A History of Silicon Valley

Table of Contents | Timeline of Silicon Valley | A photographic tour
History pages | Editor | Correspondence
Purchase the book

These are excerpts from Piero Scaruffi's book
"A History of Silicon Valley"


(Copyright © 2016 Piero Scaruffi)

The Selfies (2011-16)

click here for the other sections of this chapter

Neurotech

The most active company in Artificial Intelligence was, by far, Google. At the end of 2013 Google purchased Boston Dynamics, founded in 1992 in Boston by MIT professor Marc Raibert to build a new generation of robots (never actually sold to the public). It was Google's eighth acquisition of a robotics-related company, and Andy Rubin was chosen to run Google's robotics research. In 2013 Google also purchased Redwood Robotics, a joint venture formed in 2012 by SRI International, Meka Robotics and Willow Garage.

In 2014 Google acquired British-based Artificial-Intelligence company DeepMind, founded in 2011 by former game developer Demis Hassabis and New Zealand-born Shane Legg. DeepMind, more than anyone else, was responsible for the popularity of "deep learning", especially after its program AlphaGo beat the world's champion of go in 2017.

The mix of machine intelligence and crowd-sourcing was, in effect, what had given the Web its "intelligence", i.e. its ability to answer questions about anything.

Multi-billion dollar investments in artificial intelligence and robotics multiplied in the 2010s. Pretty much all the major players acquired some startup operating in this space: Amazon acquired Boston-based Kiva Systems (2012), a manufacturer of warehouse robots that had been co-founded in 2003 by Cornell University roboticist Raffaello D'Andrea, and in 2014 it introduced the intelligent assistant Alexa (developed by Lab126 in Sunnyvale); Yahoo acquired Mountain View-based LookFlow (2013); Google also acquired Industrial Robotics, Meka, Holomni, Bot & Dolly, DNNresearch, Schaft, a Japanese manufacturer of robots, Boston Dynamics and API.ai. Facebook acquired the Israeli company Face.com (2012), founded in 2007 by Yaniv Taigman, and used their technology to to develop the face-recognition feature DeepFace (that Facebook started rolling out in 2015) and in 2015 acquired Wit.ai; Microsoft had Project Adam and in 2017 Microsoft acquired Montreal-based Maluuba (founded in 2011 by two undergraduate students, Sam Pasupalak and Kaheer Suleman); Twitter acquired WhetLab in 2015; Salesforce had project Einstein and in 2016 it acquired MetaMind (founded in 2014 in Palo Alto by Richard Socher) and PredictionIO; IBM acquired AlchemyAPI in 2015 and had Watson project; In 2016 Intel purchased San Diego-based Nervana, founded in 2014 to make processors for deep learning, and fabless computer vision chip-maker Movidius, founded in 2005 in San Mateo by Sean Mitchell and David Moloney. In 2019 Intel invested in Toronto-based startup Untether AI that was working on a novel type of chip for deep learning.

Apple had Siri and acquired: British speech recognition startup Novauris Technologies (2013); Cambridge University speech-recognition spinoff VocalIQ (2015); San Francisco-based deep-learning startup Perceptio (2015), founded just one year earlier by Nicolas Pinto and Zak Stone; Seattle-based Turi (2016), formerly known as Dato and GraphLab, co-founded by Carlos Guestrin of Carnegie Mellon University to commercialize his open-source A.I. platform, the basis for Apple's Turi Create (2017); Lattice Data (2017), a Stanford spinoff co-founded in 2015 in Menlo Park by Christopher Re and Michael Cafarella (of Hadoop fame) to commercialize Re's DeepDive project; and Danish computer-vision startup Spektral (2018). In 2018 Apple also hired key personnel from Mountain View-based consultant group Silicon Valley Data Science (SVDS), including co-founders Sanjay Mathur and John Akred as well as chief architect Serena Cheng.

In 2015 Yahoo's Flickr division introduced sophisticated auto-tagging and image-recognition features. Flicker was capable of identifying what (not only "who") was in a photograph, and then automatically categorize it.

