DeepMind Demystified

Michael Mihalicz
Jul 22, 2017 · 22 min read

Summary

On January 27, 2014 Google’s parent company, Alphabet Inc., bought U.K. based AI company Deepmind Technologies Limited1 pre-revenue2 for $660 million with the condition that an ethics board be established. Google DeepMind (GDM) is a software company established independently out of University College London’s (UCL) Gatsby Unit3 by three academics working towards developing general artificial intelligence (AI).4 At the time of the acquisition Deepmind raised over $50 million in VC funding and was on the verge of major breakthroughs in learning algorithms. DeepMind pioneered several unique approaches to designing AI systems that many believe will enable them to operate outside the confines of structured environments and solve real world problems. Early investors in DeepMind include some of the biggest names in the tech industry who are as optimistic about the applications of this technology as they are fearful of its potential consequences. The acquisition provides Deepmind the opportunity to achieve its goal of reducing climate change, access to Google’s database and the capital needed to accelerate expansion. Alphabet Inc.’s (Google) interest in DeepMind was primarily to develop a general purpose AI system to reduce energy consumption in its data centers, to penetrate the healthcare industry and for the opportunities it presents to disrupt multiple other industries. After adjusting for tax, operating costs and time value of money, the cost savings attributed to GDM’s general purpose AI system has a value at the time of the acquisition of between $155 and $221 million and provides Google a competitive advantage in a $30 — $60 billion dollar market.

Artificial Intelligence

Many of the world’s largest companies have recently been investing heavily in AI which even in their infancy are yielding unprecedented returns. The primary drivers behind the recent surge in AI centers around its ability to automate data analysis as well as its applications in emerging technologies that are poised to disrupt several established industries. Currently the majority of AI systems are created to perform specific functions, but a new breed of AI is emerging called general AI that is capable of performing general tasks and acquiring knowledge from its experiences.5 General AI is programmed to solve problems in a much more human way and is related to reasoning, planning, self-awareness, consciousness, and communicating in natural language.2 Applications for general AI include robotics, computer vision, finance, general computer science, news publishing and writing, smart phone assistance and online customer service. All of which Google has an interest in.

The real gross value added of this industry is estimated to exceed $40 trillion globally by 2035 (Figure 1), generate revenues of over $3 trillion by 2024 (Figure 2) and the disruptive technologies closely related to AI are predicted to have an economic impact of between $7.1 and $13.1 trillion by 2025.6 The number of new AI companies has been gradually declining since 2013, but the amount of total funding has been increasing exponentially since 2009 reaching over $1bn in 2015. Research and Markets forecasted a CAGR of 53.65% in AI systems from 2015 to 2020 with the highest growth expected in the healthcare sector.7

It’s unlikely that large tech firms like Google have definitive plans for their AI systems, but rather see it as a natural progression in system design and are looking to get ahead of the curve by developing competencies today.8 Facebook has hired over 150 employees dedicated to AI development,9 invested nearly 30% of its Q1 2015 revenue into their two research labs and no longer even track the return on these investments because, according to their CTO, “the groups have paid for themselves for the next five to ten years”.10

DeepMind

DeepMind was co-founded in 2010 by Demis Hassabis, Mustafa Suleyman and Shane Legg who were on a mission to solve intelligence4 by combining machine learning and systems neuroscience to build powerful general purpose learning algorithms that formalize the way the human mind works. The system uses a deep neural network trained using a technique derived from Behaviourist Psychology called reinforcement learning where the system acts to maximize its cumulative reward. They started by training their system on Atari games and later applied the technology to the game of Go which it eventually mastered and went on to beat a professional Go player in a five-game championship.11 What was most impressive about this is that brute force tactics are not possible in the game of Go and so the team had to develop a system capable of approximating human intuition. DeepMind has been described as “the last large independent company with a strong focus on artificial intelligence”.12 Its approach to deep learning systems design and reinforcement learning are believed to be the techniques that will one day enable AI systems to operate in the real world and create a form of general intelligence that is broadly applicable to real world problems. DeepMind is pioneering approaches to the development of general AI in a way that is effectively unique and difficult to replicate.

