From AI Dependence to Independent Judgement

4 days ago

4 min read

There is a natural concern that excessive use of AI may cause our dependency on it. But I am becoming increasingly sure that the problem is not the amount of AI we use. It is the place where we stop thinking for ourselves.

Suppose you have some idea on the topic you do not know very well. You ask the AI to investigate it. In just a few minutes, it finds the related concepts, possible applications, research topics and sources. It is useful. The problem starts when AI does not only find out the knowledge but also explains it, chooses the evidence, interprets it, structures the argument and even writes down the text. You can approve the result, but how much of the thinking is now yours?

I will call it AI-first dependence, and it is the first position in the infographic. AI becomes the natural starting point, and its structure or answer is accepted without much criticism. The latest research on cognitive offloading has made an important distinction between the use of AI as an instrument and its substitution of a part of our cognitive activities. [1]

It does not mean switching off AI. The second step – guided reliance – starts when we start checking the crucial results, questioning the sources and realising that the answer sounds convincing and goes far beyond our true understanding. We are relying on AI much, but we are starting to choose when such reliance is justified. [2]

The role changes with the third position called active partnership. Instead of asking AI for an answer, we start using it to enlarge the perspective. What other interpretations are possible? What evidence contradicts my point? What previous concept is similar to this one? Where can my argument fall down? At this step, AI is not just providing the knowledge to us. It creates many options for us to consider. The study on AI-supported learning shows that it matters whether AI reduces or increases the human agency in the process. [3]

The fourth position is independent interpretation. It is especially important if we start from the idea but know little about the subject. AI can still do a lot of things. It can build the initial knowledge map, find the papers and theories, summarise the results and explain the unusual terms. But some of the knowledge has to go through us. It means reading the sources, understanding their results, realising their limitations and making sure that the evidence supports the argument we want to make. We do not have to repeat all the work that AI has already done. Finding fifty papers to prove independence is useless if AI has already found fifty papers. What matters is that we get enough knowledge to make an independent interpretation. It is connected with the new research on cognitive ownership – whether people experience AI-supported thinking and output as their own. [4]

The simple test is here. Close the AI conversation and explain the idea, why you believe it, the evidence for it, and what could change your mind. If you can do it, something important happened. AI has done much, but the knowledge is not staying outside of you anymore.

The fifth position is responsible autonomy. It does not mean being independent of AI. On the contrary, AI can perform more actions at this stage than any previous. It can look for the unknown territories, compare the evidence, find adjacent concepts, generate the scenarios, challenge the assumptions, check the originality, reorganise the draft and improve the final text. The main thing is that the human has the ability to interpret, verify, challenge and make decisions. It is close to the recent research on meaningful human oversight, which states that meaningful oversight requires more than just naming the human at the end of the process. The person must have real evaluative capability. [5]

This is the reason why I do not think that we can divide people into heavy and light AI users. One person can spend several hours a day using AI but still keep strong independent judgement. The other person uses AI less but becomes dependent if he starts every difficult question with asking the AI for an answer and accepts its conclusions without further discussion. Five positions described are not the fixed personalities. They are the states of use. I can use the responsible autonomy in the area I know well and switch to guided reliance entering the unknown territory.

Here is the challenge. Look at the infographic and try to understand what place you occupy in the model for the type of AI work you use most often: AI-first dependence, guided reliance, active partnership, independent interpretation or responsible autonomy. But do not stop with the level. Try to realise why you occupy it. Is it because of saving time? Because of the unknown subject? Because of trusting the system? Because of not checking anymore? Because of using AI to challenge my thoughts? Or because of being able to explain and defend the result without the help of AI?

Where are you in the model – and why?

[1] Zhu et al. (2026)

[2] Pearson et al. (2026)

[3] Wang et al. (2026)

[4] Su et al. (2026)

Where are you in the model – and why?

Comments