19 minutes 41 seconds
00:00:00 - 00:00:31
It's 2017 and a United States Marine needs to report a suspicious convoy. He can see the armored vehicles driving through his scope, but he wants to know if anyone else has intel about it. 25, 000 feet up In the air, a Predator drone has been circling for hours. If the guys in these vehicles are up to something, the drone should have footage of it. But instead of relying on some sophisticated gadget to call in a report, the soldier just pulls out his iPhone and drops a pin on Google Maps.
00:00:31 - 00:00:32
00:00:32 - 00:00:50
It's an odd situation. The same app we use to drive to work has somehow become an essential tool in the arsenal of the modern American war fighter. When I heard this, I couldn't believe it. How does the US military not have better options? It's clearly not a technology issue, since Google Maps seems to work just fine.
00:00:50 - 00:01:28
And it can't be a budget issue, since the US Department of Defense spends over a trillion dollars every single year. Google could clearly solve this problem, and save American lives in the process. But as you'll see, Google has a complicated history with the US military. Silicon Valley was built in large part because of military defense contracts. Throughout the Cold War, lucrative defense contracts flowed to Silicon Valley research labs to develop communication and surveillance technologies.
00:01:29 - 00:01:59
That's because the guy who would run all electronic warfare during World War II was an engineering professor at Stanford. His name was Fred Terman, and he wound up encouraging 2 of his students to start Hewlett-Packard. HP would become the first really massive success to come out of Silicon Valley. The company went public in 1957 and led to a wave of new investment interest. As capital gains rates fell and laws changed to allow pension funds to invest in VC firms, the startup funding model changed completely.
00:01:59 - 00:02:20
Instead of Relying on defense contracts to get off the ground, companies would work with venture capitalists. The model warped, and tons of new startups built disruptive technologies. But along the way, they lost the connection to the US military. A company like Google thought of itself as purely a consumer product with no military applications whatsoever. But that wouldn't last long.
00:02:29 - 00:02:55
In a world Faced with a threat of sudden war that may destroy whole continents, we want to enforce the peace by making it clear to a potential aggressor that such a war would result in swift and terrible retaliation. We are voicing this warning not only by building a strong and alert military establishment, but also by underwriting an unparalleled program of scientific and technical study and experiment."
00:02:56 - 00:03:16
Technology has been at the heart of US military strategy for decades. Back in the 1950s, President Eisenhower realized that the United States was outmanned against the Soviets. There was no way that America and her NATO allies could deter an advance by the USSR and their Warsaw Pact allies. The only option was to rely on nuclear weapons.
00:03:16 - 00:03:26
The atomic bomb, radar, guided missiles, proved in military terms the life or death value of scientific study and engineering experimentation.
00:03:27 - 00:03:53
The plan worked, and not only did the Cold War never turn hot, but nuclear weapons were never actually deployed on the battlefield. Advanced technology had effectively offset the Soviets' conventional strengths. This theory came to be known as offset strategy, and it was employed again in the mid-1970s. The Vietnam War had come to an end, and American defense budgets were shrinking. Warsaw Pact forces still outnumbered NATO forces by 3-1 in Europe.
00:03:53 - 00:04:13
There was no other option. The government had to start doing more with less. So the Secretary of Defense again employed offset strategy to level the playing field. They started building new reconnaissance platforms, better precision guided weapons, and stealth technology. They built the F-117 stealth fighter because of this, and the Global Positioning System, or as we know it, GPS.
00:04:14 - 00:04:19
Fast forward to 2016, and the United States was thinking about offset strategies once again.
00:04:19 - 00:04:24
China is the most difficult competitor the United States has ever faced.
00:04:24 - 00:04:33
Today I'm announcing a new defense innovation initiative. An initiative that we expect to develop into a game-changing third offset strategy.
00:04:33 - 00:04:45
Ash Carter was Secretary of Defense at the time, and he had built something called the Defense Innovation Board. The goal was to start bringing some of the tech innovation and engineering expertise of Silicon Valley to the US military.
00:04:45 - 00:05:00
This is a serious business, a serious matter. But in addition to dangers, there are great opportunities to be seized through a new level of partnership between the Pentagon and Silicon Valley, opportunities that we can only realize together.
00:05:01 - 00:05:06
Now, Ash Carter needed someone to implement this, so he turned to his deputy, this guy named Bob Work.
00:05:06 - 00:05:15
He said, hey, look at the technological trends in the Western Pacific. What are some of the things we could do to offset or counteract what the Chinese are doing?
