Amazon ran a commercial on this year's Super Bowl that pretended its digital assistant Alexa had temporarily lost her voice. It featured celebrities like Rebel Wilson, Cardi B and even the company's chief executive, Jeff Bezos. While the ad riffed on what Alexa can say to users, the more intriguing question may be what she and other digital assistants can hear -- especially as more people bring smart speakers into their homes. Amazon and Google, the leading sellers of such devices, say the assistants record and process audio only after users trigger them by pushing a button or uttering a phrase like "Hey, Alexa" or "O.K., Google." But each company has filed patent applications, many of them still under consideration, that outline an array of possibilities for how devices like these could monitor more of what users say and do. That information could then be used to identify a person's desires or interests, which could be mined for ads and product recommendations.

For years, Swami Sivasubramanian's wife has wanted to get a look at the bears that come out of the woods on summer nights to plunder the trash cans at their suburban Seattle home. So over the Christmas break, Sivasubramanian, the head of Amazon's AI division, began rigging up a system to let her do just that. So far he has designed a computer model that can train itself to identify bears--and ignore raccoons, dogs, and late-night joggers. He did it using an Amazon cloud service called SageMaker, a machine-learning product designed for app developers who know nothing about machine learning. Next, he'll install Amazon's new DeepLens wireless video camera on his garage.

It is hard to imagine a more scathing indictment of our ability to read another's thoughts and intentions than our inability to predict Donald Trump's next move. From the gross pre-election misjudgments to postelection bafflement, the best pundits are at a loss to accurately anticipate his response to matters like North Korean military aggressiveness or his moment-by-moment political gyrations and opinion reversals. Labeling Trump a narcissist, psychopath, megalomaniac or attention-impaired, or all of the above, might feel explanatory, but even when armed with the best psychoanalytic insights, we have no idea what he will do when presented with a new or unforeseen circumstance. If conventional psychology isn't up to the task, perhaps we should step back and consider a tantalizing sci-fi alternative -- that Trump doesn't operate within conventional human cognitive constraints, but rather is a new life form, a rudimentary artificial intelligence-based learning machine. When we strip away all moral, ethical and ideological considerations from his decisions and see them strictly in the light of machine learning, his behavior makes perfect sense. Consider how deep learning occurs in neural networks such as Google's Deep Mind or IBM's Deep Blue and Watson.

An American military drone strike over the weekend in southern Libya killed a top recruiter and logistics specialist for Al Qaeda's branch in northwest Africa, the Pentagon said on Wednesday, and a senior military official warned of more attacks on extremists there. The military's Africa Command said in a statement that the attack killed two militants, one of whom was identified as Musa Abu Dawud, a high-ranking official in Al Qaeda in the Islamic Maghreb, known as AQIM. Mr. Dawud trained Qaeda recruits in Libya for strike operations in the region, and provided logistics, money and weapons that enabled the group to threaten and attack American and Western interests, the military statement said. Until now, the Pentagon had focused its counterterrorism strikes in Libya -- eight since President Trump took office -- almost exclusively on Islamic State fighters and operatives farther north. Over several months in 2016, the military conducted nearly 500 airstrikes in the coastal city of Surt to destroy the Islamic State's stronghold there.

A look under the hood of any major search, commerce, or social-networking site today will reveal a profusion of "deep-learning" algorithms. Over the past decade, these powerful artificial intelligence (AI) tools have been increasingly and successfully applied to image analysis, speech recognition, translation, and many other tasks. Indeed, the computational and power requirements of these algorithms now constitute a major and still-growing fraction of datacenter demand. Designers often offload much of the highly parallel calculations to commercial hardware, especially graphics-processing units (GPUs) originally developed for rapid image rendering. These chips are especially well-suited to the computationally intensive "training" phase, which tunes system parameters using many validated examples.

With some limited exceptions, robots are terrible at doing almost everything that humans take for granted. For people who work with robots, this is normal and expected, but for everyone else, it's not immediately clear just how terrible robots are, especially if the robot in question looks human-like enough to generate expectations of human-like capability. Bimanual mobile manipulators like PR2 are particularly bad, because with heads and bodies and arms, it's easy to look at them and think that they should have no problem doing all kinds of things. And then, of course, comes the inevitable disappointment when you realize that (among other things) round doorknobs make for an impassable obstacle. At the ACM/IEEE International Conference on Human Robot Interaction (HRI) earlier this month, researchers from Cornell and UC Berkeley presented some work on how robots can effectively express themselves when they're incapable of doing a task.

More than a week after a self-driving Uber hit and killed a woman crossing the street in Tempe, Arizona, the company is facing the consequences. Today, on the orders of Arizona governor Doug Ducey, the Arizona Department of Transportation commanded Uber to suspend its testing of autonomous and highly automated vehicles on the state's roadways. It's an obvious setback for Uber's embattled self-driving program, which does much of its testing in Arizona, but the kibosh job also signals how local and state politicians elsewhere will be looking to control a new technology that comes with the promise of great safety and economic benefits--but also the potential to destroy jobs and, when it fails, to kill. And it's an unexpected blow from Ducey, who until the crash had championed the technology and encouraged companies like Uber to do their testing work in Arizona, where virtually no rules dictate what they can do where and when, and where they face no requirements to report or disclose anything about their programs, including crashes. In 2015, Ducey signed an executive order telling all state agencies to "undertake any necessary steps to support the testing and operation of self-driving cars."

Passengers have been carried across London by the first self-driving train on a mainline railway in the UK. Govia Thameslink Railway promised that it would not spell the beginning of the end for drivers, who remain responsible for safety and can take control of the train at any time. Automated operation using a new digital signalling system will allow many more trains to pass through the congested tracks between St Pancras and Blackfriars in central London, giving space for an additional 60,000 passengers to commute at peak hours daily. After almost 18 months of testing, the first commuter train in automatic operation was Monday's 9.46am Thameslink service from Peterborough to Horsham. Shortly after 11.08am, the driver, Howard Weir, pressed the yellow button in the cab that allowed the train's computer to do the driving between St Pancras and Blackfriars.

The horizontal lines indicate the ℓ1.

Google's machine-learning tool is being used to detect and combat illegal deforestation The news: Rainforest Connection, a San Francisco nonprofit, has developed a cheap, rigorous acoustic monitoring system made from modified cell phones and solar panels. An app on the so-called Guardian devices, which can be hidden in trees throughout forests, continuously listens for the telltale signs of illegal logging and animal poaching. On March 21, the organization announced that it will be using Google's TensorFlow, a free tool that makes it simpler for other companies and groups to develop machine-learning software (see "Google stakes its future on a piece of software"). Rainforest Connection says it will enable the organization to more accurately detect troubling sounds in the uploaded audio, such as chainsaws, vehicles, and gunshots. Deforestation reduces biodiversity, increases erosion, and promotes desertification.