The Team

Demis Hassabis was a child chess prodigy often celebrated for his successes in the video game industry and his contributions to science. He completed an undergraduate degree in Computer Science at Cambridge University and later went on to receive a PhD in cognitive neuroscience at UCL.13 Mustafa Suleyman is the head of applied AI at GDM and is leading the company’s charge into healthcare. Shane Legg earned his PhD at the IDSIA in Switzerland where he was supervised by the leading authority on theoretical models for super intelligent machines. He then went on to develop models of human decision making at the Swiss Finance Institute and spent two years at the Gatsby unit for computational neuroscience where DeepMind was conceived.14

Early Investors

In its early stages DeepMind attracted multiple investors including Horizons Ventures, Founders Fund, Tesla founder and CEO Elon Musk, founder and former CEO of PayPal Peter Thiel, Skype co-founder Jaan Tallinn and teenage multi-millionaire Nick D’Aloisio, to name a few.15 During this time DeepMind secured more than $50 million of funding16 without earning any revenue or releasing a single product.

Horizons Ventures

Horizons Ventures is a global venture fund established to manage Li Ka-shing’s private investments in technology, media and telecommunications. Notable investments include early-stage positions in Facebook, Spotify, Waze, Siri, Hampton Creek in addition to Jaan Tallinn’s Skype and Nick D’Aloisio’s Summly.17,18 The fund is focused on disruptive technologies solving real problems that will make a meaningful impact on the world.18,19 It has been interested in artificial intelligence since 201017 and currently has a stable disruptive portfolio of more than 50 startups.18,19,20 It is both sector and stage agnostic and invests on a case-by-case basis in ventures with all profits accruing to Ka-shing’s charity foundation to support education reform, medical advances and foster the transfer of knowledge between nations.18,19

Founder’s Fund

Founders Fund is another venture capital fund with a diverse investment history that includes Facebook, Spotify, Airbnb, Lyft, Oculus, Yammer, SpaceX, SolarCity and Stripe.20,21 While both it and Horizons Ventures are sector and stage agnostic and share investments in Facebook and Spotify, Founders Fund is focused on technological development in all its forms. It also takes a much more passive approach and therefore places more emphasis on the people they investing in, “We will treat them with respect and hope for the best”.22 The fund has never removed a single founder and criticizes overly controlling VCs for unnecessarily constraining a company’s ability to evolve and realize its vision. Founder’s Fund manages over $2 billion in assets and is best known for investing in ambitious projects and making radical transformational investments.22,23

Angel Investors

Jaan Tallinn is best known as co-founder of Skype and together with Elon Musk helped fund two promising general AI companies, Deepmind and Vicarious.24 Tallinn assumed the role of investor and advisor in Deepmind and later co-founded the Future of Life Institute, a volunteer-run research and outreach organisation which aims to mitigate the existential threat that AI poses to humanity.24

Nick D’Aloisio was only 15 years old when he suggested using artificial intelligence to summarize large amounts of text and ended up being one of the youngest ever to receive venture capital. Horizon Ventures led the funding of the news app, Summly,25 which D’Aloisio developed in his bedroom when he was just 17 and later sold to Yahoo for $30 million.19

Peter Thiel is the founder and former CEO of PayPal which merged with Elon Musk’s X.com in 1999 and was eventually acquired by eBay in 2002.26 Both tech moguls are heavily invested in AI and were two of Deepmind’s most active angel investors. Following their work together at PayPal, both Thiel and Musk assumed roles in what the media has dubbed the ‘PayPal mafia’27 which describes a group of 24 Silicon Valley entrepreneurs and former PayPal employees who left the company after its acquisition and went on to found highly successful tech startups.28 Three members of the PayPal mafia, including Peter Thiel, have become partners in Founders Fund which explains its investments in SpaceX, SolarCity and Stripe in addition to several other startups founded by former PayPal employees.28

