During the opening keynote of Google I/O, Demis Hassabis, CEO of Google DeepMind, made a statement intended to create impact: “We are currently at the foothills of the singularity”. The statement refers to the theoretical moment when artificial intelligence would rapidly surpass human intelligence and drastically transform the world. However, the context in which he made this remark opened up a more complex discussion: he was not only talking about hypothetical futures but also about concrete scientific applications like WeatherNext.
Hassabis was wrapping up a segment dedicated to scientific AI, which centered around a video about Google's weather prediction system. WeatherNext was presented as a tool capable of providing early warnings for extreme events, such as the impact of Hurricane Melissa in Jamaica last year. If that software allowed people to evacuate earlier, reinforce their homes, or make better decisions in the face of the storm, it represents a huge advancement, although not necessarily proof that the singularity is near.
The scene highlighted a central tension in the current debate about artificial intelligence. On one hand, there are tools designed to solve specific scientific problems, such as weather prediction, computational biology, or complex data analysis. On the other hand, there is growing expectation for agentic systems based on language models that, one day, could conduct complete research with minimal human intervention.
WeatherNext: the AI that doesn't promise the future but can improve decisions today
WeatherNext represents the most tangible and verifiable side of artificial intelligence applied to science. Unlike grand statements about machines capable of surpassing humans, this type of tool has a concrete function: processing weather data and improving the anticipation of dangerous phenomena. In that sense, its value lies not in announcing an abstract revolution, but in providing better alerts in situations where every hour can save lives.
Weather prediction is one of the fields where AI can have a direct impact. Hurricanes, severe storms, floods, and heatwaves require models that are increasingly fast, accurate, and capable of interpreting enormous volumes of data. If WeatherNext can better anticipate trajectories, intensity, or risk areas, it can help governments, emergency teams, and vulnerable populations prepare with greater lead time.
This type of application showcases a less spectacular but more socially significant artificial intelligence. It does not need to “do science” on its own or replace researchers to produce value: it is sufficient to improve a critical task that already exists. The question, then, is not just how long it will be until superintelligent AI, but what concrete problems current AI can solve more efficiently than traditional methods.
