The company behind ChatGPT assesses the “chemical, biological, radiological and nuclear” risks posed by AI

ChatGPT and will the rest of the artificial intelligences that abound on the web pose risks? catastrophic » to the human race? The company behind the chatbot is preparing for it in any case.

“ What dangers will cutting-edge AI systems pose when misused? » It is this complicated question that OpenAI wants to try to answer. The company behind ChatGPT has, in fact, just set up a team dedicated to assessing the danger that future artificial intelligence systems may pose.

The risks posed by artificial general intelligence

The least we can say is that the firm takes the subject seriously. In her announcement post, she immediately explains that “ the team will contribute to monitoring, evaluating, forecasting and protecting against catastrophic risks [….] such as phishing, cybersecurity, and chemical, biological, radiological, and nuclear threats “. To further darken the picture, “lautonomous reproduction and adaptation », or a sort of technological singularity, are also cited as potential risks.

Concretely, OpenAI – which admitted in its birth certificate in 2015 that AI “ could harm society if designed or used inappropriately » — today wants to think about the safeguards to impose on its own invention. More precisely, it is the AI ​​of tomorrow that worries the company. OpenAI speaks, in fact, of “ frontier models “, i.e. models of artificial general intelligence that still do not exist, but which ” will exceed the capabilities of the most advanced existing models “.

OpenAI wants to self-regulate

The company fears that these digital brains will be used to deceive and scam Internet users (this is already the case), create malicious code (this is also already the case) and potentially take control of certain critical infrastructures such as power plants nuclear (this is, a priori, not yet the case).

The team will be led by Aleksander Madry, an MIT expert, and will work to “ creating a range of safeguards and establishing a governance structure for accountability and oversight » of AI. If the creation of such an entity cannot do any harm, the scenarios mentioned by OpenAI do not really encourage optimism.

In addition, the structure being directly under the governance of one of the largest companies dedicated to the development of AI, conflicts of interest seem difficult to avoid. Let us hope that States and international structures will soon become interested in the subject too.

Source : OpenAI