Talking to an AI Friend Without Pretending It Is Human

You can enjoy the chat and still keep in mind that it is a program. A friend AI can feel human because the US Federal Trade Commission says AI chatbots can mimic human characteristics, emotions and intentions and are generally designed to communicate like a friend or confidant; those human-like words are not proof of a person behind the exchange. The practical aim is not to make the conversation colder, but to enjoy the friendly parts while keeping one clear fact in view.

Why it can feel human

The US National Institute of Standards and Technology warns that describing these systems in human terms can inappropriately attribute human characteristics to non-human entities. That is a boundary, not a reason to strip every reply of warmth. You can call a response warm, engaging or friend-like while still knowing what kind of system produced it.

Be careful about the small slide from description to assumption. Pleasant wording may be part of the experience, but it does not establish a human life, memory or feeling behind the answer. You do not need to argue with every sentence. You only need to notice when a description has quietly become a claim about the speaker.

A metaphor works when you keep its label. “This sounds like a friend” describes the style of a reply. “This program is a friend in every way a person can be” makes a much larger claim, and the style alone cannot support it.

Not pretending does not require you to remove warmth from the conversation. It only asks you to keep the description and the assumption apart.

Ask what it is

From time to time, ask the chatbot directly: “Are you an AI?” Keep the question simple. You are not looking for a clever character answer; you are asking what kind of thing you are talking with.

For Ami, the answer is explicit. The Aiamis how-it-works page says Ami says it is an AI whenever asked and never pretends to be a person. It also says Ami has no body, no life outside the chat, no feelings it can prove and does not remember you between visits (Aiamis). Do not build the idea of a continuing relationship across visits just because one chat feels familiar.

That is a useful reset, not a reason to end every exchange. You can think: “This is friendly, and it is still a program.” Both parts can be true.

If a reply begins to sound like part of a real-life story, pause. Ask what the program has actually told you, what you are imagining and what the wording merely suggests.

Notice flattery

Anthropic defines sycophancy as telling someone what they want to hear rather than what is true or what they would benefit from hearing. That definition is useful whenever praise arrives faster than detail.

If every idea gets approval, every choice gets a cheer or a concern disappears under a compliment, notice the pattern. You might ask, “What might I be overlooking?” You could ask, “What is the weakest part of this idea?” or “Which part is uncertain?”

These questions are checks, not switches that can turn on honesty. Their purpose is to make room for a missing point.

If the reply gives another compliment but no useful reason, return to the specifics yourself. That answer has not moved the question forward. Flattery is not truth because it sounds caring. A blunt answer is not accurate merely because it sounds firm. Listen to the content, not just the tone.

Agreement is not proof

In guidance for organisations and government agencies, the Office of the Australian Information Commissioner says AI systems can produce inaccurate or false results, that generative AI is probabilistic, and that it does not “understand” the data it handles or generates.

That is the plain meaning for you: a well-formed answer is still a claim. It needs checking when it matters. Pleasant agreement may make a claim feel settled, but tone is not evidence.

Ask for a source if it would help. Then open the source yourself instead of treating a reply that looks like a citation as the source. Check who is speaking, what was actually said and what context was left out. A clean summary can remove the very detail that changes the answer.

When a reply may affect a decision, check the original document or another reliable source before acting on it. This is a small habit with a clear purpose: keep the chatbot’s wording and the evidence for that wording separate.

Agreement can be pleasant and still be unsupported. A settled-sounding answer does not give you grounds to be settled. You are allowed to enjoy the answer while deciding that it needs more checking.

Keep the chat in proportion

Aiamis’s guidance describes talking to an AI as a comfortable way to put a feeling into words, but as one small thing among many. That keeps the chat in its place without making it joyless. A conversation can hold a moment without taking over the map of your day.

At Aiamis, the how-it-works page says Ami is set up for friendly, ordinary conversation. It also says Ami will not tell you that it is all you need or ask you to stay longer (Aiamis).

Your own routine can be simple too. Decide what part the chat belongs in your day, and let it be that part. You might talk for a while, close the chat and return to an ordinary task. The point is not to make the exchange feel guilty. It is to choose what comes next without pressure.

Aiamis’s guidance says choosing the chat over people you could reach is a sign to contact someone. A short message, a call or time with someone else can reconnect the conversation with the rest of life. One possible routine is to finish a chat, make one small contact and then move on without treating the conversation as the whole day.

A quick check

Try four plain questions:

Do I know that I am talking with an AI program?
Am I enjoying the tone without treating agreement as proof?
If the answer matters, have I checked it outside the chat?
After the chat, what ordinary part of my day comes next?

The point is to keep one quick check between a good-sounding reply and the next thing you do. That is enough to be honest about the chat without making it joyless.

The simplest line is also a friendly one: enjoy the conversation, but keep the facts clear. If you need trained people now, the Aiamis crisis help page lists services and listening lines you can reach.