If a smart home detects the unmistakable whiff of cannabis smoke in a teenager’s bedroom, should it tell their parents? Or even the police?
Situations like this are on the horizon with digital assistants, like Amazon Echo or Google Home, making it into many people’s dwellings. One proposal for reaching resolutions is that a handful of artificially intelligent bots should debate the possibilities before reaching a decision.
The moral AIs each represent one of the stakeholders, such as the device owner, the police or a guardian. They have individual priorities according to who they represent: to be lawful, to operate safely or to prioritise individual autonomy. A digital assistant faced with a drug-taking teenager would weigh up the demands of the different stakeholders and try to find a course of action that pleases them all.
Advertisement
The system maps out the various arguments from each stakeholder, noting which ones clash (“involve the police” versus “respect individual autonomy”, for example). Conflicting demands are removed, and the system decides on a course of action based on the remaining instructions.
However, the team behind the idea has yet to work out what the moral AIs should do if they can’t reach a consensus.
Moral debate
Ethical behaviour is not consistent across societies or individuals, so the AIs would be flexible, allowing them to be geared to better reflect local law and the preferences of the owner, says Marija Slavkovik at the University of Bergen, Norway. She and her colleagues presented the idea at the ACM conference on in Hawaii.
Most digital assistants lack smell sensors, but they do have microphones and some have cameras. The information they collect has already been sought by police: Amazon released audio recordings from an Echo device present at a murder scene to Arkansas police in 2017.
Google’s Nest home security system was also found this week to contain a microphone, even though this wasn’t disclosed in any of the material for the product.
“Humans and human situations are far messier than this method makes out,” says Beth Singler at the University of Cambridge. For example, the proposed moral AIs treat parents as a single unit, but parents may disagree on what to do about a teen’s use of marijuana.
Some might want it dealt with within the family, while others may take a hard line and seek police involvement. This disparity is likely to be found in all the groups of people the artificial moral agent (AMA) seeks to replicate, says Singler.
What’s more, people change their opinion, especially when faced with a tough choice. “There is no guarantee humans will not behave differently, and may disagree with what they expressed previously to the AMA,” says Singler.
Still, the question remains about whether we should be placing this responsibility on robots in the first place. “One critical consideration is of to make a decision,” says Jason Millar at the University of Ottawa, Canada.
“That question is very important, much more important, I would argue, than seeking agreement or consensus,” he says. Future smart devices may know right from wrong, yet have to watch us indulge in bad behaviour anyway.
