Guide
Mysteries AI helped solve: from cold cases to ancient scrolls
AI isn't only for chatbots. It has helped catch suspects who were on the run for 19 years, read scrolls nobody could open, and find a lost play hidden in a library. Here are real cases, what the AI actually did, and where people still had the final say. Filter by type to explore.
- AI made the breakthrough
- Without the AI, this probably would not have been solved yet.
- AI helped experts
- The AI found leads fast; people checked them and made the final call.
Crime & lost faces
AI made the breakthrough2025
A 19-year-old triple murder in Kerala
The mystery: In 2006 a young mother and her 17-day-old twins were killed in Anchal, Kerala. The two suspects, both army men, vanished and lived for years under new names.
What AI did: Kerala Police used AI to "age" the suspects' 2006 photos, trying different hairstyles and features, then compared the results with photos on social media. One matched a Facebook wedding photo, which led the CBI to the pair in Puducherry. They were arrested in January 2025.
Keep in mind: Police stress that AI can make mistakes: the match was only a lead, checked in other ways before any arrest.
AI helped experts2022
Putting names to faces in Holocaust photos
The mystery: Archives hold hundreds of thousands of Holocaust-era photos of people nobody can name anymore.
What AI did: From Numbers to Names, a free face-matching tool built by Google engineer Daniel Patt, compares a family photo with faces across these archives and suggests possible matches. Rush singer Geddy Lee used it to find a photo of his mother at the Bergen-Belsen camp, which led him to more photos of her family.
Keep in mind: The tool only suggests look-alikes with a similarity score; families and historians decide whether it is really the person.
AI helped experts2018
Who are the unknown Civil War soldiers?
The mystery: Most surviving portraits of American Civil War soldiers have no name attached, often only a uniform and a face.
What AI did: Civil War Photo Sleuth, from Virginia Tech, combines face recognition with clues people tag, such as uniform details, and compares a photo against thousands of identified portraits. In its first month, users linked over 100 unknown photos to named soldiers.
Ancient texts & places
AI made the breakthrough2025
The title of a scroll no one could open
The mystery: Which work, and by whom, was hidden inside another sealed Herculaneum scroll?
What AI did: Using the same X-ray and AI approach, researchers found the title at the very center of the roll: "On Vices" by the philosopher Philodemus. It was the first time a title had been read from a still-closed scroll.
AI helped experts2025
How old are the Dead Sea Scrolls, really?
The mystery: Most of the scrolls carry no date, and scholars dated them mainly by the style of the handwriting.
What AI did: An AI model called Enoch learned handwriting styles from scrolls with known radiocarbon dates, then estimated dates for others. Many came out older than thought, and two biblical fragments may date to the lifetime of their presumed authors.
Keep in mind: Experts judged most of the estimates realistic, but they are estimates, not proof.
AI helped experts2025
Filling the gaps in Roman inscriptions
The mystery: Thousands of Latin inscriptions on stone are broken or worn, and their date and origin are unknown.
What AI did: Google DeepMind's Aeneas suggests missing words, dates a text to within a few decades and guesses which Roman province it came from, by finding similar inscriptions among almost 150,000 others. It follows Ithaca, a similar tool for ancient Greek.
AI made the breakthrough2024
Reading scrolls burned by Vesuvius
The mystery: Hundreds of papyrus scrolls were carbonized by the eruption that buried Herculaneum in 79 AD. Unrolling them destroys them, so for 270 years nobody could read them.
What AI did: Scientists X-rayed the rolled scrolls in 3D. Then AI models trained to spot faint traces of ink found the letters inside. The first word, "purple", came out in 2023; by early 2024 a team of three students had read over 2,000 characters of a lost text about pleasure.
AI made the breakthrough2024
Hundreds of hidden Nazca Lines
The mystery: Giant figures drawn in the Peruvian desert 2,000 years ago are hard to see from the ground. It took about a century to find around 430 of them.
