Go directly to contents.
Good practices for Service Developers

Using AI responsibly

  1. Start
  2. Introduction to data ethics
  3. Assess the capabilities of your organisation
  4. Organise ethical practices
  5. Get citizens involved
  6. Define a data policy
    1. Well-working AI requires high-quality data
    2. Biases have to be identified by humans
    3. The algorithm determines the quality of AI outputs
    4. Big data creates a black box
    5. Citizens have the right to the transparency of information
    6. Challenges of recommender systems in the public sector
      1. Even the public system can recommend and tailor services
      2. The amount, quality and use of information pose ethical challenges
      3. Reflect on these when considering a recommender system
  7. Pay attention to security
  8. Checklist

Challenges of recommender systems in the public sector

Even the public system can recommend and tailor services

In addition to automated decision-making, AI systems can be used for customised, personalised services where the system produces recommendations for individuals instead of issuing binding decisions.

Recommender systems are common in the private sector, such as the “you may also be interested in these products” feature in online stores, or music and programme recommendations in streaming services.

In principle, the same technology can also be used in the public sector to direct clients to services that the system deems useful and to tailor service content to individuals’ needs and situations.

Updated: 9/11/2023

The amount, quality and use of information pose ethical challenges

One ethical challenge of recommender systems in the public sector is the quality and quantity of the information required from people and the transparency of its use. Unlike a music streaming service, a public service may require quite a large amount of people’s data that has to be correct, up-to-date and relevant.

This requirement may be emphasised in ecosystem service models where many service providers from different sectors process a person’s data simultaneously.

Updated: 9/11/2023

Reflect on these when considering a recommender system

If you are considering a recommender system, note that

  • A person’s data does not equal the entirety of a person and their experience of life in the real world. How certain can we be that individual data points create a picture and understanding of a person that is so reliable that, in accordance with good governance, we can generate recommendations and proposals for them mechanically?
  • AI systems are trained with data that always reflects the past. We very rarely have real-time data on people. Old data may still be useful, but it is also possible that it no longer reflects a person’s situation and needs, which means that any generated proactive recommendations will not be appropriate.
  • Recommender systems may also use proxy data for profiling if not all of the necessary information on a person is directly available. Such data may include the location of the service user, user-created connections to other people and searches made in the service. Proxy data can generate a surprisingly accurate picture of a person without the person’s permission or there being a real purpose for it in the service.
Updated: 9/11/2023

Are you satisfied with the content on this page?

Checklist