AI for Science has moved past the tool-building phase and is now forcing a fundamental redesign of how research gets done, according to E Weinan, an academician of the Chinese Academy of Sciences and academic committee chair of the AI for Science Institute, Beijing (AISI). Speaking at the 2026 AI for Science Congress in Beijing on August 21-22, E said the technology's real impact will be structural, not incremental.
"The value of AI goes far beyond speeding up research. It will structurally reshape the production model and operational system of basic research," E said.
From assistant to infrastructure
Most researchers still treat AI as a smarter assistant - handling literature searches, running simulations, and automating routine writing within the existing research framework. E called that only the first stage. The harder question, he said, is what comes next: once AI's capabilities leap forward, how should research organization, evaluation criteria, and resource allocation be redesigned to match?
E envisions an open, universal research infrastructure that combines computing power, simulation engines, domain knowledge bases, and automated laboratory interfaces. Routine work would be handled by platforms, making original and disruptive discoveries more visible and more valuable. He drew an analogy to carpentry: apprentice-level tasks get mechanized, but master craftsmanship stands out even more.
Under this model, research would no longer depend on heavy laboratory hardware. Scientists who can pose clear questions would run full research cycles on shared platforms, replacing the traditional model of small, independent labs. The difficulty of original innovation, E said, will not diminish.
Infrastructure before models
AISI has built its first "four-beam, N-pillar" infrastructure system. Unlike conventional platforms built around instruments and computing clusters, it integrates knowledge parsing engines, large-scale simulation modules, and automated closed-loop experimental systems. According to E, simulation capabilities have improved by orders of magnitude in just a few years, and the core modules are self-developed and fully under domestic control.
E also warned against a common mistake in scientific large model development: stacking scientific capabilities onto general-purpose models through incremental patches. That approach cannot solve difficult basic science problems. Research-adapted models require fundamental architectural reconstruction, not feature add-ons.
The infrastructure is scheduled for multiple rounds of key technical iteration this year. Independently developed intelligent laboratories are expected to reach engineering breakthroughs between late 2026 and 2027.
Talent and institutional reform
A shortage of interdisciplinary talent - people fluent in both basic science and AI - remains the key bottleneck. E said many current projects are led by AI specialists, but valuable scientific questions should originate from fundamental researchers, followed by cross-disciplinary collaboration.
He cautioned against copying Internet industry playbooks directly. "The Internet's core is information and communication; AI's core is intelligent decision-making. The underlying logic is fundamentally different - directly applying existing industry experience can lead to technical path deviations."
E predicted that the democratization of research will be a long-term trend. "Only by seizing this historic opportunity of research paradigm transformation and tackling the challenges of technical infrastructure, scientific data, interdisciplinary talent and institutional mechanisms can we translate first-mover advantage into original innovation leadership," he said.
The two-day congress gathered nearly 50 academicians and over 80 young scholars. For researchers working in the trenches, the implications are practical: the tools being built now will change what skills matter, how labs are staffed, and which questions are worth asking. Those who can frame sharp scientific questions and work across the AI divide will be the ones who benefit most - see the AI Learning Path for Research Scientists for a structured entry point. The broader shift, as E described it, is toward a research environment where shared infrastructure handles the routine and human judgment decides what deserves attention - a reordering that touches every lab, from small academic teams to large national facilities. For a fuller picture of how this transformation is taking shape across disciplines, the AI for Science & Research tag tracks related developments.
Why this matters for researchers
If E's forecast holds, the next few years will bring a shift in how research is organized: shared AI infrastructure will absorb routine simulation, literature work, and standardized writing, while the premium moves to original problem formulation and cross-disciplinary fluency. Researchers who adapt early - by learning to work with AI-driven experimental loops and interdisciplinary teams - will have an advantage over those who wait for the infrastructure to mature on its own. The transition also raises practical questions about evaluation criteria and resource allocation that are likely to affect funding decisions well before the decade is out.
Your membership also unlocks:
