Artificial intelligence has become a sensitive topic. Not because some see it as a future source of unemployment for many professionals, or even talk about human extinction. In a more practical sense, scams have grown enormously because of them. But if all this were not enough, there is also another much less known problem: invisible orders.
It is true that, from a media point of view, invisible orders or commands (prompt injection are often called in English, «rapid injection») have hardly any impact. But more and more experts are speaking on the subject. Basically, because they see a significant threat in the security of artificial intelligence. What are they and how can they really affect users?
Invisible orders, a danger for AI
As stated on the specialized portal Axios, the invisible orders of these «injections» somehow exploit the probabilistic and stochastic nature of language models. Said like that, it may sound Chinese. But if we say that what attackers want is to “hijack” AI systems, things change. And how do they do it? Well iIntroducing malicious and undetectable instructions.
For example, writing white text on a white background and entering it into the data that an artificial intelligence model processes. With something so simple, AI can be made to reveal sensitive information, or even more disturbingly, to carry out unauthorized actions. Thus, any AI can become a danger to the privacy and security of users.
And aren’t those responsible for AI doing anything to avoid this problem? Of course yes. But as is often the case in almost all these cases, hackers always seem to be one step ahead. Which does not mean that, in fact, people like OpenAI are already training and adjusting their models to resist these types of tricks. But the theory is much easier than putting it into practice.
Experts openly acknowledge it: facing these invisible orders for rapid injections is a challenge. No matter how much coded rules are being carried out to block malicious actions, or also techniques to mark external data. Yet, The problem continues, and there is not much optimism that it will disappear just like that overnight.precisely.
A risk that is increasingly greater, say experts
Currently, more than 2,000 artificial intelligence tools are used. If you think about it, it’s truly outrageous. But the matter will not stop there. According to experts, its market will grow to 1.3 trillion (with a “b”) between now and 2032, more or less. The translation is very simple: AI is going to be everywhere, from everyday apps to professional tools.
Therefore, if these AI systems are not safe, there is going to be a big problem. Not to mention what would happen if the much-feared artificial general intelligence were achieved. In that case, the consequences could be even more difficult to foresee.
Get to know how we work in NoticiasVE.
Tags: Artificial intelligence