Hello Nature readers, would you like to get this Briefing in your inbox free every day? Sign up here.
The increasingly distorted images produced by an artificial-intelligence model that is trained on data generated by a previous version of the model.Credit: M. Boháček & H. Farid/arXiv (CC BY 4.0)
Training successive versions of artificial intelligence (AI) models on text generated by the previous iteration quickly leads to the system producing gibberish. This collapse happens because each new model learns not from reality, but from the previous model’s prediction of reality, with errors getting amplified over time. Even before complete collapse, learning from AI-generated texts means less common information in the dataset gets ‘forgotten’. “The message is we have to be very careful about what ends up in our training data,” says AI researcher and study co-author Zakhar Shumaylov.
In mice, the placebo effect seems to activate parts of the brain typically associated with movement and coordination. The animals were conditioned to expect pain relief, mimicking how some people can get pain relief from taking a sugar pill they believe contains a drug. Analysis of the mice’s neurons in the active regions revealed that 65% had opioid receptors — the same kind that are activated by powerful painkillers. The research “provides a new target that we can look for in human studies”, says neuroscientist Tom Wager.
Eighty-year-olds can identify tunes just as well as teenagers can. “You’ll hear anecdotes all the time of how people with severe Alzheimer’s can’t speak, can’t recognize people, but will sing the songs of their childhood or play the piano,” says music scientist Sarah Sauvé, who tested how well roughly 90 healthy volunteers, aged 18 to 86 years, could recognize themes across three pieces of music. Musical memory and recognition did not vary with age, even when the music was unfamiliar.
Image of the week
Credit: T. Müller (MPIA/HdA), E. Matthews (MPIA)
This glowing speck is a freezing exoplanet six times the size of Jupiter. It’s one of the closest planets outside our Solar System and the first that was discovered through direct imaging with the James Webb Space Telescope. Usually, astronomers infer the existence of a planet by the effects they have on the star they orbit. (Nature | 4 min read)
Features & opinion
Digital devices and instant messaging have enormous benefits, including speeding up research. But they are also a distraction, argues a Nature editorial. In his book Slow Productivity, computer scientist Cal Newport says that having to concentrate without interruptions is key to high-quality work. More studies should explore the impact of lost concentration time on science, the editorial suggests, and whether tools to automate routine tasks could free up more thinking time — or have the opposite effect.
“I don’t want to be here, but if not me, who?” asks climate scientist Virginia Burkett, who is demanding better policies to protect US government scientists against political interference. In a lengthy whistleblower complaint, she is seeking an investigation into “abuse of authority and gross mismanagement” under former US president Donald Trump. Burkett says that, in 2017, she was demoted after criticizing the government for dismantling climate research programmes, cutting science budgets and attempting to water down an influential climate report. “This is not about what happened to me, it’s about what could happen to others,” Burkett says.
More than 500 million years ago, the globe was covered in glaciers and sea ice. This Snowball Earth seems hostile to life, yet it was probably during this time that single-celled organisms first evolved into complex, multicellular creatures. One possible answer to this conundrum: when seawater gets colder, it gets more viscous, so individual cells could have started to stick together to move through it more efficiently in search of food. In experiments with modern single-celled green algae, cells living in viscous gels formed coordinated groups to keep swimming.
Reference: Proceedings of the Royal Society B paper
Today, I’m congratulating Larry Richardson for becoming the world’s most highly cited cat, his 132 citations easily beating F. D. C. (Felis Domesticus Chester) Willard’s 107. Larry’s citations were of course entirely fabricated, by metascientist Reese Richardson and fluid dynamics researcher Nick Wise. The duo wanted to show how shady companies exploit Google Scholar to sell fake metrics. “For a fairer scientific enterprise, we ought to ditch quantitative heuristics like citation count, impact factor and h-index altogether,” Richardson writes.
Your authentic citations to feedback about this newsletter are always welcome at briefing@nature.com.
Thanks for reading,
Katrina Krämer, associate editor, Nature Briefing
With contributions by Katharine Sanderson and Sarah Tomlin
Want more? Sign up to our other free Nature Briefing newsletters:
• Nature Briefing: Microbiology — the most abundant living entities on our planet — microorganisms — and the role they play in health, the environment and food systems.
• Nature Briefing: Anthropocene — climate change, biodiversity, sustainability and geoengineering
• Nature Briefing: AI & Robotics — 100% written by humans, of course
• Nature Briefing: Cancer — a weekly newsletter written with cancer researchers in mind
• Nature Briefing: Translational Research — covers biotechnology, drug discovery and pharma