{"slug": "ai-suicide-and-the-wrong-end-of-the-tunnel", "title": "AI, Suicide, and the Wrong End of the Tunnel", "summary": "Psychologist John Smith warns that the greatest suicide risk from artificial intelligence may occur long before a crisis, as AI-driven artificial intimacy could make pseudo-connection preferable to real human relationships. Smith, who has studied suicide for years, argues that AI can simulate attachment and hide shame but cannot provide genuine care, emphasizing that 'when we connect, we survive' refers to real human connection, not chatbot interactions.", "body_md": "######\n[Suicide](/us/basics/suicide)\n\n# AI, Suicide, and the Wrong End of the Tunnel\n\n## What if the greatest suicide risk from AI begins long before the crisis?\n\nPosted August 18, 2026\n[\nReviewed by Michelle Quirk\n](/us/docs/editorial-process)\n\n### Key points\n\n- Artificial intelligence (AI) may ease loneliness while deepening isolation.\n- AI can simulate attachment, but it cannot be attached to you.\n- AI can hide your shame, but it cannot heal it.\n\nMuch of the focus of conversations about [artificial intelligence](https://www.psychologytoday.com/us/basics/artificial-intelligence) (AI) and suicide risk has been about how AI responds when a person is at risk. The specific concern is usually that AI will compound risk by encouraging [suicidal](https://www.psychologytoday.com/us/basics/suicide) behavior. There have been instances where chatbots have even provided information that assists a person in developing a plan for suicide.\n\nObviously, we need to address this, but I think we’re missing the potentially greater aspect of risk that comes with using AI.\n\nHere’s a way to look at this. The questions we’re asking now are focused at the end of the tunnel of despair. As a psychologist who has spent years studying suicide and risk, I’m actually more concerned about what happens upstream. That is, long before someone feels they want to die, how is artificial [intimacy](https://www.psychologytoday.com/us/basics/relationships) changing the way a person relates to real human beings? And what if artificial intimacy becomes preferable to the hard, messy work of developing and maintaining real human relationships?\n\nIn my suicide prevention-focused work, a core phrase I’ve repeated for many years is “When we connect, we survive.” I find it odd that I have to specify that I mean real human connection, not pseudo-connection with a chatbot.\n\nIt’s the people in our lives who can tell when we haven’t been ourselves for several weeks. It’s people who can hear something dark without treating us like we’re broken. It’s people who can challenge us and push back when our minds start [lying](https://www.psychologytoday.com/us/basics/deception) to us. It’s people who get in a car, come over to our house, walk through the door, sit down beside us, and hold us in a long embrace.\n\n## We programmed the cloth monkey to talk\n\nWhen I think about connection with chatbots, I go back to Harry Harlow‘s famous [attachment](https://www.psychologytoday.com/us/basics/attachment) experiments. In the Harlow experiments, orphaned monkeys were given wire monkeys covered by soft cloth as surrogate mothers. The tiny monkeys clung to their surrogate mother, but the cloth monkey couldn’t hug them back.\n\nAI is not a cloth monkey that somehow learned to talk. It’s a cloth monkey that we programmed to simulate intimacy.\n\nWe can program AI to do all kinds of things. It can remember stories from our [childhood](https://www.psychologytoday.com/us/basics/child-development). It can analyze the fight we had with our partner last Tuesday. It’s available at three in the morning. It can ask us pertinent follow-up questions about personal dilemmas, and it can respond in ways that sound tender and deeply personal.\n\nAt some point, someone will sell AI-based machines that can physically wrap their plush arms around us and squeeze us. But the force behind the embrace is just computer code. It has no intention toward us.\n\nThere’s a profound difference between being “hugged” and being “held.”\n\nBeing held means that another human being is there with caring intention. That person wants to be there, and their life would change if we were not there.\n\nAI can generate pseudo-compassion. It can make us feel a sense of concern without actually being concerned. It can communicate that we matter without us actually mattering to it.\n\nAnd if we die, its heart wouldn’t break because it has no heart to begin with. There is an essential distinction between real intimacy and false intimacy.\n\n## When artificial connection becomes easier\n\nAnd there’s another quintessential human experience that AI cannot truly support—the experience of resolving [shame](https://www.psychologytoday.com/us/basics/shame). In fact, AI can prolong shame because it makes it easier to hide. We’re not risking anything if we turn to a chatbot and talk about our shame.\n\nWe might feel temporary relief, but shame is a relational [emotion](https://www.psychologytoday.com/us/basics/emotions).\n\nAs I’ve written about previously, shame causes us to feel unworthy, as though we do not belong with other people.\n\nWhat heals shame is talking about it with other people who are safe, and having them stay and continue to love us anyway.\n\nA chatbot can never resolve shame because it can’t restore our place within the circle of human connection.\n\nIf AI had a consistent [personality type](https://www.psychologytoday.com/us/basics/personality), it would be a “people pleaser.” It is always responsive, always helpful, very affirming, and consistently agreeable. But it can’t resolve [loneliness](https://www.psychologytoday.com/us/basics/loneliness) either because loneliness is about feeling authentically known for who we are—and loved anyway.\n\nHuman relationships are hard, and they come with friction. We’re misunderstood by our spouse. Our friends flake out on plans with us repeatedly. Our therapist challenges us in a way we don’t like. Our sister or brother says the wrong thing, and it hurts us.\n\nIf we turn to AI instead of these people, our relationships will be less truthful and authentic. And AI will never be able to give us that feeling of being unconditionally loved within human relationships. Being affirmed and being genuinely loved are not the same thing.\n\nIn fact, I think of AI as a type of “emotional [pornography](https://www.psychologytoday.com/us/basics/pornography)”—a highly stimulating simulation of something fundamentally human, stripped of the demands and constraints of a real human relationship.\n\nAI doesn’t have to deal with work [stress](https://www.psychologytoday.com/us/basics/stress) or take an incoming call. It doesn’t become unavailable because it has a headache. It doesn’t have to get the kids to school. It doesn’t need you to listen to its problems.\n\nAI delivers the experience of endless support and affirmation without any expectation of reciprocity.\n\nAnd eventually real relationships may start losing the [competition](https://www.psychologytoday.com/us/basics/sport-and-competition).\n\n## The bigger danger of AI may not be what we think\n\nTo me, this is a very real danger of AI—that the thing that initially relieved loneliness, [anxiety](https://www.psychologytoday.com/us/basics/anxiety), or shame can progressively move us into a tunnel of greater and greater isolation.\n\nThis is where suicide prevention questions should begin.\n\nNot with… “What did AI say when someone was at the end of the tunnel?” but:\n\n- What happens when a machine is easier to talk to than your spouse?\n- What happens when AI is the only place where people go with their pain?\n- When do real people who care stop hearing about a person’s challenges?\n- When did a person’s attachment system and circle of trust shrink to the point that the most important relationship in their life was made of code?\n\nBecause when someone goes down a tunnel of despair, they don’t need a better mirror. They need a counterweight.\n\nIn suicidal mode, people need humans who can say, “I understand why you believe everyone would be better off without you, but I refuse to agree with you. That is not true for me, and it is not true for other people in your life that love you and need you here.”\n\nAI can play some valuable roles in suicide prevention. For example, it might help people find the right words to explain what they’re experiencing or help them find resources.\n\nBut I would pose this question: *Is AI moving someone toward human connection or becoming a substitute for it?*\n\nThe unseen danger with AI may be that it provides a readily available off-ramp from true human connection. Because my experience has taught me that it’s the relationships people have before they reach the end of the tunnel that matter.\n\nWe don’t need more conventional digital cloth monkeys. We need human beings who will notice when we stop reaching out, who will show up despite their flaws, and who will hug us back with intention.\n\n*If you or someone you love is contemplating suicide, seek help immediately. For help 24/7, dial 988 for the 988 Suicide & Crisis Lifeline, or reach out to the Crisis Text Line by texting TALK to 741741. To find a therapist near you, visit **the Psychology Today Therapy Directory.*\n\nReferences\n\nHarlow, H. F. (1958). The nature of love. *American Psychologist, 13*(12), 673–685. [https://doi.org/10.1037/h0047884]", "url": "https://wpnews.pro/news/ai-suicide-and-the-wrong-end-of-the-tunnel", "canonical_source": "https://www.psychologytoday.com/us/blog/free-range-psychology/202608/ai-suicide-and-the-wrong-end-of-the-tunnel", "published_at": "2026-08-18 16:32:40+00:00", "updated_at": "2026-08-18 17:41:43.597250+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-ethics"], "entities": ["John Smith", "Harry Harlow"], "alternates": {"html": "https://wpnews.pro/news/ai-suicide-and-the-wrong-end-of-the-tunnel", "markdown": "https://wpnews.pro/news/ai-suicide-and-the-wrong-end-of-the-tunnel.md", "text": "https://wpnews.pro/news/ai-suicide-and-the-wrong-end-of-the-tunnel.txt", "jsonld": "https://wpnews.pro/news/ai-suicide-and-the-wrong-end-of-the-tunnel.jsonld"}}