China’s first emperor broke the mold when he had himself buried with a terra-cotta army. Now insight into the careful crafting of those soldiers is coming from the clays used to build them. Custom clay pastes were mixed at a clay-making center and then distributed to specialized workshops that cranked out thousands of the life-size figures, new research suggests.

Roughly 700,000 craftsmen and laborers built Emperor Qin Shihuang’s palatial mausoleum in east-central China between 247 B.C. and 210 B.C. A portion of those workers gathered clay from nearby deposits and prepared it in at least three forms, researchers propose in the August Antiquity. On-site or nearby workshops used different signature clay recipes for terra-cotta warriors, parts of mostly bronze waterfowl figures and paving bricks for pits in which the soldiers originally stood.
Around 7,000 ceramic foot soldiers, generals and horses — equipped with a variety of bronze weapons — make up the army, which was accidentally discovered in 1974 by farmers digging a well. The emperor would have regarded the ceramic statues as a magic army that would protect him as he ruled in the afterlife, many researchers suspect.

Building and assembling the multitude was an enormous task. Workers poured clay mixtures into casts of torsos, limbs and other body parts, and then assembled the bodies, taking care to create different facial features for each soldier. Finished statues, now mostly gray, were covered in colored lacquers and likely fired in kilns. Most figures were placed inside one giant pit. Earthen walls formed 11 parallel corridors where statues stood in battle-ready rows.

Still, no workshops or debris firmly linked to the statue-making process have been found. As a result, the number, size, location and organization of workshops involved in producing the emperor’s ceramic troops remain uncertain.

Archaeologist Patrick Quinn of University College London and three Chinese colleagues studied the composition of clay samples from the site. The pieces were taken from 12 terra-cotta warriors, two acrobat statues found in a second pit, five clay bricks from the floor of the largest pit, clay fragments from inside three bronze waterfowl statues found in a third pit and part of an earthen wall in the acrobat pit.

Microscopic analysis of the samples revealed that the clay came from deposits near the tomb’s location, the scientists say. But the recipes for different parts varied. Paving bricks contained only a mixture of dark and light clays, while the clay used for warriors and acrobats had sand worked in. Sand and plant fragments were folded into a clay mixture that formed the core of the bronze waterfowl.
Sand may have made the clay more malleable for shaping into ornate figures and increased statues’ durability, the researchers speculate. Plant pieces may have helped reduce the weight of birds’ clay cores. A clay-processing site at or just outside the emperor’s mausoleum must have doled out the appropriate clay pastes to an array of workshops where potters made statues, bricks or other objects, the scientists propose.

What’s more, many statue and waterfowl samples show signs of having been slowly heated in kilns at maximum temperatures of no more than 750˚ Celsius. That’s lower by 150˚ C or more than some previous estimates, the investigators say. Fires set in an attack on the tomb after the emperor’s death may have refired some of the clay, accounting for the temperature discrepancy, the researchers say.

“I’m not at all surprised by the new findings,” says East Asian art historian Robin D.S. Yates of McGill University in Montreal. Legal and administrative documents previously found at two other Qin Empire sites describe workshops that specialized in various types of craft production, Yates says.

In some cases, artisans’ stamps and inscriptions on terra-cotta warriors match those on excavated roof tiles from Emperor Qin’s mausoleum. The markings suggest that some workshops made several types of ceramic objects, says East Asian art historian Lothar Ledderose of Heidelberg University in Germany. Inscriptions on statues also indicate that artisans working at off-site factories during the Qin Empire collaborated with potters at local workshops to produce the terra-cotta army, Ledderose says.

Early in Inferior, science writer Angela Saini recalls a man cornering her after a signing for her book Geek Nation, on science in India. “Where are all the women scientists?” he asked, then answered his own question. “Women just aren’t as good at science as men are. They’ve been shown to be less intelligent.”

Saini fought back with a few statistics on girls’ math abilities, but soon decided that nothing she could say would convince him. It’s a situation that may feel familiar to many women. “What I wish I had was a set of scientific arguments in my armory,” she writes.
So she decided to learn the truth about what science really does tell us about differences between the sexes. “For everyone who has faced the same situation,” she writes, “the same desperate attempt to not lose control but have at hand some real facts and a history to explain them, here they are.”

