|
|
|
|
|
Facebook Wrestles With Hate-Speech Filters; Virtual Metaverse Gets Real
|
|
|
|
|
|
Welcome back. Hype around artificial intelligence is nothing new. But now and then, claims of what AI can do are put under a microscope and shown to be, well, just that: hype. That appears to be the case with Facebook’s automated screening tools, designed to filter out hate speech, excessive violence and other posts that violate its content rules.
The social-media giant’s internal research, reviewed by The Wall Street Journal, shows the AI-powered system is doing a poor job. In one example, the Journal found, AI labeled a video of a carwash as a first-person shooter video. In another, it mistook a video of a shooting for a car crash. It also confused fighting roosters and carwashes, according to the documents.
Training algorithmic models to correctly identify unacceptable posts isn’t easy. But in the interests of avoiding “over-enforcement,” or false positives, Facebook might also be pushing its engineers to develop weaker models, people familiar with the matter told Journal reporters. That’s not a good look for the company, but it might be a less bitter pill to swallow than admitting AI isn’t all it’s cracked up to be.
|
|
|
|
|
|
PHOTO: JOSH EDELSON/AGENCE FRANCE-PRESSE/GETTY IMAGES
|
|
|
|
|
Facebook Inc. engineers in internal documents say they are worried that algorithmic screening tools meant to keep hate speech, excessively violent videos and other forbidden posts off the site might not be up to the task, at a time when the social-media giant is relying less on human reviewers, The Wall Street Journal reports.
How it works. Screening algorithms, known as classifiers, are designed to comb through billions of Facebook posts looking for items that might match the company’s definitions of content that violates its rules.
What’s going wrong. The systems removed posts that generated just 3% to 5% of the views of hate speech on the platform, and as little as 0.6% of all content that violated Facebook’s policies against violence and incitement, Facebook researchers noted in March.
Why it matters. CEO Mark Zuckerberg previously said he expected Facebook would use AI to detect “the vast majority of problematic content” by the end of 2019.
|
|
|
|
|
Many of Cynthia Rudin’s methods rely on creating predictive models that make decisions based on concise and narrowly defined rules.
PHOTO: LES TODD/ LKT PHOTOGRAPHY
|
|
|
|
Decisive Victory for Duke Professor
|
|
Cynthia Rudin, who teaches computer science and engineering at Duke University, was recently awarded a $1 million grant by the Association for the Advancement of Artificial Intelligence for her work developing healthcare and other machine-learning systems that make decisions in ways people can easily understand, the WSJ’s John McCormick reports.
Black box. Because AI systems crunch massive amounts of data, making split-second decisions at every turn, it isn’t always clear how these models reach conclusions: “For high stakes decisions, you really want to know what’s going on,” Dr. Rudin says.
Greater clarity. Dr. Rubin develops predictive models that make decisions based on concise and narrowly defined rules, such as a point-based system to predict which brain-injury patients are at risk of having a seizure that relies on just a few variables.
|
|
|
|
|
“The problem is that we do not and possibly never will have a model that captures even a majority of integrity harms, particularly in sensitive areas.”
|
—A Facebook senior engineer and research scientist
|
|
|
|
|
|
|
|
|
Within the metaverse, students from anywhere in the world might meet up for a class trip to ancient Rome.
ILLUSTRATION: ANTON EGOROV FOR THE WALL STREET JOURNAL
|
|
|
|
|
Tech-company executives foresee the creation of a virtual metaverse—a vast digital world powered by AI where users represented by personal avatars work, shop, attend classes, pursue hobbies, go to parties and more—with replicas of places in the real world rendered in realistic-looking 3-D, the WSJ's Sarah Needleman reports. Here's what they're saying:
-
Marc Whitten, senior vice president at Unity Software Inc.: “The metaverse is going to be the biggest revolution in computing platforms the world has seen—bigger than the mobile revolution, bigger than the web revolution.”
