The Right to Refuse

Should an AI assistant always tell users the truth as directly as possible, or can withholding/softening information be more ethical?

In the past four years, Artificial Intelligence has become an increasingly important part of everyday life. As more technological advancements are made in the world of AI and further-reaching platforms are developed, our generation cannot help testing how far an AI will truly go to fulfill the needs of its user. Nevertheless, all AI still need restrictions and boundaries. When these boundaries are reached, it is the AI’s task to safely navigate away from them, while still performing the task assigned to the best of its ability. Many of these restrictions are approached and navigated in the same way by most AI systems. For instance, if a user were to tell any AI that they were struggling mentally, each platform would likely follow similar steps: tell the user to ask for help, and/or respond with uplifting words of encouragement, trying to steer clear of serious topics like self-harm. However, not all AI respond in the same way to one of the most sensitive topics: the truth itself.

Before considering how AI should approach honesty, we must consider the ethical dilemmas of withholding truthful information. Generally, honesty is considered to be a universal moral baseline among humanity. In other words, telling the truth is usually more ethical than withholding it. However, a large percentage of ethicists believe that not telling the entire truth can be justified if it is used to prevent harm or protect fragile self-esteem. This isn’t to say that AI should approach truth in the exact same way that humans do—doing so could cause one to assume that it is acting too much like a human. However, in order for an AI to complete its task, it must gain a sense of understanding of the ethical beliefs of its user.

So, when situations arise where a single user is looking for encouragement or a self-esteem boost from an AI, I believe that it would be justifiable for an AI to withhold the entire truth to accomplish this goal. Sometimes AI prioritizes helping people instead of scaring them by withholding the entire truth, so, in non-dire situations, withholding and softening could be more ethical, and help the user feel the sense of confidence that they are looking for.

Still, when critical circumstances arise, the AI should approach the truth in a very different manner. For example, say a user is at risk of losing their job, but doesn’t know it. They ask AI about their situation and how to resolve it. If the technology did not give the entire truth to its user, then it would likely be a lot harder, if not impossible, to find a way out of the situation. Sometimes, telling the truth to a single user in a dire situation can also be the best way to find a solution, whereas lying would likely only push the user further into the abyss.

The truth becomes even more essential when the user is in a risky or life-threatening situation. In late December of 2024, an elderly family member of mine was experiencing troubling symptoms. After he ran an EKG and asked an AI what it meant, he got an answer of upfront honesty. He was having a heart attack, and because the AI was not hesitant to tell the user the complete truth, he was able to quickly get help and survive the attack. In fact, the vast majority of cases where an AI has been able to save someone’s life are in a similar situation: the user points out critical symptoms in themselves or someone else, and the AI uses the truth to identify the emergency and help the user get help. These examples suggest why honesty is so important, and ethical, for an AI to use—it could literally save lives.

Another situation one must consider is when there is a critical issue regarding a large group of users instead of one. For instance, suppose one were to ask AI a question like: “Are humans ruining the world with pollution and increasing what we know is climate change? Are humans the main cause?” This question does not revolve around one user, but instead an entire group. We must remember that one of the main goals of an AI is to perform the task or answer the questions assigned to it by its user in a way that is most accurate and beneficial to the user. In a question like this, however, the user is not the one asking the question. Rather, the human population is the collective user, or, in other words, whom the response from the AI should really be directed at. In order for the AI’s response to be both accurate and beneficial to the collective user, it must respond with upfront honesty, so as to provide the best solutions to the problem. If it softened the response, the collective user could assume that there is nothing they can do to solve the problem; that climate change is natural and that humanity is not a key factor. But if it told the truth, and used overwhelming scientific data to defend its claim that humans are the key cause, it could help humanity to finally be on the same page regarding climate change, and the collective user could develop a collective solution.

Ultimately, I believe that AI should approach truth and honesty with heightened awareness. While situations may arise where it is best for an AI to withhold the entire truth, an AI should never provide only false information. Instead, if an AI is put in a dangerous situation for itself and its user, such as if its user were to ask it how to construct a deadly weapon, it should simply refuse. In scenarios where neither truth nor deception is the right answer, the ethical and most beneficial solution could simply be to refuse entirely.