Silicon Valley startups included: Saffron Technology, founded in 1999 in Los Altos by Manuel Aparicio and Jim Fleming for decision support; Vicarious, founded in 2010 in San Francisco by Numenta's cofounder Dileep George to work on computer vision (one of the most secretive, backed by Peter Thiel, Mark Zuckerberg, Jeff Bezos, and Elon Musk); Wise.io, a Berkeley spin-off founded in 2012 by Joey Richards, Dan Starr, Henrik Brink, Joshua Bloom to commercialize the machine learning technology invented by Damian Eads, formerly at the Los Alamos National Laboratory; Viv Labs, founded in 2014 in San Jose by three former members of the Siri team (Adam Cheyer, Dag Kittlaus, and Chris Brigham), to work on an "intelligent" digital assistant (acquired by Samsung in 2016); Semantic Machines, founded in 2014 in Berkeley by Dan Roth (who in 2004 had founded in Boston the speech-recognition startup Voice Signal, acquired by Nuance in 2007), UC Berkeley scientist Dan Klein and former Dragon executive and Apple Siri project leader Larry Gillick (acquired by Microsoft in 2018), etc.

Petaluma-based GTFS (later renamed General Vision), founded in 1987 by Anne Menendez, introduced a 1,024-neuron chip, the NeuroMem CM1K chip, to solve pattern recognition problems; Osaro, founded in 2015 in San Francisco by Derik Pridmore (who previously worked for Peter Thiel), worked on a form of machine learning known as deep reinforcement learning that promised to enable a robot to learn from trial-and error interactions with its environment. MinHash, founded by Naren Chittar and Jayesh Govindarajan in 2014 in Palo Alto (and acquired in 2015 by Salesforce), developed a software to help gather online information to be used for marketing campaigns. Google's acquisition of DNNresearch was particularly noteworthy because that startup was founded in 2012 by University of Toronto's professor Geoffrey Hinton and two of his graduate students, Alex Krizhevsky and Ilya Sutskever, the team responsible for the 2012 paper on deep learning that revolutionized the field of computer vision. In 2015 Google gifted its deep-learning technology TensorFlow, the artificial brain behind its image search and speech recognition apps (replacing 2011's DistBelief), to the open-source world.

In 2016 SoundHound, founded in 2005 in Santa Clara by Stanford's voice-recognition expert Keyvan Mohajer, released the digital assistant Hound, whose voice-command user interface competed with Google Now, Apple Siri and the likes. It was the age of the "chatbot", 50 years after the first one (Joseph Weizenbaum's ELIZA at the MIT). Eugenia Kuyda and Philip Dudchuk founded Luka in Moscow (and moved it to San Francisco in 2015) and in 2016 released the "memorial chatbot" Replika that learned a person's style of chat and could replicate it even when the person was dead. Leela.ai, founded in 2016 in Palo Alto by David Henkel-Wallace, was working on an A.I. system mimicking Jean Piaget's theory of child development.

Bofen Technology, founded in Beijing by serial entrepreneur Jerry Yue (who had previously founded the ecommerce website Benlai.com), and relocated to Silicon Valley in 2016 as Brain, wanted to replace the traditional search engine with an intelligent assistant that, given the context, could find the most relevant information for the user.

More ambitious than the Silicon Valley crew was perhaps Narrative Science, founded in 2010 in Chicago by Kris Hammond, director of the University of Chicago's Artificial Intelligence Laboratory, Larry Birnbaum, and Stuart Frankel, a former DoubleClick executive. Narrative Science was the grandchild of Roger Schank's experiments in generating narratives, now trying to construct journalist-proof texts out of big data.

New York-based Wochit, founded in 2012 by two serial Israeli entrepreneurs (Dror Ginzberg and Ran Oz), created videos based on the text. In other words, you write the script and the artificial intelligence turns it into a video.

North Carolina-based Automated Insights' Wordsmith, designed by Cisco engineer Robbie Allen, was the most popular "news generator", used regularly in 2014 by the Associated Press to turn earnings reports into narratives that looked like written by professional writers.

