Eye on A.I.— Why Hardware Is So Important to A.I.’s Future
Using artificial intelligence isn’t just about software. Companies are quickly realizing that the hardware it runs on and is trained on is also critical.
Take the example of Google, a leader in A.I. Last week, during its annual developer conference in Mountain View, Calif., the company debuted its new line of Internet-connected home products that are all linked to its voice-controlled Google Assistant.
Executives bragged on stage about how one of the new devices, the Nest Hub Max, can use its camera to immediately recognize individual family members. In that way, the technology can better respond to requests like playing music from someone’s song lists or showing photos that they are more likely to be interested in.
In the past, Google marketed its A.I. as equally compatible with any Android device. But increasingly, it’s tailoring its A.I. to its own hardware so that it operates more smoothly.
“The pendulum swings both ways,” Rishi Chandra, Google’s head of Home and Nest products told Fortune, explaining Google’s philosophical shift.
Facebook and Amazon are following a similar strategy to Google in their growing focus on customizing A.I to their hardware. They’re designing their own Internet-connected devices and data center chips tailored for their own machine-learning tasks.
What’s the lesson here for businesses that are trying to incorporate A.I. into their operations? Using A.I. isn’t as simple as feeding numbers into fancy software and waiting for a result that will lead to lots of profits. In fact, leading A.I. companies have deployed legions of employees to work on complex software while also fine tuning the hardware that runs and trains it. Some companies are also building their own data centers to handle some of the work, as Walmart recently did inside its futuristic store in Levittown, N.Y., rather than using a cloud service.
It’s all a huge and expensive undertaking. Don’t believe anyone who says differently.
Apple and SAP are A.I. buddies. Under an expanded partnership, German business software giant SAP is updating its software development toolkit for building iPhone and iPad apps to include support for Apple’s Core ML A.I. tools. The agreement appears to be similar to 2018 deal between Apple and IBM that linked IBM’s Watson data crunching service with Apple’s Core ML technology.
Come play with Cisco’s voice. Cisco has made its MindMeld digital assistant and voice technology available in open source so other companies and developers can modify and improve it. Cisco bought MindMeld in 2017 for $125 million to enhance its work-collaboration products.
Ericsson plants A.I. flag in Canada. Networking giant Ericsson debuted an A.I. research hub in Montreal and plans to hire 30 data scientists, machine learning engineers, and other software developers to work in the new unit. Several other big tech companies like Google, Facebook, and Microsoft also have A.I. research labs in Montreal, a leading city for deep learning talent.
Facebook has to “label” data somehow. Facebook uses Indian contract workers from IT firm Wipro, among other consulting groups, to hand-label people’s photos and other content in order to train its A.I. systems, Reuters reported. Facebook told the news service that it tells its users in its data policy that the company uses people’s information to “improve their experience.” But, the report noted that users are “not offered the chance to opt out of their data being labeled.”
Companies can’t do machine learning well if they don’t understand or properly track all of their corporate data, according to tech news site TechRepublic. Ryan Johnson, the data science head of education tech company GoGuardian, told the publication that when it comes to data crunching, “A lot of companies are putting the cart before the horse there.”
Ghost Locomotion, a startup specializing in self-driving car technology, has named David Purdy as chief scientist. Purdy was previously a senior data science manager for Uber’s safety data science team.
Data analytics company Alteryx hired Alan Jacobson as chief data and analytics officer. Jacobson was formerly the director of global analytics for Ford.
United Kingdom postal-service company Royal Mail has chosen Kat James as its new head of data science. James replaces Ben Dias, who is now the data science director of airline company EasyJet.
A.I. as a census tool. Researchers from Stanford University, Dartmouth College, and the World Bank published a paper about using deep learning to estimate the population in rural areas of India. The researchers trained their neural networks on satellite imagery taken from above rural villages and said that their systems performed better than traditional efforts, and that there “may still have room for improvement if images with higher resolution are available.”
A.I.’s privacy problems. Researchers from organizations like U.C. Berkeley, Duke University, and Chinese retail giant JD.com, Inc. published a paper exploring potential privacy problems that can occur with reinforcement learning—in which computers learn through trial and error. The researchers discovered that they could learn certain details within so-called simulated “training environments” (like a warehouse floor where a robot learned to navigate around) that are crucial for reinforcement learning.
A.I. as a “climate change” issue. The Financial Times looks at how countries racing to be the world’s leader in A.I. could result in governments and companies failing to consider some of A.I.’s ethical dilemmas, such as increasing existing societal biases. Deep learning “godfather” Yoshua Bengio suggests an “international order” for A.I. that would involve governments creating the norms. “Just like with climate change, we have to stigmatise (sic) countries which don’t want to play by the rules necessary for the benefit of the whole planet,” he told the publication.