In Inferior, Saini marshals plenty of facts and statistics contradicting sexist notions about women’s bodies and minds. She cites study after study showing little or no difference in male and female capabilities.

But it’s the book’s historical perspective that makes it most compelling. Only by understanding the cultural context of the men whose studies and ideas first pointed to gender imbalances can we see how deeply biases run, Saini argues.

Charles Darwin’s influential ideas reflected his times, for instance. In The Descent of Man, he wrote that “man has ultimately become superior to woman” via evolution. To a woman active in her local women’s movement, Darwin wrote, “there seems to me to be a great difficulty from the laws of inheritance … in [women] becoming the intellectual equals of man.”

If that idea sounds absurd now, don’t fool yourself into thinking it has vanished. Saini’s book is full of examples right up to today of scientists who have started from this and other flawed premises, which have led to generations of flawed studies and results that reinforce stereotypes. But the tide has been turning, as more women have entered science and more scientists of both sexes seek to remove bias from their work.
Saini does an excellent job of dissecting research on evolution, neuroscience and even the long-standing notion that women’s sexual behavior is driven by their interest in stable, monogamous relationships. By the end, it’s clear that science doesn’t divide men and women; we’ve done that to ourselves. And as scientists become more rigorous, we get closer to seeing ourselves as we really are.

A tiny, shimmying cantilever wiggles a bit more than expected in a new experiment. The excess jiggling of the miniature, diving board–like structure might hint at why the strange rules of quantum mechanics don’t apply in the familiar, “classical” world. But that potential hint is still a long shot: Other sources of vibration are yet to be fully ruled out, so more experiments are needed.

Quantum particles can occupy more than one place at the same time, a condition known as a superposition (SN: 11/20/10, p. 15). Only once a particle’s position is measured does its location become definite. In quantum terminology, the particle’s wave function, which characterizes the spreading of the particle, collapses to a single location (SN Online: 5/26/14).
In contrast, larger objects are always found in one place. “We never see a table or chair in a quantum superposition,” says theoretical physicist Angelo Bassi of the University of Trieste in Italy, a coauthor of the study, to appear in Physical Review Letters. But standard quantum mechanics doesn’t fully explain why large objects don’t exist in superpositions, or how and why wave functions collapse.

Extensions to standard quantum theory can alleviate these conundrums by assuming that wave functions collapse spontaneously, at random intervals. For larger objects, that collapse happens more quickly, meaning that on human scales objects don’t show up in two places at once.

Now, scientists have tested one such theory by looking for one of its predictions: a minuscule jitter, or “noise,” imparted by the random nature of wave function collapse. The scientists looked for this jitter in a miniature cantilever, half a millimeter long. After cooling the cantilever and isolating it to reduce external sources of vibration, the researchers found that an unexplained trembling still remained.

In 2007, physicist Stephen Adler of the Institute for Advanced Study in Princeton, N.J., predicted that the level of jitter from wave function collapse would be large enough to spot in experiments like this one. The new measurement is consistent with Adler’s prediction. “That’s the interesting fact, that the noise matches these predictions,” says study coauthor Andrea Vinante, formerly of the Institute for Photonics and Nanotechnologies in Trento, Italy. But, he says, he wouldn’t bet on the source being wave function collapse. “It is much more likely that it’s some not very well understood effect in the experiment.” In future experiments, the scientists plan to change the design of the cantilever to attempt to isolate the vibration’s source.

The result follows similar tests performed with the LISA Pathfinder spacecraft, which was built as a test-bed for gravitational wave detection techniques. Two different studies found no excess jiggling of free-falling weights within the spacecraft. But the new cantilever experiment tests for wave function collapse occurring at different rate and length scales than those previous studies.
Theories that include spontaneous wave function collapse are not yet accepted by most physicists. But interest in them has recently become more widespread, says physicist David Vitali of the University of Camerino in Italy, “sparked by the fact that technological advances now make fundamental tests of quantum mechanics much easier to conceive.” Focusing on a simple system like the cantilever is the right approach, says Vitali, who was not involved with the research. Still, “a lot of things can go wrong or can be not fully controlled.”

To conclude that wave function collapse is the cause of the excess vibrations, every other possible source will have to be ruled out. So, Adler says, “it’s going to take a lot of confirmation to check that this is a real effect.”