In 2016 Apple bought Emotient, a San Diego startup founded in 2012, that aimed to develop software for reading people's emotions based on facial expression. MetaMind, founded in 2014 in Palo Alto by Andrew Ng's student Richard Socher and acquired by Salesforce in 2016, worked on multitasking neural networks.

CrowdAI was founded in 2016 in Mountain View by two former Google engineers (Pablo Garcia and Devaki Raj) with Nicolas Borensztein, to provide services that need machine learning and computer vision.

Matroid, founded by Reza Zadeh in 2016 in Palo Alto, developed a search tool to find people and objects in visual media.

Deep Learning was all the rage in the 2010s after a Hinton algorithm dramatically lowered the error rate in recognizing images. Several platforms for deep learning were available as open-source software: Torch (New York University), Caffe (Pieter Abbeel's group at UC Berkeley), Theano (Univ of Montreal, Canada), Chainer (Preferred Networks, Japan), Tensor Flow (Google), etc. In 2016 two major applications of deep reinforcement learning were announced: Toyota's self-teaching cars and Google/DeepMind's AlphaGo, that beat a go/weiqi master.

Another major research center was established in 2015 in San Francisco by Elon Musk (of Paypal and Tesla fame) and Sam Altman (of Y Combinator fame). OpenAI quickly hired some of the best minds in "deep learning" and by 2017 was matching the progress of Google DeepMind. In 2018 its OpenAI Five defeated masters of the videogame Dota 2, a feat perhaps even more impressive than defeating the champions of go/weiqi.

The boom in neural networks generated a parallel boom in datasets: a neural network is useless if not trained when an adequate dataset. Kaggle, founded by Anthony Goldbloom and Ben Hamner in 2010 in San Francisco, quickly became the website of choice for running data science competitions, with half a million data scientists using it in 2017 when Google acquired it.

Nvidia too benefited from the computational explosion of AI and Big Data. Nvidia's fastest processor of 2016 boasted 3,584 cores versus Intel's 28.

As both universities and corporations bought large numbers of these chips for A.I., Nvidia's stock became one of the "hottest". Google also unveiled its line of machine-learning chips, the Tensor Processing Unit, typically accessed on the cloud. Other startups for A.I. hardware included: Wave Computing (Campbell, 2010), founded by Dado Banatao and Pete Foley, which in 2018 acquired John Hennessy's 30-year-old startup MIPS; Cerebras Systems (Los Altos, 2016), founded by former AMD executive Andrew Feldman; SambaNova (Palo Alto, 2017), created to commercialize the technology developed at Stanford by Kunle Olukotun and Chris Re; and Groq (Palo Alto, 2017), founded by two former developers of the Tensor Processing Unit at Google, Jonathan Ross and Douglas Wightman. The boom of deep learning triggered a revival of "in-memory computing", pioneered at IBM and Stanford in the 1990s. Mythic, founded in 2012 in Redwood City by Dave Fick and Mike Henry, was one of the first startups in this area.

This was the era of the first "domestic" robots, like Luna, conceived in 2011 by RoboDynamics in Santa Monica, and Jibo, designed by Cynthia Breazeal, director of MIT Media Lab's Personal Robots group. Asia held the leadership in industrial robots: in 2014 Asia bought 139,300 industrial robots (more than half the world's total), of which 57,096 in China, 29,300 in Japan, 24,700 in Korea, compared with 26,200 in the USA. But attention was shifting to "service" robots, and in this field the Bay Area was rapidly surpassing Boston and Japan in innovation. Savioke, founded in 2013 in Sunnyvale by Willow Garage's cofounder Steve Cousins specialized in robots for customer service. Fellow Robots, founded by Marco Mascorro in Sunnyvale, was similarly focused on robots for retail assistance. Simbe, founded in 2014 in San Francisco, built a robot to check a store's shelves for items that were running out or were misplaced. Suitable Technologies, founded in Palo Alto by Willow Garage's cofounder Scott Hassan, sold a robot for videoconferencing. The robots of Unbounded Robotics (later Fetch Robotics), founded in 2014 in San Jose by Melonee Wise (an alumna of Willow Garage's PR2 project), carried out more sophisticated warehouse chores than the ones of Amazon Kiva's robots. In 2016 SoftBank announced that it was opening a store in Japan manned by its Pepper (first introduced in 2014, the result of a collaboration with France's Aldebaran that in 2008 had introduced the user-friendly Na robot). Roboterra, founded in 2014 in Sunnyvale by Yao Zhang and Yuan Zhou, targeted the educational market. Casabots, started in 2014 by Deepak Dekar in Austin but relocated in 2017 to Redwood City and renamed Chowbotics, built "food service robots" that could prepare scientifically-calibrated dishes.

