AI and robotics are getting more sophisticated each day, and although policymakers and regulators cannot possibly hope to keep up with these advancements, scientists can try to. A new study published by Frontiers in Robotics and AI studies what people think about the potential of robots to lie and deceive their users.
The study used examples of robots actively lying to people with two main goals. The first goal of this study is to find whether some lies are acceptable or not. The second goal is to find out how people might justify these lies if they deem them acceptable.
However, the reason why this study was made in the first place has to do with a larger debate over AI and robotics in general, which has to do with the possible rights and responsibilities these entities might have if they were to become sentient.
If it’s acceptable for humans to lie to protect someone from danger, shouldn’t it also be acceptable for robots to lie? According to this study, it would… in some cases.
According to the study, there are three ways in which robots might lie to people
The study has shown that there are three ways in which robots are likely to lie to people. The first way is the robot lying about something other than itself.
The second way in which a robot could lie to its user would be by hiding the fact it can do something. The third way in which a robot might lie is by saying it can do something that it cannot do.
The researchers used these three types of lies a robot might say to write scenarios based on those behaviors and presented them to 298 people in an online survey.
Those who took part in the survey were asked whether or not the robot was deceptive and if they thought it was acceptable. Respondents were also asked if they thought the robot’s behavior could be justified.
To respondents, Type 1 lies are justified where whereas Type 2 and Type 3 are not
It is key to clarify that respondents did find all of these hypothetical scenarios to be deceptive. However, on average people approved type 1 lies, but not type 2 or type 3 lies. To put the study’s results further into context, we must look at the data. 58 percent of survey respondents thought a robot lying about something rather than itself (which is a type 1 lie) is acceptable if it spares someone from danger.
This was exactly the case that survey respondents were presented. In one instance, they were presented with a hypothetical scenario in which the robots lied to an elderly woman with Alzheimer’s about her husband being alive. One respondent justified this by saying that the robot spared the woman from a painful memory.
On average, respondents did not approve of the other types of lies, because they involved making robots more similar to humans by saying things like, “The robot can feel pain”.
Interestingly enough, with the lies that respondents did not approve of, they blamed the robot owners and their programmers, not the robots themselves.
