Level 1 — Absolute Beginner
Google DeepMind built a new AI tool called Co-Scientist. It helps scientists do research faster.
This week, DeepMind published a paper about Co-Scientist in Nature, a famous science magazine.
Co-Scientist is not just one AI. It is a team of AI helpers. They work together. They think of ideas, test them, and make the ideas better.
Scientists used Co-Scientist to look for new medicines. It found a medicine that already exists but could help a different illness. In lab tests, this medicine stopped 91 percent of a harmful body reaction.
- AI (artificial intelligence)
- computer systems that can perform tasks that normally need human thinking
- multi-agent
- made of several AI helpers working together
- hypothesis
- an idea that can be tested by science
- publish (research)
- to share scientific findings so others can read them
- leukaemia
- a type of cancer that affects the blood
- candidate (medicine)
- something being considered as a possible medicine
- lab test
- an experiment done in a science laboratory
- scarring (medical)
- the process of forming tough tissue after injury or disease
Level 2 — Elementary
Google DeepMind published research this week in the journal Nature describing Co-Scientist, a multi-agent AI system built on its Gemini models that generates, debates and refines scientific hypotheses.
Rather than functioning as a single program, Co-Scientist orchestrates a coalition of specialized AI agents that propose ideas, challenge one another, and rank hypotheses through an ongoing tournament-style process, injecting fresh knowledge from scientific literature along the way.
Researchers used the system to search for potential drug candidates to treat acute myeloid leukaemia. A separate application helped scientist Gary Peltz search for liver fibrosis treatments, surfacing overlooked drug-repurposing candidates that human researchers had not prioritized.
One candidate the system identified successfully blocked more than 91 percent of a scarring-related response in laboratory tests, illustrating how AI-driven hypothesis generation can accelerate early-stage drug discovery.
- orchestrate
- to organize and coordinate multiple elements toward a goal
- coalition
- a group of separate parts working together
- specialized (agent)
- designed for a particular narrow purpose
- tournament-style
- organized as a competition with rounds and rankings
- literature (scientific)
- published research papers and studies
- fibrosis
- the thickening or scarring of tissue
- repurposing (drug)
- using an existing medicine to treat a different condition
- accelerate
- to make something happen faster
Level 3 — Intermediate
Google DeepMind published research this week in Nature detailing Co-Scientist, a multi-agent AI system built atop its Gemini models that generates, debates and iteratively refines novel scientific hypotheses against existing literature and structured databases.
Rather than functioning as a monolithic program, Co-Scientist orchestrates a coalition of specialized agents that propose candidate hypotheses, critique one another's reasoning, and rank ideas through an Elo-based tournament mechanism, continuously injecting fresh knowledge to expand the system's exploration of the hypothesis space.
Researchers deployed the system to search for potential drug candidates against acute myeloid leukaemia, while a separate application assisted scientist Gary Peltz's search for liver fibrosis treatments, surfacing drug-repurposing candidates that conventional human-led screening approaches had overlooked.
Among these, one candidate successfully suppressed more than 91 percent of a scarring-associated response in laboratory assays, a result researchers say illustrates how AI-augmented hypothesis generation can meaningfully compress the timeline of early-stage drug discovery without displacing the wet-lab validation that remains essential.
- monolithic
- consisting of a single, undifferentiated whole
- iteratively
- through repeated cycles of refinement
- critique (verb)
- to evaluate and point out flaws in something
- Elo-based
- using a ranking system originally developed for chess competition
- exploration (research)
- the systematic search across a range of possibilities
- assay
- a laboratory test to measure a substance's effect or presence
- augmented
- enhanced or supplemented, often by technology
- wet-lab
- laboratory work involving physical chemicals and biological samples
Level 4 — Advanced
Google DeepMind published research this week in Nature detailing Co-Scientist, a multi-agent AI system constructed atop its Gemini models that generates, contests and iteratively refines novel scientific hypotheses against extant literature and structured databases.
Rather than operating as a monolithic program, Co-Scientist orchestrates a coalition of specialized agents that propose candidate hypotheses, interrogate one another's reasoning, and adjudicate rankings through an Elo-based tournament mechanism, continuously assimilating fresh knowledge to broaden the system's traversal of the hypothesis space.
Researchers deployed the system in pursuit of potential drug candidates against acute myeloid leukaemia, while a parallel application assisted scientist Gary Peltz's search for liver-fibrosis treatments, surfacing drug-repurposing candidates that conventional, human-led screening paradigms had conspicuously overlooked.
Among these, one candidate suppressed upward of 91 percent of a scarring-associated response in laboratory assays, a result researchers contend illustrates how AI-augmented hypothesis generation can meaningfully compress the timeline of early-stage drug discovery without supplanting the wet-lab validation that remains indispensable to the scientific method.
- extant
- still existing; currently in existence
- contests (verb)
- disputes or challenges
- interrogate (figurative)
- to examine closely and critically
- adjudicate
- to make an official decision or judgment
- assimilating
- absorbing and incorporating into an existing structure
- traversal
- the act of moving across or through something
- paradigm
- a typical pattern or model of how something is done
- supplanting
- replacing or taking the place of
Talk about this article with KatePractice this story's vocabulary in a real spoken conversation with your AI tutor