Google says Circle to Search can now solve physics and math problems, and will expand later this year to diagrams and graphs, powered by LearnLM models
Context & Ripple Effects
Circle to Search began as an Android gesture layer for searching from any screen, initially on select Samsung and Pixel phones through its Galaxy S24 and Pixel launch. This update makes that interface a vehicle for educational problem-solving rather than only visual lookup.
It also extends Google’s longer effort to improve AI performance on quantitative tasks, from Bard’s math and logic upgrade to more specialized DeepMind reasoning work. The important shift is delivery: those capabilities are being placed inside a consumer search interaction.
First-order effects
- Circle to Search users can submit physics and math problems through the existing gesture workflow, with LearnLM supplying the new answer capability.
- Google gains a clearer education-oriented use case for Circle to Search, while planned diagram and graph support broadens the kinds of on-screen questions it can handle later in the year.
Second-order effects
- Android device partners that market Circle to Search gain a more tangible AI utility, increasing pressure on rival mobile assistants and search tools to support multimodal academic queries.
- As more queries are answered in-context, the feature can shift some homework and reference activity away from manually opening a browser search result, an example of Google’s parallel investment in math-reasoning models.
Third-order effects
- If Google can reliably expand from text problems to diagrams and graphs, mobile search may increasingly become an answer layer embedded over whatever a user is viewing rather than a destination for links.
- That raises the value of domain-specific models and evaluation in consumer AI: educational usefulness will depend on accuracy and explanation quality, not merely on adding an AI label to search.
The trend: Consumer search interfaces are being remade into multimodal, task-specific answer engines powered by specialized AI models.