Why business and academia need each other for better A.I.

Take self-driving cars. Advances in neural networks, the software that recognizes and acts on patterns by sifting through huge quantities of data, have let companies like Google’s Waymo and General Motors’s Cruise develop autonomous vehicles that are better than a few years ago.

Still, self-driving cars are years, and possibly decades, from widespread use. The best way to accelerate the needed innovation is cooperation between academics and business, explained Martial Hebert, Carnegie Mellon University’s (CMU) dean of computer science.

Researchers have “200 years of engineering science” to draw from when developing complicated machinery like automobiles. This rich history helps researchers certify and explain how their technologies work. This is important so that when people “take an elevator,” it’s not a mystery to engineers, and by extension, the general public, as to why they are moving up or down, Hebert said. 

But with machine learning, “we have basically none of that,” said Herbert said, creating a big challenge for companies developing self-driving cars.

“How do you validate the performance of a system whose performance depends not just on the correctness of the code or the hardware, but all of the data it used for training,” Herbert said about proving how A.I. systems work. “How do you do that when the data can evolve over time in ways that you cannot predict ahead of time?”

Autonomous vehicles still have trouble navigating through hailstorms, fog, and snow. And researchers still can’t foresee every road condition that can confuse a car’s neural network.

Herbert said that self-driving car companies need universities to create the mathematical tools and computing techniques for testing and certifying their A.I. systems. Likewise, universities need companies to provide them with the real-life data that is required to create these technology development standards and methods. 

He cited a partnership between CMU and the Ford-backed self-driving company Argo AI as a prime example of industry working with academia. Under the partnership, Argo AI lets CMU researchers access its driving data, which will help the university develop A.I. tools and techniques for explaining how self-driving cars work. This kind of technology, although basic, is critical and could be used to test other kinds of A.I. systems, making it a worthy academic pursuit, Herbert explained.

Other CMU corporate partners include Amazon, Apple, Google, Facebook, and German engineering giant Bosch. These companies also have major A.I. and engineering hubs in CMU’s hometown of Pittsburgh, partly to help with their work on A.I. with local schools.

Herbert said such partnerships are for the long term, just like improving A.I. is a never-ending job. “When is computer vision going to be solved?” said Herbert, referring to the technology used to teach computers to see like humans. “The answer is never.” 

Still, not every corporate and academic partnership is as rosy a picture as Herbert paints.

In 2015, Uber poached several professors and researchers from CMU after partnering earlier with it. At the time, Uber was trying to develop self-driving cars, but it longer plans to do so. In December, Uber CEO Dara Khosrowshahi sold the research unit to self-driving car company Aurora; as part of the deal, Uber took a stake in Aurora. Herbert minimized the impact of the raid on his school’s staff by saying that Uber only hired a small number of CMU researchers relative to how many the university now employs.

“Word of our demise was largely exaggerated,” Herbert said.

Jonathan Vanian 
@JonathanVanian
jonathan.vanian@fortune.com

A.I. IN THE NEWS

Qualcomm gobbles some neurons. Qualcomm said it acquired the assets of the startup Twenty Billion Neurons for an undisclosed amount, The San Diego Union-Tribune reported. The startup, which raised about $10 million from Microsoft’s venture capital arm, specializes in computer vision technology and more capable digital assistants. Considering that the startup had less than 20 employees and that its products didn’t really catch on in the marketplace, it appears that Qualcomm acquired the company for its staff and data.

What’s up with IBM Watson? IBM is no longer marketing its Watson data crunching technology as the world-changing A.I. it originally pitched, capable of “tackling cancer and climate change,” according to a New York Times report. The article said that after several setbacks, including a “costly failure” that involved MD Anderson Cancer Center in Houston, IBM is instead pitching Watson as a useful tool to “mainly streamline and automate basic tasks in areas like accounting, payments, technology operations, marketing and customer service.”

OpenAI says goodbye to robots. OpenAI, the high-profile A.I. firm led by entrepreneur and investor Sam Altman, has “disbanded its robotics team after years of research,” according to a report by tech publication VentureBeat. The article said that OpenAI is pursuing other “domains, where data is more readily available,” implying that robotics research is a more difficult task than the firm originally believed. OpenAI is now focused on A.I. services that “have obvious and immediate business applications,” the report said.

