PREDICTIVE policing that aims to work out when and where a crime will take place promises a future of data-driven law enforcement. But a flaw found in the design of the software used suggests that instead of fixing biases in policing, predictive algorithms are to blame for a whole new set of problems.

Pre-crime tech is catching on in the US. PredPol – a market-leading system – is already used by police departments in places such as California, Florida and Maryland. Their hope is that such systems will bring down crime rates while simultaneously reducing human bias in policing.

But when researchers in the US examined how PredPol predicts crime, they found something disturbing. Their that the software merely sparks a “feedback loop” that leads to officers being repeatedly sent to certain neighbourhoods – typically ones with a high number of racial minorities – regardless of the true crime rate in that area ().

The problem stems from the logic that PredPol uses to decide where officers should be sent. If an officer is sent to a neighbourhood and then makes an arrest, the software takes this as indicating a good chance of more crimes in that area in future.

What this means, says Matt Kusner at the Alan Turing Institute in London, is that the PredPol system seems to be learning from reports recorded by the police – which may be higher in areas where there are more police – rather than from underlying crime rates.

“That’s how dangerous feedback loops are,” says Joshua Loftus at New York University, who wasn’t involved in the study. Although these loops are only part of how PredPol makes its predictions, he says they may explain why predictive policing algorithms have sometimes seemed to recreate exactly the kind of racial biases their creators say they overcome.

“A ‘feedback loop’ in software leads to officers being repeatedly sent to certain neighbourhoods”

To better understand how the system comes to its conclusions, the study team created a simplified mathematical model of the PredPol software. The algorithm chooses how to distribute a certain number of officers between two locations. If more are sent to one location, they tend to make more arrests there. The team found that this feeds back into the system and leads it to send even more officers to that same place.

That means the software ends up overestimating the crime rate in one neighbourhood, without taking into account the possibility that more crime is observed there simply because more officers have been sent there – like a computerised version of confirmation bias.

There might be a way to stop the feedback loop. The authors also modelled a different system, in which the algorithm only sent more officers to a neighbourhood if the area’s crime rate was higher than expected. This led it to distribute officers in a way that much more closely matched the true crime rate.

Loftus says that many more problems need to be solved before policing algorithms can be truly called fair. “Human decisions affect every aspect of the design of the system,” he says. The algorithm could be thrown off if officers are more likely to arrest racial minorities, for example.

This article appeared in print under the headline “A flaw in the pre-crime system”

Robots might not be stealing our jobs after all – at least, not if you’re German. An analysis of more than 20 years of labour automation in Germany found no evidence that robots caused job losses in the country as a whole.

Automation has changed the kinds of work people do, but hasn’t taken jobs away permanently, says at Heinrich Heine University DĂŒsseldorf. Although there are plenty of reports that predict future job losses caused by automation, SĂŒdekum’s study is only the second to look at how many jobs robots have already taken from us.

The first – a – painted a much gloomier picture. Researchers at Massachusetts Institute of Technology and Boston University found that workplace automation squeezes wages and takes away between three and six jobs per robot.

Given that Germany has about eight robots for every thousand workers – four times as many as in the US – SĂŒdekum and his colleagues expected to see more job losses there. “Our initial impulse was that it should be much worse in Germany,” he says. So their finding that the country’s workforces have actually faced less of an impact from the worst effects of automation was a surprise.

One reason for this is that while robots have stopped companies creating new jobs in manufacturing – SĂŒdekum estimates that 275,000 fewer jobs were created between 1994 and 2014 because of automation – this has been offset by new posts created in other sectors, particularly the service industry.

People that already worked in industries prone to automation, such as the automotive industry, were also more likely to stay employed after robots were introduced into their workplaces. “We don’t see people getting fired,” says SĂŒdekum.

It’s not all good news, however. Robots have tended to lead to lower wages in lower and medium-skilled manufacturing jobs, . “Some of the workers do have to swallow wage cuts because of robots,” says SĂŒdekum. But for more highly skilled workers and those outside manufacturing, robots had no impact on wages.

Shielding the impact

Germany’s unions might be to thank for reducing the impact of automation on jobs, says SĂŒdekum, explaining that they are generally more willing to accept wage cuts for workers if it means people keep their jobs.

But at Massachusetts Institute of Technology says that might not be enough to keep people in work in the long run. “We know the only way to ameliorate the problem is through continuing education,” he says. Cebrian co-authored a study that found smaller cities – which tend to have higher proportions of low-skilled workers – will be hit by automation much harder than big ones.