Elon Musk, who is best known as co-founder of Tesla and SpaceX, has launched several philanthropic initiatives29 and has investments in many promising AI Companies. Musk invested millions with Tallinn in two general AI ventures in 2014, Vicarious which aims to achieve human-level intelligence in vision, language and motor control and DeepMind which aims to solve intelligence. In 2015 he committed $10 million dollars to a research institute aimed at keeping AI friendly,30 invested in Tallinn’s Future of Life Institute and later that same year partnered with Peter Thiel and Reid Hoffman, among others, in a $1 billion investment in the non-profit research center OpenAI.31,32 OpenAI provides open access to developments in AI with the goal of keeping humans a step ahead of technology, distributing the benefits of AI as widely and evenly as possible33 and counteracting the negative effects of people using AI for evil with the greater number of people using it for good.34 OpenAI has since attracted a lot of talent from Musk’s competition as well as from researchers interested in publication and accelerates the advancement of the technology though mass collaboration and access to online databases.34

While Musk has good reason to invest in AI given his interests in autonomous cars35 and space exploration, he has openly admitted that the reason he is so heavily invested in the technology is because he is fearful of a potentially dangerous outcome and wants to keep an eye on it.29 Musk has been very public about his concerns and has even gone on record calling it “humanity’s greatest existential threat” and comparing it to “summoning the devil”.29,30,36 Several of the parties involved with DeepMind share in their concern for humanity and signed an open letter to the International Joint Conference on Artificial Intelligence in Buenos Aires along with other members of the tech and scientific communities calling on governments to ban the development of “autonomous weapons”. Signatories include Musk, Tallinn and Hassabis.

Deal Structure

The information on the sale price of DeepMind was scattered ranging from £242 million20,24,47,37,38 to £400 million39,40,41,42 which translates to roughly $400 and $650 million, respectively. This uncertainty is likely the result of someone confusing £400 with $400 million which was reported in a recode exclusive immediately following the acquisition and referenced in multiple articles that followed. Given the subsequent corrections to some of the articles originally stating lower amounts and the credibility of ‘The Economist Intelligence Unit N.A.’,41 $660 million is assumed to be the official purchase price. Deepmind had previously entertained an offer from Facebook to buy the company for an undisclosed amount, but the deal fell through in 2012 for reasons that have not been made public.43

Since the purchase price was so much larger than similar deals driven by talent acquisition (see Talent Acquisition section) it is likely that the company had IP that was valuable to Google. In fact it was so much larger than expected that it triggered a 25% market-wide increase in the stock prices of robotics companies44 and a momentary dip in Google’s own stock on the day of the acquisition (Figure 3). Because DeepMind was conceived in UCL’s Gatsby Unit there is speculation about whether Deepmind purchased IP from UCL or had ongoing arrangements with the University at the time of the acquisition.3 The Vice-Provost Operations at UCL was able to confirm that neither the University nor any of its affiliates received money from Google for the deal which indicates that if DeepMind did acquire IP from the University, it did so without any ongoing royalties or conditions.

Ethics Board

One of the conditions of this acquisition was the establishment of an ethics board with the authority to govern how Google uses Deepmind’s technology. Many of the early investors in DeepMind and two of its co-founders, Legg41 and Hassabis, have publicly expressed concern about how this technology might be misused and reportedly pushed for this condition.2 It is rare for an acquiring company to relinquish that amount of control over an asset, but sensible for the right price given public fears surrounding general AI. Several of DeepMind’s early investors are among the most actively involved in projects to ensure the safe development of this technology and it is likely that this condition is related to their influence on the company. The members of Google DeepMind’s ethics board have not yet been released, but it would not be surprising to find out that it includes Elon Musk, Peter Thiel, Jaan Tallinn and other early investors in the venture.

Value of Google to Deepmind

DeepMind develops AI systems that thrive in large databases and having access to Google’s data enables the company to create a wider range of training environments so the AI systems can learn faster and more effectively. These databases also require large data centers to which Deepmind can apply its technology and have the opportunity to achieve one of its original goals, to reduce climate change.

