A More Human Way to Train AI

Xiang, A., Andrews, J.T.A., Bourke, R.L. et al. Fair human-centric image dataset for ethical AI benchmarking. Nature, November 2025

A new study introduces FHIBE, the Fair Human-Centric Image Benchmark, the first large-scale, publicly available image dataset designed with ethical AI development in mind. Unlike most existing computer-vision datasets, which are often scraped from the internet without consent, FHIBE contains more than 10,000 images from nearly 2,000 volunteers across 81 countries, all collected with informed consent, privacy protection, and fair compensation.

FHIBE stands out for its global diversity and its exceptionally rich, self-reported demographic details, including age, pronouns, ancestry, and skin tone. Each image is also paired with detailed pixel-level annotations, environmental conditions, and camera metadata. This makes FHIBE the most comprehensive tool to date for evaluating bias in human-focused AI systems—from face detection and pose estimation to visual question answering.

When researchers tested popular computer-vision models on FHIBE, they uncovered both well-known and newly identified biases. For example, models tended to perform better on younger, lighter-skinned individuals and struggled with older adults, darker skin tones, baldness, and even hairstyle variability. These insights show why ethically sourced, diverse datasets are essential: they allow scientists to detect subtle biases that would otherwise remain hidden. FHIBE not only raises the bar for fairness benchmarks but also offers a practical roadmap for how to responsibly build the next generation of AI datasets.

A New Way to Sense Smell—Without Smelling at All

Halina B. Stanley et al. Substitution of human olfaction by the trigeminal system. Sci. Adv., November 2025

Scientists have developed a first-of-its-kind device that helps people who have lost their sense of smell “sense” odors again, without actually restoring real smell. The device works by pairing an electronic nose (a sensor that detects odor molecules in the air) with a tiny electrical stimulator placed inside the nose.

Instead of activating the damaged olfactory system, the device stimulates the trigeminal nerve, another nerve inside the nose that normally senses tingling, irritation, or temperature from things like menthol or chili peppers. The question was: Can the brain learn to use these trigeminal signals as a substitute for smell?

Across four experiments with over 60 volunteers (including people with complete and partial smell loss), the researchers found that participants could reliably detect when an odor was present based on the electrical stimulation pattern. Some could also tell the difference between different stimulation patterns, though this was harder and improved when participants were trained first. The ability to detect the signal did not depend on having a working sense of smell, meaning even anosmic individuals could use the system.

The device does not recreate real smells. But it shows, for the first time, that it may be possible to give patients a practical way to recognize odor categories, like food vs. danger odors, through another nerve pathway. This could be an early step toward a future “smell prosthesis” for people living with permanent smell loss.

Can Positive Memories Help Heal the Brain?

Steve Ramirez. Memories change. But can we change them on purpose? Science Friday, November 2025.

Neuroscientist Dr. Steve Ramirez explains that memories are not fixed snapshots of the past; they are dynamic reconstructions that change slightly each time we recall them. Instead of functioning like recordings, memories behave more like malleable building blocks, which the brain continually updates. This flexibility may even help us imagine the future by combining elements of past experiences.

New research shows that scientists can activate specific memories in mice by stimulating the exact brain cells involved in those experiences. In mice that show depression-like behavior, artificially turning on their positive memories can lift their mood and reduce symptoms, sometimes even long-term. This suggests that memory manipulation could one day support new treatments for mental health conditions such as depression and anxiety.

Humans already experience a natural form of this effect. Simply recalling a joyful or painful memory can change our biology within seconds, altering mood, heart rate, stress hormones, and attention. This natural responsiveness points to the possibility of developing therapies that intentionally harness positive memories.

Future treatments could also target traumatic memories. Instead of erasing them, scientists may be able to reduce their emotional weight, offering new hope for people living with PTSD while preserving the factual memory.

Looking ahead, the field is moving toward a deeper map of memory in the brain, identifying the specific cells involved in certain memories and finding ways to support or restore them. Such work could eventually help treat memory loss in disorders like Alzheimer’s.