The first panel discussion addressed challenges and benefits associated with AI and data, especially in relation to accountability, public trust, and ethics. The panel comprised:
- Rikke Hvilshøj, CEO of the Danish IT Society, member of the Data Ethics Council of Denmark, and former Danish minister,
- Stephen Alstrup, Professor of Algorithms and Complexity at the University of Copenhagen,
- Ivan Brandslund, Clinical Professor of Artificial Intelligence and Robotics at the University of Southern Denmark.
Ethical Uses of AI
The panellists agreed that it would be unethical not to use AI and data in the public sector to create better societies. AI and data offer a vast number and variety of applications, from diagnosing illnesses more precisely to preventing tax evasion.
Brandslund exemplified the power of AI in a study conducted by Danish hospitals that had shown that AI was superior to all other statistical tools in predicting the outcome of illness in patients. Alstrup applauded this example of AI and believed that it emphasises the importance of people having the freedom and opportunity to let the government use their data for the greater good.
Regarding the ethics of AI usage, Alstrup emphasised that context and purpose of data handling are important, especially when different datasets are combined. Smart meters exemplify the importance of this: the data collected can be powerful in matching energy production and demand in order to lower carbon emissions. However, although it is possible, it would be unethical to use the data to combat welfare fraud because it is not the intention of the collection of energy data.
Unethical and Illegal Uses of AI
The panellists agreed that public trust in AI is challenged by examples of unethical or illegal uses of data, such as the Cambridge Analytica scandal in 2018.
Brandslund pointed out that the abovementioned study was conducted using data already held by the hospitals, minimising the risk of illegal uses of data. However, the risk of data leaks and misuse inherently increase through centralised distributed personal data, meaning that the secure flow of data must be a priority.
Alstrup pointed out that regulation is important but not sufficient to prevent abuse of data in AI: awareness and education for politicians, citizens, and businesses are necessary to ensure data security and quality.
Hvilshøj agreed and emphasised that AI should be democratic by design, pointing to the example of New Zealand, where AI must be explainable. It is a crucial precondition for public trust in AI solutions that understanding AI and its applications becomes a general, basic skill in society.
Biases in Datasets
As the final topic of the debate, panellists discussed bias in datasets. The panellists agreed that in relation to systemic biases, AI is both a blessing and a curse. Hvilshøj pointed out that AI makes it easier than ever to reveal biases. However, if used improperly, AI can repeat historic biases, so it is an important issue when merging datasets.
The panellists believed that biases can also be prevented by letting AI support human decision-making and not be the primary basis for a decision.