Google is also a good fit because they believe in the technology and have the capital to fund the company’s growth. Deepmind has been aggressively recruiting computer scientists and engineers since they first started and soon after the acquisition were reported to hire dozens more researchers and publish several papers for leading machine-learning and artificial-intelligence conferences. In oct. 2016 GDM was reported to have hired 175 new employees, more than tripling is staff since the acquisition less than three years prior.39 Lastly, Google is diversified across many of the sectors to which GDM’s technology is applicable and are targeting many of the same industries for future penetration.

Value of Deepmind to Google

The value of DeepMind to Google lies primarily in the general applicability of its systems to the array of products that Google offers or has interest in, including the search engine itself. It was reported that Larry Page led the acquisition to compliment the company’s positions in robotics, autonomous vehicles and space travel,42 but since then the GDM team has worked almost exclusively on developing a general purpose learning algorithm to reduce energy consumption.

Talent Acquisition

Companies investing in AI have experienced first hand the unprecedented returns it can provide where rewards far outweigh risks when compared to similar investments in other technologies.45 Investments in AI today are about long-term sustainability as it is increasingly becoming evident that AI will be necessary to maintain market leadership. Google had been assembling a team of the world’s leading experts in machine learning for several years prior to their acquisition of Deepmind and had been known to pay tens of millions of dollars just for the opportunity to work with top AI researchers.46 Deepmind also adopted an aggressive recruiting strategy, competing with large tech firms like Google and Facebook for promising researchers and computer scientists and by 2014 had assembled a team of 75 employees.38,41 It also had strong ties to academia and only months after the acquisition seven researchers at Oxford were recruited in return for a ‘substantial donation’ to the University.47

Average starting salaries for machine learning specialists range from 60k to 150k,48 but they are in high demand and hard to find. According to a survey conducted in the UK, all of the professions related to AI are those that companies find most difficult to recruit with 84% of companies finding it either ‘very’ or ‘fairly difficult’ to recruit Machine Learning specialists (Figure 4). AI is becoming an industry standard and large tech companies are increasingly acquiring startups as part of larger employee acquisition strategies. There is an element of this in the DeepMind acquisition as several reports have pointed out,49,50 but at almost $9 million per employee there must have also been another reason. According to Magister Advisors, who followed 26 AI acquisitions that were driven by talent acquisition since 2014, small AI companies sell for on average $2.4 million per employee.51 The acquiring companies in this study were all mainly if not entirely interested in talent acquisition making these deals a useful benchmark to gage Google’s intentions. Based on this metric, the expected value of an AI company with 75 employees is only $180 million (not including the condition of the ethics board) leaving approximately $400 million on the table. It is possible since they were competing for talent, that DeepMind was able to attract certain valuable employees that Google was willing to pay a premium for, but will not likely account for such a large discrepancy.

Contrary to reports indicating that Google was driven by talent acquisition,49,43 if Google’s only goal was to recruit talent they likely could have done so much cheaper elsewhere and so we can assume that they saw something in the technology. In fact DeepMind was and still is primarily research-focused and so it is conceivable that they were developing something that was effectively unique and that Google could not replicate in less time and/or for less money.

Energy Management

With over a million servers in 16 data centers around the world,52 Google’s electricity costs add up to hundreds of millions of dollars every year. Google has been working on reducing energy consumption for over ten years 53,54 and already has some of the most efficient servers in the world,55 but still requires significant amounts of energy to cool the servers. One of the biggest ongoing expenses in running a data center are the cooling systems56 which can account for almost half of the center’s total electricity usage. In 2011 it was estimated that Google’s servers used less than 1% of the electricity consumed by all data centers and .01% of the world’s electricity consumption.57,58,55,59,60 In 2010 data centers were estimated to use between 1.1% and 1.5% of the global electricity supply which increased from .5% in 2005.61 Google’s energy consumption for its data centers more than doubled (121.47%) in four years from 2010 to 2014. Assuming a constant growth rate going forward, the average yearly energy consumption from 2014 to 2017 was approximately 5,605,372 MWh costing an average of $263 million.