Air pollution is a drag for renewable energy. Dust and other sky-darkening air pollutants slash solar energy production by 17 to 25 percent across parts of India, China and the Arabian Peninsula, a new study estimates. The haze can block sunlight from reaching solar panels. And if the particles land on a panel’s flat surface, they cut down on the area exposed to the sun. Dust can come from natural sources, but the other pollutants have human-made origins, including cars, factories and coal-fired power plants.

Scientists collected and analyzed dust and pollution particles from solar panels in India, then extrapolated to quantify the impact on solar energy output in all three locations. China, which generates more solar energy than any other country, is losing up to 11 gigawatts of power capacity due to air pollution, the researchers report in the Aug. 8 Environmental Science & Technology Letters. That’s a loss of about $10 billion per year in U.S. energy costs, says study coauthor Mike Bergin of Duke University. Regular cleaning of solar panels can help. Cleaning the air, however, is harder.

A growing band of digital characters that converse, read faces and track body language is helping humans to communicate better with one another. While virtual helpers that perform practical tasks, such as dealing with customer service issues, are becoming ubiquitous, computer scientist M. Ehsan Hoque is at the forefront of a more emotionally savvy movement. He and his team at the University of Rochester in New York create software for digital agents that recognize when a person is succeeding or failing in specific types of social interactions. Data from face-to-face conversations and feedback from professional counselors and interviewers with relevant expertise inform this breed of computer advisers.

One of Hoque’s digital helpers grooms people to be better public speakers. With words on a screen, this attentive app notes, for example, how many times in a practice talk a person says “um,” gestures inappropriately or awkwardly shifts vocal tone. With the help of Google Glass, the app even offers useful reminders during actual speeches. Another computerized helper, this one in the form of an avatar, helps people hone their job interviewing skills, flagging long-winded responses or inconsistent eye contact in practice interviews. In the works are computerized conversation coaches that can improve speech and communication skills among people with developmental conditions such as autism and mediate business meetings in ways that encourage everyone to participate in decision making.

“There has been some progress in artificial intelligence, but not much in developing emotional aspects of AI,” Hoque says. “We’re just cracking through the surface at this point.”
The U.S. Department of Defense and the U.S. Army have taken notice. With their financial support, Hoque is developing avatars that collaborate with humans to solve complex problems, and digital observers that monitor body language to detect when people are lying.
This is heady stuff for a 35-year-old who earned a doctoral degree just four years ago. Hoque, who was born in Bangladesh and immigrated to the United States as a teenager, did his graduate work with the MIT Media Lab’s Affective Computing research group. The group’s director, Rosalind Picard, helped launch the field of “affective computing” in the 1990s, which focuses on the study and development of computers and robots that recognize, interpret and simulate human emotions.

Hoque’s approach puts a service spin on affective computing. As a grad student, he developed software he dubbed MACH, short for My Automated Conversation coacH. This system simulates face-to-face conversations with a computer-generated, 3-D man or woman that sees, hears and makes decisions while conversing with a real-life partner. Digital analyses of a human partner’s speech and nonverbal behavior inform the avatar’s responses during a session. A simulated coach may, for instance, let a user know if smiles during an interview look forced or are mistimed. After a session, users see a video of the interaction accompanied by displays of how well or poorly they did on various interaction skills, such as keeping eye contact and nodding at appropriate times.

MACH got its start in trials that trained MIT undergraduates how to conduct themselves during interviews with career counselors. First, Hoque analyzed smiles and other behaviors that either helped or hurt the impressions job candidates left on experienced counselors in mock interviews. In a series of follow-up studies, his team developed an automated system that recognized impression-enhancing behaviors during simulated interviews. That pilot version of MACH was then put to the test. Women, but not men, who received MACH training and got feedback from their digital coach while watching videos of their initial interviews with a counselor displayed substantial improvement in follow-up interviews. MACH trainees who watched interview videos but got no feedback showed minimal improvement. Testing with larger groups of men and women is under way.
As he developed MACH, Hoque consulted MIT sociologist and clinical psychologist Sherry Turkle. That was a bold move, since Turkle has warned for 30 years that, despite its pluses, digital culture discourages person-to-person connections. Social robots, in particular, represent a way for people to escape the challenges of forging authentic relationships, Turkle contends.

