October 22, 2015      

Large and small technology companies worldwide are racing to develop artificial intelligence through machine vision, speech recognition, and the algorithms to understand and use massive amounts of data. As a leader in AI research, Google Inc. is aggressively pursuing developments in all of these technologies.

Google has taken a minority stake in AI company Mobvoi as part of a $75 million Series C funding round. Beijing-based Mobvoi (short for “mobile intelligence voice search”) is developing voice-controlled software and hardware.

Google’s search engine, e-mail, and mapping services are blocked in China, so the company has been looking for ways to expand in that large user market. Five years ago, Google withdrew its operations from China because of the censorship dispute. Google’s parent company recently named itself Alphabet Inc.

Mobvoi co-founders and CEO Zhifei Li and Chief Technology Officer Mike Lei were previously researchers at Google U.S. The company’s Chumenwenwen is a virtual concierge similar to Google Now or Apple’s Siri and gives users access to services from more than 100 content providers. The technology is also similar to that used in emerging social robots.

Google had already partnered with Modvoi for voice-powered search in Chinese on the Moto 360, a smartwatch running the Android Wear operating system. The Moto 360 is a rival to Apple’s iWatch and Mobvoi’s own Ticwatch.

TechCrunch speculated that Google’s investment was about $60 million, but Mobvoi responded that the value was less but declined to specify an amount. Mobvoi will retain a controlling stake and is valued at $300 million.

Mobvoi plans to use the latest round of funding for its AI and robotics research, and it is working on voice-search capabilities for vehicles and a household robot for the Chinese market, according to co-founder Li Yuanyuan.

Alphabet advances on multiple fronts
Google’s YouTube unit is using “deep neural networks” to make the algorithm that picks thumbnail images for videos more relevant. Rival Baidu’s Minwa supercomputer recently beat Google and Microsoft’s records in image recognition.

In addition, Google’s voice-recognition systems, which Google Now uses, have switched from deep neural networks to recurrent neural networks to create models that better understand human speech. Its “chatbot” is capable of having conversations on abstract questions reminiscent of (but far from passing) a Turing test.

Google's Deep Dream software generates hallucinatory images.

Google’s Deep Dream combines deep learning and machine vision.

Mac users can upload images to the Dreamscope app in the RealMac beta software, and Google’s Deep Dream software will create a surrealistic version of it to rival the visions of Hieronymus Bosch.

These efforts connect machine vision, human-computer interaction, and deep learning, which have applications across a wide range of robotics.

Thanks to these neural networks, big data in the cloud, and testing against real-world problems, AI is coming closer to reality, said Alphabet Chairman Eric Schmidt in a BBC op-ed.

Geoff Hinton, a professor at the University of Toronto who has worked with Google, goes further still. He said he believes that a new “thought vector” algorithm could lead to human-level abilities to reason.

Last year, Google DeepMind demonstrated that a computer could learn to play video games better than most humans.

Ray Kurzweil, a futurist, keynote speaker at RoboBusiness 2015, and director of engineering at Google, noted the challenges of machine learning. “Mastering intelligence is so difficult that we need to throw everything we have at it,” he said.

Even if true AI is a ways off, Google is readying for robots to interact with and follow humans. This past spring, Google won a patent for “methods and systems for robot and user interaction are provided to generate a personality for the robot.”

Such personalities could be downloaded and transferred between robots, allowing for customizable personal assistants, celebrity avatars, and a way to talk with the deceased. United Therapeutics Corp. in Silver Spring, Md., is working on a similar technology, which collects data about a person from social media to create a software simulation.

AI won’t be evil, say Google chiefs
Mustafa Suleyman, head of applied AI at London-based Google DeepMind, has dismissed the widespread fears of AI, even as Google set up an ethics and safety board, partly to address privacy concerns.

Kurzweil is more optimistic than the AI doubters such as Tesla founder Elon Musk. “The best way to keep [AI] safe is in fact widely distributed, which is what we are seeing in the world today,” he said.

In another recent patent, Google said it’s developing ways to allocate “tasks to a plurality of robotic devices.” Multiple robots could coordinate chores around a warehouse, retail store, or household.

The race to strong AI
In the meantime, however, Alphabet’s units face stiff AI competition from Apple, Microsoft, and Facebook, among other tech heavyweights. “This technology revolution will help how humans and intelligent machines interact in complex, yet complementary ways,” Microsoft engineer Patrice Simard told Fortune.

Smaller businesses are also pursuing piecemeal AI breakthroughs and applications. Publicis.Sapient has acquired an unspecified minority stake in software developer Lucid to develop AI services for energy, financial services, healthcare, and marketing clients.

London-based Navmii Holding Ltd. has launched a crowdfunding campaign to raise £1 million ($1.54 million) to combine machine vision, voice-guided navigation, and AI in the next version of its driver-assist app.

Rainbird Technologies Ltd., a Techstars accelerator recipient in Norwich, England, has raised £28,000 ($43,000) to develop cloud-based software that learns and that can be applied to the healthcare and financial industries.

U.K.-based Riverview Law Ltd. has acquired New York-based Clixlex for an undisclosed amount to develop AI applications for the legal market.

“The race is on to provide virtual assistants to knowledge workers, powerful tools that help people and organizations make quicker and better decisions and allow knowledge workers to focus on the key tactical and strategic matters,” said Karl Chapman, CEO of Riverview Law. “The legal market is not immune from this trend, which will impact all sectors of the economy.”

A general-purpose robot in the works?
In addition to DeepMind’s work on artificial intelligence, Alphabet’s Boston Dynamics unit has been developing mobile robots for the U.S. military and consumers, surgical robotics, and drone deliveries. The company is also working on self-driving vehicles, space exploration, and possibly a general-purpose robot.