AI作为科学合作者

From biology to black holes, ChatGPT is accelerating research January 2026 Introduction: Why AI for Science Matters The number of monthlyadvanced science messagesgrew nearly 50%last year Roughly1.3 millionweeklyChatGPT users 8.4 millionaverage weeklymessages on advanced topicsin the hard sciencesand mathematics focus on advanced math &science topics worldwide OpenAI is building tools to help researchers generate insights, accelerate scientific discovery andtranslate those insights into real-world impact. Across ChatGPT, researchers, students, STEM facultyand engineers already use AI to read and synthesize technical literatures, debug and write code, analyzedata, and plan experiments. Each week, ChatGPT sees almost 8.4 million messages on advancedtopics in the sciences and mathematics. These now come from roughly 1.3 million weekly usersworldwide. Only about 0.1 percent of the global population identifies as scientists, according toUNESCO, and yetthey have an outsized impact. Scientific research drives the engine of progress toward a healthier, moreprosperous, and more resilient future. New medicines, new technologies, and new industries comefrom new knowledge put to practical use. A small group of early twentieth-century physicists laid thefoundations of quantum mechanics through abstract research that, decades later, would underpinmuch of the modern digital economy, now measured in the tens of trillions of dollars. Basic research –work done before the payoffis clear – was the source of that knowledge. In 1947, scientists at BellLabs created thefirst working transistor,based on insights from quantum physics. The transistorbecame a building block of computers, phones, and today’s digital technology. The Global PositioningSystem (GPS) relies on Einstein’sinsights into relativityto guide our cars and keep atomic clocksaligned. Yet in many domains, it is getting harder to keep making progress. Economists and research analystspoint to falling “research productivity,” meaning more people, time, and money are required to producethe same number of insights. Semiconductors offer a well-known example: sustaining Moore’s Law todouble the number of transistors in an integrated circuit every two years has required a dramaticincrease in effort, with the number of researchers needed today estimated at more than18 timeswhatwasneeded in the early 1970s.As knowledge grows more complex,each new generation ofresearchers faces a heavier burden just to reach the frontier, which lengthens their training time andnarrows their specializations. Institutionally, research has shifted toward larger teams, with growingoverhead for grant proposals, compliance, reporting, and coordination costs. In medicine, scientific advances have saved countless lives. Worldwide life expectancy rosefromroughly 32 years in 1900 to about 73 years in 2023 (and to more than 78 years in the United States).Butthe remaining burden of disease is heavy.But the World Health Organization reports thatnoncommunicable diseases such as stroke, heart disease, cancer, and diabetes still account for about74% of global deaths. Even when progress is rapid, turning new ideas into available treatments takestime. On average, it takes10-15 yearsfrom target discovery to regulatory approval of a new drug in theUnited States, a lag imposed on patients who need new and better treatments. Makingprogress faster will save lives and improve them.AI is already helping to address thebottlenecks that slow science down. Modern research is fragmented across disciplines and constrainedby limits that are both cognitive and logistical: reading and digesting enormous literatures to determinewhat is known, translating ideas into mathematics and code, setting up analyses and simulations,checking calculations, searching huge design spaces, and deciding which future experiments are themost promising. Used well, AI can serve as a high-throughput partner for thought, computation, andstructured reasoning, shortening the cycle from hypothesis to test and increasing the capacity ofresearchers working alone and in teams, even across disciplinary barriers. Kevin Weil, VP of OpenAI for Science, describes the opportunity this way:“AI is increasingly being usedas a scientific collaborator, and we’re seeing its impact grow in real research settings. More researchersare using advanced reasoning systems to make progress on open problems, interpret complex data,and iterate faster in experimental work. That usage has been growing quickly over the past year, and theresults are starting to show up acrossfields. We’re still early, but the pace of adoption and the quality ofthe work suggest science is entering a new acceleration phase.” OpenAI is proud to work with research partners across government agencies, national laboratories,academia, and medicine, including the U.S. Department of Energy, Lawrence Livermore NationalLaboratory, the U.S. Centers for Disease Control and Prevention, Harvard Univers