Hello everyone! AI saves time, but haven't we borrowed that time from our own competence? If you remember, in the previous episode we talked a good deal about this, and in general our episode was devoted to the question: if AI saves time, who gets that saved time? Does a person become freer? Do they start producing more? Or does that time convert into someone else's tasks, possibly into a larger amount of work? Today I want to continue this theme and ask one more question. It may sound somewhat provocative.
And the question is this: haven't we borrowed that time from our own competence? What is this about? I want to give a disclaimer straight away and say that I in no way diminish what technology has brought us, and what technology keeps bringing us over the past three years. I always want to look more widely and to answer honestly the questions we ask ourselves. Besides, much of what pushed me towards this theme was my son, who came to me and said: listen, mum, that was an interesting podcast.
Could you talk about how, perhaps, with AI none of us are getting smarter? That resonated strongly with some of my own thoughts, my own concerns of late. So then, if AI makes the work faster, then perhaps in some cases it makes the specialist weaker. Is that so? Does it take away the practice that our professional depth is built from, that our professional competence is built from? We have already said that faster is not always more productive, and more productive with AI is perhaps not always more competent without AI.
So I suggest we talk about this today. There is a fairly fresh update, and fresh research from my favourite lab at Stanford University. The Stanford University lab in California studies the influence of digital technology on people's lives, on people's work, on the economy, on the labour market. And right now a colossal amount of fresh research is coming out of there, which is very motivating for thinking this theme through. So then, one of the latest studies, which was done on millions of American workers, including data through June of twenty-twenty-six, says the following.
And the first thing they say is that there are no mass layoffs of any kind across the whole economy caused by the use of artificial intelligence. But what is interesting is that employment among people aged twenty-two to twenty-five in occupations more strongly affected by artificial intelligence came out nineteen percent lower than employment among young people of the same age in occupations that make less use of artificial intelligence today. It seems to me we have already spoken about this contradiction in relation to programmers. Alexander has repeatedly drawn attention to the shrinking of entry-level tasks for junior specialists.
We have several podcasts on this theme. And that immediately raises a question for me: if a young person cannot have this fast way into a profession,
then how are they going to acquire professional depth? How are they going to form their skill? In essence, an expert does not become an expert straight away. An expert is not born an expert; they emerge from a large number of routine tasks, from a large number of mistakes, repetitions, from a large amount of feedback. And now imagine that we remove this training ladder, and in effect, after some time, we may end up with a very strange, specific market. Seniors and highly qualified specialists are obviously still needed, and there are fewer and fewer people who managed to become them.
If we take this line of thought further, another question arises: is practice really needed to form a skill?
Obviously, yes. And the question is fairly rhetorical. But are we losing that practice when we use AI? In real work we are indeed more and more often starting to assess the result. How the letter is written, how the code is generated. And in effect, what does it matter how many mental operations a person performed to get there? I began to notice this even in myself. Let me share this personal experience. I used to write very quickly, and I could express any thought in writing.
That was overall a fairly well-developed and confident skill of mine. And now I have a different process. For several years now, under heavy load, I have dictated a large number of letters by voice. Then a model helps package them along my logic, using my wording, using my style, filling them with my meanings and phrasing. But the speed, without question, is entirely different. And in effect this is a completely different skill from putting a letter into text myself straight away.
I began to feel the difference. It seems to take me longer now to write letters on my own. I want to be precise and honest here: perhaps the skill itself did not weaken. Perhaps I am simply not ready to spend more time on a task I can do much faster. I even went looking for research specifically on adult professionals who regularly dictate letters and hand them to AI. I found nothing substantial or interesting, but in general I came across an interesting thing: dictating and typing text yourself are very different processes.
In them, planning, pauses, the forming of thoughts and the conclusion are distributed differently. And in effect this is not about losing a skill but about redistributing the skill and its components inside it. But along with that, once again, a hypothesis appeared for me: what we are losing is not so much the ability; perhaps we are losing the readiness to stay inside a hard task for a long time.
But, everyone, it is exactly this ability to stay inside a task for a long time that shapes deep professional skill. It is the ability to write and compose something at length, to create something at length, to think one more time. In effect, it is within that time that our ability is trained, within that time that our skill is formed — that is, being inside the task itself. So, on the one hand, it turns out that the absence, the shrinking of the way in for young specialists without skills into a profession, does not let professional skill form in a way that lets you truly own it deeply and well. Second, the use of external tools leads to us no longer wanting to work long and hard on some task, which also cuts the time needed to form a skill. Everyone, do share how you see it and what you think, who has what experience?
Who has become stronger with artificial intelligence, who feels that they are starting to lose some skills and urgently needs to restore them and pay attention? Let us look further along this line of thought and look at what happens if this additional help is taken away from a person — additional help in the form of artificial intelligence, in the form of various tools. There is quite a lot of interesting research on learning right now, run in different contexts, with different designs, on American students and schoolchildren, on Chinese students and schoolchildren, which certainly cannot be framed as some universal law, but they all show a very important distinction. There is a person's result together with artificial intelligence, and there is the knowledge that stays with the person if you take artificial intelligence away.
