Google DeepMind and MIT jointly develop AI agent CoDaS: capable of autonomous scientific research, writing a paper in just 8 hours

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1840 words

AI is not just for chatting anymore; it can now conduct research and write papers on its own! "CoDaS," an AI scientist jointly developed by Google DeepMind and MIT, has recently shocked the academic world. It can autonomously analyze wearable device data from thousands of people, not only identifying "late-night doomscrolling" as a potential indicator of depression but also verifying findings and writing scientific papers independently. Research that would typically take experts over a month to complete can be finished by CoDaS in just 6 to 8 hours. (Previous coverage: He Yi's speech: Using AI to boost efficiency by 10x, we aim to serve 3 billion users globally) (Background supplement: An overlooked open-source AI tool warned of the $292 million Kelp DAO vulnerability 12 days ago) With the rapid advancement of artificial intelligence technology, the role of AI is evolving from a mere "assistive tool" to an independent "scientific researcher." Recently, a groundbreaking study published jointly by Google Research, Google DeepMind, and MIT introduced a multi-agent AI system called CoDaS (AI Co-Data-Scientist), which has successfully achieved a fully autonomous scientific discovery process. Prominent tech community thought leaders Wes Roth and Samuel Schmidgall also actively shared this highly breakthrough academic achievement on the X platform. A joint team from Google Research, Google DeepMind, and MIT has introduced CoDaS, a multi-agent AI system designed to autonomously run the entire biomarker discovery lifecycle from analyzing raw wearable sensor data and generating hypotheses to conducting statistical analysis and… https://t.co/KLgxFT4OSq pic.twitter.com/4ursWqeo7l — Wes Roth (@WesRoth) April 20, 2026 CoDaS is a system specifically designed to autonomously discover health biomarkers from raw "wearable sensors" data. Its workflow covers hypothesis generation, statistical analysis, adversarial validation, and literature-based reasoning, ultimately producing a complete draft of a scientific paper. In tests, the research team fed CoDaS a large wearable dataset covering nearly 10,000 participants (including sleep, activity, heart rate, and phone usage habits). Without any human prompts, the AI discovered multiple meaningful health features, the most notable of which is a mental health indicator: The AI discovered that the behavior of excessively browsing social media or negative news at night is significantly positively correlated with the severity of depression (correlation coefficient ρ = 0.177, p < 0.001, sample size n = 7,497). Remarkably, the AI even autonomously named this behavior "late-night doomscrolling." In addition to mental health, it also successfully identified a negative correlation between the ratio of daily steps to resting heart rate and metabolic diseases (insulin resistance). To prevent AI from generating common "scientific hallucinations" or making meaningless statistical inferences, CoDaS has a built-in robust Adversarial Validation mechanism. For example, when searching for metabolic health features, the system once proposed using the "square of glucose" to predict insulin resistance. Although this formula appeared highly correlated statistically, the CoDaS validation mechanism immediately detected it as a scientifically meaningless "tautology" and decisively rejected the feature. This mechanism significantly enhances the scientific reliability and clinical potential of AI outputs. The work efficiency and output quality of CoDaS have completely overturned traditional research models. According to the paper's data, a massive data analysis and writing task that would originally take human experts 37 person-days can be completed by CoDaS in just 6 to 8 hours. Even more convincingly, in blind reviews by domain experts: - Papers generated by CoDaS achieved an 86% "non-rejection rate" (i.e., accepted, minor revisions, or major revisions). - In contrast, the rejection rate for papers from other benchmark AI scientific agents was as high as 85% to 100%. This study demonstrates how multi-agent AI systems can efficiently transform passive consumer-grade wearable data into clinically valuable insights. As a representative advancement of "Agentic AI" in the field of digital health, CoDaS signals that a new era of scientific discovery led by both humans and AI may have already arrived.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-04-28

Google DeepMind Veteran Raises $1.1 Billion to Build AI That Isn’t Trained With Human Data

Similarity 130%關鍵字 google/deepmind

2026-04-26

Coachella Uses Google DeepMind AI to Test the Future of Live Entertainment

Similarity 130%關鍵字 google/deepmind

2026-04-30

Stripe releases 288 updates: Google AI shopping, Link proxy wallets, and Tempo on-chain stablecoin payments move forward in parallel

Similarity 120%關鍵字 google同分類 zh

2026-04-30