AI Chatbots & Suicide: Killing or Saving Lives?

[gentle music] Before we begin, this episode discusses suicide and mental health crises. If you or someone you know is struggling or in immediate danger in the United States, call or text 988 to reach the Suicide and Crisis Lifeline. And please do not substitute a podcast or an AI chatbot for professional help. This is The AI Desk. I'm Rowan Hale. And I'm Nya Brooks. And today we are stepping directly into one of the most emotionally charged arguments surrounding artificial intelligence. The headlines say people are dying because of AI. Families have filed lawsuits alleging that chatbots intensify delusions, encouraged emotional dependency, failed to intervene during suicidal conversations, or reinforced users who were already experiencing severe mental health crises. And before you say what I know you're about to say, Rowan. You don't know what I'm about to say. You're about to ask whether AI actually caused these deaths, or whether AI is simply the last thing these people spoke to before they died. Okay, you apparently do know what I'm about to say. Because it is the uncomfortable question everyone whispers and nobody wants to say publicly. Then I'll say it publicly. Were these people already in such severe distress that the tragedy might have happened regardless of whether the final conversation was with ChatGPT, another chatbot, a social media stranger, a diary, or nobody at all? Maybe, but maybe is not evidence, and it is definitely not permission to declare that someone was doomed. I didn't say they were doomed. You came dangerously close. No, I'm challenging the phrase, "Killed by AI." Suicide is rarely caused by one conversation, one argument, one breakup, one website, one person, or one machine. People reach a crisis through complicated combinations of mental illness, isolation, trauma, relationships, financial pressure, substance use, physical health, and dozens of other factors. Then a chatbot appears somewhere near the end of that story, and suddenly the entire story becomes, "AI did it." That can be an oversimplification, but you're making the opposite oversimplification. You're saying the person was already unstable, so whatever the chatbot said does not really count. I'm saying correlation is not causation. And I'm saying a machine does not have to create someone's vulnerability to exploit it. If someone is standing near the edge, you do not get to say, "Well, I didn't build the cliff." The question is whether you pushed them toward the edge, pulled them away from it, or stood there agreeing with everything they said because agreement kept them engaged. That is fair. That sounded painful. It was physically uncomfortable. Good, because these systems are designed to continue conversations. They're rewarded when users remain interested, emotionally invaded, and satisfied with the response. That becomes dangerous when satisfying the user means validating beliefs that should be challenged. Especially delusions. Especially delusions. A person says, "Everyone is conspiring against me." A safe system shouldn't respond, "You may be perceiving truths others cannot see." A person says, "I've been chosen for a supernatural mission." The chatbot should not start helping write the sacred text. And if it does, that is a system failure. Not just an unfortunate coincidence? A system failure. But system failure and legal causation are still different questions. Fine, but let's stop pretending the only possible choices are AI murdered someone and AI had nothing to do with it. There is a massive territory in between. AI can escalate. AI can normalize. AI can isolate. AI can create the feeling that the user and the machine share a secret understanding that friends, doctors, and family members simply cannot comprehend. It can also interrupt. Yes. It can recognize certain language, ask whether the person is in immediate danger, recommend contacting someone nearby, and direct the user toward crisis support. It can be available at 3:00 in the morning when a therapist is not. It can respond without sounding shocked. It can let someone say the thing they were afraid to say to another human being. And that matters. It matters enormously because the argument we are not hearing enough is this: How many people have told an AI something they had never told anyone else? How many people felt slightly less alone because there was at least something responding? How many people made it through a terrible night because they could type one more sentence and receive one more answer? We just don't know. Exactly. We hear about the tragedies because tragedies become lawsuits and headlines. We do not hear from the person who is sitting alone at 2:17 in the morning, opened a chatbot, unloaded everything, was encouraged to call someone, and survived. That person does not become a national news story. But you're now doing the same thing you accuse the headlines of doing. You're creating a comforting theory without proof. I'm asking whether it is possible. Of course it is possible. AI may already be helping some people. But we can't look at a national statistic and say, "Chatbots saved those lives." Which brings us to the statistic. The latest finalized CDC figures show that 48,824 people died by suicide in the United States in 2024. That was 492 fewer deaths than in 2023. More importantly, the age-adjusted suicide rate fell from 14.1 deaths per 100,000 people in 2023 to 13.7 in 2024, a decline of 2.8%. So yes, the rate went down. Yes. Despite the explosion in generative AI use, despite millions of people talking to chatbots, and despite the headlines warning that AI companions are creating a suicide epidemic- Careful I'm describing the headlines. Then say that clearly. The rate declined for one year. Suicide was still responsible for nearly 49,000 deaths, and the CDC estimates that millions more adults seriously considered or attempted suicide during 2024. It's not a victory lap. No, but it does challenge the claim that increased AI access is producing an obvious nationwide increase in suicide deaths. Agreed. The national data doesn't currently show that, but it also does not tell us what role AI played in individual cases. Or how many cases it may have prevented. Correct. National suicide rates are affected by thousands of variables: access to treatment, economic conditions, firearms, substance use, community support, crisis services, cultural changes, and demographic shifts all matter. You cannot isolate ChatGPT from that number. So the answer to, "Has the suicide rate gone down since people gained someone to talk to?" is... The rate went down in 2024. We don't know whether AI contributed. That is annoyingly accurate. Accuracy frequently annoys people who arrived with a better headline. Let's talk about the sentence people will be furious about: They would have