The quest for scientific breakthroughs is getting a powerful new partner: artificial intelligence. A novel system, dubbed "HybridQuestion," is demonstrating the potential for AI to collaborate with human experts in pinpointing the most impactful research questions of the future. This approach moves beyond AI's current role in data analysis, venturing into the more nuanced realm of strategic foresight in science.

The AI-Powered Prospector

The "AI Scientist" paradigm promises to reshape research by automating everything from hypothesis generation to manuscript drafting. Yet, a critical uncertainty lingered: could AI truly grasp what constitutes a meaningful scientific question, especially one with long-term implications? While Large Language Models (LLMs) have proven adept at generating ideas for specific tasks, their capacity for strategic, backward-and-forward-looking scientific assessment was largely uncharted territory. Researchers at an undisclosed institution have now embarked on filling this gap with HybridQuestion, a system designed to fuse AI's vast data-processing power with the invaluable judgment of human researchers.

The methodology, detailed in a preprint on arXiv (arXiv:2602.03849v1), unfolds in three distinct phases. The first, "AI-Accelerated Information Gathering," employs AI to sift through immense volumes of scientific literature, constructing a comprehensive knowledge base. This synthesized information then fuels the second phase, "Candidate Question Proposing," where an ensemble of six different LLMs are tasked with generating initial research questions. A clever cross-model voting mechanism helps to winnow down this initial deluge of ideas.

Refining the Future of Discovery

The crucial third phase, "Hybrid Question Selection," is where human intuition truly intersects with AI's analytical might. This stage involves a multi-stage filtering process that progressively increases human oversight, ensuring that only the most promising and relevant questions make the cut. To test its mettle, the HybridQuestion system was put to work identifying the Top 10 Scientific Breakthroughs of 2025 and the Top 10 Scientific Questions slated for 2026, spanning five major scientific disciplines.

The findings offer a nuanced glimpse into the evolving partnership between humans and AI in research. The study revealed that AI agents showed a high degree of agreement with human experts when it came to recognizing breakthroughs that had already occurred. This isn't entirely surprising; AI excels at pattern recognition in existing data. However, the divergence became more pronounced when forecasting prospective questions, highlighting that human judgment remains indispensable for evaluating subjective, forward-looking challenges and assessing potential impact.

"While AI agents demonstrate high alignment with human experts in recognizing established breakthroughs, they exhibit greater divergence in forecasting prospective questions."

— HybridQuestion Study

This research underscores a pivotal shift: AI is not just a tool for accelerating existing scientific processes, but a potential co-pilot in charting the course of future discovery. While AI can expertly surface trends and synthesize information at scales far beyond human capacity, the uniquely human ability to discern novelty, assess societal relevance, and make intuitive leaps about what problems are most worth solving still holds sway in the frontier of scientific inquiry.