What AI did: A Yamagata University team, working with IBM, trained AI to flag likely figures in aerial images. Archaeologists then checked the spots on foot and confirmed 303 new geoglyphs in about six months, nearly doubling the known number.
Keep in mind: Hundreds of AI-flagged spots have not been checked yet.
AI helped experts2023
Missing pieces of the Epic of Gilgamesh
The mystery: The world's oldest epic survives on clay tablets broken into thousands of fragments scattered across museums. Matching them by hand takes a lifetime.
What AI did: The Fragmentarium project at LMU Munich uses machine learning to suggest which fragments belong together. It has turned up new pieces of Gilgamesh, a lost hymn to the city of Babylon, and hundreds of missing lines from other texts.
Art & literature
AI made the breakthrough2023
A lost play by Lope de Vega
The mystery: A handwritten comedy had sat in Spain's National Library since 1882, catalogued as anonymous.
What AI did: Researchers used AI to transcribe over a thousand old Spanish play manuscripts, then compared word habits against hundreds of playwrights. "La francesa Laura" matched Lope de Vega, one of Spain's greatest writers. Experts then confirmed it, and the play has since been staged in Madrid.
AI helped experts2023
A face Raphael did not paint
The mystery: Art historians had long debated whether the Renaissance master painted all of the "Madonna della Rosa".
What AI did: An AI trained to recognize Raphael's brushwork, color and shading checked each figure separately. The Madonna, Jesus and John came out as Raphael's; the face of Joseph most likely did not.
Keep in mind: The AI cannot say who painted Joseph. One of Raphael's pupils is a guess, nothing more.
AI helped experts2019
Who really wrote Shakespeare's Henry VIII?
The mystery: Since the 1850s, scholars suspected Shakespeare co-wrote Henry VIII with John Fletcher, but disagreed on who wrote what.
What AI did: A machine-learning model trained on each writer's word and rhythm habits went through the play line by line. It found that Fletcher wrote roughly half, and that the switch between writers sometimes happens in the middle of a scene.
Keep in mind: It mostly confirmed what one scholar had proposed in 1850, with finer detail.
Science & nature
AI helped experts2024
Do sperm whales have an alphabet?
The mystery: Sperm whales talk in bursts of clicks called codas. Researchers had sorted them into types, but did not know if the clicks had any deeper structure.
What AI did: Project CETI used AI and statistics on nearly 9,000 codas recorded over a decade. They found the clicks vary in rhythm, tempo and small extras that whales combine, a bit like sounds in a "phonetic alphabet".
Keep in mind: Still open: nobody knows yet what the whales are saying.
AI made the breakthrough2020
The 50-year protein folding puzzle
The mystery: Proteins fold into 3D shapes that decide what they do. Working out one shape in the lab could take years, and predicting it from the recipe alone had defeated scientists since the 1970s.
What AI did: Google DeepMind's AlphaFold2 predicts a protein's shape in minutes, close to lab accuracy. It has since predicted the shapes of nearly all 200 million known proteins, and its creators shared the 2024 Nobel Prize in Chemistry.
AI made the breakthrough2020
A new antibiotic hiding in plain sight
The mystery: Bacteria are becoming resistant to our drugs, and few truly new antibiotics have been found in decades.
What AI did: MIT researchers trained an AI on about 2,500 molecules, then had it screen about 6,000 existing compounds. It picked one once studied as a diabetes drug that kills many drug-resistant bacteria in new ways. They named it halicin, after HAL from "2001: A Space Odyssey".
Keep in mind: Halicin worked in lab and mouse tests; it is not an approved medicine.
How we chose these cases
A case makes the list only if a reliable source (the researchers, a journal, a museum or a trusted news outlet) confirms that AI played a real part. Headlines often say “AI solved it” when the real hero was DNA testing or old-fashioned detective work, so we leave those out. Each card says whether the AI made the breakthrough itself or helped experts get there faster. Know a case we’ve missed? Tell us.