How do you know if you're talking to AI?

Part of my work includes providing text based therapy. Increasingly I get asked the question "Are you AI?"

I typically provide a brief response saying how AI responds immediately all the time, whereas the work I do involves me responding on a specific day of the week where I don't typically respond at the same time the person is logged in to the program.

I did previously work providing instant chat responses, but there would be times where people would ask if I was still there because I would need to write my response, which would typically come in small paragraphs so that I can communicate that I understand and am responding.

However, after seeing a youtube video talking about a teen who committed suicide after interactions with a non-therapist AI, I thought it was important to help people understand more about AI. While the incident described in the video was describing interactions with a general AI I am focusing my response on distinguishing between humans and therapist AI. The principles should still apply to other forms of AI as well.

So where to start? I asked Chat GPT. The response is in italics.

Identifying whether you're talking to AI or a person can be tricky, but here are some tips:

1. Response Patterns: AI often provides quick, structured responses. If the conversation feels overly formal or lacks personal anecdotes, it might be AI.

2. Consistency: AI typically maintains a consistent tone and style, while humans may exhibit more variability in their language and emotion.

3. Depth of Understanding: Ask complex or nuanced questions. AI may struggle with topics that require deep personal experience or emotional insight.

4. Repetitiveness: AI can repeat phrases or concepts, while a person is more likely to offer varied responses.

5. Contextual Awareness: AI might have limitations in recalling previous parts of the conversation, especially over longer exchanges.

6. Personal Experience: If you ask about personal feelings or experiences, a human will share stories, while AI might provide generalized information.

7. Errors and Quirks: Humans make typos or mistakes that are contextually aware, while AI errors are often more systematic.

8. Open-ended Questions: Engage in open-ended discussions. A human might elaborate and explore tangents, while AI may steer back to predefined topics.

Using these cues can help you gauge whether you're conversing with a human or an AI!

What else would I add?

A Therapist will tell you if you're wrong. While I tend to do this in a gentle way, I do still point out if I think someone is wrong. If someone has diagnosed themselves with a problem I will point out things they missed. I often challenge people to consider other perspectives.

Responding takes time. While I do have scripts for responding to common issues I still have to take time to make those scripts relevant to the person I'm talking to. Human response takes time. AI can respond with paragraphs in less than 10 seconds consistently.

Dependency on the therapist is discouraged. With any client I know that our relationship is temporary, to improve their function, and increase their ability to address their own needs.

Consent forms. A human therapist involves communication regarding consent for therapy. We have to explain limits to confidentiality, which includes the duty to report for concerns of harm to oneself or someone else.

Fees. Who is paying for it? Have you paid for the service? Is it paid for through public or government funding like a crisis line? Is it paid for through your work benefits like an EAP? When a professional is spending consistent time providing a service then someone is likely paying for their time if it's a real person.

There's probably more that could be added, but this should be enough to help people distinguish between AI and people.