public sectors of the world.
Overall, the study encompasses many important points, much that are already known, yet still it is important to emphasise these. Without any attention on these problem areas, it is easy to not focus on them and keep the status quo. Read the full study here.