I want to give the example on students specifically for a reason, since we are talking about the process of learning, and I want to think about what actually stays here. That knowledge which stayed with the person is much lower, and students and schoolchildren solve faster and better, without question, with artificial intelligence, do their homework faster, show a more
fast and high-quality result. But what is interesting is that without the use of artificial intelligence, in supervised exams, this same group, this same audience, began to show results twenty percent or more weaker. That is, in effect I want to say the following: this productive human-plus-AI system grows, and grows very fast, but the human capital inside this system may be shrinking, or at least is not growing as fast as we expect. Next, a conclusion appears for me which is, to my mind, extremely important. Perhaps we are not only losing and missing this ability to form our skill in
completely different areas of our life, but also doubt — because in relation to artificial intelligence we very often talk about critical thinking, about how important it is to think critically and assess the result that one model or another produces. We talk a great deal about how every answer has to be checked. But to check it, a person has to allow that they are wrong. And a person has to have this inner doubt and feeling, this sense, on the basis of which they understand that an answer needs checking. On PubMed, a medical portal that today holds a large number of papers, clinical trials and medical research, there is an interesting study from twenty-twenty-six which was run not on practising doctors, fortunately, but on medical students. The students were split into three equal groups and each given twenty-five hard diagnostic questions, which were built in the style of clinical tasks from the American medical exam. So then, the first group saw only the question.
The second group saw the question and a very correct AI explanation, which had been verified by medical specialists. The third group received a deliberately created convincing explanation, deliberately created by artificial intelligence, but one that was incorrect. So then, what is interesting. The probability of a correct answer fell by more than half in the group that received the incorrect correct explanation from artificial intelligence. Here I even came across a very interesting scientific term: a confidence calibration failure.
That is, in plain words, a person stops sensing a wrong answer as less reliable, and starts relying entirely on the answer the system gives them. This is a very live theme and a very live question specifically within the medical community. Well, at the very least because there is quite a lot of similar research, and I will not dwell on it here. And it is obvious that such things are better tested in a learning situation, not where the question already touches on life and death.
But for me this is a case that once again shows this slowly shrinking ability of ours to begin doubting the output a model gives us, and
the way it presents that result to us. After all, the most dangerous specialist is not the one who does not know; the most dangerous specialist, as you and I know perfectly well, is the one whose confidence has stopped telling knowledge from error at all. We have already built up a multi-level chain here. I will sum it all up now, but I want to say that the market is in effect already sending us two opposite signals.
On one side, companies are looking for AI-first, AI-enabled specialists, people who can competently use AI in their professional work, get results fast, pass that knowledge to others, build new processes, manage the tool. That is one side. On the other side, it is getting harder and harder for an employer to understand which professional skill they are actually buying. What lies behind the competence of one specialist or another. A survey was run: two hundred and fifty British employers were asked, and sixty-nine percent of them said that they would rather see even some grammatical error in a CV or an application, because it lets them think that a real person wrote that CV and that it was not entirely artificially generated by artificial intelligence. That is, in effect, the company itself wants to see the person without an intermediary.
And the manager themselves, taking the decision, wants to be clearly sure of what exactly they are buying. And here a double qualification arises, because I can work with artificial intelligence, but I have to prove that my professional judgement, that my professional skills, are independent of artificial intelligence. A week or a week and a half ago, on one of the channel's podcasts, Alexander talked about labelling AI-generated content. And during that podcast I thought that a hundred percent, after some time, there will appear ways to get around such labelling.
And alongside that, a hypothesis came to me that after some time, lines in CVs saying “I work without AI” or “I can write without AI”, or “I can generate code without artificial intelligence”, will become quite relevant. Perhaps we will come to that. But, in fact, I want to close today's podcast with Anthropic's policy for candidates as well. And Anthropic is the company that created Claude, and Claude is used inside the company itself every day. And without question, Anthropic looks for candidates who definitely know how to collaborate with and use artificial intelligence competently.
But then a very interesting separation of modes begins. And first, the CV and the answers to questions the candidate has to prepare themselves, and the test task, if there were no additional permissions, candidates have to take without using models. For preparing for the interview, and during the live interview, no help from artificial intelligence, unless the company has specifically stated a requirement for some other approach. That is, in effect, one of the most AI-native companies already tests two different qualifications at the door.
Whether a person can collaborate with artificial intelligence, and whether they have their own proven, convincing skills and evidence. If you put together everything we are talking about today, a very interesting picture emerges.
Artificial intelligence definitely does not lay people off en masse, but, as we can see, it narrows the way into some professions. It also hinders the forming and growth of seniors inside those professions — people who have deep skills, which is exactly what they are valued for. Artificial intelligence gives a finished result and saves time, but it can take away the practice from which our professional depth is formed. Artificial intelligence gives a fast finished result and saves time, but along with that saved time it takes away the time for the practice in which our professional depth is formed.
Artificial intelligence is able to make the overall human-plus-AI system more productive. But that does not mean that the person inside this system, the human component inside the system, becomes more competent in themselves. And a convincing AI answer may contain an error. But the point is not even that it may contain an error; it may make us less able to feel that we are making one. So, probably, if we sum all this up, our main professional asset becomes not even AI literacy but a kind of double competence.
I can amplify myself with the help of artificial intelligence, but alongside that I keep my own depth, I keep my own competence, I keep my own practice, so that I can take responsibility for the result. So if in the previous episode we talked and reasoned about who gets the time AI saved, to that question I want to add one more: did we not trade that saving for our own competence? And this is not about going back to some world without AI. This is about learning to manage this process properly, so that in a world with artificial intelligence there remains a person who manages that artificial intelligence.
We live in interesting times. Stay in touch.