New tool can measure how much an AI knows about something

artificial intelligence (AI) ChatGPT or Google’s BardGain knowledge from databases, pages, and web browsing by being fed massive amounts of information. But how much can dangerous or harmful in terms of digital security, privacy and system integrity?

This question prompts researchers University of SurreyIn the UK, to develop a new program checks how much an AI actually knows about a company or platform.

The software does a search within the artificial intelligence itself to verify the amount and depth of its knowledge about a company and to know exactly what kind of data is extracted while feeding that platform. Privacy concerns led to the temporary suspension of ChatGPT in Italy.

just to be sure

In practice, this service could be: Integrated with digital security protocol For example, to check if there is any AI in the market, a company had access in education to access datasensitive information or details about security vulnerabilities in code.

So the software can know if there is AI potentially dangerous if it is used by malicious users or vice versa, already have enough knowledge It will be implemented alongside the company’s services, such as in collaboration with a chatbot with other systems such as autonomous cars, smart homes or corporate intranets.

In other possible cases, the software can tell if an AI has learned enough already. to “break” some systemshow to exploit codes and loopholes in online games to illegally ensure the user’s victory.

Research is conducted by the researcher Solofomampionone Fortunat Rajaona and won an award at an international symposium. The full research paper can be read here (in English), but has not yet been published in engineering journals or peer-reviewed.