Supreme Tech Quisling

Good for China and Google, Bad for America.
Google is sharing a military technology with America's monstrous enemy.
By Peter Thiel
DeepMind drew attention in 2016, when its AlphaGo software beat Lee Sedol, champion of the game Go.

A “Manhattan Project” for artificial intelligence is how Demis Hassabis, the founder of DeepMind, described his company in 2010, when I was one of its first investors. 

I took it as figurative grandiosity. 

I should have taken it as a literal warning sign, because that is how it was taken in foreign capitals that were paying close attention.
Now almost a decade later, DeepMind is the crown jewel of Google’s A.I. effort. 

It has been the object of intense fascination in East Asia especially since March 2016 when its AlphaGo software project beat Lee Sedol, a champion of the ancient strategic board game of Go.
Such feats notwithstanding, DeepMind, having now gone on three times longer than the original Manhattan Project, is not clearly any closer to its core goal of creating an “artificial general intelligence” that rivals or replaces humanity. 

But it is finally becoming clear that, as with nuclear fission before it, the first users of the machine learning tools being created today will be generals rather than board game strategists.
A.I. is a military technology. 

Forget the sci-fi fantasy; what is powerful about actually existing A.I. is its application to relatively mundane tasks like computer vision and data analysis. 

Though less uncanny than Frankenstein’s monster, these tools are nevertheless valuable to any army — to gain an intelligence advantage, for example, or to penetrate defenses in the relatively new theater of cyberwarfare, where we are already living amid the equivalent of a multinational shooting war.
No doubt machine learning tools have civilian uses, too; A.I. is a good example of a “dual use” technology. 

But that common-sense understanding of A.I.’s ambiguity has been strangely missing from the narrative that pits a monolithic “A.I.” against all of humanity.
A.I.’s military power is the simple reason that the recent behavior of America’s leading software company, Google — starting an A.I. lab in China while ending an A.I. contract with the Pentagon — is shocking. 

As President Barack Obama’s defense secretary Ash Carter pointed out last month, “If you’re working in China, you don’t know whether you’re working on a project for the military or not.”
No intensive investigation is required to confirm this. 

All one need do is glance at the Communist Party of China’s own constitution: Xi Jinping added the principle of “civil-military fusion,” which mandates that all research done in China be shared with the People’s Liberation Army, in 2017.
That same year, Google decided to open an A.I. lab in Beijing. 

According to Fei-Fei Li, the executive who opened it, the lab is “focused on basic A.I. research” because Google is “an A.I.-first company” in a world where “A.I. and its benefits have no borders.” All this is part of a “huge transformation” in “humanity” itself. 

Back in the United States, a rebellion among rank and file employees led Google last June to announce the abandonment of its “Project Maven” A.I. contract with the Pentagon. 

Perhaps the most charitable word for these twin decisions would be to call them naïve.
How can Google use the rhetoric of “borderless” benefits to justify working with the country whose “Great Firewall” has imposed a border on the internet itself? 

This way of thinking works only inside Google’s cosseted Northern California campus, quite distinct from the world outside. 

The Silicon Valley attitude sometimes called “cosmopolitanism” is probably better understood as an extreme strain of parochialism, that of fortunate enclaves isolated from the problems of other places — and incurious about them.
A little curiosity about China would have gone a long way, since the Communist Party is not shy about declaring its commitment to domination in general and exploitation of technology in particular. 

Of course, any American who pays attention and questions the Communist line is accused by the party of having a “Cold War mentality” — but this very accusation relies on forgetfulness and incuriosity among its intended audience.
Since 1971, the American elite’s Cold War attitude toward China’s leaders has been one of warm indulgence. 

In the 1970s and 1980s, that meant supporting China against a greater adversary, the Soviet Union. What is extremely strange is that this policy of indulgence continued and even deepened after the Soviet Union’s collapse in 1991.
A few years after the Cold War ended, American leaders started treating China the way they had treated West Germany and Japan. 

We tolerated punishing trade deficits in the 1970s and 1980s to support those two allies, and we had strategic reasons to do it. 

As for building up China in the 1990s and 2000s, America’s generosity was supposed to somehow lead to China’s liberalization. 

In reality, it led to the transfer of our industrial base to a foreign rival.In this sense, a zombie “Cold War mentality” never went away — though it certainly stopped making sense. 