But she came away impressed with Hoque, whose goals she calls refreshingly modest and transparent. “His avatars will be helpers and facilitators,” she says, “not companions, friends, therapists and pretend people.”

Hoque’s approach grew out of personal experience. He is the primary caregiver for his 16-year-old brother, Eshteher, who has Down syndrome and does not speak. Eshteher can make sounds to refer to certain things, such as food, and has limited use of sign language. “I’ve spent a lot of time with him and can read what he’s experiencing, like when he’s frustrated or repentant,” Hoque says.
So it’s not surprising that Hoque’s next-generation MACH, dubbed LISSA for Live Interactive Social Skill Assistance, is an avatar that conducts flexible, “getting acquainted” conversations while providing feedback on users’ eye contact, speaking volume, smiling and body movements via flashing icons.

LISSA has shown promise in preliminary tests aimed at improving the conversational chops of college students attending speed-dating sessions and individuals with autism spectrum disorders. Hoque plans to expand this technology for use with people suffering from social phobia and post-traumatic stress disorder. He’s also working on an avatar that trains doctors to communicate clearly and compassionately with patients being treated for life-threatening cancers.

Hoque’s work on emotionally perceptive avatars may eventually transform the young industry of digital assistants, currently limited to voices-in-a-box such as Apple’s Siri and Microsoft’s Cortana, says cognitive scientist Mary Czerwinski, a principal researcher at Microsoft Research Lab in Redmond, Wash. Avatar research “could lead to more natural, personable digital assistants,” Czerwinski predicts. Hoque agrees.

“In the future, we’ll all have digital, personalized assistants,” he says. If he gets his way, emotionally attuned helpers will make us more social and less isolated. That’s something to applaud — if we can manage to put down our smartphones.

Pregnant women who do not have enough folic acid — a B vitamin — in their bodies can pass the deficiency on to their unborn children. It may lead to retarded growth and congenital malformation, according to Dr. A. Leonard Luhby…. “Folic acid deficiency in pregnant women could well constitute a public health problem of dimensions we have not originally recognized,” he says. — Science News. December 9, 1967

Update
Folic acid — or folate — can prevent brain and spinal cord defects in developing fetuses. Since the U.S. Food and Drug Administration required that all enriched grain products contain the vitamin starting in 1998, birth defects have been prevented in about 1,300 babies each year. But some women still don’t get enough folate, while others may be overdoing it. About 10 percent of women may ingest more than the upper limit of 1,000 micrograms daily — about 2.5 times the recommended amount, a 2011 study found. Too much folate may increase a woman’s risk for certain cancers and interfere with some epilepsy drugs.

On astrophysicists’ charts of star stuff, there’s a substance that still merits the label “here be dragons.” That poorly understood material is found inside neutron stars — the collapsed remnants of once-mighty stars — and is now being mapped out, as scientists better characterize the weird matter.

The detection of two colliding neutron stars, announced in October (SN: 11/11/17, p. 6), has accelerated the pace of discovery. Since the event, which scientists spied with gravitational waves and various wavelengths of light, several studies have placed new limits on the sizes and masses possible for such stellar husks and on how squishy or stiff they are.
“The properties of neutron star matter are not very well known,” says physicist Andreas Bauswein of the Heidelberg Institute for Theoretical Studies in Germany. Part of the problem is that the matter inside a neutron star is so dense that a teaspoonful would weigh a billion tons, so the substance can’t be reproduced in any laboratory on Earth.

In the collision, the two neutron stars merged into a single behemoth. This remnant may have immediately collapsed into a black hole. Or it may have formed a bigger, spinning neutron star that, propped up by its own rapid rotation, existed for a few milliseconds — or potentially much longer — before collapsing. The speed of the object’s demise is helping scientists figure out whether neutron stars are made of material that is relatively soft, compressing when squeezed like a pillow, or whether the neutron star stuff is stiff, standing up to pressure. This property, known as the equation of state, determines the radius of a neutron star of a particular mass.

An immediate collapse seems unlikely, two teams of researchers say. Telescopes spotted a bright glow of light after the collision. That glow could only appear if there were a delay before the merged neutron star collapsed into a black hole, says physicist David Radice of Princeton University because when the remnant collapses, “all the material around falls inside of the black hole immediately.” Instead, the neutron star stuck around for at least several milliseconds, the scientists propose.

