|Google, Amazon Find Not Everyone Is Ready for AI|
Author: Tom Simonite
December 1, 2107
Executives at ascendant tech titans like Amazon and Google tend to look down on their predecessor IBM. The fading giant of Armonk, New York, once sustained itself inventing and selling cutting-edge technology, but now leans heavily on consulting. Renting out people to help other companies with tech projects is a messier and less scalable business than selling computing power on a distant cloud server, and leaving the customer to do the grunt work.
Yet as Amazon and Google seek greater riches by infusing the world with artificial intelligence, they’ve started their own consulting operations, lending out some of their prized AI talent to customers. The reason: Those other businesses lack the expertise to take advantage of techniques such as machine learning.
Many companies use cloud platforms for tasks like data storage or powering websites and mobile apps. Market leader Amazon and its rivals are now trying to convince their customers to also buy AI services to mine insights from the hordes of data they amass. But AI experts are in short supply, in no small part because big tech companies compete fiercely to hire them.
Amazon launched several new cloud services for tasks such as understanding audio and training machine-learning models at its AWS re:Invent cloud conference in Las Vegas this week. An executive from the NFL came onstage Wednesday to boast how the league tapped Amazon’s machine-learning tools to determine how far players run, and how fast they accelerate.
But the NFL couldn’t do that work itself. It got hands-on help from Amazon’s elite machine-learning experts through a new consulting operation called Amazon ML Solutions Lab. Lab staffers examine a company’s data and systems, brainstorm ideas for how to improve them using AI, and help implement the plans.
AWS made its first big push into AI services last year. Swami Sivasubramanian, who leads AI initiatives at AWS, says the consulting shop was launched in response to requests from customers for help building AI systems. “We consistently heard they wanted to learn from the machine-learning scientists who built these capabilities for Amazon.com,” he says. Companies pay to tap Amazon’s experts, but Sivasubramanian declined to detail the menu they are offered or the prices, saying it varies depending on the project.
Google launched its own consulting AI shop late last year. The Machine Learning Advanced Solutions Lab, as it is called, lets customers such as insurer USAA work on projects with Google AI engineers at a dedicated facility at the company’s campus in Mountain View, California. It also offers a four-week training program to help customers’ engineers improve their AI chops.
That such prized and well-compensated employees are now being put to work for others suggests that selling AI is more complex than executive keynotes imply. “The gating factor is people don’t know how to do this stuff,” says Rob Koplowitz, who tracks cloud AI for Forrester. “There needs to be some hand-holding here in the early stages.”
The hand-holding stage may continue for a while. Amazon’s Sivasubramanian believes it will take a few years before machine-learning expertise is as widely shared as knowledge of distributed systems, the practice of using networked computers to solve problems at the heart of cloud computing.
Google CEO Sundar Pichai said this fall that there are only “a few thousands” of people capable of creating sophisticated machine-learning models. He has a team trying to make machine-learning software create machine-learning software, but it’s so far just a research project.
The expertise shortage upsets the usual dynamic of the cloud market, where Amazon, Google, and others mostly compete on price and technical features. “If you’re a random manufacturing company in the midwest you may have money, but it’s hard to attract a $250,000-a-year Stanford PhD to work for you,” says Diego Oppenheimer, whose Google-backed startup provides tools that help companies deploy machine-learning software. Companies in that situation may be more swayed by an offer of help building AI, than pricing and performance, he says.
Cloud companies have made AI more accessible. Amazon launched a new service that transcribes speech from audio or video this week, for example. A company that wants to transcribe meetings or calls can very easily ship off files to Amazon’s servers and get text back. Amazon and Google both offer services that identify common objects and scenes in photos.
The most powerful use cases for AI aren’t one-size-fits-all. Machine learning software typically is trained to solve a very specific problem. “If I need to figure out how much rust is on my industrial boiler, a cat and dog recognizer is not going to help,” says Chris Nicholson, CEO and cofounder of Skymind, which sells machine-learning tools and has helped organizations including the Department of Homeland Security use them in machine learning projects. Nicholson says that by creating consulting services, Amazon and Google “basically showed the Achilles heel of their business model.”
A Microsoft vice president said at a conference this spring that many cloud AI systems are too complex for many companies to reap the same benefits from machine learning as big tech companies. Microsoft is trying to help customers for its AI services with a suite of online courses marketed as AI School. It was part of a $102 million investment round this summer into Element AI, a startup that will offer AI consulting services.
Amazon launched its own education initiative this week. A new $250 camera called DeepLens is designed to give developers an easy way to learn about machine learning—and Amazon services. Carnegie Mellon University plans to use the device with students, and other colleges are expected to do the same.
Many attendees of Amazon’s conference this week are getting a DeepLens for free. Some started hacking with the device on Wednesday. Sivasubramanian says people with little or no experience with machine learning were soon building copies of the Hotdog detector from the TV show Silicon Valley, or apps that use object and face recognition. “We’re going to make machine learning a normal part of programming,” he says. Until it is, expect to see leading cloud companies aping IBM.