Only recently, with help from Xi Jinping’s decision last year to, in effect, declare himself potential leader for life, has Donald Trump become the first president since Richard Nixon to pay attention and run a reality check on China.
Silicon Valley is not alone in its inattention to geopolitical reality; Wall Street has been eager to make excuses for Google’s naïveté. 

The timing is not coincidental; just this week American officials met their Chinese counterparts in Shanghai to negotiate a trade deal.
The flip side of China’s huge trade surplus has been America’s huge current account deficit. 

All of the dollars we send abroad that never get used to buy American goods have to go somewhere, and most go through New York’s money center banks on their way to buying financial assets. 

Since upsetting this imbalance is a threat to profits, Wall Street would prefer to cave on trade and keep Google’s stock price high while they’re at it.
But the banks’ experience of the last few decades of globalization has not been representative. 

The trade deficits that brought flows of money to Wall Street took jobs and bargaining power away from the median worker.Wages have been stagnant since the 1970s. 

The difference between our post-1971 era of globalization and the post-1945 midcentury boom is a breakdown in the relationship between the parts and the whole: An archipelago of inward-looking, parochial places like Wall Street and Silicon Valley have done exceedingly well for themselves while their fellow citizens have been left behind in a stagnant economy.
In the 1950s, the cliché was that “what’s good for General Motors is good for the country.” 

Google makes no such claim for itself; it would be too obviously false. 

Instead, Google says it is “committed to significantly improving the lives of as many people as possible”— a standard so vague as to defy any challenge.
By now we should understand that the real point of talking about what’s good for the world is to evade responsibility for the good of the country.

Chinese Peril

AI CAN HELP HUNT DOWN MISSILE SITES IN CHINA
By Jeremy Hsu

A surface-to-air missile is seen through a doorway in Zhuhai, China.

INTELLIGENCE AGENCIES HAVE a limited number of trained human analysts looking for undeclared nuclear facilities, or secret military sites, hidden among terabytes of satellite images. 

But the same sort of deep learning artificial intelligence that enables Google and Facebook to automatically filter images of human faces and cats could also prove invaluable in the world of spy versus spy. 

An early example: US researchers have trained deep learning algorithms to identify Chinese surface-to-air missile sites—hundreds of times faster than their human counterparts.
The deep learning algorithms proved capable of helping people with no prior imagery analysis experience find surface-to-air missile sites scattered across nearly 90,000 square kilometers of southeastern China. 

Such AI based on neural networks—layers of artificial neuron capable of filtering and learning from huge amounts of data—matched the overall 90 percent accuracy of expert human imagery analysts in locating the missile sites. 

Perhaps even more impressively, the deep learning software helped humans reduce the time needed to eyeball potential missile sites from 60 hours to just 42 minutes.
"The algorithms were used to find the locations where they said there is a high confidence of a missile site, and then humans reviewed the results for accuracy and figured out how much time the algorithms saved," says Curt Davis, a professor of electrical engineering and computer science, and director of the Center for Geospatial Intelligence, at the University of Missouri. 

"To my knowledge that’s never been studied before: How much time did you save, and how does that ultimately impact the human performance?"
The University of Missouri study, published on October 6 in the Journal of Applied Remote Sensing, comes at a time when satellite imagery analysts are figuratively drowning in a deluge of big data. DigitalGlobe, a leading commercial satellite imagery company, generates about 70 terabytes of raw satellite imagery each day, never mind all the imagery data coming from other commercial satellites and government spy satellites.
Davis and his colleagues showed how off-the-shelf deep learning models—heavily trained and modified for satellite imagery analysis—could identify objects of potentially great interest to intelligence agencies and national security experts. 

The deep learning models, including GoogleNet and Microsoft Research's ResNet, were initially created to detect and classify objects in traditional photo and video imagery. 

Davis and his colleagues adapted such models to the challenges and limitations of interpreting satellite imagery, such as training some deep learning models to interpret both color and black-and-white imagery, in case only black-and-white images of SAM sites were available.
They did so with satellite imagery representing a huge swath of Chinese territory, not that much smaller than the entire country of North Korea.
And in fact, analysts rely extensively on satellite imagery to keep track of how North Korea's weapons programs evolve. 

Human analysts have already likely identified most, if not all, existing SAM sites within the relatively small country. 

