Helsinki commits to explainable AI and human oversight

Helsinki has created “ethical principles” for the use of data and artificial intelligence (AI).

The eight principles are: human-centredness, transparency, explainability, fairness and non-discrimination, accountability and trust, privacy, safety, and human oversight over AI operations.

A statement from the city said: “The principles were drawn up because our city wants to use data and artificial intelligence responsibly and act as a trailblazer for others.

“There is tremendous potential in the use of data and artificial intelligence, but the implementation must be ethically sustainable.”

AI systems already used by the city are listed in its AI Register, launched in 2020. There are currently eight listings related to chatbots and library services, and the city sees opportunities for further applications.

The statement said: “Artificial intelligence can create opportunities to provide services that are predictive, personalised and more efficient. By using the ethical principles, the city aims to minimise the ethical risks related to the use of data and artificial intelligence.”

Explainable AI

Mikko Niemi, Data Scientist for the City of Helsinki, told Cities Today that the principles were developed by internal staff with other stakeholders over the course of around 16 months.

Under the principles, the city has pledged to communicate as clearly as possible about how and where data and AI are being used.

”We will actively seek to share all the relevant information, unless prohibited by law. We disclose the identified shortcomings in the data. When possible, we will also share the source code,” they say.

To ensure accountability, a responsible party will be designated for each service that uses AI. They will ensure that the principles are followed, act as a point of contact for residents, and oversee AI systems where necessary.

The principles state: ”In some situations, managing system risks requires human oversight. In these cases, we will ensure that the responsible person is able to monitor and supervise the operation of the system and intervene if necessary.”

Similarly, when there are notable risks or legal impacts involved with using AI, the principles require outputs to be explainable. This means that the city would need to be able to explain an individual result or the algorithm’s general operating logic in a clear way.

“Explainability requires some special measures if the algorithm is complicated and risks involved are relatively high, and I don’t think that we have encountered such a case so far,” said Niemi.

As well as launching the AI Register, the city has also developed a contract element specifically for AI systems.

“Other processes regarding monitoring and upholding the principles are not fully formulated, but subject to ongoing work,” Niemi said.