My unhealthy relationship with AI

The unhealthy relationship I want to examine is a habit: allowing a conversation with AI to take the place of a decision, a difficult discussion or the effort needed to understand something.

AI can be useful precisely because it is available when another person is busy. You can use a conversation to organise a thought, practise an explanation or work through a problem. The question I want to keep beside that usefulness is whether the conversation helps you return to the task or becomes a reason to postpone it.

When another answer stops helping

In a hypothetical example, you ask for advice about a difficult email. You get a reasonable draft, request a warmer version, then a firmer version, then an assessment of what the recipient might think. Eventually you have spent longer discussing the email than deciding what you need to say.

The system cannot know the recipient's private thoughts. Asking it to generate more interpretations does not turn those interpretations into evidence. A useful stopping point might be a draft you understand, a clear reason for sending it and one check that it says what you intend.

I would judge the exchange by what it allows you to do afterwards. Look at whether you resolved the issue, can explain the reasoning without reopening the chat and spoke to the person whose answer you needed.

What the research can tell us

The workplace example above is a question about task avoidance; the following study examined social and emotional outcomes instead. Fang and colleagues studied 981 participants over four weeks, assigning different chatbot interaction modes and conversation types. Every group used a chatbot, so this was not a comparison of chatbot use with no chatbot use. They measured loneliness, social interaction, emotional dependence and problematic use through surveys. Their revised paper reports: No significant effects were detected from experimental conditions.

Participants who voluntarily spent more time using the chatbot showed worse psychosocial outcomes. That association does not establish that longer use caused the outcomes: people with different needs may choose different amounts of use. The experiment also cannot settle what happens over years or across every type of AI product.

Keep some thinking outside the chat

For a decision you need to own, write your first view before asking the system. Include what you know, what you are unsure about and what would change your mind. You then have something to compare with its response. Otherwise, you may end up polishing a position you never independently considered.

Ask for a counterargument when it would help, but do not mistake generated disagreement for an independent witness. Check the information that matters outside the conversation. If the issue concerns another person, their account may matter more than several pages of plausible interpretation.

There should also be an end to the task. When the draft or explanation meets your agreed purpose, act or deliberately pause. A new version should solve a remaining problem, rather than supply another moment of reassurance.

Notice what the tool is displacing

Watch the consequences for your time, sleep, relationships and ability to work without the assistant. Those observations are more useful than blaming yourself for enjoying a convenient tool. If use is causing distress or displacing essential parts of your life, speak with someone you trust or a qualified professional rather than asking the chatbot to be your only judge of the problem.