But similar deep learning tools could help automatically flag new SAM sites that appear in North Korea or other countries. 

Knowing the location of existing and new SAM sites can sometimes lead analysts to other locations of interest, because countries often place SAM sites in specific areas to defend valuable nearby assets from air attack.
The latest study also illustrates the challenges of applying deep learning AI to satellite imagery analysis. 

One major problem is the relative lack of large training datasets that include the hand-labeled examples needed to train deep learning algorithms to accurately identify features in satellite imagery. The University of Missouri team combined public data on the worldwide locations of about 2,200 SAM sites with DigitalGlobe satellite imagery to create their training data, and then tested four deep learning models to find the best-performing one.
The researchers ended up with only about 90 positively identified Chinese SAM site examples to train their AI. 

Such a puny training dataset might normally fail to yield accurate deep learning results. 

To get around that problem, Davis and his colleagues transformed the 90-odd training samples into about 893,000 training samples by shifting the original images slightly in different directions.
The impressive deep learning performance in the study likely benefited from SAM sites being fairly large, and having distinctive patterns when viewed from above in satellite images. 

Davis cautioned that deep learning algorithms face a much greater challenge when trying to analyze smaller objects such as mobile missile launchers, radar antennas, mobile radar systems, and military vehicles, because the available satellite imagery data will have fewer pixels to work with in extracting identifying features.
"It is an open question in our mind how well convolutional neural networks will work on smaller scale objects like this, especially when tested against large area datasets like we did with the China study," Davis says.
Even imperfect AI tools could prove incredibly helpful for intelligence gathering. 

For example, the International Atomic Energy Agency has the unenviable task of monitoring all declared nuclear facilities and also searching for undeclared facilities among nearly 200 countries. 

Deep learning tools could help the IAEA and other independent organizations use satellite imagery to monitor development of nuclear power and related weapons of mass destruction, says Melissa Hanham, a senior research associate in the East Asia Nonproliferation Program at the Middlebury Institute of International Studies at Monterey, Calif.
"We're in a world where there is just so much data that the best way to approach it is to do a good job on a lot of it rather than a perfect job on a small bit of it," Hanham says. 

"I'm looking forward to automating all the tedious and redundant parts of my job."

Chinese Peril

U.S. weighs restricting Chinese investment in artificial intelligence
By Phil Stewart | WASHINGTON

An MQ-9 Reaper remotely piloted drone aircraft performs aerial maneuvers over Creech Air Force Base, Nevada, U.S., June 25, 2015. 

U.S. Defense Secretary James Mattis testifies before the Senate Armed Services Committee on Capitol Hill in Washington, D.C., U.S., June 13, 2017.

The United States appears poised to heighten scrutiny of Chinese investment in Silicon Valley to better shield sensitive technologies seen as vital to U.S. national security, current and former U.S. officials tell Reuters.
Of particular concern is China's interest in fields such as artificial intelligence and machine learning, which have increasingly attracted Chinese capital in recent years. 

The worry is that cutting-edge technologies developed in the United States could be used by China to bolster its military capabilities and push it ahead in strategic industries.
The U.S. government is now looking to strengthen the role of the Committee on Foreign Investment in the United States (CFIUS), the inter-agency committee that reviews foreign acquisitions of U.S. companies on national security grounds.
An unreleased Pentagon report, viewed by Reuters, warns that China is skirting U.S. oversight and gaining access to sensitive technology through transactions that currently don't trigger CFIUS review. 

Such deals include joint ventures, minority stakes and early-stage investments in start-ups.
"We're examining CFIUS to look at the long-term health and security of the U.S. economy, given China's predatory practices" in technology, said a Trump administration official, who was not authorized to speak publicly.
Defense Secretary Jim Mattis weighed into the debate on Tuesday, calling CFIUS "outdated" and telling a Senate hearing: "It needs to be updated to deal with today's situation."
CFIUS is headed by the Treasury Department and includes nine permanent members including representatives from the departments of Defense, Justice, Homeland Security, Commerce, State and Energy. 

