The EU’s executive—the European Commission—has published its ‘Ethics Guidelines for Trustworthy AI.’ With the help of 52 experts, it has worked out three components, broken down into four ethical principles and seven requirements. It has more than just the number three in common with Isaac Asimov’s laws of robotics.
AI must be lawful, ethical, and robust in order to be trustworthy, says the Commission. The guidelines are not going to be laws, so they steer clear of the first component beyond acknowledging its importance. The idea is to provide a framework for achieving components two and three.
Get the BREAKERMAG newsletter, a weekly roundup of blockchain business and culture.
The biggest overlap with Asimov is in the four ethical principles that the EU wants developers to respect. One of them is “Prevention of harm”—pretty much Asimov’s first law, that robots should never harm a human or let one come to harm by inaction.
Another is “Respect for human autonomy”, which is explained as “robots should not … coerce, deceive, manipulate” but should “leave meaningful opportunity for human choice”. It’s not far from Asimov’s second law: a robot must obey human orders, unless they would harm a human.
“Robots should not … coerce, deceive, manipulate” but should “leave meaningful opportunity for human choice”.
The principles also say robots must be fair and explicable: they have to respect proportionality of means and ends, and their own means and ends must always be explained as far as possible.
Asimov’s third law, that robots must protect themselves as long as they do not break the other laws, is notable for its absence.
Seven so-called “requirements” follow, which are reflected to some extent in law already. What is not already legally binding is a “requirement” in name only. They include things like non-discrimination and environmental sustainability.
A number of assessment points are provided for each requirement. Early next year, a selection of companies will sign up to the assessments and provide feedback. The Commission says it will then consider “next steps.”