The optimistic view is a historic winner. Even today, it argues that artificial intelligence is a productivity-enhancing complement to the human employee, not a competitor. However, humans need qualifications to work with AI.

In practice, what is the success rate of retraining employees to work with AI? This is the question that four American researchers attempted to answer over the summer.

They looked at a dataset of 1.6 million completed retraining sessions in the US for the period 2012 to 2023 and compared pre- and post-training income levels. The results are positive - though it has a catch.

AI-focused training boosts the average salary in a new AI-intensive job by 14 percent over comparable employees who only received help finding a new job after quitting. In monetary terms, this represents a difference of $6,225 per year.

The catch is that if a job seeker takes a general retraining course (not AI-focused) and finds a job in an industry with a lower rate of AI use, they will earn 29 percent more.

There are more logical explanations. AI-focused training will increase an employee's general awareness of the subject, but it's often not enough to handle complex and specific tasks in an environment with a high rate of AI use.

It does not equip employees with the skills to correctly apply individual tools to specific problems. The authors point out that training can also play a strong signalling role for employers about an employee's personality (e.g. that they are willing to learn new things), but that this plays a more significant role where there is less need for practical application of AI.

Just over a quarter of employees moved into jobs where they had higher wages after AI training and also used AI to a greater extent than in their previous job. Thus, the authors conclude that retraining focused on this area pays off. However, it may be even more profitable to "escape" AI and use retraining to move into an industry where the pressure of AI is lower.

That doesn't sound like an ideal solution. On the other hand, it should be remembered that ChatGPT will not celebrate its third birthday until the end of November. The real reinvention of the human role in the world of AI for employees is just beginning.