The CFIUS panel is so secretive it normally does not comment after it makes a decision on a deal.
Under former President Barack Obama, CFIUS stopped a series of attempted Chinese acquisitions of high-end chip makers.
Senator John Cornyn, the No. 2 Republican in the Senate, is now drafting legislation that would give CFIUS far more power to block some technology investments, a Cornyn aide said.
"Artificial intelligence is one of many leading-edge technologies that China seeks and that has potential military applications," said the Cornyn aide, who declined to be identified.
"These technologies are so new that our export control system has not yet figured out how to cover them, which is part of the reason they are slipping through the gaps in the existing safeguards," the aide said.
The legislation would require CFIUS to heighten scrutiny of buyers hailing from nations identified as potential threats to national security. 

CFIUS would maintain the list, the aide said, without specifying who would create it.
Cornyn's legislation would not single out specific technologies that would be subject to CFIUS scrutiny. 

But it would provide a mechanism for the Pentagon to lead that identification effort, with input from the U.S. technology sector, the Commerce Department, and the Energy Department, the aide said.
James Lewis, an expert on military technology at the Center for Security and International Studies, said the U.S. government is playing catch-up.
"The Chinese have found a way around our protections, our safeguards, on technology transfer in foreign investment. And they're using it to pull ahead of us, both economically and militarily," Lewis said.

"I think that's a big deal."

China made the United States the top destination for its foreign direct investment in 2016, with $45.6 billion in completed acquisitions and greenfield investments, according to the Rhodium Group, a research firm. 

Investment from January to May 2017 totaled $22 billion, which represented a 100 percent increase against the same period last year, it said.

AI'S ROLE IN DRONE WARFARE
Concerns about Chinese inroads into advanced technology come as the U.S. military looks to incorporate elements of artificial intelligence and machine learning into its drone program.
Project Maven, as the effort is known, aims to provide some relief to military analysts who are part of the war against Islamic State.
These analysts currently spend long hours staring at big screens reviewing video feeds from drones as part of the hunt for insurgents in places like Iraq and Afghanistan.
The Pentagon is trying to develop algorithms that would sort through the material and alert analysts to important finds, according to Air Force Lieutenant General John N.T. "Jack" Shanahan, director for defense intelligence for warfighting support.
"A lot of times these things are flying around (and)... there's nothing in the scene that's of interest," he told Reuters.
Shanahan said his team is currently trying to teach the system to recognize objects such as trucks and buildings, identify people and, eventually, detect changes in patterns of daily life that could signal significant developments.
"We'll start small, show some wins," he said.
A Pentagon official said the U.S. government is requesting to spend around $30 million on the effort in 2018.
Similar image recognition technology is being developed commercially by firms in Silicon Valley, which could be adapted by adversaries for military reasons.
Shanahan said he was not surprised Chinese firms were making investments there.
"They know what they're targeting," he said.
Research firm CB Insights says it has tracked 29 investors from mainland China investing in U.S. artificial intelligence companies since the start of 2012.
The risks extend beyond technology transfer.
"When the Chinese make an investment in an early stage company developing advanced technology, there is an opportunity cost to the U.S., since that company is potentially off-limits for purposes of working with (the Department of Defense)," the report said.

CHINESE INVESTMENT
China has made no secret of its ambition to become a major player in artificial intelligence, including through foreign acquisitions.
Chinese search engine giant Baidu Inc launched an AI lab in March with China's state planner, the National Development and Reform Commission. 

In just one recent example, Baidu Inc agreed in April to acquire U.S. computer vision firm xPerception, which makes vision perception software and hardware with applications in robotics and virtual reality.
"China is investing massively in this space," said Peter Singer, an expert on robotic warfare at the New America Foundation.
The draft Pentagon report cautioned that one of the factors hindering U.S. government regulation was that many Chinese investments fall short of outright acquisitions that can trigger a CFIUS review. Export controls were not designed to govern early-stage technology.
It recommended that the Pentagon develop a critical technologies list and restrict Chinese investments on that list. 

It also proposed enhancing counterintelligence efforts.
The report also signaled the need for measures beyond the scope of the U.S. military, such as changing immigration policy to allow Chinese graduate students to stay in the United States after completing their studies, instead of returning home.
Venky Ganesan, managing director at Menlo Futures, concurred about the need to keep the best and brightest in the United States.
"The single biggest thing we can do is staple a green card to their diploma so that they stay here and build the technologies here – not go back to their countries and compete against us," Ganesan said.

Google 2, China 0

Google’s AlphaGo Continues Dominance With Second Win in China
By Cade Metz

A book analyzing the play of AlphaGo Master, the Go-playing machine built by researchers at Google's DeepMind lab.

