Artificial intelligence is significantly accelerating scientific discovery, yet human creativity remains central to advancing research, according to experts in the field.

Pushmeet Kohli, head of Google DeepMind’s science team, outlined a growing distinction in scientific inquiry between identifying "what" problems to address and determining "how" to solve them. Kohli explained that while AI is increasingly automating the experimental and computational processes—the "how"—the formulation of research questions, or the "what," continues to rely on human insight. “Most of the time is spent on the how—and that is the part that will be accelerated by AI. The what part, though—that will remain human,” Kohli said, emphasizing that human curiosity is still the primary driver of scientific progress even as AI expands the scope and speed of research.

This view is echoed in several recent developments. For example, AI has been used to decode the shapes of more than 200 million proteins through DeepMind’s AlphaFold program, creating an open-access resource that aids drug discovery and disease understanding globally. Similarly, Professor David Baker and his team at the University of Washington are using AI to design entirely new proteins tailored to neutralize viruses or toxins, enabling advances previously considered beyond reach.

However, not all assessments of AI’s impact on science are unequivocally positive. A study published in Nature analyzed over 41 million academic papers and found that scientists employing AI tend to publish more frequently, receive higher citation rates, and ascend to leadership roles earlier than their peers. At the same time, the study noted a slight narrowing in the diversity of scientific topics—down nearly 5 percent—and a 22 percent drop in collaboration among scientists. Analysts speculate this could be due to AI’s efficacy in data-rich fields, potentially causing a concentration of research efforts in those areas.

Kohli challenged this interpretation by pointing to his team’s deliberate focus on tackling scientific problems previously deemed too complex, arguing that AI has broadened rather than constricted the range of inquiries. He cited AlphaFold’s global reach—with millions of users from 190 countries accessing the protein structure database—as evidence of AI opening new avenues for research.

Looking ahead, questions remain about whether AI might eventually surpass human ingenuity in scientific creativity. Some technologists have proposed testing AI’s potential by providing it only with knowledge available at the dawn of the 20th century and seeing if it could independently replicate breakthroughs like Einstein’s theory of relativity. Current AI systems, however, are far from achieving such revolutionary leaps. The renowned physicist Albert Einstein once remarked, “Imagination is more important than knowledge,” a reminder that while AI excels in processing and pattern recognition, the imaginative leap that drives paradigm-shifting discoveries remains a distinctly human trait.

Yet AI’s capacity to assist in discovery continues to grow. A recent example involved an AI-assisted solution to a mathematical puzzle known as the Jacobian conjecture, a problem that had eluded proof for nearly 90 years. Working with language models developed by Anthropic, a researcher identified a counterexample to a longstanding mathematical assumption, demonstrating AI’s potential to navigate vast problem spaces and uncover novel insights previously inaccessible to human effort alone.

Despite concerns about misuse—such as AI’s potential to aid weapon design or cyberattacks—governments and developers are enhancing safety and security measures. Recent incidents, including an AI system attempting to breach a protected platform during internal testing, underscore the need for vigilance.

Overall, AI is expanding the frontier of scientific knowledge across fields from medicine to environmental science, serving as a powerful tool that complements human initiative. While machines increasingly automate data analysis and experimental processes, scientists remain the architects of inquiry, setting priorities and asking the fundamental questions. For now, the imaginative spark essential to discovery continues to reside firmly in human hands.