AI is rapidly advancing beyond its current role as an assistant, with OpenAI’s Chief Scientist Jakub Pachocki suggesting that in the near future, AI models could engage in original scientific research autonomously. In a report by Nature.com, Pachocki highlights the evolving capabilities of AI, specifically noting that models are moving towards a point where they will require minimal human input.
Although tools like ChatGPT still need prompts and guidance from humans, OpenAI has developed systems like Deep Research, which can already process and analyze large data sets independently for several minutes. This system can produce useful results, even with limited computing resources. As computing power increases, Pachocki envisions AI making strides in scientific research, automated software development, hardware engineering, and potentially contributing to novel scientific discoveries.
A core aspect of this progression is reinforcement learning—a process where AI learns and improves by receiving feedback from its actions. While earlier AI models relied heavily on pre-training with massive datasets, modern models are starting to use reinforcement learning to develop strategies for solving complex tasks independently. This shift marks a significant departure from earlier models, which mainly mimicked human output based on patterns in the data.
Pachocki notes, however, that AI models don’t think like humans. They don’t have memory of past experiences or an understanding of time. But their ability to simulate logical steps and solve problems through structured thinking makes them valuable tools for tackling challenges that require organized reasoning.
Looking ahead, Pachocki is focusing on building more powerful AI models and pushing the boundaries of what AI can do in fields like science and technology. While the debate about artificial general intelligence (AGI) continues, Pachocki believes the true test will be how well these AI systems can operate independently and tackle unsolved problems—a key milestone in AI’s evolution.
