August 08, 2019
Thanks for reading Axios Science. Please send me your feedback at [email protected] or just reply to this email.
1 big thing: Looking to AI to understand how we learn
Two parallel quests to understand learning — in machines and in our own heads — are converging in a small group of scientists who think that artificial intelligence may hold an answer to the deep-rooted mystery of how our brains learn, Kaveh Waddell and I write.
Why it matters: If machines and animals do learn in similar ways — still an open question among researchers — figuring out how could simultaneously help neuroscientists unravel the mechanics of knowledge or addiction, and help computer scientists build much more capable AI.
The big picture: For decades, researchers compared human and machine learning and largely rejected the notion that they are closely linked.
- At the center of the question is the credit-assignment problem: the enigma of how the brain knows which parts of itself need to change in order to better accomplish a task.
- In AI, a major method for credit assignment is known as error backpropagation.
- After backprop fueled major advances in AI image recognition, some scientists started revisiting whether the brain could be doing something similar.
"There is a big undercurrent in neuroscience [saying] we should go back to neural networks," says Konrad Kording, a neuroscientist at UPenn, referring to a reigning AI technique that relies on backprop.
- Backprop allows machines to learn from their mistakes. If an actual outcome differs from the computer's predicted outcome, information about what went wrong gets passed back through layers in the neural network, adjusting the system accordingly.
- One central debate is over whether neurons, which communicate through chemical signals, can simultaneously transmit information to another neuron while receiving feedback from that same neuron about what went wrong.
A trio of scientists in Toronto and a DeepMind researcher are searching for that evidence in the brains of mice. In their experiment, animals watch patterns on a screen as their brain activity is recorded.
- The animals see a consistent pattern of moving shapes for hours — then, an aberration, like a square going the wrong way.
- Preliminary results suggest there is in fact a specific, measurable signal that passes between neurons only when the animals witness an "error."
- "We know the brain has to have some mechanism of credit assignment," says Joel Zylberberg, a professor at York University in Toronto. "The most promising candidate still seems to be these top-down feedback signals."
But, but, but: The brain doesn't just learn from error. Some of our knowledge is based on intuition and some is acquired throughout our lives.
- "The general thing that I think is being missed by the field is that there's a huge disconnect between the diversity and complexity of the brain and the relative simplicity of the models people are using," says Gary Marcus, an NYU psychology professor and vocal critic of AI's dependence on deep learning.
The bottom line: Researchers know learning hinges on the strengthening and weakening of the synapses between individual neurons. But how that change plays out globally among the roughly 100 trillion synapses in the human brain — so we can recognize someone's face, for example — is unknown.
- "I want to be inspired by neural networks but I don’t want to take them too literally, as if the only way to do this is to have the literal algorithm for backprop implemented in the brain," says neuroscientist Aaron Batista from the University of Pittsburgh.
2. Happy 7th birthday, Curiosity
NASA's Curiosity rover just completed its 7th year on Mars, during which the car-sized spacecraft has changed the way we understand the Red Planet, Axios' Miriam Kramer writes.
By the numbers: Mission controllers experienced "7 minutes of terror" during its 2012 landing when a complex series of maneuvers brought Curiosity to the surface.
- Since then, Curiosity has traveled about 13 miles on Mars, according to NASA.
- The rover has drilled 22 samples of Martian dirt, chemically analyzing it to figure out what Mars may have looked like in the past.
- Curiosity is now studying Mount Sharp — an 18,000-foot mountain that has deposits dating back to the time when this part of Mars was covered in lakes and rivers.
- NASA expects the rover can keep trucking on Mars for about 5–7 more years.
What's next: Curiosity discovered that Mars was likely habitable for millions of years. The rover will use its remaining years to try to solve the mystery of why that habitability ended.
- "Did the climate change dramatically? The answer may be recorded in the rocks ahead," Curiosity project scientist Ashwin Vasavada told Axios via email.
- Vasavada also said that the rover will be on the lookout for methane — a gas that, on Earth, can indicate life — to try to figure out where the intermittent bursts of methane on the Red Planet are coming from.
3. Axios stories worthy of your time
Where the taps could run dry (Dave Lawler — Axios)
America's mental health problem isn't mass shootings (Caitlin Owens — Axios)
A sedative epidemic is taking root (Nambi J. Ndugga, Elsa Pearson, Melissa Garrido — Axios Expert Voices)
The unknown risks of radiation in space (Miriam Kramer — Axios)
The climate peril from land degradation (Ben Geman — Axios)
4. A microbiome discovery
Researchers say they've discovered more than 4,000 new, small protein families generated in the human microbiome, Axios' Eileen Drage O'Reilly writes.
- The function of these newly found tiny proteins still needs to be determined, but the authors believe their size could allow them to be leveraged to deliver drugs.
What they did: Over a 4-year period, Ami Bhatt of Stanford and her colleagues collected large amounts of microbiome data from humans, animals and the environment and sequenced their DNA, they report in a study published Thursday in the journal Cell.
What they found: They discovered more than 4,000 protein families, 90% of which have no known function and almost half have not been previously catalogued.
- They estimate 30% of the proteins may be involved in cell-to-cell communication.
What they're saying: Nicola Segata, associate professor and principal investigator at University of Trento's Centre for Integrative Biology, tells Axios the study offers a first, "crucial step" but "it is premature to speculate on whether and how this discovery will be relevant for future therapies."
5. What we're reading elsewhere
Gender-neutral pronouns reduce biases (Ian Sample — The Guardian)
Pediatricians warn racism has devastating effects on children (William Wan — Washington Post)
A vicious disease is wiping out China's pig population (Chris Baraniuk — OneZero)
A tale of elephants, ants, trees and fire shows how complex nature is (The Economist)
6. Something wondrous
About 773,000 years ago, Earth's magnetic north and south poles reversed — a complex process that took about 22,000 years, according to new research.
The big picture: Throughout our planet's history, the poles have flipped every several hundred thousand years or so. How long that process takes, what's involved and when it might happen again aren't well understood, but they are much debated.
How it works: During "excursions," Earth's magnetic field intensity decreases globally and compasses start to point in intermediate directions.
- The field can bounce back and return to its polarity.
- But the poles can also flip entirely if the magnetic field's strength wanes to 10%–20%.
What's new: In research published this week in Science Advances, geologist Brad Singer of the University of Wisconsin-Madison and his colleagues homed in on the last reversal event in search of the steps leading up to it.
- They compared volcanic records from 6 different locations, Antarctic ice cores and sediments from the ocean floor — all of which captured markers of the direction of Earth's magnetic field during the last reversal.
- The team found the process took about 22,000 years — about 3 times longer than previous estimates reported — and involved 2 excursions (about 795,000 and 784,000 years ago) before it flipped.
But, but, but: The geological record has its limits — samples can be disrupted and the details of the reversal smeared out by the slow deposition of material over time. Singer says data from multiple volcanoes and other observations strengthen the findings but some researchers disagree.
What's next: Some scientists think we may be going into a reversal now as Earth's magnetic field strength is declining about 5% per century.
- At the same time, magnetic north is drifting toward Siberia.
- But "the strength of the field and directional changes seen today are not mimicked by what lead to an excursion in the paleontological record," says Singer. "The evidence we are headed into reversal — it’s not out there."
Still, researchers want to understand the potential effects of a magnetic reversal on life on Earth.
- Because the planet's protective magnetic field would be powered down, the electric grid and satellites could be at greater risk of disruption from the sun's normal solar storms, says Scott Bogue, a geologist at Occidental College.
- But, he adds, the Golden Gate Bridge would not melt.