WUZHEN, CHINA — Ke Jie, the number one Go player in the world, spent much of the game playing with the hair on his head. 

Time and again, he pinched the short strands between his thumb and index fingers, twisting the hair around one and then the other. 

His opponent, AlphaGo, the machine built by researchers at Google’s DeepMind lab, merely played the game. 

And in the end, as seemed inevitable, it won.
With the win, AlphaGo claimed victory in its three-game match with the Ke Jie, taking a 2-0 lead. The victory confirmed that modern AI techniques have already exceeded the talents of even the best humans when playing the ancient game of Go—something that didn’t seem possible just a few years ago.
Last year, AlphaGo became the first machine to beat a leading Go professional when it topped the Korean grandmaster Lee Sedol. 

Considering the extreme complexity of Go—there are more potential positions in the game than atoms in the universe—the win was a turning point in the progress of artificial intelligence. Underpinned by technologies that are already changing everything from internet services to healthcare to robotics, AlphaGo is a harbinger of so many things to come.
This week’s match underlined the powers of the machine, but for Google, the event isn’t the necessarily the success the company hoped it would be. 

The match was a chance for Google to raise its profile in China, where it hasn’t offered online services since 2010, and in the weeks before the event, Google seemed to enjoy the cooperation of local authorities. 

But two days before the first game, Chinese state TV pulled out of the match. 

Then, about a half hour into the game, broadcasts from other media local went dark. 

Local news outlets did cover the event, but most did not mention Google, apparently under orders from local authorities.
In 2010, after Chinese hackers broke into Google’s internal systems and grabbed lifted information about Chinese human-rights activists from their Google Gmail accounts, the company moved its internet servers to Hong Kong, saying it would not obey Chinese internet policies, and the Chinese government banned the company’s services. 

Clearly, Google now wants back into this very large—and potentially lucrative—market, but this week shows that navigating Chinese politics and culture is enormously difficult for American companies.
The gate to the ancient Chinese city of Wuzhen, where AlphaGo is playing the current world number one, Ke Jie.
In any event, AlphaGo won the first game against Je Kie, taking hold of play rather early in the match. 

“It is like a god of a Go player,” the Chinese grandmaster said during the post-game press conference, through an interpreter. 

But for the machine, the second game was a slightly different prospect. 

Unlike in the first game, AlphaGo played the black stones, which means it played first, something it views as a small handicap. 

“It thinks there is a just a slight advantage to the player taking the white stones,” AlphaGo’s lead researcher, David Silver, said just before the game. 

And as match commentator Andrew Jackson pointed out, Ke Jie is known for playing well with white.
That said, AlphaGo typically overcomes that small handicap. 

After the match in Korea last year, the DeepMind team rebuilt the system, significantly improving the architecture, and in January, when it played several top players over the internet under the pseudonym “Master,” it won all 60 of its games.
The new AlphaGo tends to play what experts previously viewed as an unusual opening, a strategy called “3-3 point.” 

And indeed, it played the opening in today’s game. 

In this way, the contest became a mirror image of game one. 

Ke Jie played “3-3 point” in the opening game, mimicking AlphaGo’s new approach to this ancient game—though, for him, the gambit was not successful.
In the early stages of game two, however, Ke Jie seemed to hold his own. 

“Incredible,” DeepMind founder and CEO Demis Hassabis tweeted about an hour into the game. “According to #AlphaGo evaluations Ke Jie is playing perfectly at the moment.” 

But as the match continued, according to match commentators, he seemed to lose ground in the lower half of the board. 

And soon, he began twisting the hair on his head.
By the game’s third hour, the 19-year-old had used up about twice as much playing time as AlphaGo. He was on the verge of losing the lower half of the board. 

And the hair-twisting continued. 

Early in the contest, Ke Jie had worked to create an enormously complicated game, but with a typically swift move as the four hour approached, AlphaGo made it simple again. 

“The fact that AlphaGo has simplified the game is a bad sign for the human player,” said match commentator Michael Redmond. 

Within 15 minutes, Ke Jie resigned.
The last game of the series is set for Saturday. 

But before then, on Friday, the machine will play two other game: one against at team of top human players and alongside human players. 

And frankly, that’s where the interest lies. 

Given AlphaGo’s dominance, there’s little mystery left when it goes head-to-head with the single grandmaster.