A Milestone in Achieving Artificial General Intelligence—DeepMind Beats the Pros at StarCraft II
Follow James on Twitter @jwangARK
DeepMind’s latest AI system—AlphaStar—has beaten two professional players in Blizzard’s real time strategy game StarCraft II. While it received some press coverage, this milestone deserves far more attention because we believe it is more impressive than the breakthroughs achieved by IBM’s DeepBlue and DeepMind’s AlphaGo.
Unlike classic board games such as chess and go, StarCraft II embodies the complexity, random nature, and chaos associated with real world problems. Games like chess are turn-based, limited in scope, simple and fully visible on a board. StarCraft II is a full-blown war simulation requiring the player to make real time decisions on resource allocation, construction, unit control, and battle. Unlike chess players who always play on the same board, StarCraft II players face different maps each season, making their moves unlimited in scope.
Similarly to AlphaGo, DeepMind’s AlphaStar was trained with supervised learning and reinforcement learning. This time, however, the training took place on hundreds of Google TPUs (Tensor Processing Units) and cost an estimated $3.8 million in computing power.
Arguably, AlphaStar is the most important step forward thus far toward the goal of artificial general intelligence—the holy grail of AI. Perhaps because it has yet to beat the world’s best StarCraft II player, the press has not zeroed in on the importance of AlphaStar’s recent milestone. Having learned from the rapid and relentless progress of AlphaGo in 2017, we believe that AlphaStar will defeat the world champion of StarCraft II by the end of the year.
Remember, you heard it here first!
How Will Unicorns Find the Exit?
Follow Max on Twitter @
Since 2014, unicorns have been exiting the private world at an exceedingly slow pace. The ratio of their annual exit value to their total unrealized value has averaged just 15% during the past three years, leaving most of the nearly $1.2 trillion in unrealized value hostage to a liquidity event, as shown in the graph below.
Unicorn investors probably are eager for a return to strong exit years like 2012 and 2014. Each of those years, however, was distorted by Facebook’s and Alibaba’s IPOs tallied at $100 billion and $160 billion, respectively. Now that the market for tech deals is cooling down and valuation levels appear highly questionable, such blockbuster exits are unlikely to repeat.
Source: ARK Investment Management LLC, 2019; CrunchBase, CBInsights, Wallstreet Journal, Company Information
Mobility sector unicorns highlight the imbalance between low exit volumes and record high valuations, as shown below. Only 3% of the unrealized value exited over the past six years. During the same time, the valuations of mobility startups soared more than 20-fold from $9 billion to more than $200 billion, with 24 of them now accounting for nearly 20% of the $1.15 trillion in unrealized value of all unicorns. Because so few have been able to exit, some are lowering their valuations. Perhaps private investors should do the same with their return expectations.
Source: ARK Investment Management LLC, 2019; CrunchBase, CBInsights, Wallstreet Journal, Company Information
The Smart Home Race Is Heating Up
Follow Nick on Twitter @GrousARK
One of the largest homebuilders in the United States, KB Home (KBH), is incorporating its KB Smart Home System with a Google Assistant in two of its newest developments. The homes will come with a Google Wi-Fi network, two Google Home smart speakers, and a Nest Hello video doorbell. The Google and KB Home partnership is the first in the homebuilding industry, but probably the first of many as competition among tech firms vying for a place in the home heats up.
Google launched its smart speaker in November 2016, two years after Amazon’s Echo hit the market. Having acquired Nest Labs in January 2014, Google entered first but Amazon gets more credit for creating the smart home market with the Echo in late 2014.
Amazon’s smart speaker helped it gain majority share of the U.S market, but Google, Apple and others have ramped and are beginning to catch up. In 2017, Amazon’s Echo outsold Google’s Home 3.7 to 1, a ratio that dropped to 0.9 late last year. In other words, throughout 2017 and 2018Google outsold Amazon in the US, pushing the Echo from roughly 93% market share in 2016 to roughly 63%, as shown below.
Only a quarter of American households own smart speakers, so the market is far from saturated. Consequently, if Google strikes partnerships with more homebuilders, as well as other industries touching the home like insurance companies, it could continue to erode Amazon’s lead.
The BIS Concludes That Bitcoin Is Unsustainable. Really?
Follow Yassine on Twitter @yassineARK
A recent report by the Bank for International Settlements’ (BIS) Principal Economist, Raphael Auer, discusses the unsustainable economics of Bitcoin’s proof of work and concludes that it is destined to fail. The paper delineates two economic limitations of the proof-of-work model: “The first lies in the fatality of 51% attacks and the extreme costs of ensuring payment finality in a reasonable space of time. The second is that these systems will not be able to generate transaction fees that are adequate to guarantee payment security in the future.”
Upon deeper inspection of Auer’s arguments, we posit that the first limitation is based on a fundamental misunderstanding of proof-of-work, while the second has been debunked at length in the past. That said, we agree that Bitcoin’s security mode is critical to its success and encourage active research on its long term sustainability.
Addressing the first limitation that Auer cites, Hugo Nguyen lays out a cogent analysis on Bitcoin’s security here. Addressing the second limitation, Bitcoin’s ability to generate market-based fees, Nic Carter explains two incompatible economic theories here.
Perhaps more worthy of discussion is the author’s employer, the Bank of International Settlements. Nearly a century old, Switzerland-based BIS is governed by 60 of the world’s largest central banks, entities that might have a strong point of view about the viability of a global digital currency and the technology supporting it. Effectively, Bitcoin’s base layer is competing with central banks and BIS/Fedwire for money-issuance and final settlement. Potentially disintermediating central control in final settlement, Bitcoin would rob the BIS of one of the most important reasons it exists.