What Will Scientific Research Look Like in the coming 5 years? AI for Science Is Offering Answers

In July 2026, on the banks of the Huangpu River in Shanghai, at the Scientific Frontier Forum of the World Artificial Intelligence Conference (WAIC), a consensus was emerging among participating scientists: AI is evolving from a research tool into a research infrastructure. The industry's focus has shifted toward embedding AI deeply into laboratories and integrating it into complete research workflows, taking over a substantial portion of traditional manual scientific work.

This trend was confirmed in the WAIC "Treasure of the Hall" selection. MatwingsVenus™ (晓鹜™), a conversational protein research agent independently developed by Matwings Technology, stood out among hundreds of participating products to become the only AI for Science (AI4S) product selected, appearing alongside achievements from leading domestic technology companies. During the conference, United Nations Secretary-General António Guterres visited the Matwings Technology exhibition booth and expressed appreciation for the platform's achievements in enabling green industries through AI.

At the policy level, the national "AI+" Action Implementation Plan identifies "AI + Science and Technology" as one of six priority action areas; the Fifteenth Five-Year Plan concurrently outlines forward-looking plans in quantum technology, biomanufacturing, and other future industries, providing policy support for AI4S. According to market research institutions, the global AI for Science market was valued at $4.538 billion in 2025 and is projected to grow to $26.23 billion by 2032.

I. The Fifth Scientific Paradigm: From Experimental Trial-and-Error to Intelligence-Driven Discovery

The Fifth Scientific Paradigm

Turing Award laureate Jim Gray articulated the conceptual framework for scientific paradigm evolution, upon which later scholars built a four-paradigm framework—experimental, theoretical, computational, and data-intensive. Building on this foundation, some domestic scholars have proposed that scientific intelligence—anchored in AI-driven autonomous reasoning and dry-wet closed-loop iteration—is giving rise to a fifth scientific paradigm that fundamentally reshapes research models, transitioning science from "workshop-style" to "platform-style" operation.

Traditional scientific research has relied heavily on experience and intuition, testing hypotheses through trial and error—a process that often consumes years or even over a decade for drug discovery and novel material exploration. AI4S empowers AI to serve as a core contributor to hypothesis generation, model solving, and data analysis, systematically reconstructing the research chain. By 2026, AI4S has achieved multidisciplinary adoption, becoming a standardized capability across biology, chemistry, materials science, and beyond, dramatically compressing R&D cycles and reducing experimentation costs.

Matwings Technology's MatwingsVenus™ (晓鹜™) stands as a representative case of the fifth paradigm's translation into practice. Diverging from conventional structure-prediction models, MatwingsVenus™ (晓鹜™) leverages a proprietary database of tens of billions of protein sequences and functional labels to enable a leap from "structure observation" to "function creation"—accurately predicting protein function from sequence and achieving end-to-end delivery through dry-wet closed-loop iteration. Researchers can accomplish protein design, prediction, and validation simply through natural language dialogue, effectively lowering the barriers to protein research.

Industry reports indicate that AI is evolving into end-to-end research agents capable of independently formulating hypotheses, designing experiments, analyzing data, and drawing conclusions. Routine trial-and-error and experimental execution will be increasingly undertaken by AI, freeing researchers to focus on higher-order tasks: posing scientific questions, evaluating research outcomes, and defining future research directions.

II. AI Reshaping the Entire Research Chain: End-to-End Transformation from Literature to Experiment

Full-Chain Research Transformation

2.1 Literature and Knowledge Discovery

During WAIC 2026, a domestically coordinated scientific foundation model 2.0 was officially released, aggregating 8 million high-quality scientific reasoning data points and covering over 200 research tasks. Its "Literature Compass" agent can generate professional-level literature reviews within three hours, achieving 90% evidence attribution accuracy, and has produced nearly 40,000 reviews to date.

2.2 Simulation and Computation

AI is dramatically reducing the time cost of scientific computation. The fluid dynamics software powered by this model can produce simulation results within 10 seconds with error margins under 5%; astronomical spectroscopy agents compress simulation times from hours to minutes. A domestic platform has reduced reaction pathway scanning from several days using traditional DFT to just hours.

2.3 Laboratory Automation

A domestic university team has developed an "Intelligent Scientist" platform integrating AI with automated robotics. Its robotic chemist, guided by Bayesian optimization, identified optimal high-entropy catalysts from 550,000 theoretical compositions within five weeks, using only a limited number of experimental runs. The system has now integrated 110 research robots and 196 intelligent chemical workstations.

MatwingsVenus™ (晓鹜™) employs a multi-agent collaborative architecture capable of parsing natural language instructions, autonomously decomposing tasks, planning experimental workflows, and dynamically adjusting parameters. Once connected to automated equipment, it operates 24/7 unattended, independently completing plasmid construction, protein expression, purification, and assay workflows—establishing a complete dry-wet closed loop of "model design → automated experiment → feedback iteration."

