To make sure ethical questions are taken into account in all our future AI solutions, we wanted to discuss them in advance. That is why we decided to accept the AI ethics challenge.
For the purpose of the Finnish Tax Administration’s operations, it is essential to promote our customers’ willingness to pay taxes. When our customers are motivated and sufficiently informed to do the right thing, we can collect the taxes prescribed by law and avoid a tax gap. However, we can only maintain our customers’ willingness to pay taxes if they trust the Finnish Tax Administration to do the right thing and uphold ethical standards.
Adherence to ethical standards applies to many other areas in addition to artificial intelligence, and it involves compliance with laws and regulations, sufficient data protection, always using reliable data sources and making sure there is a responsible party behind every decision.
Admittedly, there are certain special challenges involved in the use of artificial intelligence. Strict demands must be placed on data quality and data protection. However, the biggest challenges may not be technical but ethical. Even if something is technologically possible and perhaps most efficient from the operational perspective, does it mean it’s ethical?
What is also special is the amount and content of the data available. We hold vast amounts of tax information, which is mostly confidential. Data protection and privacy must never be put at risk, in any circumstances. If, for instance, we test various technology solutions and find a best match that delivers less than a 100% sure data protection, we have to find other, alternative solutions.
Input from several levels of the organisation
A large number of specialists from various levels of the Tax Administration were invited to participate in the preparation of ethical principles for the use of artificial intelligence. For the preparation of the first draft, we invited representatives from security, analytics, communications, data protection, internal audit, technical architecture and development functions.
The Tax Administration’s senior management has given the first draft the green light, and the next step is to post it in the Tax Administration’s intranet for further discussion. After any revisions required, the principles will be presented to senior management for approval, after which they will be published within the Tax Administration, and later for external interest groups.
This is where the most important work begins: practical implementation and commitment to the principles. The principles must be included as a natural part of everyday AI development and utilisation. If the development work shows that a component is not compatible with the principles, we have to return to the drawing board.
Undoubtedly situations will arise where we are tempted to venture into the grey area to achieve the best results. This is when the ethical principles offer the necessary framework for making the right and ethical decisions.