AI will be able to reproduce itself by 2028, says former OpenAI executive

The startup CEO Anthropic and former Vice President of Research at OpenAI Dario Amodei, launched a warning that sounds like a movie scene during an interview, on Friday (12) with journalist Ezra Klein, from The New York Times, about the future of artificial intelligence (AI).

According to Amodei, between 2025 and 2028, AIs will be able to reach a level of autonomy that will enable their survival and reproduction, thus posing a significant risk to global geopolitical and military security.

Currently, Anthropic defines security levels for AIs, identified by the acronym ASL. Amodei suggests that we are at level 2, where language models can already provide dangerous information, such as construction of biological weapons , for example. However, he emphasizes that this information is still unreliable and represents a relatively low risk.

However, concern increases when considering the potential achieved at level 3, which could be reached as early as next year. At this stage, the risk of a catastrophe would be considerably greater, with the possibility of using these technologies in biological and cyber weapons.

Furthermore, Amodei highlights that level 4, still speculative, could bring features such as autonomy, which includes independent ability to reproduce systems, and ability to persuade . He estimates that models with this classification could emerge between 2025 and 2028, which raises serious questions about global security in the face of the military use of these technologies.

Given this scenario, the CEO emphasized the importance of closely monitoring the development and application of AIs, especially considering the potential of some states to increase their military capabilities with this technology.

With investments from giants like Amazon and Google, Anthropic launched its own language model, Claude – the startup claims that it develops the technology in a more responsible way than OpenAI (Anthropic's founders are dissidents from the owner of ChatGPT). Anthropic's AI is already in its third version, since March this year.

(Alice Labate is an intern under the supervision of editor Bruno Romani)

Source: CNN Brasil

You may also like

Biden with Cancer: What do we know?
World
Bruce

Biden with Cancer: What do we know?

The diagnosis of metastatic prostate cancer of former US President Joe Biden understandably raised concerns and questions: How long has