The ethics of A.I. as experienced through Anthony Bourdain. The filmmakers behind a recent documentary about the late Anthony Bourdain hired a software company to create an A.I.-generated voice of the famed chef and writer, according to a report by The New Yorker. The publication has since published another article about the ethics of creating a synthetic version of Bourdain’s voice for the documentary, a fact that the filmmakers originally did not disclose. From the article: Setting aside questions of technological ethics, the artificial voice may trouble people in large part because of the close connection they feel with Bourdain—what psychologists call a parasocial relationship.

A.I. supercomputing scam. The U.S. Securities and Exchange Commission shut down the Las Vegas-based Profit Connect Wealth Services over allegations the company, operated by a mother and son duo, had swindled investors. While the company claimed to use an “artificial intelligence supercomputer” to help trade securities and cryptocurrencies on behalf of investors, it actually “misused investor money” by depositing money into one of the owner’s bank accounts and “paying millions of dollars to promoters, and making Ponzi-like payments to other investors.” The company claimed on its website that in addition to its supercomputers, dubbed Orwell and Tesla, it also used the services of IBM Watson, Google, and Amazon to help make trades.

 

EYE ON A.I. TALENT

Apple vice president of technology Kevin Lynch will help the iPhone maker’s various automobile projects, which could include electric vehicles or self-driving cars, Business Insider reported, citing unnamed sources. Lynch helped oversee the development of Apple Watch and was once the chief technology officer for Adobe.

The New York Times Company picked Jason Sobel to be its CTO. Sobel was previously the head of infrastructure at Airbnb and was once a Facebook engineering director.

Twitch, the Amazon-owned video game streaming service, hired Christine Weber to be the firm’s CTO, according to The Hollywood Reporter. Weber was previously the interim CTO of telecommunications company Liberty Latin America and an engineering executive at Sling TV.

EYE ON A.I. RESEARCH

Deep learning comes to charts and tables. Researchers from Hong Kong University of Science and Technology and Microsoft published a paper in the IEEE Transactions on Visualization and Computer Graphics journal about using deep learning to automatically create visuals and so-called dashboards from complicated data sets. The researchers said that A.I. can help remove some of the manual grunt work required to create sophisticated visualizations based on data sets that highlight important information.

From the paper: We contribute a deep-learning-based method that assists in designing analytical dashboards for analyzing a data table. Given a data table, data workers usually need to experience a tedious and time-consuming process to select meaningful combinations of data columns for creating charts. This process is further complicated by the needs of creating dashboards composed of multiple views that unveil different perspectives of data. Existing automated approaches for recommending multiple-view visualizations mainly build on manually crafted design rules, producing sub-optimal or irrelevant suggestions. To address this gap, we present a deep learning approach for selecting data columns and recommending multiple charts.

 

FORTUNE ON A.I.

DeepMind unveils how it solved a 50-year-old scientific challenge that could speed drug discovery—By Jeremy Kahn

The facial recognition technology nightmare—By Jonathan Vanian

Is China’s Bitcoin crackdown cutting mining’s emissions—or shifting them somewhere else?—By Sophie Mellor

How low-code software paved the way for a flood of new custom apps—By Fortune Editors

Will Beijing’s tech crackdown be a windfall for Hong Kong?—By Clay Chandler

BRAIN FOOD

Come here the deepfakes sing. Musician Holly Herndon has debuted a tool that people can use to submit audio clips to have them transformed into music that’s sung in Herndon’s voice, reported music publication Pitchfork. The tool is powered by the same neural technology that’s used to create deep fakes, which are realistic but synthetic video and audio clips.

From Herndon’s post explaining her project: A Voice Model is a deep neural network that can generate raw audio of an individual voice. The network is trained on recorded speech and singing from the target voice, and can be interacted with in various ways, from text-to-speech applications to more complex interactions such as audio style transfer, where audio from one voice can be converted to resemble the target voice, a kind of vocal puppetry 🤖

The recent introduction of projects like DeepMind’s Wavenet, Google’s Tacotron and others have advanced the field of voice generation sufficient to make me confident that generating convincing spoken and sung voices will soon become standard practice for artists and other creatives, as presaged by the popularity of celebrity vocal deep fakes already found all over YouTube.