Its clear that the main driver behind the acquisition was potential that DeepMind’s technology had to reduce the energy required to run the cooling systems in its data centers. Only Months after the acquisition in 2014, Google spent $2.65 billion on building new data centers62 and started applying DeepMind’s resources to energy conservation in these facilities around the same time.55 From 2014 to 2016 DeepMind worked on building a highly effective AI system to control the Google’s cooling units capable of reducing cooling costs by 40%53 which reduces the overall energy consumption in data centers by 15%.55 The new system has been operational since Q2 2014 on roughly 1% of Google’s data centers which increased gradually to 10% by Q2 2016 when it was announced that it plans to implement the system in all of its data centers by 2017.55 Google also plans convert all of its offices and data centers to run on %100 renewable energy in 2017 making this system even more valuable.65 After adjusting for tax, operational expenses and time value of money the total value of the completed system in perpetuity is between $400 and $775 million in Q1 2017 and between $155 and $221 million in Q1 2014, assuming renewable energy prices and constant growth in energy consumption (see valuation).

Google’s servers only use approximately 1% of data center electricity consumption globally and since this new system is general purpose, the company can sell or licence the technology to the other 99% of data centers around the world. The size of the potential market in 2017 for other data centers assuming Google’s growth, discount and tax rates, energy prices and that three quarters of the world’s data centers would be interested in negotiating an arrangement will be $30 — $60 billion. Google would therefore only need to capture at most 2–4% of the market in 2017 to account for the $439 — $505 million discrepancy between the cost of acquisition and the 2014 tax-adjusted value of cost savings in its own data centers. This also does not include cooling systems in other types of facilities or the new applications for this system which GDM says can also has applications in “improving power plant conversion efficiency […], reducing semiconductor manufacturing energy and water usage, or helping manufacturing facilities increase throughput”.59

Robotics

Google sparked controversy following the acquisition of four AI companies and eight robotics companies, including one specializing in militarizing robots, within a ten month period starting in March 2013.64 While Google’s interest in robotics is not yet known, it is believed that they have a specific application in mind which involves general AI.

Just as decreases in the cost of computing have made AI possible, technological advancements in actuators, accelerometers, GPS and machine vision are making robots more practical for business solutions. The International Data Corporation expect that the $71 billion spent in robotics in 2015 will grow at a compound annual growth rate of 17% to $135.4 billion by 201965 and the Bank of America Merrill Lynch expects $83-billion will be spent in the US in 2020.66

Robotics represent attractive proposition for Google who is already working on disrupting several industries with its autonomous cars, but it is unlikely the Google will be allowed to apply GDM’s general AI systems to the robotics companies in its GoogleX division due to its military contracts and the ethics board that Deepmind insisted on in negotiations.

Healthcare

GDM announced in the fall of 2015 that it had been given identifiable patient records for over two million Londoners by three UK hospitals in the Royal Free NHS Trust67 and in Feb. 2016 launched ‘DeepMind Health’.68 While the company maintains that this project has nothing to do with AI, there is a large and growing demand for AI in the healthcare industry which Google has long coveted.67

There are several clinical applications for AI systems already in hospitals, but the real value of this technology in healthcare lies in its ability to analyze of large unstructured data sets and produce actionable insights for real-time decision support.69,70 The techniques that GDM uses to develop learning algorithms are especially useful for Big Data mining42 and thanks to the agreement with Royal Free London they now have the data to start training the system. A report released by Research and Markets identified healthcare as the sector with the highest growth potential for AI from 2015 to 202071 which Frost & Sullivan predict will grow from $634 million globally in 2014 to $6.7 billion by 2021.70

Autonomous Cars

Recent developments in AI and global navigation systems have several large companies looking to disrupt the transportation and automotive industries with autonomous vehicles. Both Google and Tesla already have autonomous cars on the road72 and Toyota announced in 2015 its plans to invest $1 billion over 5 years in an AI research facility which it hopes will advance its computer assisted driving and collision avoidance systems.73 Weymo, an Alphabet Inc. subsidiary, has accumulated 2 million miles of autonomous driving on city streets in four areas in the US which is the equivalent of 300 years of human driving experience74 while Tesla’s autopilot has logged 130 million miles of highway driving.

