Artificial intelligence is increasingly gaining access to personal information, accounts, and work data, but not every service should be trusted with everything. Vladimir Leshchenko, an expert in AI and knowledge management, believes that the acceptable level of risk depends прежде всего on who may be affected by the neural network’s decisions.

Even a calendar contains more information than it seems

The very fact of using AI for personal tasks does not necessarily mean a serious threat. Leshchenko said that he himself uses neural networks, including for planning his family’s schedule.

However, to do this, the AI gains access to accounts and calendars. As a result, the neural network may learn not only about the user’s plans, but also about the daily routines of other family members.

According to the expert, in such a situation the risk primarily concerns the person himself. He decides for himself how willing he is to give the machine access to his information and accept the consequences associated with it.

When personal risk turns into public risk

The situation looks entirely different if AI begins making decisions that affect other people.

In Leshchenko’s view, in such cases simple user consent is not enough. If a neural network’s mistake can affect the safety of others, strict limitations should apply.

This is especially true for critical infrastructure — systems whose failure or incorrect decisions can affect entire cities and their residents. Handing over such processes to AI without human oversight creates not a personal risk, but a public one.

The final decision must remain with a human

The expert considers full automation of critical infrastructure unacceptable. Even if a neural network is used for data analysis, forecasting, or preparing a decision, the final word in situations related to people’s safety must remain with a human.

Essentially, the dividing line lies not so much between “safe” and “dangerous” data as between the different consequences of their use. AI access to a personal calendar is a risk that the user to some extent assumes personally. Access to systems on which other people’s safety depends requires a fundamentally different level of control.