AI that recognizes passwords by voice: Be careful when typing them in public

The development of artificial intelligence has reached another milestone, something that may worry many users.

In recent months, the development of artificial intelligence has raised many concerns about the future of humanity. Among these main reasons is the fear that it will replace entire sectors in the world of work, and that it will direct decision-making processes that have hitherto been the preserve of humans. Unfortunately, there are threats far more pressing than these: In fact, AI is already being used in many IT sectors, including by hackers.

A recent study conducted by British researchers, for example, revealed that there is software equipped with artificial intelligence that can do this Decipher what we type simply by listening to the sound of our keys. This can be very dangerous, but before giving in to panic, it is important to understand how this model works and assess the risks realistically.

We may also have to worry about what the AI ​​is listening to

Using a deep learning model, a team of British researchers from Cornell University was able to achieve this Teaching programs how to recognize the sound of keys pressed on the keyboard. By pressing the keys of a modern MacBook Pro 25 times and recording the sounds made, the researchers obtained a set of data that they then used to train the artificial intelligence, teaching it to recognize differences in sound between one key and another.

“If you get enough data, the model can be built easilyOllie Buckley, professor of cybersecurity at the University of East Anglia, told BBC Science Focus. The potential danger is quite obvious: if you were to use such a tool near someone typing on their computer, You can decrypt everything that person types, including passwords.

While this all sounds like something out of a dystopian movie, there may be reasons why it’s not as unsettling as it seems. The main reason, as previously explained, is this AI always needs training data. To know someone’s writing patterns, they must have access to a data set about that specific person’s writing habits.

The only way is to hack into the computer or smartphone and access the user’s microphone to capture data with which the model can be trained. But all it takes is a little background noise, like music or chatter, and this becomes much more difficult.

Another problem concerns, for example, the case of the “Shift” key, which is mainly used to type uppercase letters or use symbols. In these cases, it will be very difficult for the software to understand when the user uses the key combination.

In short, this new aspect of AI is important because it once again demonstrates the potential of these tools, but also as Professor Buckley mentioned For now, this should not become a widespread problem.