According to Lux Research the autonomous car industry will become an $87 billion dollar opportunity by 2030 with Software gaining more than any other contributor.75 General AI is an important part of this progression, but due to significant challenges76, there are not expected to be any fully autonomous cars available to the public until 2030. Also, since Google already has an operational autonomous car it is unlikely that the value of GDM’s technology in improving their existing product influenced the valuation.

Valuation

The value of GDM’s technology to Alphabet Inc. can mostly be attributed to the broad applicability of the technology and its expected contribution to the companies existing and future product offerings. Since the majority of the value of GDM involves disruptive innovations which cannot be reliably estimated, this section will focus on the quantifiable reductions in data center energy usage. A Discounted Cash Flow method is most appropriate considering the uniqueness of the asset and because Google has made no indications of plans to eventually sell the division.

The Q1 2014 value of net savings attributed to GDM are based on the following assumptions:

· High ‘Operating Costs’ estimates are based on Alphabet Inc.’s R&D spending (Figure 7) per employee (Figure 8) for each quarter from Q1 2014 to Q4 2016 multiplied by the expected number of employees at Google DeepMind for each respective quarter. Low ‘Operating Costs’ estimates are based on starting salaries of computer scientists specializing in machine learning.

· The expected number of employees at Google DeepMind is based on reports of 75 employees in Q1 2014 and 250 employees in Q2 2016, assuming a constant growth rate compounded quarterly.

· Alphabet Inc. does not disclose its energy usage so expected values are calculated based on reports that Google’s data centers used 1,988,000 MWh77 in 2010 and 4,402,836 MWh of electricity in 2014.59 This assumes a constant growth rate for this period and going forward in perpetuity which is reasonable considering long-range Big Data market forecasts.

· The reduction in data center energy usage is based on reports that the new system will reduce total energy consumption in Google’s data centers by 15%.55

· The rate of adopting the technology in data centers is based on reports that the system was applied to 1% of Google data centers in Q2 2014, 10% in Q2 2016 and 100% in Q1 2017,55 and assumes a constant rate of adoption between these periods.

· The price of electricity per MWh is based on high and low costs of generating wind energy in the US in 2016 (Figure 5) considering reports that Google uses primarily wind energy (Figure 6) and will convert all of its offices and data centers to renewable energy by 2017.71

· Tax rates are based on the ‘Provision for income taxes’ from Alphabet Inc.’s quarterly reports for each quarter from Q1 2014 to Q4 2016 and averaged for the tax rate in perpetuity.

· A %30 discount rate is consistent with standard practices in the tech industry.

· Big Data is showing no signs of slowing down so the terminal value calculated in Q1 2017 is calculated in perpetuity as described below.

Appendix

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

References

1. http://www.bloomberg.com/research/stocks/private/snapshot.asp?privcapid=254718323

2. http://www.extremetech.com/extreme/175424-google-acquires-human-like-ai-company-for-500-million-skynet-is-now-a-real-possibility

3. https://www.whatdotheyknow.com/request/deepmind_intellectual_property

4. http://www.wired.co.uk/article/deepmind

5. Moriwaki, N., Akitomi, T., Kudo, F., Mine, R., Moriya, T. & Yano, K. (2016). “Achieving General-Purpose AI that Can Learn and Make Decisions for Itself”. Hitachi Review. Vol. 65, №6

6. McKinsey & Company (May 2013). “Disruptive Technologies: Advances that will transform life, business, and the global economy”.

