Where AI Ends and the Human Begins: An Analysis of a Review of 27 Scientific Papers

On October 30, 2026, Habr published an analysis of a Q&A on a review of 27 scientific papers. The topic is where the boundary between AI and humans lies. Below are the main points in plain language.

What AI literacy is and why there is no single definition

AI literacy is a set of skills to critically evaluate AI, work with it together, and apply it at home and at work. There is no single definition: different authors put different things into it.

Some reduce literacy to knowledge and application of AI. Others add emotions, values, self-confidence, and awareness of personal responsibility. The authors of the review on schools believe that literacy depends on the tool: separate “GenAI literacy” and even “ChatGPT literacy” for language teaching have emerged. They also distinguish “AI competence” — the ability to use AI easily and confidently.

An important conclusion: a self-assessment questionnaire shows confidence, not knowledge. A person may consider themselves literate but be unable to verify a model's answer.

Example: an HR specialist is designing a training program “AI literacy for employees.” Before ordering a course, they write down what exactly is meant: knowledge of how AI works, the skill of working with a specific tool, or the ability to verify its answers. This determines both the content of the course and how to measure it.

Why people start using chatbots

Literacy, attitude, and use of AI go together, but the data do not show what causes what. People start using AI when they see benefit or interest. The weight of convenience and the opinion of those around them depends on the tool.

This means: one cannot say that a person first becomes literate and then starts using AI. It may be the other way around — they started using it because it was convenient, and literacy caught up later.

Example: a support employee tries a chatbot for drafting responses. If the bot saves time, the employee continues to use it. If the answers take a long time to verify, interest drops. Colleagues' opinions also influence, but not equally for all tools.

Can AI collect employee experience

AI helps collect what employees can explain with the right questions. It does not extract fully inexpressible experience.

In other words, if a person can formulate their experience in words, AI can help structure it. If the experience is inexpressible — for example, an intuitive decision in a complex situation — AI will not extract it.

Example: an engineer describes how they diagnose a fault. The chatbot asks clarifying questions and collects the answers into an instruction. But if the engineer themselves cannot explain why they chose this particular solution, the bot will not help.

Where a human expert is indispensable

A good result on an exam or test does not equal reliability in real work. In rare and complex cases, a human verifies the AI's answer.

This is a key limitation. AI may score high on a test but make a mistake in an atypical situation. Therefore, in complex cases an expert is needed to verify the answer.

Example: AI helps a doctor with a draft conclusion, but the final decision is made by the doctor. If the case is rare, AI may suggest an incorrect option, and a specialist is indispensable.

What to do about this

1. Define what you mean by AI literacy before launching training. Write down: is it knowledge about AI, the skill of working with a tool, or the ability to verify answers.

2. Do not rely on self-assessment. To test knowledge, use practical tasks, not just questionnaires.

3. Start with tools that provide quick and clear benefits. Then employees will use AI and learn along the way.

4. To collect experience, formulate precise questions. AI will help structure what employees can explain in words.

5. In complex and rare cases, leave verification to a human expert. A high test score does not guarantee reliability in real work.