2.4 Full-Chain Research Platforms

During WAIC 2026, a domestic organization launched a scientific discovery platform covering the entire workflow from "hypothesis formulation" to "experimental validation," now deployed across six frontier fields: life sciences, critical materials, semiconductors, nuclear fusion, quantum science, and earth meteorology.

MatwingsVenus™ (晓鹜™) similarly bridges the full chain from design to validation. Its underlying protein models consistently rank among the top performers on the Harvard Medical School ProteinGym leaderboard, delivering one-stop protein design, experimentation, and iteration. According to official information, the platform has completed dozens of successful engineering projects, with over 10 projects already in production capacity, spanning innovative drugs, health and wellness, and the circular economy.

III. AI4S Industrial Deployment: From Laboratory to Production Line

Cross-Industry Applications

3.1 Biopharmaceuticals: AI Accelerates Drug Discovery

Biopharmaceuticals represent the most mature sector for AI4S deployment. AI covers the entire pipeline—protein structure elucidation, antibody design, target screening, and toxicity prediction. While a new drug traditionally takes over a decade from target identification to market, AI has cut early-stage R&D timelines by half.

Matwings Technology, in partnership with Genescience Pharmaceutical, leveraged MatwingsVenus™ (晓鹜™) to achieve an industrial breakthrough in alkali-resistant single-domain antibody design, completing a protein product designed by a domestic AI large language model, scaled up to 5,000-liter industrial production, and commercialized. Within less than a year, the platform increased the alkali resistance of a single-domain antibody by 4-fold. While traditional R&D cycles for similar proteins require 2–5 years, 晓鹜™ compresses this to 2–6 months, reducing experimental samples from tens of thousands to around one hundred, and elevating the screening success rate for candidates meeting preset performance metrics from approximately 5% to over 30%.

3.2 Advanced Materials: AI "Design" Replaces "Trial-and-Error"

The advanced materials field faces broad compositional space and high trial-and-error costs. AI, by learning from existing materials data, precisely extracts "composition–structure–performance" relationships and intelligently screens high-potential formulations. Domestic research teams are building single-atom catalyst databases to enable AI-assisted rational design of catalysts.

3.3 Circular Economy and Carbon Neutrality: AI Empowers Green Manufacturing

Matwings Technology, leveraging MatwingsVenus™ (晓鹜™), has directionally optimized natural PET-degrading enzymes through intelligent amino acid sequence redesign, achieving a 97-fold improvement in degradation efficiency. The high-performance enzyme was developed within just six months, converting plastic waste into renewable feedstocks. Its AI-designed enzymatic lactulose project has entered pilot-scale trials.

3.4 China's Distinctive AI4S Development Pathway

AI4S pathways differ between China and other countries. Overseas approaches tend to prioritize foundational scientific exploration using general-purpose large language models; China pursues a "bottom-up" pathway, with Beijing and Guangdong having issued provincial-level AI4S implementation plans promoting research digitization infrastructure and large-scale deployment of autonomous laboratories.

Matwings Technology, established in Shanghai in 2021, has focused intensively on intelligent protein R&D, with its self-built protein sequence and functional label database selected as a national exemplary high-quality dataset. Its trajectory—from foundational data accumulation to WAIC "Treasure of the Hall" selection—embodies China's AI4S development logic, grounded in real industrial demand and committed to tangible deployment.

IV. Challenges and the Future: The Path Forward for AI4S

Parallel Timelines Before and After AI

AI4S remains in a rapid development phase. Industry consensus holds that while technical capabilities continue to improve, adoption is accelerating, with more research teams and enterprises beginning to deploy AI4S tools at scale.

Key challenges remain: insufficient data standardization—laboratories employ inconsistent data standards, limiting reusability; limited model generalization—specialized models still require optimization in their understanding of general scientific principles; and high computing barriers—GPU resource requirements create entry barriers for smaller teams.

There is broad agreement that AI4S serves as a powerful enabling tool but cannot fully replace wet-lab validation—computational predictions must be complemented by experimental verification, and human-AI collaboration remains the most scientifically sound model. In response, domestic efforts continue to expand public computing infrastructure and shared databases, while lightweight conversational platforms lower adoption barriers and promote AI4S accessibility.

At WAIC 2026, a Turing Award laureate noted that AI will continue to raise the bar for research evaluation standards, and future core competitiveness will shift toward higher-order capabilities such as critical thinking and complex collaboration. Human-machine collaboration will become the dominant mode of scientific discovery.

From computational assistance to autonomous iteration, AI4S has completed its transition from a tool to a research infrastructure. Leveraging domestic AI4S achievements represented by MatwingsVenus™ (晓鹜™), China—grounded in real industrial needs and committed to tangible deployment—has forged a pragmatic and efficient localized pathway, establishing a digital research foundation for new quality productive forces, autonomous biomanufacturing capabilities, and green low-carbon industrial transformation.