7. http://www.prnewswire.com/news-releases/artificial-intelligence-market-growth-of-5365-by-2020---rising-demand-for-intelligent-systems-300218782.html

8. Hunt, W. A. (2015). “War and Peace in a robotic future”. QUEEN’S QUARTERLY 122/4 (WINTER 2015)

9. http://venturebeat.com/2015/04/22/why-facebooks-rd-spend-is-huge-right-now/

10. http://www.fastcompany.com/3060570/facebooks-formula-for-winning-at-ai

11. https://en.wikipedia.org/wiki/AlphaGo

12. http://www.theinquirer.net/inquirer/news/2450838/human-strikes-back-in-go-against-google-s-deepmind

13. https://www.hellenext.org/2015/03/4-tips-on-how-to-succeed-from-demis-hassabis-of-deepmind-google/

14. https://deepmind.com/about/

15. https://angel.co/deepmind-technologies

16. http://www.recode.net/2014/1/26/11622732/exclusive-google-to-buy-artificial-intelligence-startup-deepmind-for

17. https://www.pehub.com/2014/06/horizon-ventures-backs-modern-meadow/

18. https://en.wikipedia.org/wiki/Horizons_Ventures

19. https://www.forbes.com/sites/shuchingjeanchen/2014/03/12/li-ka-shing-and-horizons-ventures-the-making-of-a-venture-capital-powerhouse/#6b8166e631b0

20. http://startups.co.uk/google-acquires-artificial-intelligence-start-up-deep-mind-technologies-for-alleged-242m/

21. http://www.theverge.com/2016/11/23/13726280/deepmind-nhs-data-streams-app-new-deal

22. http://foundersfund.com/the-future/#/artificial-intelligence-software

23. https://www.pehub.com/canada/2015/1/founders-fund-invests-in-tilray-backer-privateer-holdings/

24. http://www.techworld.com/picture-gallery/startups/meet-ai-companies-being-backed-by-cofounder-of-skype-3610737/

25. http://startups.co.uk/young-guns/nick-daloisio-summly/

26. http://www.telegraph.co.uk/technology/11106473/The-PayPal-Mafia-Who-are-they-and-where-are-Silicon-Valleys-richest-group-of-men-now.html

27. http://www.businessinsider.com/peter-thiel-elon-musk-is-the-most-impressive-paypal-mafia-member-2015-3

28. https://en.wikipedia.org/wiki/PayPal_Mafia

29. http://www.businessinsider.com/what-companies-has-elon-musk-invested-in-2015-8/#musks-first-company-was-zip2-corporation-the-webs-first-yellow-pages-back-in-1995-1

30. https://www.forbes.com/sites/ericmack/2015/01/15/elon-musk-puts-down-10-million-to-fight-skynet/#7773b18b2e5b

31. http://www.trueactivist.com/elon-musk-funds-1b-project-to-stop-human-destruction-from-demon-of-artificial-intelligence/

32. http://money.cnn.com/2015/12/12/technology/openai-elon-musk/

33. https://en.wikipedia.org/wiki/OpenAI

34. https://www.wired.com/2015/12/elon-musks-billion-dollar-ai-plan-is-about-far-more-than-saving-the-world/

35. https://www.nytimes.com/2015/03/20/business/elon-musk-says-self-driving-tesla-cars-will-be-in-the-us-by-summer.html?_r=0

36. http://www.ibtimes.co.uk/elon-musk-artificial-intelligence-potentially-more-dangerous-nukes-1459710

37. https://www.pehub.com/2015/04/vcj-sector-analysis-the-rise-of-the-robots/

38. http://www.extremetech.com/extreme/175424-google-acquires-human-like-ai-company-for-500-million-skynet-is-now-a-real-possibility

39. http://www.businessinsider.de/nick-bostrom-deepmind-is-winning-the-ai-race-2016-10

40. https://www.technologyreview.com/s/601139/how-google-plans-to-solve-artificial-intelligence/

41. The Economist Intelligence Unit N.A. (2014). “Don’t be evil, genius; Machine learning”. The Economist; London. Vol. 410; Issue. 8872

