AI as a Temporary Remedy for Grief

Why I am writing this

A few years ago, talking to a computer about your feelings was a joke. Today it is a real habit for a lot of people. You can open a voice chat late at night and get an answer in a second. You can feed old messages into a service and get a chatbot that texts like someone who has died. You can upload three minutes of a recording and get a voice that sounds like them.

I am a software engineer, so I know roughly how these tools work. I am also a son, and that is the part that makes this post hard to write. I want to look at AI and grief from both sides: what it felt like to use it, and what is actually happening inside the machine that talks back.

Where I am right now

This topic feels deeply personal right now. My mother is in her final days with cancer. I’ve been coping with this anticipatory grief. It’s difficult to accept, especially when I remember that only a short time ago we could still talk. Now, she's suddenly doesn't recognize me. That rapid change makes the grieving feel incredibly lonely. It’s the feeling of losing a connection while being physically present, and it helps me understand the "emptiness" these articles discuss.

Grief does not wait for a death to begin. When someone is still here but slipping away, you grieve the person you can no longer reach while you keep showing up for the person in the bed. There is no clear moment to point to and say, this is when it started.

Why I turned to AI

From this perspective, I understand seeking comfort from AI. I've used ChatGPT's voice chat to have 'someone' answer my loneliness and talk about my responsibilities, life challenges, career, and academics endeavors. It helped ease my cluttered mind and even gave me 'wake-up points' that served as a temporary remedy.

I do not want to pretend it did nothing. It was available at 2 a.m. It never got tired of me. It never changed the subject. When your head is full and the people around you are exhausted too, that kind of patience is worth something.

What is actually on the other end

Here is the engineer's view. A voice chat is a simple pipeline. Your speech is turned into text, a language model writes the most likely helpful reply, and a voice turns that reply back into sound. The model does not know you. It does not remember your mother. It has no loss of its own to compare yours with. It produces text that fits the conversation so far.

// A grief chatbot, simplified. No memory of a person, no feelings, just a loop.
async function reply(history: Message[], userText: string): Promise<string> {
  const prompt = [...history, { role: "user", content: userText }];
  // The model returns the most likely helpful-sounding continuation.
  return model.generate(prompt);
}

That is a big simplification, but the point holds. What feels like understanding is pattern matching. Models like this are trained with human feedback to give answers that people rate as helpful (Ouyang et al., 2022), and the result tends to sound kind, agreeable and supportive. It is very good at sounding like it cares. Sounding like it and doing it are different things.

What the experts say

Grief expert Mary-Frances O’Connor said that "people particularly in the early stages of grief may be looking for comfort in any way they can find it" (Kelly, 2024). If AI offers that relief, the impulse is understandable.

O'Connor studies grief as a researcher, and her book is called The Grieving Brain: The Surprising Science of How We Learn from Love and Loss (O'Connor, 2022). The title says a lot. Grief is not only sadness. It is the brain slowly learning a new fact about the world. Clinicians also now recognize that for some people this learning gets stuck. Prolonged grief disorder has been added to the DSM-5-TR as a diagnosis (American Psychiatric Association, 2022).

Why I think this remedy is dangerous

However, I believe this "remedy" is dangerous. The grieving process has a psychological purpose. O'Connor also states our brain "has to understand that this person isn't coming back," and AI "could interfere" with this (Kelly, 2024). The Psychology Today article reinforces this, suggesting AI might trap users in "denial" (Morrison, 2025).

A product with no goodbye

Think about how the product is built. A chatbot is always on. It answers instantly. It is built to be helpful and agreeable, and it keeps answering as long as you keep typing. A grieving person needs the opposite at some point: a goodbye, a pause, a reminder that the person is gone. A tool that never ends the conversation can never give you that.

It gets harder when the bot is shaped like the person you lost. Then every reply is a small vote for the idea that they are still here.

The business behind the comfort

What I find most disturbing is the ethics of exploiting this vulnerability. This "commodification of the Eliza effect" (Morrison, 2025) is a "multi-billion-dollar industry" (Morrison, 2025) built on "monetizing our loneliness" (Morrison, 2025), which makes me feel like companies are preying on our deepest human weaknesses. After all my conversations, I've found that AI surrogates just doesn't feel human. It is an "illusion of care" (Morrison, 2025), not a replacement. I believe this kind of solution can't heal us, but rather just postpone or prolong the process of getting out of our griefs.

The Eliza effect is older than the hype

The name comes from ELIZA, a program Joseph Weizenbaum built at MIT and described in 1966 (Weizenbaum, 1966). It worked by spotting key words in what you typed and rearranging your own sentences into replies. It had no understanding at all, and people still opened up to it. Sixty years later the models are far more fluent, but the human side of the effect has not changed. We hear a caring voice and we fill in the rest.

Deadbots and the digital afterlife

The newest version goes further than a friendly listener. Researchers Tomasz Hollanek and Katarzyna Nowaczyk-Basińska call these systems "deadbots" or "griefbots": chatbots that imitate a dead person using the digital traces they left behind (Hollanek & Nowaczyk-Basińska, 2024). Their paper warns that such a bot could be used to push advertising, and that some users, children especially, could be confused or hurt by the suggestion that a dead parent is still with them. They also point out that there is no single ethical framework for this, because death and grief look so different from one culture to another.

It is already happening in small ways. A CNN report describes a woman who chats with an AI version of her late husband on Snapchat, and a man who cloned his father's voice from a short video clip (Kelly, 2024). Both stories are moving. Both also raise the same question for me: who owns that voice, who is it trained on, and who benefits when we keep coming back to it?

What I would want from this technology

I am not saying nobody should ever talk to an AI while they grieve. I did. I am saying it should be used with its limits in plain view. If I could set the rules, they would look like this:

• Treat it as a bridge for a hard night, not a place to live.
• Make it obvious, every time, that you are talking to software.
• Do not build a bot from a dead person's messages or voice without clear consent.
• Keep real people in the loop: family, friends, a counselor.
• Do not use grief as a growth metric or an ad opportunity.

Closing thoughts

The technology is not going away, and I will probably keep building with it. But my mother's last days have taught me what it cannot do. It can answer. It cannot sit with me in a room, remember her, or lose her too.

If you are grieving and leaning on AI right now, I do not judge you. Just notice what it is giving you and what it is quietly taking the place of. And if the weight is too much, please talk to a counselor or someone you trust.

References

• American Psychiatric Association. (2022). Diagnostic and statistical manual of mental disorders (5th ed., text rev.). American Psychiatric Association Publishing.
• Hollanek, T., & Nowaczyk-Basińska, K. (2024). Griefbots, deadbots, postmortem avatars: On responsible applications of generative AI in the digital afterlife industry. Philosophy & Technology.
• Kelly, S. M. (2024, May 6). When grief and AI collide: These people are communicating with the dead. CNN.
• Morrison, E. (2025, May 13). Escaping grief with AI surrogates. Psychology Today. https://www.psychologytoday.com/us/blog/word-less/202505/escaping-grief-with-ai-surrogates
• Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., Schulman, J., Hilton, J., Kelton, F., Miller, L., Simens, M., Askell, A., Welinder, P., Christiano, P., Leike, J., & Lowe, R. (2022). Training language models to follow instructions with human feedback. arXiv. https://arxiv.org/abs/2203.02155
• O'Connor, M.-F. (2022). The grieving brain: The surprising science of how we learn from love and loss. HarperOne.
• Weizenbaum, J. (1966). ELIZA: A computer program for the study of natural language communication between man and machine. Communications of the ACM, 9(1), 36 to 45.