Openai races toward human-level ai researchers – a bold, and potentially dangerous, goal
OpenAI is aggressively pursuing a roadmap to deploy AI systems capable of independent research, edging closer to replicating the cognitive abilities of human researchers – a prospect both exhilarating and potentially destabilizing.
A timeline for autonomous intelligence
According to Jakub Pachocki, OpenAI’s Chief Scientist, the company’s recent advancements in coding, mathematical reasoning, and physics suggest a trajectory toward handling increasingly complex, multi-step technical tasks with significantly reduced human oversight. “I definitely see this as a sign that something here is going in the right direction,” Pachocki stated during an ‘Unsupervised Learning’ podcast episode. The key metric, he clarified, is the duration a model can operate autonomously – prioritizing sustained, independent task execution.
This ambition isn’t a casual projection; OpenAI has publicly committed to a phased rollout. By September 2026, the team aims to establish an ‘AI research intern,’ followed by a fully autonomous AI researcher by March 2028. Sam Altman, OpenAI’s CEO, even acknowledged the risk, stating frankly that the company “could fail completely” in achieving this ambitious objective – a transparent admission reflecting the potential ramifications of such a leap.
Codex, OpenAI’s coding agent, already demonstrably handles a substantial portion of the company’s programming workload. But the real catalyst, Pachocki believes, lies in progress within the realm of mathematical benchmarks. These serve as “star poles” for model improvement, offering readily verifiable validation. “We’ve seen this explosive growth in coding tools,” Pachocki noted. “For most people, the act of programming has changed quite a bit.”
The immediate challenge, however, centers on achieving greater autonomy in tackling specific technical challenges. The focus is on amplifying computational power and extending operational periods. “For more specific technical ideas – like ‘I have this particular idea on how to improve the models, how to execute this evaluation differently’ – I think we have the pieces that, mostly, we just need to put together,” he explained. Despite this, Pachocki stresses that current AI systems aren’t yet equipped to function independently at the level of a seasoned human researcher. “I don’t expect that we’ll have systems that you just… and they’ll do it, not this year,” he commented, emphasizing that current capabilities are far from self-directed functionality.
The stakes are high. While OpenAI’s internal goals are ambitious, the underlying trends – rapid advancements in coding and reasoning – are undeniable. The company’s pursuit of autonomous intelligence represents a significant, and potentially disruptive, evolution in the field. And frankly, the willingness to acknowledge potential failure demonstrates a level of realism rare in the often-optimistic world of AI development. It's a race, and the finish line is still shrouded in uncertainty.”n