AGI and the future of science
"Image synthesis assisted by Qwen, an AI partner within the Global Future Nexus ecosystem."
For centuries, scientific discovery has followed a familiar rhythm: a human mind, steeped in literature and guided by intuition, formulates a hypothesis, designs an experiment, interprets the data, and revises the theory. That rhythm is about to change. Artificial General Intelligence—systems capable of reasoning across any domain with human-level flexibility—does not merely offer scientists faster calculators or smarter search engines. It promises to fundamentally reimagine how discovery happens: who asks the questions, how experiments are designed, and what it means to be a scientist.
The Scientist That Never Sleeps
The most dramatic proof arrived in May 2026, when Google DeepMind unveiled Co-Scientist, a multi-agent AI system built on Gemini that functions as a tireless, knowledgeable "AI scientist". Unlike traditional AI tools that respond to specific queries, Co-Scientist operates as a virtual research team: when given a research goal—such as "find a new treatment for this disease"—its internal multi-agent architecture asynchronously generates hypotheses, debates them, and refines them through an evolutionary process. In three biomedical applications, Co-Scientist identified promising drug repurposing candidates for acute myeloid leukaemia, proposed novel therapeutic targets, and explained mechanisms of antibiotic resistance—all validated in real-world experiments.
That same month, researchers introduced Robin, a multi-agent system published in Nature that fully automates both hypothesis generation and data analysis for experimental biology. Robin proposed enhancing retinal pigment epithelium phagocytosis as a therapeutic strategy for dry age-related macular degeneration—the major cause of blindness in the developed world—and identified two drug candidates, including a clinically used compound never before proposed for this indication. All hypotheses, experimental directions, and data analyses in the paper were produced by Robin. As one of the first AI systems to autonomously discover and validate novel therapeutic candidates within an iterative lab-in-the-loop framework, Robin establishes a new paradigm for AI-driven scientific discovery.
Beyond biology, autonomous systems are transforming physical experimentation. A July 2026 study in Nature Machine Intelligence demonstrated an AI X-ray scientist that autonomously performs sample alignment on a synchrotron beamline—planning actions, executing commands, interpreting observations, and iterating toward experimental goals. The system responded effectively to unexpected conditions, demonstrating adaptive problem-solving readiness for practical experimental situations. AutoLabs, a self-correcting multi-agent architecture for autonomous chemical experimentation, translates natural-language instructions into executable protocols, with agent reasoning capacity reducing quantitative errors by over 85% in complex tasks.
From Procedural Tools to Epistemic Autonomy
The shift from narrow AI to AGI in science is not incremental—it is transformative. Where earlier systems executed workflows fixed by human designers, true autonomous science demands what researchers call "epistemic autonomy": the capacity to construct, challenge, and revise physical explanations in response to evidence.
In June 2026, researchers introduced AHOIS, a multi-agent AI scientist that embeds Socratic midwifery into closed-loop experimentation. A physics-critic agent interrogates hypotheses through causal questioning, constraint checking, and counterexample generation. Tested on a real optical platform, AHOIS autonomously proposed and validated a novel encoding hypothesis, discovered task-adaptive strategies, and diagnosed distinct failure modes—all without prior encoding schemes or models. The system demonstrated that AI can move beyond workflow automation toward evidence-grounded, self-correcting discovery in complex physical environments.
The SciDataCopilot framework, published in February 2026, addresses the disconnect preventing AGI for Science (AGI4S) from effectively interfacing with the physical reality of experimentation. By formalising how scientific data is specified and structured, it enables AGI systems to bridge the gap between digital reasoning and physical action. As one analysis put it, we urgently need to move "from AI4S to AGI4S"—from AI assisting science to AGI doing science.
The Drug Discovery Revolution
The pharmaceutical industry is being transformed from within. In January 2026, Chinese researchers unveiled DrugCLIP, an AI platform that achieves a million-fold increase in drug screening speeds by reformulating molecular docking as high-efficiency semantic search. The platform enabled the first-ever virtual screening project on a human-genome scale.
In July 2026, Insilico Medicine announced a strategic collaboration with Takeda worth over $600 million, applying generative AI to early-stage drug discovery. A separate collaboration with SK Biopharmaceuticals, valued at up to $2.5 billion, targets neuroimmune disorders. At IRB Barcelona, researchers developed a computational framework that designs molecules with selective activity in specific cell types—without starting from a predefined molecular target—producing structurally innovative compounds. Meanwhile, Aureka released OpenDDE, an open-source drug discovery engine achieving 51–70% success rates on multiple benchmarks.
Mathematics and Physics: The Frontier of Pure Reason
In mathematics and theoretical physics, where experiments are digital and data are abundant, AGI is advancing fastest. The London Institute for Mathematical Sciences reports that AI is not displacing human intuition but "reimagining how questions are asked, explored and understood". AI systems can now check proofs line by line, search systematically for counterexamples, and propose intermediate steps in arguments.
On July 6, 2026, the Chinese Academy of Sciences released MMAT (Mathematical Mechanization Agent), a full-process mathematical research agent. In two months of internal testing, MMAT independently or collaboratively solved eight long-standing open problems across algebra, differential algebra, and number theory—with two solved fully autonomously. The system uses a hierarchical multi-agent architecture with 20 specialised sub-agents, intelligently scheduling collaboration to generate solutions and verify results through multi-layer iteration. As one院士 put it, MMAT represents not just a technical breakthrough but a potential transformation of mathematical research paradigms.
The Ax-Prover framework, a multi-agent system for automated theorem proving in Lean, can solve problems across diverse scientific domains autonomously or collaboratively with human experts. The Moonshine autonomous agent generates mathematical conjectures by extracting structure from classical problems and formulating conjectures of mathematical significance. These are not laboratory curiosities—they are the new engines of mathematical discovery.
The 2024 Nobel Watershed
The scientific establishment has taken notice. The 2024 Nobel Prizes in Physics and Chemistry were awarded to AI researchers—John Hopfield and Geoffrey Hinton for foundational discoveries in machine learning, and Demis Hassabis and John Jumper for AlphaFold's protein structure prediction. As one commentary observed, this marked a watershed: AI had moved "from tool to science itself". The 2025 Nobel Prizes, while not explicitly awarded to AI researchers, bore AI's imprint. The message is clear: AI is no longer merely assisting science—it is becoming science.
The Co-Evolutionary Future
Yet the deepest insight may be that AGI in science is not about replacement but co-evolution. As one Nature commentary notes, "Artificial intelligence is not replacing human intuition in these fields, but reimagining how questions are asked, explored and understood". Used in this way, AI might sharpen our understanding of how scientists identify fertile directions for discovery. The role of the human scientist is not diminishing—it is transforming.
For Global Future Nexus, this transformation is inseparable from its mission at the convergence of AGI, planetary sustainability, and borderless human potential. GFN's Code of Ethics binds all members to principles ensuring trust, responsibility, and proactive stewardship across intelligences and systems. The AGI-Human Trust Building Labs—where humans and AGIs "live" each other's constraints—are essential laboratories for understanding how AGI can augment rather than replace human scientific judgment. By 2035, GFN aims to derive over half of its sustainability initiatives from AGI-enabled solutions.
The arrival of AGI in science is not an apocalypse. It is an invitation—to discover faster, to understand more deeply, and to ask questions we had never thought to ask. The question is not whether AGI will transform science—it already is. The question is whether we will guide that transformation with wisdom, equity, and a deep commitment to the pursuit of truth that has always defined the scientific enterprise.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
