Visits: 2
Hi. I’m Jamie Condliffe. Greetings from London. Here’s a look at the week’s tech news:
Machines are more than ever controlled by software, not humans. Occasionally it goes fatally wrong.
On March 10, 157 people died when an Ethiopian Airlines Boeing 737 Max 8 jet crashed. Five months earlier, another crash of the same model of airplane killed 189 people. There are indications that software intended to prevent the jets from stalling may have played a role in both accidents.
Reporting by The New York Times suggests that the software didn’t receive a detailed review by the Federal Aviation Administration before it entered use: Under new rules, the agency delegated much of the responsibility to Boeing. If the software was at fault, and the problem did slip through the regulatory net, it raises questions about how safety-critical technology is vetted.
Those questions will become more important over the next few years.
A year ago, an Arizona woman was struck and killed by one of Uber’s autonomous cars. The vehicle’s autonomy systems failed to brake, as did the safety driver behind the wheel.
Companies like Uber and Waymo, along with most of the auto industry, expect autonomous cars to proliferate over the next decade. Such advances aren’t limited to cars — they will further automate everything from air travel to food delivery. They are built on technologies, like artificial intelligence, that will make split-second decisions for humans.
The Boeing software was designed to perform a simple task: Nudge the airplane’s nose down based on sensor readings if a stall was anticipated. As harder tasks are handed to software, the stakes rise.
“As you create more advanced A.I. systems, the harm that can result from them failing can be really large,” said Jade Leung, a researcher at Oxford University’s Center for the Governance of Artificial Intelligence.
But increasing the complexity of systems makes checking them more difficult. Hardware, from chips to special sensors, can be difficult to test. And it can be difficult for humans to understand how some A.I. algorithms make decisions.
Ms. Leung said regulators needed to be more aware of tail-end risks — the highly unlikely but catastrophic events that could occur if something malfunctioned. That might mean introducing more conservative rules that relax as technology matures, ideally developed in tandem with technologists who understand deeply how the systems work.
“Verifying the performance and safety of software is a really, really hard technical challenge,” Ms. Leung said. Nonetheless, it’s a challenge that has to be addressed.
Big Tech’s new hustle: Cloud gaming
When a company spends billions on world-leading digital infrastructure, it naturally wants to wring every last cent of revenue out of it. That’s partly what is driving Google’s new Stadia gaming service, announced on Tuesday.
Google’s pitch is straightforward: Think of it as Netflix for gaming. As long as they have a fast internet connection, users can pay a subscription to play high-definition games, akin to what they’d find on current top-end consoles, on any computer, phone or tablet.
The company’s promise: that its cloud infrastructure makes that achievable. It will add racks of gaming-specific chips to existing server farms to essentially give people an on-demand, remote gaming computer. And Google officials believe that since most users are now so close to its pervasive hardware, lag won’t be a problem — an issue that held back earlier game streaming platforms, like the now-defunct OnLive.
Google is not alone in the push into what some people see as the future of gaming. Microsoft had already announced that it planned to offer a trial of a similar service for Xbox consoles, computers and mobile devices this year. Amazon, which owns the game-watching service Twitch, is widely believed to be planning something similar, built on its Amazon Web Services cloud infrastructure.
Those three companies happen to be the world’s largest cloud providers. It’s not surprising that they’re enamored of the idea of taking a slice of the $135 billion gaming industry, when all it could take is the flex of an existing muscle.
The U.S. struggles to supercompute
About $500 million should buy a lot of computer. This past week, we found out how much. Writing for The Times, Don Clark explained what the Department of Energy would get for dropping that sum on a supercomputer:
Lab officials predict it will be the first American machine to reach a milestone called “exascale” performance, surpassing a quintillion calculations per second. That’s roughly seven times the speed rating of the most powerful system built to date.
The device, called Aurora, will be used to figure out everything from how drugs work to the impact of climate change. It’s also a useful indicator of the nation’s competitiveness in science and technology — or, at this point, whether it’s leading or lagging behind China. On that front, Mr. Clark reports that it has been a mixed bag for the United States:
An IBM system called Summit, built for the Oak Ridge National Laboratory in Tennessee, took back the No. 1 position last year on a twice-yearly ranking of the world’s 500 most powerful systems — a spot held by China for five years. But China leads by another key measure: It accounted for 227 systems on the Top 500 list, compared with 109 for the United States.
China is expected to have its own exascale supercomputer running as soon as 2020 — a full year before Aurora boots up.
And some stories you shouldn’t miss
Google received its third antitrust fine from the European Union since 2017. This one, for 1.5 billion euros, or about $1.7 billion, was for imposing unfair terms on the search service it offers to other websites.
Facebook will stop targeting some of its ads. It will no longer allow advertising of housing, jobs or credit to be aimed at those of a certain race, gender or age group.
The Pentagon’s giant cloud contract has a one-man holdup. Deap Ubhi, a little-known entrepreneur, is at the center of a legal battle between Amazon and Oracle over the $10 billion project.
Take a look at an early iPhone prototype. It’s red, about the size of an old computer motherboard and helped engineers to program the first breakthrough smartphone.
A.I. researchers could give computers a little more credit. Rich Sutton, a pioneer of some of today’s most effective A.I. techniques, argues that a “bitter lesson” in artificial intelligence is that “the only thing that matters in the long run is the leveraging of computation.”
Why is there still relatively little tech regulation? “Lawmakers are reluctant to disrupt the enormous wealth creation machine that technology has turned out to be,” according to the security expert Bruce Schneier.
Screen sharing can be an easy route to professional humiliation. But don’t worry, the aftermath usually isn’t as bad as you might think. (Even so, here are some tips for avoiding future catastrophes.)
Jamie Condliffe is editor of the DealBook newsletter. He also writes the weekly Bits newsletter. Follow him on Twitter here: @jme_c.