“The major cities are disproportionately attracting the types of jobs that are more resilient to automation,” he says. But he thinks it’s still too early to know whether enough of those jobs are being created to make up for losses in other industries.

Another option, SĂŒdekum says, would be to make sure that the benefits of automation – which saves manufacturers money that would otherwise be spent on wages  – are shared with all workers. Earlier this week, Jeremy Corbyn, leader of the , joined Bill Gates in calling for higher taxes on companies that replace human workers with robots.

But SĂŒdekum thinks that talk of robots stealing all of jobs is a touch hyperbolic. If the automotive industry is anything to go by, we’re more likely to see a gradual shift in the kinds of jobs that people move into rather than permanent losses, he says. “I just don’t believe the horror stories.”

Read more: Automation will have a bigger impact on jobs in smaller cities

Jigsaw puzzle

Unknown unknowns

Who are you calling a sucker? Underwater robots could soon hitch rides on sharks and whales thanks to a fish-inspired suction cup that clamps on to shark skin and other surfaces.

“żìĂš¶ÌÊÓÆ”s could record data by attaching this robot to animals without hurting them,” says Li Wen at , whose team developed the sucker. It is designed to cling to a moving surface, like a shark, even as it twists and turns at high speed.

The design for Wen’s robotic suction cup is inspired by the – a marine fish that attaches itself to sharks, rays and turtles using a sucking disc on its head. In the wild, the sharksucker hitch-hikes rides on hosts so it can snack on food scraps, and the crustaceans that live on marine animals, all while expending minimal energy.

Hitch-hiking like this on sharks and whales could offer an improvement on existing ways of tracking and tagging animals, which are criticised for possibly causing harm, or for being ineffective as sensors fall off.

Slow swimmers

Saving energy while swimming is big deal for robots too. The handful of existing types of swimming robots aren’t very fast on their fins. Earlier this year, researchers developed a robotic stingray that can reach top speeds of 6 centimetres per second. While slower than even very small fish, it is a recording-breaking pace for robots of its kind.

But robots attached to live fish could experience a real speed boost. Wen’s artificial sucker can withstand the kinds of forces that sharksuckers face when attached to sharks cruising at more than 1.5 metres per second.

at the University of Bristol, UK, is impressed with the sucker’s grip. “Trying to grip underwater is a big problem,” he says, so this fish-inspired sucker could also be put to good use helping robots to maintain underwater infrastructure like submerged oil pipes.

Attached to a smooth surface, the robotic sucker has serious sticking power, withstanding pull-off forces equal to more than 340 times its own weight. In the lab it has also been attached to shark skin, glass, a carton of orange juice and an iPhone.

Its gripping abilities are largely down to the sucker’s fish-inspired design, which combines a large suction pad with some 1000 tiny carbon fibre spinules that help the sucker stay attached. The spinules, arranged in rows along the sucker, can be raised and lowered to help the device attach and detach.

It takes Wen and his colleagues in the US about two days to make a swimming robot with a sucker attached. The next step will be to attach a hitch-hiking robot to swimming sharks or dolphins, and see how well it can hang on for in the real world. “That is certainly interesting and challenging,” says Wen.

Journal reference: Science Robotics, DOI:

Read more: Bioinspired tube robot can sneak round corners and turn on taps

Next steps

Providing comfort

Facebook allows organisations to target adverts at people who declare an interest in topics such as “how to burn Jews” and “Hitler did nothing wrong”.

The investigative news organisation ProPublica y that it was able to use Facebook’s ad-buying service to direct adverts to almost 2300 people who expressed an interest in anti-Semitic topics. Other categories included “Jew hater” and “why Jews ruin the world.” Facebook’s self-service ad-buying system describe all of these categories as “fields of study”.

ProPublica journalists paid Facebook $30 to place three adverts in the newsfeeds of users who had expressed an interest in the above. All three ads were approved within 15 minutes, although Facebook later removed the categories after ProPublica contacted the social media giant for comment.

The anti-Semitic ad categories seem to have been generated because Facebook users had listed those themes as either an interest, an employer or a “field of study” in their profiles. Facebook’s ad-placing algorithms automatically turn users’ self-declared interests into advertising categories.

The anti-Semitic groups had very small audiences, indicating that only a handful of Facebook users had declared an interested in those topics. There were so few potential viewers that Facebook automatically suggested that ProPublica add related interest groups, including “German Schutzstaffel”, more commonly known as the Nazi SS; and “Nazi Party”. These two groups had a combined audience of almost 6000 users.