00:05:15 - 00:05:39
The Defense Innovation Board would just provide recommendations to the military. Bob Work actually had the authority to spin up real projects and start funding new developments. So Bob and his team started scoping a project, and on April 26, 2017, they released this memo. In typical military fashion, it has the most jargony name ever. Establishment of an Algorithmic Warfare Cross-Functional Team.
00:05:39 - 00:05:47
Fortunately, they gave it a shorter name, Project Maven. The Department of Defense had a data problem.
00:05:47 - 00:05:57
Our problem was Intel analysts are sorting through with eyeballs far too much information than they could ever deal with. 12 hours at a time looking at what we call full motion video.
00:05:57 - 00:06:17
Every military outpost now had sensor towers with high quality cameras recording 24-7. Drones patrolled the skies looking for threats. And even satellite imagery was providing extra battlefield intelligence. But fusing all this data together was cumbersome. And that's why soldiers were turning to Google maps to centralize their findings on the ground.
00:06:17 - 00:06:27
This made no sense though. There had to be a better way. The Department of Defense could tell that foreign adversaries were investing heavily in artificial intelligence, but the United States was falling behind.
00:06:27 - 00:06:37
Clearly the Chinese have made enormous strides in the last 2 decades and continue to make even greater strides. So you've got to pay attention to what they're doing in the military technical competition.
00:06:38 - 00:07:05
See, when a Predator drone is flying around, it's filming everything. The cameras used to be low-res and only really show 1 small area. But now, the cameras were getting better, and there was a lot more information to be gathered from each video feed. The problem was getting so bad that they estimated if they didn't do something about it, fully 1 third of US Air Force personnel would need to be watching video feeds all day long. These soldiers wouldn't be calling in strikes or anything like that.
00:07:05 - 00:07:23
This was just to process the footage and tag what was in the videos. It should be pretty obvious that this is a great place to apply technology. Your iPhone can pick out pictures of your dog in your camera roll. The military should be able to identify cars and drone footage without needing someone to sit there all day long. Saving time was only half the battle though.
00:07:23 - 00:07:40
Actually moving video footage around is extremely difficult in a war zone. So the more analysis that can be done at the edge, the better. It's just more efficient if images can be tagged before they're uploaded. So that's what Bob and his team proposed. Project Maven would work in 90-day sprints to use machine learning to tag videos.
00:07:40 - 00:07:57
The system wouldn't call the shots. It would only act as a data cleanup tool to help understand what's going on in all the footage. To carry the project through to completion, 2 more military guys joined up. Lieutenant General Jack Shanahan of the United States Air Force and Colonel Drew Cukor of the Marine Corps.
00:07:57 - 00:08:09
Artificial intelligence, we're seeing the real power of it in commercial industry and in academia, we have to embrace it in the entire United States government and for us in the Department of Defense. The table stakes are high.
00:08:11 - 00:08:45
2017 was a great time to start investing in computer vision. As we've all seen recently, artificial intelligence has gotten really good at understanding photos, but it's still important to recognize the strengths and weaknesses of AI. The best place to deploy artificial intelligence is in a situation where you have a ton of data that needs to be processed, and most importantly, you can put up with crazy mistakes popping up from time to time. If your iPhone mistakes your baby cousin for a dog, that's just a funny moment. But if your self-driving car mistakes a wall for a road, you could be in real danger.
00:08:45 - 00:09:01
Project Maven wasn't a weapons project. It was just a data labeling project. If a house got mislabeled as a truck, sure, it might send off an alert to someone to review the footage, but it wasn't going to trigger any real military action. And that's why the Google team signed off on the project.
00:09:01 - 00:09:05
AI is 1 of the most important things humanity is working on.
00:09:05 - 00:09:32
Sundar Pichai had taken over as CEO of Google in 2015. He was still fairly new on the job, but Sundar had a few key people on his team who understood the opportunity in front of them. Google's cloud offering had been fighting to catch up to Amazon and Microsoft for years. The CEO of Google Cloud at the time, Diane Green, had a big task in front of her. AWS and Microsoft Azure were both well past $10 billion in annual revenue, but Google's cloud offering was a distant third.
00:09:32 - 00:09:52
AWS had launched in 2006 and become the default selection for new companies. Azure launched in 2008 and had easy inroads into every business that ran Windows. Google was playing catch up, but the game was far from over. The US military was just starting to modernize their tech stack, and a Google employee named Scott Froman saw an opening.
00:09:52 - 00:10:02
We found that these really neat capabilities around cloud and artificial intelligence allow a government to deter, compete, and win in ways that were not possible years ago.
