The most powerful AI stories right now are not chatbots. They are the quiet algorithms reading burnt Roman scrolls, trawling through millions of galaxies, and finding things hidden in data no human team could ever finish searching.

In the realm of artificial intelligence, much of the public fascination centers around chatbots capable of generating essays, responding to inquiries, and engaging in conversations. However, a different breed of AI is quietly revolutionizing scientific research by tackling complex problems that are beyond human capacity. This AI operates in specialized domains, analyzing vast datasets to uncover insights that would otherwise remain hidden. From deciphering ancient texts to identifying astronomical phenomena, these algorithms are reshaping how researchers approach their work.

The Vesuvius Challenge: Unveiling Ancient Knowledge

One remarkable example of this AI-driven innovation is the Vesuvius Challenge, initiated in March 2023 by Nat Friedman, Daniel Gross, and computer scientist Brent Seales. The challenge aims to read the charred scrolls from Herculaneum, a Roman town buried by the eruption of Mount Vesuvius in 79 CE. The library of a villa in Herculaneum contains over 1,800 surviving papyri, many of which are carbonized and too fragile to unroll. Utilizing high-resolution X-ray scans from a synchrotron, the challenge employs machine-learning models to detect faint traces of carbon ink against the carbonized papyrus.

In October 2023, a contestant successfully read the first word, which was the Greek term for purple. By February 2024, a team managed to recover over 2,000 characters from one scroll, believed to be an Epicurean text by the philosopher Philodemus, discussing themes of pleasure, music, and food. By May 2025, the title and author of another scroll, PHerc. 172 from Oxford’s Bodleian Libraries, were identified as Philodemus’ On Vices, marking a significant milestone as it was the first time the title of one of these scrolls had been read. Importantly, while the algorithm identifies potential areas of interest, it is the papyrologists who confirm the letters and contextual meaning, illustrating the collaborative nature of this research.

Advancements in Astronomy: The Role of AI in Cosmic Discoveries

Similarly, in the field of astronomy, AI is proving invaluable in managing the overwhelming volume of data generated by space missions. The European Space Agency's Euclid mission, which aims to study dark matter and cosmology, released its first batch of data in March 2025. Deep-learning models were employed to rank approximately one million galaxies in a small section of the sky, a mere fraction of the planned survey. Following this, around 1,800 citizen scientists and 61 professional astronomers worked together to vet the top candidates, resulting in a catalogue of 497 galaxy-galaxy strong lens candidates within just six weeks of data analysis.

A separate initiative utilized similar AI tools to sift through the Hubble Space Telescope's archive, examining 99.6 million image cutouts and identifying nearly 1,400 anomalous objects, with over 800 of these previously undocumented in scientific literature. This effort, reported in Astronomy & Astrophysics in December 2025, revealed 138 new candidate gravitational lenses, along with jellyfish galaxies and numerous mergers or interacting galaxies. Again, the pattern remains consistent: the AI sorts through the data, while human experts validate the findings.

The Value of Narrow AI in Scientific Discovery

The scientific community has begun to recognize the significance of these narrow AI systems as essential tools for discovery rather than mere software demonstrations. The 2024 Nobel Prize in Chemistry was awarded in part to David Baker for his work in computational protein design and to Demis Hassabis and John Jumper of Google DeepMind for AlphaFold, an AI system that predicts protein structures from amino acid sequences. AlphaFold has successfully predicted structures for around 200 million proteins, nearly encompassing the entire catalog of proteins known to researchers. This recognition underscores the value placed on AI that addresses specific scientific challenges rather than general-purpose applications like chatbots.

These AI systems are not designed to possess general intelligence; instead, they excel at processing specific types of labeled data, such as identifying ink on papyrus or distinguishing between lensed and unlensed galaxies. Their primary function is to act as filters, narrowing down vast datasets to highlight the most promising areas for further investigation. It is crucial to understand that the results produced by these algorithms are often candidates for further study rather than definitive answers. For instance, a lens candidate requires additional spectroscopic follow-up, a reconstructed scroll passage necessitates verification by a papyrologist, and a predicted protein structure comes with a confidence estimate rather than an absolute guarantee.

The distinction between these narrow AI systems and chatbots is significant. While chatbots generate coherent text and can present incorrect information in a convincing manner, lens-finders and ink-detectors are focused on ranking data to direct human attention where it is most likely to yield results. As the volume of data collected continues to grow, the gap between available human resources and the need for analysis widens. Future data releases from the Euclid mission, ongoing scans of Herculaneum scrolls, and the anticipated outputs from the Vera C. Rubin Observatory will further emphasize the necessity of ranking models to manage the influx of information.

In conclusion, while chatbots may dominate headlines, it is the specialized AI systems that are fundamentally transforming scientific discovery. By efficiently processing and ranking vast datasets, these algorithms are enabling researchers to uncover new knowledge, one ranked list at a time.