Simulations indicate that if neutron stars are soft, they will collapse more quickly because they will be smaller than stiff neutron stars of the same mass. So the inferred delay allows Radice and colleagues to rule out theories that predict neutron stars are extremely squishy, the researchers report in a paper published November 13 at arXiv.org.
Using similar logic, Bauswein and colleagues rule out some of the smallest sizes that neutron stars of a particular mass might be. For example, a neutron star 60 percent more massive than the sun can’t have a radius smaller than 10.7 kilometers, they determine. These results appear in a paper published November 29 in the Astrophysical Journal Letters.

Other researchers set a limit on the maximum mass a neutron star can have. Above a certain heft, neutron stars can no longer support their own weight and collapse into a black hole. If this maximum possible mass were particularly large, theories predict that the newly formed behemoth neutron star would have lasted hours or days before collapsing. But, in a third study, two physicists determined that the collapse came much more quickly than that, on the scale of milliseconds rather than hours. A long-lasting, spinning neutron star would dissipate its rotational energy into the material ejected from the collision, making the stream of glowing matter more energetic than what was seen, physicists Ben Margalit and Brian Metzger of Columbia University report. In a paper published November 21 in the Astrophysical Journal Letters, the pair concludes that the maximum possible mass is smaller than about 2.2 times that of the sun.

“We didn’t have many constraints on that prior to this discovery,” Metzger says. The result also rules out some of the stiffer equations of state because stiffer matter tends to support larger masses without collapsing.

Some theories predict that bizarre forms of matter are created deep inside neutron stars. Neutron stars might contain a sea of free-floating quarks — particles that are normally confined within larger particles like protons or neutrons. Other physicists suggest that neutron stars may contain hyperons, particles made with heavier quarks known as strange quarks, not found in normal matter. Such unusual matter would tend to make neutron stars softer, so pinning down the equation of state with additional neutron star crashes could eventually resolve whether these exotic beasts of physics indeed lurk in this unexplored territory.

Galileo’s most famous experiment has taken a trip to outer space. The result? Einstein was right yet again. The experiment confirms a tenet of Einstein’s theory of gravity with greater precision than ever before.

According to science lore, Galileo dropped two balls from the Leaning Tower of Pisa to show that they fell at the same rate no matter their composition. Although it seems unlikely that Galileo actually carried out this experiment, scientists have performed a similar, but much more sensitive experiment in a satellite orbiting Earth. Two hollow cylinders within the satellite fell at the same rate over 120 orbits, or about eight days’ worth of free-fall time, researchers with the MICROSCOPE experiment report December 4 in Physical Review Letters. The cylinders’ accelerations match within two-trillionths of a percent.

The result confirms a foundation of Einstein’s general theory of relativity known as the equivalence principle. That principle states that an object’s inertial mass, which sets the amount of force needed to accelerate it, is equal to its gravitational mass, which determines how the object responds to a gravitational field. As a result, items fall at the same rate — at least in a vacuum, where air resistance is eliminated — even if they have different masses or are made of different materials.

The result is “fantastic,” says physicist Stephan Schlamminger of OTH Regensburg in Germany who was not involved with the research. “It’s just great to have a more precise measurement of the equivalence principle because it’s one of the most fundamental tenets of gravity.”
In the satellite, which is still collecting additional data, a hollow cylinder, made of platinum alloy, is centered inside a hollow, titanium-alloy cylinder. According to standard physics, gravity should cause the cylinders to fall at the same rate, despite their different masses and materials. A violation of the equivalence principle, however, might make one fall slightly faster than the other.

As the two objects fall in their orbit around Earth, the satellite uses electrical forces to keep the pair aligned. If the equivalence principle didn’t hold, adjustments needed to keep the cylinders in line would vary with a regular frequency, tied to the rate at which the satellite orbits and rotates. “If we see any difference in the acceleration it would be a signature of violation” of the equivalence principle, says MICROSCOPE researcher Manuel Rodrigues of the French aerospace lab ONERA in Palaiseau. But no hint of such a signal was found.