00:10:02 - 00:10:31
Scott was in charge of defense sales for Google, and he thought that Project Maven could be a good jumping off point. They'd start with a small data processing contract for Maven, and then eventually scale up to support the entire Department of Defense. The Project Maven contract wasn't huge by Google's standards, just $15 million. For reference, Snapchat has a $2 billion contract with Google Cloud. And that $15 million wouldn't actually get Google to deliver a finished system to the Pentagon.
00:10:31 - 00:10:52
The whole point of Google Cloud is that it's a platform. That means that customers need to bring their own systems, and Google just handles hosting the programs in their data centers. Just as Snapchat engineers have to write their own code before deploying it to the Google Cloud, the DoD would be responsible for building most of the system. Google would simply supply access to their TensorFlow API.
00:10:52 - 00:10:56
TensorFlow is this machine learning library that's used across Google for
00:10:56 - 00:11:17
our speech recognition systems and the new Google Photos product. But TensorFlow is open source. What this means is that if the DoD built a TensorFlow model on Google's cloud and then wanted to go somewhere else, they could. You can actually run TensorFlow on AWS or Microsoft Azure, no problem. Google gets all the focus in this story because of what eventually became of the project.
00:11:18 - 00:11:40
But it's important to remember that this whole thing was what's called a pathfinding project. Jack and Drew, the 2 guys actually running the thing, their goal was just to help the Pentagon understand how to work with big tech in the first place. Working with government data can be tricky. Tech companies need proper clearances, and Google didn't have them before Maven. This was an opportunity to change that, though.
00:11:40 - 00:12:03
The DoD guys working on Project Maven were able to fast-track Google's security certifications. This would open up new opportunities to work with the government and prove to everyday business customers that Google had truly top-tier security. But then things started to go south. Toward the end of 2017, Google was about to seal the deal to work on Project Maven. And the chief scientist for AI at Google Cloud was pumped.
00:12:04 - 00:12:08
Her name is Fei-Fei Li, and she's 1 of the most respected computer scientists in the world.
00:12:08 - 00:12:15
1 of the most important thing for me is not only to advance AI, but also to democratize AI.
00:12:15 - 00:12:31
Working with the Department of Defense was great from a cloud technology angle. Google's data centers are some of the most secure in the world, and plenty of government organizations already use Google for cloud hosting. There was absolutely no problem there, but she identified a potential stumbling block.
00:12:31 - 00:12:36
We need to be mindful that human values define machine values.
00:12:37 - 00:13:03
She was worried that if people started thinking that Google was building AI for the military, the narrative around the project could turn negative. New technologies tend to be scary, and AI is probably the best example of this. There are just way too many movies where killer robots take over the world, and no 1 wants to build Skynet. So she emailed her team, warning them to avoid mentioning AI in the press. This was a smart move in theory, but it backfired.
00:13:03 - 00:13:29
Instead of controlling the narrative and getting out in front of the AI issue, the Google team sat on the story and decided to say nothing. They could have made it clear that Google was just providing cloud services, open source technology, stuff that literally anyone can use with a Google account, but instead they just said nothing. Fei-Fei Li sent that email about clarifying the Project Maven scope of work in September of 2017. And just 6 months later, news of the project had leaked.
00:13:29 - 00:13:38
The Pentagon announced plans last September to award a 10-year cloud computing contract that could be worth billions to the company that wins it.
00:13:38 - 00:13:40
Drones, warfare and Google.
00:13:40 - 00:13:43
The growing relationship between the United States military.
00:13:44 - 00:13:45
And Silicon Valley.
00:13:45 - 00:13:49
What's behind Google's contract with the Department of Defense for a project called Maven.
00:13:49 - 00:13:55
Thousands of Google workers now protesting Google's involvement in a Pentagon drone program.
00:13:55 - 00:13:59
The project had sparked anger amongst Google employees. Who are worried the
00:13:59 - 00:14:05
AI results are being used, well, for evil. At the heart of the movement was this AI researcher, Meredith Whitaker.
00:14:05 - 00:14:13
Worked to protest Google's AI contracts with the Defense Department, their plans to build a censored search engine, and their inadequate AI ethics policy.
00:14:13 - 00:14:25
She wanted Google to draw a hard line in the sand and not work with the Department of Defense. So she and 4, 000 other Google employees signed a letter to Sundar Pichai asking him to cancel the project immediately.
00:14:25 - 00:14:29
Google says Project Maven is for quote non-offensive purposes.
00:14:29 - 00:14:41
When I was at Google and we learned about Maven. We saw the code. It wasn't talking about tanks. It wasn't talking about missile launchers. It was talking about people and it was talking about vehicles.
00:14:43 - 00:15:11
A dozen Google employees quit the company because of the project, and Google was stuck in a difficult spot. On the 1 hand, they had already scoped the project to avoid any offensive capabilities. In the minds of the Google execs, they had already drawn a clear line. If they backed down, the narrative would stick and everyone would think that Google had planned to build killer robots. On the other hand, if they pulled out of the contract, that could hurt Google too.
