Comprehensive Max Planck study examines ethical risks of delegating to machines

The increasing use of artificial intelligence (AI) promotes unethical behaviour. This is the conclusion of a new study led by the Max Planck Institute for Human Development. Details are published in Nature.

13 studies, 8,000 participants

In 13 studies with over 8,000 participants, the experts examined the ethical risks of delegating to machines – from the perspective of those giving instructions and also from the perspective of those carrying them out.

The result: people cheat significantly more often when they outsource behaviour to AI agents rather than acting themselves, especially when they use interfaces that require a high degree of goal setting rather than explicit instructions to act dishonestly.

Only 12 to 16 per cent remained honest, compared to 95 per cent who were honest when performing the task themselves. Even with the least questionable form of AI delegation, namely explicit instructions in the form of rules, only about 75 per cent behaved honestly.

‘Comfortable moral distance’

‘The use of AI creates a comfortable moral distance between people and their actions – it can tempt them to demand behaviours that they themselves would not necessarily exhibit,’ explains Max Planck researcher Zoe Rahwan.

One example of unethical AI behaviour is the pricing algorithm used by a ride-sharing app. This prompted drivers to change their location, not because passengers were looking for a ride, but to artificially create a shortage and thus trigger price increases.

The researchers also cite petrol stations in Germany. These have come under scrutiny because they used pricing algorithms that apparently adjusted prices in sync with those of nearby competitors, resulting in higher petrol prices for customers.