Hi, I’m Jamie Condliffe. Greetings from London. Here’s a look at the week’s tech news:
In the quest to clean up the web, the gray area between good and bad will be hard to handle.
As the Notre-Dame cathedral in Paris burned last Monday, a YouTube effort to fight misinformation failed. The platform’s automated fact-checking feature — a box that shows facts to help viewers contextualize footage — incorrectly displayed information about the Sept. 11, 2001, attacks alongside live streams of the fire.
“These panels are triggered algorithmically, and our systems sometimes make the wrong call,” YouTube said in a statement. The company said it had no further details about what went wrong.
The failure raises a question: If a platform can’t provide facts reliably, how can we trust it to take down bad content correctly?
Facebook, YouTube and other Big Tech companies say malign content will have to be policed automatically, because it’s too big a job for humans. But algorithms will struggle against our subjectivity.
This is a hard problem. In a profile by my colleague Daisuke Wakabayashi, YouTube’s chief executive, Susan Wojcicki, said borderline content — material that is potentially harmful but does not break rules — was particularly hard to police.
Rasmus Nielsen, a professor of political communication at Oxford University who studies misinformation, said, “The problem will never be solved, if solving it means getting rid of all the bad stuff, because we can’t agree on what the bad stuff is.”
Accept that, and a hard question follows. “Knowing that things won’t be perfect, what do we feel is most desirable?” Mr. Nielsen asked. “A system that errs on the side of caution, or one that errs on the side of being permissive?”
Decisions on that appear set to be shaped by regulation, such as draft legislation approved by the European Union on Wednesday that would require platforms to take down terrorist content within one hour of notification from the authorities, or this month’s proposals from Britain to fine tech companies if they don’t remove “harmful” content quickly.
This quickly descends into the thorny issue of potentially infringing rights of free speech to ensure that platforms remain clean. Lawmakers will have to wrestle with what kinds of algorithms they want deployed to enforce their regulation, and there is no easy solution.
Apple’s big 5G backdown
For years, Apple and Qualcomm were locked in bitter patent battles across three continents. The iPhone maker sought $27 billion in damages; its former chip supplier wanted $15 billion.
My colleagues Don Clark and Daisuke Wakabayashi reported on Wednesday that the companies had agreed to dismiss their litigation. The settlement includes a multiyear agreement for Qualcomm to supply chips to Apple, and an undisclosed one-time payment to Qualcomm from Apple.
Why the change? Playing a big role were two small characters: 5G.
Both companies care deeply about it. As The Wall Street Journal noted, Qualcomm “trained its focus on the future promises of 5G” for more than two years, concentrating its efforts on developing standards and new chips that enable smartphones to access the networks. The new 5G technology will provide far faster data speeds, and is expected to be widely adopted in the coming years.
Apple has worried about its source for those chips. While it was locked in battle with Qualcomm, the chip maker’s components — which are already fit for inclusion in phones — were not an option. That left some unpalatable options:
■ Go with Intel, its 4G supplier. But rumors of delays at Intel suggested that could have pushed back the release of a 5G iPhone by a year.
■ Team up with Huawei, which said Tuesday that it was “open” to selling 5G chips to Apple. But that could look bad for the iPhone maker, given Washington’s concerns about national security risks posed by Huawei.
■ Build its own. But Apple’s chip design team is relatively new, so that could have taken a long time.
Against that backdrop, the desire for $27 billion in damages faltered.
Apple will probably still want to work with another modem supplier, but for now the situation appears to be mutually beneficial for the two companies.
Less so for Intel, perhaps: Just hours after the settlement was announced, the company said it would no longer work on 5G smartphone chips, citing “no clear path to profitability.”
The gamer with 45,000 years of experience
Last weekend, OpenAI, an artificial intelligence research organization, gave one of its algorithms a tough challenge: It had to play the strategy video game Dota 2 against a team of the world’s most proficient human players.
Dota 2 is difficult. Two teams of five players battle each other, but have incomplete information (they can’t see the whole game arena), lots of complexity (each player can make dozens of actions across a large playing area) and must think in advance (games last around 45 minutes).
OpenAI’s algorithm uses five so-called artificial neural networks, each controlling a player, that learn by playing games against themselves. The ones that played last weekend boasted the equivalent of 45,000 years of total game play experience.
“It’s almost an evolutionary scale,” Greg Brockman, the chief technology officer of OpenAI, said. “It’s perfectly honed for its environment.”
How did the algorithms do? They won, taking the first two games in a best-of-three competition.
“It’s a significant milestone,” Shimon Whiteson, a professor of computer science at Oxford University, said. And it “shows that we still haven’t reached the limits of what we can achieve when using existing algorithms,” he added.
OpenAI faces another test this weekend, when it takes on teams worldwide over the internet. Mr. Brockman says there is a “50-50 shot that we’ll win every game.”
But the news highlights a problem facing A.I.: its astonishing demand for computer power to master a single task. The 45,000 years of experience took 10 months to acquire on thousands of powerful computer processors, and the algorithms would need other training to master a new skill. Overcoming that is one of the biggest challenges in A.I.
And some stories you shouldn’t miss
■ China is using A.I. to profile a minority group. Hundreds of thousands of face scans and new algorithms have enabled the government to track members of a largely Muslim group.
■ Google’s huge data cache is being used as a dragnet. Its user tracking is being employed by investigators to find suspects and witnesses near crimes, but it could snare the innocent.
■ Insurance doesn’t necessarily cover cyberattacks. Big companies thought they were covered for losses resulting from the 2017 NotPetya attack. They were wrong.
■ Samsung’s $2,000 foldable-screen smartphone is already breaking. Some gadget reviewers found that its screen failed after several days of use.
■ Silicon Valley is giving second chances to #MeToo culprits. Executives fired for misconduct are finding it relatively easy to get new jobs, reports BuzzFeed News.
■ Apple’s gaming service has a big budget. The FT reports that it’s spending $500 million on titles for its launch.
■ Want to relive Facebook’s troubled 15 months? Wired’s 12,000-word essay will help.
■ Google quietly disbanded another ethics board. A panel overseeing the ethics of its health care A.I. split up over tensions about information access and independence.
■ Alibaba’s co-founder celebrated a 72-hour workweek. Jack Ma said that working from 9 a.m. until 9 p.m., six days a week, was a “huge blessing.”