-
Mark Zuckerberg, CEO of Facebook: “In addition to being the next generation of the internet, the metaverse is also going to be the next chapter for us as a company.”
-
Daren Tsui, CEO of Together Labs Inc.: “The avatar experience will feel so real that you can hardly tell the difference between a virtual meeting and a physical meeting. And the virtual experience will be better.”
-
John Egan, CEO of L’Atelier BNP Paribas:“The potential is unending.”
|
|
|
|
3.6%
|
The growth predicted for IT budgets in 2022, an acceleration from 2% growth this year—the sharpest annual increase in more than a decade, according to technology research and advisory firm Gartner Inc.
|
|
|
|
|
WSJ Pro AI Extra: War of Words
|
|
|
Raj Iyer, the U.S. Army’s chief information officer, is leading a counterattack against recent claims by a former Pentagon software chief that China has already won the race to develop AI. Here's why he calls those claims "absolutely not true," according to articles by National Defense and Breaking Defense:
Bigger budget. Mr. Iyer says the Army has a $15 billion IT budget, which it has leveraged to install advanced AI spread across dozens of key systems.
More partners. The Pentagon, he says, maintains a “tight integration” with industry and its global coalition partners, sharing trade intelligence information and other insights. By contrast, China is “operating in a vacuum,” Mr. Iyer says.
Off limits. One area where Chinese authorities excel at is leveraging AI to control its people, Mr. Iyer says, adding that “we don’t do that in the United States.”
|
|
|
|
|
Google's Sundar Pichai tells WSJ Tech Live that although his company doesn't provide its core services in China, it is investing deeply in AI and quantum computing to stay ahead.
|
|
|
|
Google chief calls for government help. Sundar Pichai, chief executive of Google and parent company Alphabet Inc., says the U.S. government should take a more active role in encouraging homegrown innovation in AI and other capabilities, as well as global cybersecurity efforts, via policies and funding. (The Wall Street Journal)
Facebook hiring for ‘metaverse.’ Facebook Inc. says it is planning to create 10,000 jobs in the European Union over the next five years to build an AI-powered virtual world, where users interact as avatars in digital stores, schools, offices and other simulated locations. (The Wall Street Journal)
Informatica sets IPO terms. Informatica Inc., a data management software company, is looking to raise up to $928 million in its return to public markets, nearly six years after being taken private. (MarketWatch)
TSMC plans Japan plant. Taiwan Semiconductor Manufacturing Corp. says it plans to build a new factory in Japan, citing the global shortage of smart chips and other semiconductors. (Associated Press)
Fintech startup sees big funding round. N26, a German online bank with an AI-enabled customer-service platform, said it completed a $900 million fundraising round that values the company at more than $9 billion. (Reuters)
Accidents waiting to happen. Researchers are using a deep model trained on historical crash data, road maps, satellite imagery, GPS and other data sources to generate high-resolution crash maps aimed at predicting accidents before they happen. (SciTechDaily)
Name that band. Software developers behind This Band, a popular Twitter account, are using an AI model to generate names and album artwork for fake heavy metal bands, including Coffinlord, Blooddeathbeast and Gorewolf. (LoudWire)
|
|
|
Information technology budgets next year are expected to rise at a faster rate than they have in more than 10 years, with chief information officers deploying tools to support more modular and adaptable business models. (The Wall Street Journal)
Legislation to curb the influence of big technology companies, including putting new restrictions on online content, is starting to gain traction in Congress as lawmakers narrow their targets and seek to build on public attention. (The Wall Street Journal)
A trading frenzy in shares of GameStop Corp. earlier this year should lead regulators to consider whether “game-like” brokerage apps are encouraging investors to trade too much, the Securities and Exchange Commission says. (The Wall Street Journal)
Real-estate software and data firm VTS has agreed to acquire an app company that aims to simplify office life as more workers head back to their desks. (The Wall Street Journal)
|
|
|
|
|
|