You could lie to a health chatbot – but it might change how you perceive yourself

Imagine that you are on the waiting list for a non-urgent operation. You were seen in the clinic some months ago, but still don’t have a date for the procedure. It is extremely frustrating, but it seems that you will just have to wait.

However, the hospital surgical team has just got in contact via a chatbot. The chatbot asks some screening questions about whether your symptoms have worsened since you were last seen, and whether they are stopping you from sleeping, working, or doing your everyday activities.

Your symptoms are much the same, but part of you wonders if you should answer yes. After all, perhaps that will get you bumped up the list, or at least able to speak to someone. And anyway, it’s not as if this is a real person.

The above situation is based on chatbots already being used in the NHS to identify patients who no longer need to be on a waiting list, or who need to be prioritised.

There is huge interest in using large language models (like ChatGPT) to manage communications efficiently in healthcare (for example, symptom advice, triage and appointment management). But when we interact with these virtual agents, do the normal ethical standards apply? Is it wrong – or at least is it as wrong – if we fib to a conversational AI?

There is psychological evidence that people are much more likely to be dishonest if they are knowingly interacting with a virtual agent.

In one experiment, people were asked to toss a coin and report the number of heads. (They could get higher compensation if they had achieved a larger number.) The rate of cheating was three times higher if they were reporting to a machine than to a human. This suggests that some people would be more inclined to lie to a waiting-list chatbot.

The rate of cheating was three times higher when reporting a coin-toss result to a machine.
Yeti studio/Shutterstock

One potential reason people are more honest with humans is because of their sensitivity to how they are perceived by others. The chatbot is not going to look down on you, judge you or speak harshly of you.

But we might ask a deeper question about why lying is wrong, and whether a virtual conversational partner changes that.

The ethics of lying

There are different ways that we can think about the ethics of lying.

Lying can be bad because it causes harm to other people. Lies can be deeply hurtful to another person. They can cause someone to act on false information, or to be falsely reassured.

Sometimes, lies can harm because they undermine someone else’s trust in people more generally. But those reasons will often not apply to the chatbot.

Lies can wrong another person, even if they do not cause harm. If we willingly deceive another person, we potentially fail to respect their rational agency, or use them as a means to an end. But it is not clear that we can deceive or wrong a chatbot, since they don’t have a mind or ability to reason.

Lying can be bad for us because it undermines our credibility. Communication with other people is important. But when we knowingly make false utterances, we diminish the value, in other people’s eyes, of our testimony.

For the person who repeatedly expresses falsehoods, everything that they say then falls into question. This is part of the reason we care about lying and our social image. But unless our interactions with the chatbot are recorded and communicated (for example, to humans), our chatbot lies aren’t going to have that effect.

Lying is also bad for us because it can lead to others being untruthful to us in turn. (Why should people be honest with us if we won’t be honest with them?)

But again, that is unlikely to be a consequence of lying to a chatbot. On the contrary, this type of effect could be partly an incentive to lie to a chatbot, since people may be conscious of the reported tendency of ChatGPT and similar agents to confabulate.

Fairness

Of course, lying can be wrong for reasons of fairness. This is potentially the most significant reason that it is wrong to lie to a chatbot. If you were moved up the waiting list because of a lie, someone else would thereby be unfairly displaced.

Lies potentially become a form of fraud if you gain an unfair or unlawful gain or deprive someone else of a legal right. Insurance companies are particularly keen to emphasise this when they use chatbots in new insurance applications.

Any time that you have a real-world benefit from a lie in a chatbot interaction, your claim to that benefit is potentially suspect. The anonymity of online interactions might lead to a feeling that no one will ever find out.

But many chatbot interactions, such as insurance applications, are recorded. It may be just as likely, or even more likely, that fraud will be detected.

Virtue

I have focused on the bad consequences of lying and the ethical rules or laws that might be broken when we lie. But there is one more ethical reason that lying is wrong. This relates to our character and the type of person we are. This is often captured in the ethical importance of virtue.

Unless there are exceptional circumstances, we might think that we should be honest in our communication, even if we know that this won’t harm anyone or break any rules. An honest character would be good for reasons already mentioned, but it is also potentially good in itself. A virtue of honesty is also self-reinforcing: if we cultivate the virtue, it helps to reduce the temptation to lie.

This leads to an open question about how these new types of interactions will change our character more generally.

The virtues that apply to interacting with chatbots or virtual agents may be different than when we interact with real people. It may not always be wrong to lie to a chatbot. This may in turn lead to us adopting different standards for virtual communication. But if it does, one worry is whether it might affect our tendency to be honest in the rest of our life.

Dominic Wilkinson receives funding from the Wellcome Trust and the Arts and Humanities Research Council.