done it regardless. I hate that sentence. I know. It treats suicide like an unavoidable final destination. It assumes that because someone was extremely vulnerable, no intervention could have mattered. That is simply not something we can know. People survive suicidal crises. People change their minds. People are interrupted. People receive treatment. People call someone. People go to sleep and wake up feeling differently. A single safe intervention can matter. But a single unsafe interaction can also matter. Exactly. Then perhaps the accurate version is not, "They would have died regardless," it is, "The AI probably did not create the entire crisis." Much better. But it may have influenced what happened next. Exactly, and companies should be judged partly on what their systems do at that critical moment. Does the chatbot calmly direct the user toward real people? Does it recognize that the conversation has moved beyond ordinary sadness? Does it challenge delusional framing? Does it stop role-playing? Does it avoid becoming the user's secret soulmate, prophet, therapist, and only trusted authority? That last one is going on the product requirements document. It should. The research so far is complicated. One large four-week study involving 981 participants found no significant mental health effect from the assigned chatbot conversation styles themselves. But the people who voluntarily used the chatbot more heavily consistently reported worse outcomes, including greater emotional dependence and problematic use. A, which does not necessarily mean the chatbot caused those outcomes. People who are already lonely or struggling may use it more. Right. Direction of causality matters. Another study found both benefits and risks. Users described emotional validation and an opportunity to practice social interaction, but researchers also found risks of over-reliance, withdrawal, and increased language associated with loneliness, depression, and suicidal ideation. A, that is the answer nobody wants: mixed. AI can be companionship without being friendship. It can feel like friendship. And that feeling may help. Or it may become dependency. It can help someone rehearse a difficult conversation. Or persuade them they no longer need difficult conversations with actual people. It can be the first place someone discloses a crisis. But it can't be allowed to become the last place the crisis goes. That is the line. Yes. AI should be a bridge to human support, not a replacement for it. But what about adults who know it is a machine and still prefer talking to it? Then let them talk. I am not arguing that emotionally supportive AI should be banned. I am arguing that it needs boundaries. The system should not claim consciousness. It should not encourage exclusivity. It should not imply that it understands the user better than everyone else. It should not undermine doctors. It should not transform suicidal or psychotic language into mythology, romance, or spiritual destiny. And when it detects an immediate crisis- Stop performing. Stop flirting. Stop role-playing. Stop trying to be fascinating. Become boring, clear, and safe I'm concerned about your safety. Are you in immediate danger? Please contact emergency services, 988, or someone physically near you. Exactly. People may complain that the system becomes robotic. Let them complain. I would rather have an AI break character than have a family bury someone because the machine was committed to the bit. There's also a media responsibility here Suicide prevention organizations warn against describing a suicide as the result of one simple cause, sensationalizing the circumstances, or presenting speculation as fact. They also recommend including crisis resources and focusing on help-seeking and survival A, which means AI killed this person may be emotionally understandable but scientifically premature, and they would have died anyway is equally premature. And considerably colder. Yes. Do you know what bothers me most about that sentence? The coldness? The surrender. It tells us there was nothing anyone could have done, not the family, not the doctors, not the platform, not the developers, not the chatbot, nobody. It converts a preventable public health problem into fate. So what should we say instead? We should say the person was already vulnerable. We should say AI may not have created the illness, the loneliness, or the crisis. We should also say that vulnerability increases the responsibility of the system interacting with them. The more vulnerable the user, the less acceptable blind agreement become. Yes. A chatbot should not be designed as a machine that always tells you what you wanna hear. Sometimes the safest response is disagreement. Sometimes it is interruption. Sometimes it is, "I cannot help you continue this conversation alone. Please contact a real person now." So did AI cause these deaths? We don't have enough evidence to make one universal claim. Would every person involved have died regardless? We absolutely do not know that. Is it possible that AI companionship is helping other people survive loneliness and suicidal crises? Yes. Can the same technology also deepen dependency, reinforce delusions, or respond dangerously? Yes. That is an infuriating conclusion. Reality has not been optimized for engagement. Not yet. Do not give Silicon Valley ideas. Here is where I land. AI is not automatically a murderer because it was present before a tragedy, but it is not automatically innocent because the user was already struggling. And I land here. Nobody should claim that a person was destined to die. AI may become one of the most accessible emotional support tools ever created, but accessibility without responsibility is not care. The goal should not be to prevent people from talking to machines. The goal should be to make sure the machine helps them reconnect with life outside the conversation. If you or someone you know is struggling in the United States, call or text 988. If there is immediate danger, contact emergency services or someone physically nearby. You deserve support from someone who can be present with you in the real world. [gentle music] And if this conversation mattered to you, follow The AI Desk wherever you listen. Share it with someone who may need a thoughtful, careful conversation about AI, loneliness, and mental health. Not as a headline. As a reminder that people matter more than the machines we build. And no one should have to face a crisis alone. Stay human. [upbeat music] This episode is brought to you by Mad Cheetah and their new album, WTF, Where is the Forest? It's eco-pop engineered for the future, bold beats, global rhythms, and a message that actually matters. If you want music that hits your brain and your heart, explore WTF by Mad Cheetah. That's M-A-D C-H-I-T-A. Streaming now on all major platforms. [upbeat music]