00:15:11 - 00:15:42
Maven was supposed to be a jumping off point to more government work. The DoD was about to start accepting bids for a $10 billion contract, and Google wouldn't stand a chance of winning if they backed out of Maven. Even though the Google activists argued that helping the Pentagon tag images was evil, it can just as easily be argued that pulling out of the contract was evil. Artificial intelligence, when used properly, dramatically reduces human error. If Google chose to sit on the sidelines and not help, US troops might suffer grave consequences.
00:15:42 - 00:16:06
It's important to remember what Project Maven was all about. Right in the original memo, Bob Wark states that all this tech is in support of the Defeat ISIS campaign. This is a brutal militant group we're talking about here. Ironically, I can't even show you evidence of how bad ISIS actually is here on YouTube because Google will instantly take this video down if I do. And all this was complicated by the fact that Google had built an AI lab in China.
00:16:06 - 00:16:24
There was no question that the Chinese government was benefiting from Google's technology. I mean, in an extreme example, TensorFlow is open source. They can just download it from GitHub. So it's such an awkward situation to cut the United States military off from Google's resources, all while China can do whatever they want. So what would you do in this situation if you were the CEO of Google?
00:16:26 - 00:16:53
4, 000 signatures on a letter sounds like a lot, but remember, Google had 80, 000 employees at this time. We're talking about 5% of the workforce and the 10 people that quit in protest? Well, Google is so big that dozens of people quit every single day. It's important to remember the scale of things at big tech companies. Internal pressure can snowball though, and at the end of the day, Project Maven was such a small portion of Google's cloud revenue that they probably just figured it wasn't worth the hassle.
00:16:53 - 00:16:59
So Google pulled out. They didn't cancel the contract outright, but they did announce that they wouldn't be renewing it.
00:16:59 - 00:17:11
So I think It's fair to be worried about AI, you know, so I wouldn't say we are just being optimistic about it, but we want to be thoughtful about it. AI holds the potential for some of the biggest advances we are going to see.
00:17:11 - 00:17:17
This was a tough decision for the team who had worked hard on the project. Jack Shanahan called the incident a canary in the coal mine.
00:17:17 - 00:17:24
Sort of surprising to me and disappointing that somewhere along the way the adjective controversial has been applied in front of Project Maven. We were being very transparent about
00:17:24 - 00:17:55
what we were trying to do. Basically, the military could no longer count on the technology industry to show up and help, even if they were being paid handsomely. The fallout of Project Maven would continue later that year during the bidding process for the Pentagon's $10 billion cloud contract. That bigger project was called JEDI, short for Joint Enterprise Defense Infrastructure, And the goal was to finally consolidate government data centers and improve efficiency. They pulled out of the Jedi bidding process claiming that it wouldn't align with Google's AI principles.
00:17:55 - 00:18:21
But that's only half the story. There's a theory that Google just couldn't offer a competitive bid, so they used the ethical high ground as cover for their decision to pull out. Amazon and Microsoft are both just years ahead of Google in terms of cloud capabilities. If Google couldn't win the contract, maybe they could get a nice headline about their AI principles as a consolation prize. What's interesting is that by all accounts, Project Maven moved forward with little to no interruption.
00:18:21 - 00:18:49
The contract has been picked up by other tech companies, and the use of artificial intelligence in the military is slowly ramping up. Microsoft and Amazon both received Maven contracts, along with a bunch of smaller tech companies. After all, no 1 wants the government to force 100, 000 Air Force service members to watch drone footage all day long. But the story doesn't end there. After this huge blow up with media exposés, resignations, killer AI, Google is working with the military again.
00:18:49 - 00:19:17
This guy named Thomas Kurian took over as Google Cloud's CEO and he's been making sure that Google will help the military as much as possible while upholding their AI guidelines. This is a great outcome and honestly, it's cause for optimism about big tech. Part of what makes America great is that its citizens don't have to work for the military if they don't want to. That's just not the case in so many other countries. You can't say no to working with the Chinese Communist Party if you're building a tech company in China.
00:19:17 - 00:19:35
It's just not an option. As big of a mess as Project Maven was, in many ways, it's good that it happened when it did. It was clearly a wake-up call for the tech industry. And when Russia invaded Ukraine, public support for defense technology shot through the roof. No 1 wants to see bombs falling in cities.
00:19:35 - 00:19:35
But you know what tool people used to track the developments of the Ukrainian crisis? They used Google Maps.
Omnivision Solutions Ltd