Facebook has also come under fire for allowing fake accounts to place adverts that may have sought to influence the US presidential election. Last week, Facebook admitted that ‘inauthentic’ accounts, thought to be operated out of Russia, had placed $100,000 worth of ads between June 2015 and May 2017. These adverts “appeared to focus on amplifying divisive social and political messages across the ideological spectrum – touching on topics from LGBT matters to race issues to immigration to gun rights”, wrote Facebook .

Facebook has since shut down these fake accounts, but not before they were able to post around 3000 adverts. These findings add weight to the ongoing investigation into whether Russia actively tried to sway the result of the 2016 presidential election.

Responding to a request from żìĂš¶ÌÊÓÆ” for a comment on this story, Facebook . “As people fill in their education or employer on their profile, we have found a small percentage of people who have entered offensive responses, in violation of our policies,” it wrote.

“To help ensure that targeting is not used for discriminatory purposes, we are removing these self-reported targeting fields until we have the right processes in place to help prevent this issue.”

Shrink to fit

As Hurricane Irma continues to head north, now downgraded to a “tropical depression”, it leaves behind a trail of destruction across the Caribbean and southern Florida.

Caribbean islands including St Martin, Puerto Rico, Barbuda and Cuba were hit hardest. At the time of writing, 37 people in the Caribbean and 10 in the US are thought to have been killed by the hurricane. On the island of Barbuda more than 90 per cent of buildings were destroyed.

The French and UK governments have sent aid workers, food and medical equipment to their overseas territories to start the long process of reconstruction there. Many of these islands are strongly reliant on tourism and will continue to feel the economic impact of Irma for a long time into the future.

For many places across the region, the priority will be to ensure residents have access to safe drinking water, food and shelter. After the hurricane comes an increased risk of disease, partly because stagnating water attracts mosquitoes that can carry diseases like dengue fever. In Haiti the concern is that cases of cholera will increase; in the wake of 2016’s Hurricane Matthew, the affected parts of Haiti experienced a thanks to a lack of clean drinking water.

Another priority will be to restore power to the millions of people that have been without electricity for several days. At its peak, Irma knocked out power to at least two-thirds of Florida: some 6.5 million homes. Much of Puerto Rico may be without electricity for four months.

The financial cost of Irma across the region, which came just over a week after Hurricane , is still to be calculated.

Flood mapping

In Florida, technology has helped people keep on top of the disaster. A let people use Twitter, Facebook or Telegram to crowdsource information about the depth of water and risks in different areas, allowing them to avoid the most dangerous regions.

As the waters recede and highways become accessible again, people will begin the slow process of returning to their homes. In Florida, 6.3 million people were evacuated. Some of those who find their homes and communities destroyed may decide to move away rather than face the risk of another disaster, says at Brown University in Providence, Rhode Island. Twelve years after Hurricane Katrina, the population of New Orleans is still lower than its pre-Katrina levels.

Those who do return home may find themselves forced to live in flood-prone areas because they cannot afford to pay the higher rents in safer areas. There will be “growing economic inequality” in cities hit by Irma, says Fussell.

In the long-term, Irma could reshape how cities like Miami, which are extremely exposed to rising sea levels and an increasing number of hurricanes, are developed.

“We [have] some very hard questions to ask about which places we should return to mother nature,” says at New York University. “It used to be the case that when something went down, you built it bigger and better than before.” With the threat of climate change, it might be time to rethink that mindset, he says.

Worse to come?

But the US federal government is taking the opposite stance. Just 10 days before Hurricane Harvey hit Texas, President Donald Trump that made it harder for developers to build on floodplains.

Companies looking to redevelop parts of Florida hit by Irma will want reassurance that they can get flood insurance, says at the University of Southern California. But out-of-date federal flood insurance maps mean that insurance estimates don’t take into account the increased flooding risk brought about by climate change.

Because of this, developers are likely to end up building on vulnerable areas despite the risks, knowing they will be bailed out by the government’s if disaster strikes again. Trump’s government, it seems, is already creating the conditions for the next disaster. “They’ve had more than enough chances to change their minds on this,” Lakoff says.

“I think Americans should be worrying as much about the planning disasters and wasted money that will happen now as they are about the next acute disaster,” says Klinenberg.

Posture right

Heavy lifters