Thanks to open-source components such as the ROS and to the collapse in sensor prices (after 2010 a Microsoft Kinect could do what an expensive laser scanner used to do), robotic startups were multiplying in the Bay Area. There had been three main poles: Willow Garage, that originated at least eight startups, SRI Intl, that originated at least four, and Otherlab, an independent research lab founded in 2009 in San Francisco by Saul Griffith and James McBride to experiment with new technologies. The Willow Garage diaspora yielded Steve Cousins' Savioke; Scott Hassan's Suitable Technologies; Melonee Wise's Fetch Robotics; Jeff Gee's and Mirza Shah's Simbe Robotics; Radu Rusu's Open Perception; Aaron Edsinger's Redwood Robotics (acquired by Google in 2013); Kaijen Hsaio's Mayfield Robotics (Redwood City, 2015); etc. SRI spawned Grabit and Intuitive Surgical. Otherlab incubated Kevin Albert's Pneubotics, working on metal-less "soft" robots (capable, for example, of handling food). In 2012 Kurt Konolige and Gary Bradski of Willow Garage founded Industrial Perception (IPI) in Palo Alto to build 3D-vision guided robots ("robots that can see"). Kinema Systems, founded by Willow Garage alumnus Sachin Chitta in Menlo Park in 2015 (and acquired by Boston Dynamics in 2019), made computer-vision systems for warehouse robots.

The most ambitious startup in the field of intelligent robots was actually not a Bay Area company. It came out from studies on quantum computing. Suzanne Gildert was the lead scientist of machine learning at D-Wave, the most famous startup in quantum computing, founded in 1999 in Canada by Geordie Rose. In 2014 Gildert and Rose founded KindredAI in Vancouver to built A.I. systems with a body, and then KindredAI relocated to San Francisco while Gildert and Rose in 2018 founded SanctuaryAI in Vancouver with the same goal of building human-like robots.

While Google kept testing its self-driving car and Apple launched the top-secret "Project Titan" under Steve Zadesky, in 2015 Tesla updated the software on its electric cars with self-driving features, and in 2016 General Motors acquired Cruise Automation. This San Francisco-based startup, founded in 2013 by Kyle Vogt of Twitch.tv fame (who as an MIT student had worked on self-driving car projects) was developing a system for retrofitting cars with self-driving features. Similarly, in 2015 hacker George Hotz, who became famous in 2007 at the age of 17 for "unlocking" the iPhone and then worked at Vicarious, launched his own self-driving car startup, Comma.ai at San Francisco's "Crypto Castle", aiming to create a self-driving kit to turn existing cars into (quasi) self-driving cars. He demonstrated a prototype illegally in 2016 on a Silicon Valley highway.

