Saturday Citations: Human impacts on reef systems; testing AI systems; a woman with perfect memory

Chris Packham
staff contributor

Gaby Clark
scientific editor

Robert Egan
associate editor

The week in science: UK fishermen are reporting a massive octopus bloom in the waters off southwest England. Researchers found a massive fossilized pearl in the Australian outback, the largest ever found in the country. And physicists detailed an extremely high-efficiency heat engine that challenges two centuries of thermodynamic understanding. (It's reported in 鶹Ժical Review Letters, so it's probably legit.)
Additionally, a popular oceanic reserve rebounded in the absence of humans during the COVID-19 pandemic of 2020. Researchers developed a metric to characterize the ability of AI systems to uncover domain systems from sequence predictions. And neurologists reported on a young woman with extraordinary memory recall:
Humans disruptive
Evidence from the Hanuama Bay Nature Preserve in Hawai'i demonstrates the impact of humans on oceanic reef systems. The preserve, a wildly popular tourist destination, closed completely during the 2020 COVID-19 pandemic. Researchers at the University of Hawaiʻi at Mānoa Hawaiʻi Institute of Marine Biology took advantage of the closure to study changes in water quality, seal sightings, fish abundance and sea life behavior. In the absence of human visitors, the physical and biological health of the site quickly returned to normal levels, including densities of fish populations and reef health.
Elizabeth Madin, lead author, said, "The ecosystem responded in remarkable ways. These kinds of changes happened quickly, suggesting that everyday human presence can have a real and measurable impact on reef health. It's a powerful reminder of just how sensitive and responsive coral reef ecosystems are to our activity."
Standardized testing for AI
Artificial intelligence systems can now make specific predictions that resemble Johannes Kepler's laws of motion. But researchers at MIT wanted to know whether current systems were capable of a deeper understanding of these laws of motion as represented by Isaac Newton's laws of gravity. In other words, can foundation models that make good sequence predictions actually uncover deeper domain understanding? If so, they would be able to generalize that understanding to new kinds of problems.
The goal of wasn't to determine the capabilities of AI systems; rather, the researchers sought to develop a metric for their predictive power that could be applied to any model. They developed a metric called the measurement inductive bias, describing a tendency or a bias toward responses that reflect reality based on inferences developed from a large corpus of specific data. They used a lattice model: In a one-dimensional lattice, an object can only move along a line.
Predictive models proved capable of reconstructing the world of a one-dimensional lattice; adding dimensions to the lattice increased the complexity. The models continued to perform well with two- and three-state lattices but began diverging from real-world models as the number of states increased.
Total recall
Autobiographical memory is the ability to remember life experiences serially from the present back to childhood. It includes sensory and emotional memories of people, places and events. Most people have some combination of high- and low-vividness memories depending on their importance or how long ago the related events occurred.
However, a small number of people have a condition called hyperthymesia, in which they remember everything that ever happened to them, indexed by date. Hyperthymesics are able to travel through time in their minds fluidly, with high recall, and can relate the exact events that occurred on specific days even decades later.
Researchers at the Paris Brain Institute , a 17-year-old girl called TL, who exhibits significant control over how she accesses memories. TL distinguishes her memories into two types: "black memory," corresponding to the kinds of encyclopedic knowledge accumulated through study; and a sophisticated memory palace she calls the "white room" in which she can visualize any autobiographical detail from her past.
Some of her memories are even stored in the form of text messages and photographs. Using two clinical tests—the Episodic Test of Autobiographical Memory and the Temporal Extended Autobiographical Memory Task—they determined that she remembers her life with extraordinary intensity.
Neurologist Laurent Cohen says, "Autobiographical hyperthymesia also seems closely linked to synesthesia, a neurological condition in which processing one sensory modality involves at least two senses. For example, synesthetes may hear colors, see sounds, or taste music. Even though TL is not a synesthete, several members of her family are. It would be interesting to explore this association."
Written for you by our author , edited by , and fact-checked and reviewed by —this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive. If this reporting matters to you, please consider a (especially monthly). You'll get an ad-free account as a thank-you.
Journal information: 鶹Ժical Review Letters
© 2025 Science X Network