With about 10 times the precision of previous tests, the result is “very impressive,” says physicist Jens Gundlach of the University of Washington in Seattle. But, he notes, “the results are still not as precise as what I think they can get out of a satellite measurement.”

Performing the experiment in space eliminates certain pitfalls of modern-day land-based equivalence principle tests, such as groundwater flow altering the mass of surrounding terrain. But temperature changes in the satellite limited how well the scientists could confirm the equivalence principle, as these variations can cause parts of the apparatus to expand or contract.

MICROSCOPE’s ultimate goal is to beat other measurements by a factor of 100, comparing the cylinders’ accelerations to see whether they match within a tenth of a trillionth of a percent. With additional data yet to be analyzed, the scientists may still reach that mark.

Confirmation of the equivalence principle doesn’t mean that all is hunky-dory in gravitational physics. Scientists still don’t know how to combine general relativity with quantum mechanics, the physics of the very small. “The two theories seems to be very different, and people would like to merge these two theories,” Rodrigues says. But some attempts to do that predict violations of the equivalence principle on a level that’s not yet detectable. That’s why scientists think the equivalence principle is worth testing to ever more precision — even if it means shipping their experiments off to space.

Bigwigs in a more than 600-year-old South American population were easy to spot. Their artificially elongated, teardrop-shaped heads screamed prestige, a new study finds.

During the 300 years before the Incas’ arrival in 1450, intentional head shaping among prominent members of the Collagua ethnic community in Peru increasingly centered on a stretched-out look, says bioarchaeologist Matthew Velasco of Cornell University. Having long, narrow noggins cemented bonds among members of a power elite — a unity that may have helped pave a relatively peaceful incorporation into the Incan Empire, Velasco proposes in the February Current Anthropology.
“Increasingly uniform head shapes may have encouraged a collective identity and political unity among Collagua elites,” Velasco says. These Collagua leaders may have negotiated ways to coexist with the encroaching Inca rather than fight them, he speculates. But the fate of the Collaguas and a neighboring population, the Cavanas, remains hazy. Those populations lived during a conflict-ridden time — after the collapse of two major Andean societies around 1100 (SN: 8/1/09, p. 16) and before the expansion of the Inca Empire starting in the 15th century.

For at least the past several thousand years, human groups in various parts of the world have intentionally modified skull shapes by wrapping infants’ heads with cloth or binding the head between two pieces of wood (SN: 4/29/17, p. 18). Researchers generally assume that this practice signified membership in ethnic or kin groups, or perhaps social rank.
The Callagua people lived in Colca Valley in southeastern Peru and raised alpaca for wool. By tracking Collagua skull shapes over 300 years, Velasco found that elongated skulls became increasingly linked to high social status. By the 1300s, for instance, Collagua women with deliberately distended heads suffered much less skull damage from physical attacks than other females did, he reports. Chemical analyses of bones indicates that long-headed women ate a particularly wide variety of foods.
Until now, knowledge of head-shaping practices in ancient Peru primarily came from Spanish accounts written in the 1500s. Those documents referred to tall, thin heads among Collaguas and wide, long heads among Cavanas, implying that a single shape had always characterized each group.

“Velasco has discovered that the practice of cranial modification was much more dynamic over time and across social [groups],” says bioarchaeologist Deborah Blom of the University of Vermont in Burlington.

Velasco examined 211 skulls of mummified humans interred in either of two Collagua cemeteries. Burial structures built against a cliff face were probably reserved for high-ranking individuals, whereas common burial grounds in several caves and under nearby rocky overhangs belonged to regular folk.
Radiocarbon analyses of 13 bone and sediment samples allowed Velasco to sort Collagua skulls into early and late pre-Inca groups. A total of 97 skulls, including all 76 found in common burial grounds, belonged to the early group, which dated to between 1150 and 1300. Among these skulls, 38 — or about 39 percent — had been intentionally modified. Head shapes included sharply and slightly elongated forms as well as skulls compressed into wide, squat configurations.

Of the 14 skulls with extreme elongation, 13 came from low-ranking individuals, a pattern that might suggest regular folk first adopted elongated head shapes. But with only 21 skulls from elites, the finding may underestimate the early frequency of elongated heads among the high-status crowd. Various local groups may have adopted their own styles of head modification at that time, Velasco suggests.

