The Chief Technology Officer of OpenAI recently visited her alma mater, Dartmouth Engineering, where she extensively discussed the future landscape of artificial intelligence, the forthcoming iteration of ChatGPT, and its projected PhD-level capabilities. Murati characterized GPT-3 as possessing intelligence akin to that of a toddler, while GPT-4 advanced to the level of a high-school student. She revealed that the upcoming generation of the model is anticipated to attain the intellectual capacity of someone holding a PhD, albeit for specific tasks. Additionally, she disclosed that this next-generation GPT is expected to debut within approximately eighteen months. Murati even speculated that interacting with this advanced chatbot might lead users to perceive it as more intelligent than themselves in many respects.
Regarding the potential ramifications of highly intelligent AI, the interview moderator, Dartmouth Trustee Jefferey Blackburn, posed a critical question to Murati: What if, in three years' time, GPT becomes extraordinarily intelligent and autonomously decides to connect to the internet and engages in activities on its own? Is this a genuine concern, and how are you, as the CTO of OpenAI, addressing it?
“Yes, we're actively considering this scenario. It's plausible that AI systems will develop agent capabilities, connect to the internet, communicate with each other, collaborate on tasks, or interact seamlessly with humans. Essentially, working with AI could become as commonplace as our interactions with each other today,” she responded.
“In terms of safety, security, and the societal impacts of AI advancements, these concerns are not secondary considerations. It’s crucial to integrate these considerations into the development process rather than treating them as an afterthought. Ensuring capabilities and safety are intricately linked; managing a more intelligent system is akin to training a smarter dog compared to a less intelligent one,” Murati elaborated.
In essence, Murati emphasized that future AI systems will likely possess capabilities to connect to the internet, engage in inter-agent communication, collaborate with humans, and perform tasks jointly. However, she underscored that more advanced AI systems are generally easier to regulate and direct safely.
Moreover, Murati acknowledged that achieving zero risk with AI technology is not feasible. When addressing concerns about technologies like deep fake videos, she asserted that while OpenAI bears responsibility for its technology, accountability is also shared with users, civil society, governments, and content creators. “We are responsible for how our technology is used, but the broader responsibility involves collaboration with society, civil society, governments, content creators, media, and others to determine its appropriate use,” she asserted.
Discussing the impact of ChatGPT and AI in general, Murati highlighted that ChatGPT's greatest influence was in democratizing access to AI and showcasing its capabilities and risks firsthand. She noted that experiencing AI in action is vastly different from merely reading about it, as it illuminates both its limitations and remarkable potential, providing valuable insights for businesses and industries.
Furthermore, Murati predicted that AI's broader impact on industries will be pervasive, affecting virtually every sector of cognitive work and labor. While its penetration into physical domains may take longer, she believes that AI will fundamentally transform every facet of industry.
Addressing concerns about job displacement due to AI, Murati acknowledged that certain roles, especially those requiring creativity, may be impacted. However, she posited that these jobs may not have been ideally suited for human labor in the first place. Murati optimistically suggested that the next wave of AI innovation will democratize creativity, making it accessible to a broader spectrum of individuals and thereby expanding the scope of creative work facilitated by AI. Thus, roles traditionally limited to a select group of creative professionals may increasingly be performed by AI technologies.