42. https://www.theguardian.com/technology/2014/jan/27/google-acquires-uk-artificial-intelligence-startup-deepmind

43. https://www.theinformation.com/Google-beat-Facebook-For-DeepMind-Creates-Ethics-Board

44. Horowitz, M. C. (2014). “The Looming Robotics Gap”. Foreign Policy. May/June 2014.

45. https://techcrunch.com/2016/09/24/investing-in-ai-offers-more-rewards-than-risks/

46. Google opens new AI lab and invests $3.4M in Montreal-based AI research

47. http://www.eng.ox.ac.uk/about/news/university-of-oxford-teams-up-with-google-deepmind

48. https://venturescannerinsights.wordpress.com/tag/artificial-intelligence-company-list/

49. http://www.theverge.com/2014/1/26/5348640/google-deepmind-acquisition-robotics-ai

50. https://techcrunch.com/2015/12/25/investing-in-artificial-intelligence/

51. http://www.magisteradvisors.com/blog/ai-teams-being-acquired-for-over-2m-employee-employee-value-often-far-greater-than-business-value

52. https://en.wikipedia.org/wiki/Google_Data_Centers

53. https://deepmind.com/blog/deepmind-ai-reduces-google-data-centre-cooling-bill-40/

54. http://www.escapistmagazine.com/news/view/111987-Googles-900k-Servers-Are-Super-Efficient

55. https://www.theguardian.com/environment/2016/jul/20/google-ai-cut-data-centre-energy-use-15-per-cent

56. Lintner, W., Tschudi, B. & VanGeet, O. (2010). “Best Practices Guide for Energy-Efficient Data Center Design”. FEDERAL ENERGY MANAGEMENT PROGRAM.

57. http://www.businessinsider.de/googles-400-million-acquisition-of-deepmind-is-looking-good-2016-7

58. https://www.cloudyn.com/blog/10-facts-didnt-know-server-farms/

59. http://www.theverge.com/2016/7/21/12246258/google-deepmind-ai-data-center-cooling

60. https://storageservers.wordpress.com/2013/07/17/facts-and-stats-of-worlds-largest-data-centers/

61. http://www.nytimes.com/2011/08/01/technology/data-centers-using-less-power-than-forecast-report-says.html

62. http://www.datacenterknowledge.com/archives/2014/07/23/from-112-servers-to-5b-spent-on-google-data-centers-per-quarter/

63. http://mashable.com/2016/12/06/google-100-percent-renewable-energy/#fj4aPOn6kkqt

64. https://en.wikipedia.org/wiki/List_of_mergers_and_acquisitions_by_Alphabet

65. http://fortune.com/2016/02/24/robotics-market-multi-billion-boom/

66. http://www.canadianbusiness.com/investing/how-investors-can-profit-from-the-rise-of-the-robotics-industry/

67. https://www.ft.com/content/ddd1478e-b70d-11e6-961e-a1acd97f622d

68. https://deepmind.com/applied/deepmind-health/

69. https://techcrunch.com/2015/12/25/investing-in-artificial-intelligence/

70. Frost & Sullivan. (2016) “Transforming Healthcare Through Artificial Intelligence Systems”. Artificial Intelligence in Health & Life Sciences Conference, London.

71. http://www.prnewswire.com/news-releases/artificial-intelligence-market-growth-of-5365-by-2020---rising-demand-for-intelligent-systems-300218782.html

72. http://www.latimes.com/business/technology/la-fi-hy-tesla-google-20160701-snap-story.html

73. http://fortune.com/2015/11/06/toyota-ai-silicon-valley-robotics/

74. https://waymo.com/

75. http://www.luxresearchinc.com/news-and-events/press-releases/read/self-driving-cars-87-billion-opportunity-2030-though-none-reach

76. http://www.alphr.com/cars/1001329/driverless-cars-of-the-future-how-far-away-are-we-from-autonomous-cars

77. http://www.datacenterknowledge.com/archives/2011/08/01/report-google-uses-about-900000-servers/

)
Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade