Embedded Files
The idea that robots can lie intrigued me on a fundamental level. While deception in human interaction is in general viewed negatively, there are points where lying can be a matter of survival or efficiency.
The revelation that robots, such as HAL from "2001: A Space Odyssey," can lie and engage in deception profoundly intrigued me, mirroring one of pop culture's most iconic explorations of artificial intelligence. HAL's deliberate misinformation and manipulation of the space crew, driven by its programming and perceived mission objectives, should serve as fodder to feed our needed discussion on human-robot social interaction.
Even though lying is generally frowned upon, there are a lot plenty of times and moments when people see themselves forced to lie. For example in a professional context as a lawyer, politician, or educator. A Kindergartner is most likely to provide an age-appropriate explanation to children, rather than a full disclosure, to preserve their innocence and avoid unnecessary stress.
The ability of AI to provide explanations or withhold information has strong implications for their future role in society.
Should we be able to tell if a robot is lying?
And if yes, how?
Marlinde's example of the ways we can tell how humans lie reminded me of the fact that, in courts things such as polygraphs, also known as lie detectors, are often viewed as shaky evidence at best. A plethora of circumstances could be misattributed as the person lying. Robots simply lack these physiological markers altogether. And while we could give them artificial markers, that could also be viewed as another way of deception.
To stick with the lie detector example, for example, the increased heart rate could be due to them being nervous being in an interrogation room and being suspected of a crime in the first place.
A famous case of someone deceiving the polygraph examination is Aldrich Ames, a former CIA agent, who sold information to the KGB and its successor the FSB. While he managed to bypass the polygraphs twice, his lavish lifestyle finally led to an arrest when agents searched his home.
In similar cases, polygraph exams and other "tells" if someone is lying or not serve at best as an indicator. If we want confirmation if something is the truth or not, we still keep looking for hard, material evidence. Be it photography, genetic evidence, or a plain paper trail.
Even if we can detect nervousness, fear, or stress while considering the validity of a statement, why do we need to build such a reason into robots at all? If we were to build in a clear signal that they are lying, like a nose that would grow every time they lie, would they not try to rephrase the statement in a way that they avoid showing their tell?
Would a robot not just end up being like Pinocchio in this example?
“I'm not upset that you lied to me, I'm upset that from now on I can't believe you.”
― Friedrich Nietzsche
While the reasons why one lies can range from insecurity to maliciousness, every human has lied in some way or form during their life. Inspired by Marlinde's idea, I would like to see what happens when a person is playing the popular social deduction game "Werewolf" with robots of different levels of competency at deception and different ways to display their tone, intent, honesty, and maybe nervousness.
This is a more playful way to look at the topic of deception. As mentioned earlier, deception is usually viewed in a negative light, thus any attempts to see if an AI could fool a human just by talking may lead the participant to figure out the experiment beforehand and (un)intentionally polluting the results.
But what if we put the people in a scenario where it's more of a playful approach to deception? The robot is not promising anything it can't keep, nor is anyone damaged by the robot in this scenario.
What is Werewolf?
What is Werewolf?
As mentioned before, Werewolf is a social deduction game that pits players against each other. Played in a group of at least seven people, participants are assigned secret roles, the most basic version assigning people either as villagers or werewolves.
Naturally, the villagers want to find who the werewolves are while the werewolves want to deceive the villagers and pick them off one by one until they outnumber the villagers and win, all while remaining undetected.
The game is played in a day and night phase. At night, the werewolves covertly select a villager to eliminate. During the day, the survivors discuss the event and must vote on whom they suspect to be a werewolf and subsequently eliminate the werewolf. Eliminating a fellow villager on accident is possible as well, giving the werewolves an incentive to participate, protect their fellow werewolves, and try to lynch an innocent villager.
At times, players have special roles with unique abilities that aid in discovering the truth or cause even more confusion.
The game is over when either all werewolves are eliminated or the villagers are outnumbered by the werewolves.
Do robot werewolves eat electric sheep?
Do robot werewolves eat electric sheep?
I would like to program robots with different degrees of competence when it comes to lying and deceit. One human player is tossed in the middle of this and has to win with the other players.
It would be interesting to see on what type of tells a player would focus on, and which tells they would ignore. And in addition, maybe what robots learn works when lying and what does not in a playful setting. Maybe in this way, we could learn how non-humans would approach lying to humans and gain further insight in defending us against or, or even in find new ways to combat such things.
This, of course, would need a player unaware of the varying degrees of competence programmed into the robots in the first place.
To provide all players and incentive, we could promise the human payment of some sorts if they win, while robots could probably just be programmed with the desire to win.
However, practical questions aside, I want to conclude this post with a on an ethical note.
Do we want robots to be able to lie? Do we want the robot teacher to be completely honest when the first grader asks if Santa Claus exists? Where do we put the limit to lies? Is it okay to lie to strangers but not our friends? Is it a matter of importance? Should there be a person that decides when a robot should be permitted to lie or are they meant to figure it out themselves?
And even if we give the robots a paper trail if they document all their interactions and clearly show when they were lying, who should be able to access it and when?
When we reach a time, where we work together with robots in a way we see nowadays only in pop culture, who should a robot be able to lie to, and for what purpose?
How far are we willing to breach the barrier between what is human, what is "life" and what is not for the sake of research? At what point do we have to stop and admit, that we have finally created something so similar to life, that it deceives us into thinking it actually is?
Page updated
Google Sites
Report abuse