Today in Science: How AI knows things nobody told it

September 5, 2023: Huge benefits of childhood reading, AI's emerging abilities and biologists are trying to save the endangered ocelot.
Andrea Gawrylewski, Chief Newsletter Editor
TOP STORIES

The Perks of Being a Reader

A new study found a dramatic and positive link between reading for pleasure in early childhood and better cognition, mental health and educational attainment in adolescence. The research team examined childhood reading rates of 10,000 participants, ages 9 to 13, across different ethnicities and varying socioeconomic status. Those who read for pleasure as kids had larger cortical surface areas in several brain regions that are related to cognition and mental health.

Why this matters: Early childhood is a critical period for brain development. Brain health in childhood is directly linked to better mental health, cognition and educational level later in life. Stronger brains can also provide resilience in times of stress. Amazingly, the research showed that the benefits of reading in childhood hold regardless of socioeconomic status, suggesting that reading can counteract the negative brain effects of poverty.

What the experts say: "While reading for pleasure is unlikely, on its own, to fully address the challenging effects of poverty on the brain, it provides a simple method for improving children's development and attainment," write Barbara Jacquelyn Sahakian, a professor of clinical neuropsychology at the University of Cambridge, and her colleagues.

AI's Black Box

Researchers are finding that large language models (LLMs) like ChatGPT display "emergent abilities," that is, skills they were not trained to have, and even a genuine understanding of what they have learned. Scientists who research these models have shown that some can: Keep an Othello game board in their "mind's eye" in order to make the best move; make inferences about different shades of a particular color like red (even if they've never "seen" the color); or display "in context" learning, which is learning from users' prompts. 

Why this matters: Researchers aren't quite sure how LLMs achieve these feats since the tech is not yet a true AI (with the consciousness or resourcefulness seen in animal brains). As quickly as this technology is advancing, the window to study these systems may be closing. OpenAI–the company that built ChatGPT–has not divulged the details of how it designed and trained its latest GPT-4, in part because it is locked in competition with Google and other companies, not to mention other countries.

What the experts say: A lack of transparency hinders efforts to understand the social impacts of quickly advancing AI technology, says Melanie Mitchell, an AI researcher at the Santa Fe Institute. "Transparency about these models is the most important thing to ensure safety."
TODAY'S NEWS
• The updated fall COVID booster will likely be available around mid-September. Some people will benefit from getting it more than others. | 6 min read
• Biologists are taking on an ambitious plan to try to save the endangered ocelot by artificially inseminating females born and raised in zoos using sperm extracted from wild ocelots in southern Texas. | 7 min read
• Many machines, including cellphones, data centers, cars and airplanes, become less efficient and degrade more quickly in extreme heat. Physicists explain how. | 5 min read
• Doom and gloom, or realism and hope? Here's how six climate experts describe the future to their young children. | 12 min read
More News
EXPERT PERSPECTIVES
• Oversight of AI technologies (for instance, requiring technology companies to disclose their research or technical information for all to see) is too big of a job for one agency. "We recommend that all U.S. agencies come together quickly to finalize cross-agency rules to ensure the safety of these applications," write the editors of Scientific American, in the September issue. Specific regulations should be devised by governing agencies of specific industries (ie. healthcare and automotive), we suggest. | 5 min read
More Opinion
IMAGE OF THE DAY
Credit: Jasius/Getty
The concentric circles or eyespots on butterfly and moth wings—like those seen on this Suraka silk moth—not only look like real eyes but may also appear to glare directly at predators from many directions. This optical illusion has been called the "Mona Lisa effect" (though some scientists have said that the Mona Lisa doesn't actually stare directly at anyone). But in insects the effect could scare would-be attackers and buy the bugs enough time to escape. A recent experiment tested whether the direction of this fake gaze was more or less scary to potential predators
A 2021 Gallup poll showed the lowest reading rates among adults (about a book a month) since the company began tracking reading habits more than 40 years ago. Studies have shown that reading can decrease mortality, prevent cognitive decline and reduce stress. Combined with the findings above on the benefits of reading for kids, I think we can safely say that picking up a book is one of the better choices you can make for your health. For some great September science reads, check out book recommendations from Scientific American here
Did you have a favorite read this summer? Email me and tell me about it: newsletters@sciam.com. See you tomorrow!
—Andrea Gawrylewski, Chief Newsletter Editor
Subscribe to this and all of our newsletters here.

Scientific American
One New York Plaza, New York, NY, 10004
Support our mission, subscribe to Scientific American here

Comments

Popular Posts