OpenAI, a pioneer in the artificial intelligence arena, recently revealed the establishment of a specialized team titled “Preparedness”, with the express goal of delving deep into the potential hazards associated with rapidly evolving AI technologies.
At the helm of the Preparedness initiative stands Aleksander Madry, celebrated for his work at MIT’s Center for Deployable Machine Learning. Having onboarded OpenAI in May, Madry’s core objective now is to orchestrate efforts to understand, predict, and mitigate threats that AI models may pose shortly.
The range of risks Preparedness aims to scrutinize is extensive, from concerns about AI systems deceiving humans, reminiscent of sophisticated phishing schemes to apprehensions regarding AI’s potential code-tampering aptitudes. What stands out most is OpenAI’s declaration that the team will probe into “chemical, biological, radiological, and nuclear” AI-related threats.
This announcement by OpenAI echoes CEO Sam Altman’s recurrent cautionary sentiments about AI’s potential to harbour adverse outcomes on a colossal scale. While some critics might find it an exaggeration, the commitment of substantial resources to investigate such scenarios underscores the gravity OpenAI attributes to these concerns.
Moreover, OpenAI has signalled an openness to explore even the less conventional arenas of AI-associated risks. In alignment with the unveiling of Preparedness, OpenAI has launched a community initiative, encouraging individuals to contribute their visions of AI risks. The most insightful entries can garner a $25,000 reward, with the top ten submissions even standing a chance to be absorbed into the Preparedness team.
The contest poses challenging questions to the participants, encouraging them to think like potential adversaries with access to OpenAI’s top-tier models like Whisper, Voice, GPT-4V, and DALLE·3. The emphasis is on discerning these technologies’ most plausible yet potentially devastating misuse scenarios.
In line with its mission, the Preparedness team is also tasked with drafting a comprehensive “risk-informed development policy”. This framework aims to encapsulate OpenAI’s holistic strategy, from AI model creation to monitoring, underlining the organization’s proactive risk-mitigation measures and establishing robust oversight mechanisms.
Emphasizing its dedication to the global community, OpenAI expressed in its recent blog post, “AI models… hold promise to usher in unprecedented benefits for humanity. But parallelly, they harbor escalating risks… Our mission is to ensure we’re equipped with the requisite knowledge and tools to ensure the safe deployment of highly competent AI systems.”
The curtain-raiser for Preparedness, aptly timed with a significant U.K. government summit dedicated to AI safety, follows on the heels of OpenAI’s earlier declaration. A commitment to studying and steering emergent, “superintelligent” AI forms. It’s a shared vision between Altman and Ilya Sutskever, OpenAI’s chief scientist and co-founder, that human-surpassing AI may be closer than we anticipate, necessitating earnest efforts to define its boundaries.
What is the primary objective of OpenAI’s ‘Preparedness’ team?
Answer: The ‘Preparedness’ team at OpenAI aims to assess, predict, and counteract potential hazards linked to advanced AI technologies. They are focused on understanding threats that future AI systems may pose, from deceiving humans to possible nuclear threats related to AI.
Who is leading the Preparedness initiative at OpenAI?
Answer: The Preparedness team is led by Aleksander Madry, known for his contributions at MIT’s Center for Deployable Machine Learning. He joined OpenAI in May and will direct efforts to scrutinize and address AI-related threats.
How does OpenAI involve the community in its mission for AI safety?
Answer: Alongside the announcement of the Preparedness team, OpenAI initiated a community challenge. They are encouraging individuals to share their visions of potential AI risks. The top submissions could earn up to $25,000, with standout contributors having an opportunity to join the Preparedness team.
Why is there a growing concern around “superintelligent” AI?
Answer: Sam Altman, the CEO of OpenAI, and Ilya Sutskever, the chief scientist and co-founder, believe that AI with intelligence surpassing human capacities could emerge within the next decade. Such AI models may not inherently have benevolent intentions, underscoring the need for research and preventive measures to manage and restrict their capabilities.