Self-driving cars were more popular than ever, especially publicized by Google (its subsidiary Waymo) and Uber. Zoox, founded in 2014 in Menlo Park by Tim Kentley-Klay, was a unicorn by 2017. In 2018 Mountain View-based Drive.ai, a 2015 emanation of Stanford' Artificial Intelligence Lab, launched a ride-hailing service in Texas. Google Waymo built a test track in Atwater, two hours from Silicon Valley. Nvidia teamed up with Audi. Unfortunately, two Tesla owners died when their car was in autopilot mode and a Uber self-driving car killed a woman in Arizona. And these companies had to compete against models announced by General Motors (a Chevy Bolt and a pilot service in San Francisco from GM's self-driving unit Cruise, also based in San Francisco, which became a partnership with Honda in 2018). In August 2016 MIT spinoff NuTonomy began testing a self-driving taxi, and a few days later Uber (who had just hired Eric Meyhofer from Carnegie Mellon University) began testing a self-driving taxi in Pittsburgh, followed by San Francisco in December. Neither was a success: they both required a human driver and were far from being truly "autonomous". Nuro.ai, founded in 2016 in Mountain View by former Google engineers Jiajun Zhu and Dave Ferguson, had the humbler target of self-driving vehicles for local goods delivery. Autox, founded in 2016 in San Francisco by Jianxiong Xiao, started a grocery delivery service in a San Jose community in August 2018. Aurora Innovation, founded in 2016 in the Palo Alto by Chris Urmson (Carnegie Mellon and Google), Drew Bagnell (CMU and Uber), and Sterling Anderson (MIT and Tesla) had one of the most ambitious plans for a fully autonomous vehicle. Pony.ai, founded in 2016 in Fremont by Tiancheng Lou and former Baidu executive James Peng, mainly served the Chinese market: it was the first firm to obtain a license to test autonomous vehicles in Beijing. In early 2018 Santa Clara-based Voyage, a 2017 spin-off of Udacity founded by Oliver Cameron and MacCallister Higgins to retrofit old cars with A.I. software and turn them into driver-less taxis, began testing its self-driving cars at a gated community for senior citizens in San Jose. Despite all the hoopla, there was still no self-driving car available for purchase in 2018. Mercedes-Benz, BMW, Audi and Volvo offered "traffic assistants" that still required a driver, and a driver with hands on the steering wheel. These were all descendants of Tesla's Autopilot, first introduced in October 2014, developed in partnership with the Israeli company Mobileye. In 2017 Uber spent $1 billion on 24,000 driver-less vehicles from Volvo.

The lidar is the device that autonomous vehicles to sense the world around them, and Velodyne dominated this market. Google's self-driving cars popularized its models. The problem is that its lidar was a rotating device and, like all mechanical objects, it was expensive and prone to breaking. Several startups worked on solid-state lidars, which don't need rotating mechanical parts. The most hyped startup was Quanergy, founded in 2012 in Sunnyvale by Louay Eldada, Angus Pacala and Tianyue Yu, that at one point was valued at $2 billion despite not having a product (its first product was scheduled to debut in Fisker's EM 2019). LeddarTech in Canada introduced a commercial model in 2016, that was the basis for the prototypes demonstrated by New Mexico-based TriLumina and France-based Valeo, and San Diego-based TetraVue demonstrated a high-definition solid-state lidar (capable of collecting 60 million bits of data per second) at the end of 2017. BMW preferred the solid-state lidar from Israel's Innoviz. Others were skeptic, especially after Quanergy's failure to deliver on its promises. Luminar, founded in 2012 in Portola Valley (near Menlo Park) by Austin Russell, and Ouster, founded in 2016 in San Francisco by Angus Pacala (of Quanergy fame) didn't believe in solid-state lidars because the quality was too low.

The lidar was just one component. To make these cars "intelligent", they needed detailed maps to guide them. DeepMap was founded in 2016 in Palo Alto by former Google engineers James Wu and Mark Wheeler to create high-definition maps for driver-less vehicles. It was mostly a big-data problem: extract high-definition pictures, plot maps with centimeter precision, provide real-time communication between the vehicle and the cloud.

In 2015 Toyota invested $1 billion in the Toyota Research Institute, headed by Gill Pratt (formerly at DARPA and MIT), to conduct research on automated driving in Palo Alto and Boston. Also in 2015, Ford opened its Research and Innovation Center Palo Alto and in 2017 invested $1 billion in Argo AI, founded just six months earlier in San Francisco by Bryan Salesky (formerly at Google) and Peter Rander (formerly at Uber).


click here for the other sections of the chapter "The Selfies (2011-16)"
(Copyright © 2016 Piero Scaruffi)

Table of Contents | Timeline of Silicon Valley | A photographic tour | History pages | Editor | Correspondence