In contrast, among 114 skulls from elite burial sites in the late pre-Inca period, dating to between 1300 and 1450, 84 — or about 74 percent — displayed altered shapes. A large majority of those modified skulls — about 64 percent — were sharply elongated. Shortly before the Incas’ arrival, prominent Collaguas embraced an elongated style as their preferred head shape, Velasco says. No skeletal evidence has been found to determine whether low-ranking individuals also adopted elongated skulls as a signature look in the late pre-Inca period.

In courtrooms around the United States, computer programs give testimony that helps decide who gets locked up and who walks free.

These algorithms are criminal recidivism predictors, which use personal information about defendants — like family and employment history — to assess that person’s likelihood of committing future crimes. Judges factor those risk ratings into verdicts on everything from bail to sentencing to parole.

Computers get a say in these life-changing decisions because their crime forecasts are supposedly less biased and more accurate than human guesswork.
But investigations into algorithms’ treatment of different demographics have revealed how machines perpetuate human prejudices. Now there’s reason to doubt whether crime-prediction algorithms can even boast superhuman accuracy.

Computer scientist Julia Dressel recently analyzed the prognostic powers of a widely used recidivism predictor called COMPAS. This software determines whether a defendant will commit a crime within the next two years based on six defendant features — although what features COMPAS uses and how it weighs various data points is a trade secret.

Dressel, who conducted the study while at Dartmouth College, recruited 400 online volunteers, who were presumed to have little or no criminal justice expertise. The researchers split their volunteers into groups of 20, and had each group read descriptions of 50 defendants. Using such information as sex, age and criminal history, the volunteers predicted which defendants would reoffend.
A comparison of the volunteers’ answers with COMPAS’ predictions for the same 1,000 defendants found that both were about 65 percent accurate. “We were like, ‘Holy crap, that’s amazing,’” says study coauthor Hany Farid, a computer scientist at Dartmouth. “You have this commercial software that’s been used for years in courts around the country — how is it that we just asked a bunch of people online and [the results] are the same?”

There’s nothing inherently wrong with an algorithm that only performs as well as its human counterparts. But this finding, reported online January 17 in Science Advances, should be a wake-up call to law enforcement personnel who might have “a disproportionate confidence in these algorithms,” Farid says.

“Imagine you’re a judge, and I tell you I have this highly secretive, highly proprietary, expensive software built on big data, and it says the person standing in front of you is high risk” for reoffending, he says. “The judge would be like, ‘Yeah, that sounds quite serious.’ But now imagine if I tell you, ‘Twenty people online said this person is high risk.’ I imagine you’d weigh that information a little bit differently.” Maybe these predictions deserve the same amount of consideration.

Judges could get some better perspective on recidivism predictors’ performance if the Department of Justice or National Institute for Standards and Technology established a vetting process for new software, Farid says. Researchers could test computer programs against a large, diverse dataset of defendants and OK algorithms for courtroom use only if they get a passing grade for prediction.

Farid has his doubts that computers can show much improvement. He and Dressel built several simple and complex algorithms that used two to seven defendant features to predict recidivism. Like COMPAS, all their algorithms maxed out at about D-level accuracy. That makes Farid wonder whether trying to predict crime with anything approaching A+ accuracy is an exercise in futility.

“Maybe there will be huge breakthroughs in data analytics and machine learning over the next decade that [help us] do this with a high accuracy,” he says. But until then, humans may make better crime predictors than machines. After all, if a bunch of average Joe online recruits gave COMPAS a run for its money, criminal justice experts — like social workers, parole officers, judges or detectives — might just outperform the algorithm.

Even if computer programs aren’t used to predict recidivism, that doesn’t mean they can’t aid law enforcement, says Chelsea Barabas, a media researcher at MIT. Instead of creating algorithms that use historic crime data to predict who will reoffend, programmers could build algorithms that examine crime data to find trends that inform criminal justice research, Barabas and colleagues argue in a paper to be presented at the Conference on Fairness, Accountability and Transparency in New York City on February 23.

For instance, if a computer program studies crime statistics and discovers that certain features — like a person’s age or socioeconomic status — are highly related to repeated criminal activity, that could inspire new studies to see whether certain interventions, like therapy, help those at-risk groups. In this way, computer programs would do one better than just predict future crime. They could help prevent it.