AI can lead to extinction of humans and this risk is totally real, says US govt report – India Today

author
2 minutes, 28 seconds Read
image

In October 2022, when the ChatGPT launch was still a month away, the US government commissioned Gladstone AI to work on a report to evaluate the proliferation and security threats posed by weaponised and misaligned AI. A little over a year later, the assessment is complete. The report has found that AI can possibly cause an “extinction-level threat to the human species”.

advertisement

“The rise of advanced AI and AGI [artificial general intelligence] has the potential to destabilise global security in ways reminiscent of the introduction of nuclear weapons,” the report reads.

This was first reported by Time.

AGI, or Artificial General Intelligence, refers to a concept of technology which would be capable of performing tasks equal to or surpassing human abilities. Many tech leaders like Meta CEO Mark Zuckerberg and OpenAI chief Sam Altman have repeatedly spoken of AGI being the future. While such systems are not presently in existence, it is widely anticipated within the AI community that AGI could become a reality within the coming five years or possibly even sooner.

The assessment report suggests the US government to move “quickly and decisively” to avoid “growing risks to national security” caused by AI.

The report has been authored by three researchers. In over a year of completing the report, they reportedly spoke with over 200 individuals, including government officials, experts, and employees of some of the prominent AI companies, such as OpenAI, Google DeepMind, Anthropic, and Meta.

Insights gleaned from these conversations reportedly highlight a troubling trend, suggesting that numerous AI safety professionals within advanced research labs are apprehensive about the potential negative motivations that can possibly influence the decision-making processes among company executives who hold sway over their organisations.

The big ‘Action Plan’

The report also introduces an Action Plan to address these challenges proactively.

The report proposes a comprehensive and unprecedented set of policy measures that, if implemented, would significantly disrupt the AI sector. According to the report’s recommendations, the US government should consider making it illegal to train AI models using computational power exceeding a specified limit. This threshold, as suggested by the report, should be determined by a newly established federal AI agency, with a potential guideline being slightly above the computational capacities utilised in training current models like OpenAI’s GPT-4 and Google’s Gemini.

Further, the report suggests that the new AI agency should mandate AI companies like Google and OpenAI to obtain governmental authorisation before training and deploying new models that exceed a certain lower computational threshold.

advertisement

Additionally, the report emphasises the urgent need to explore the possibility of prohibiting the public disclosure of the intricate details (known as “weights”) of potent AI models, such as through open-source licences, with potential penalties including imprisonment for violations.

Moreover, the report recommends enhanced governmental oversight over the production and exportation of AI chips, along with directing federal funding towards research initiatives focused on aligning advanced AI technologies with safety measures.

Published By:

Nandini Yadav

Published On:

Mar 12, 2024

Tune In

This post was originally published on this site

Similar Posts