After lunch, Chen had his driver take them by West Lake before they left Hangzhou on the return trip to Shanghai. They spent an hour walking along the lake shore and out onto the Su causeway, talking about the history of West Lake and the government’s efforts to expand domestic and international tourism.
When the discussion turned to business, Chen said, “Whether you are dealing with consumer marketing or political marketing, two things are clear. First, the more you know about your subject, the more you will be able to influence him to do what you want him to do, whether it’s buying your shirt or voting for your candidate. Second, AI can make this process faster and more accurate. As you know, AI needs data to learn. Our ability to access and use large amounts of personal data to teach and test our AIs has been an important factor in China’s AI leadership.”
Ward said, “How are you able to know this learning and the algorithms it has produced will apply effectively outside of China?”
Chen said, “We have access to a lot of more data from the United States than you may think. I am sure you understand my point. Even so, additional U.S. data would be helpful in fine tuning our algorithms across broader demographics. This is one of the benefits we could get from working with your company.”
On the drive back to Shanghai, they discussed next steps.
They were back in Shanghai in time for Ward to shower and change for her dinner with Zhang Xiu Ying, an old classmate from her days at Stanford.
Chapter 15
Taking the elevator down to the lobby of her hotel, Ward spotted Zhang standing in the lobby near the door to the motor court. She was easy to pick out of a crowd. Tall and slender, with long black hair, a perfect complexion and the sophisticated presence of a model, she could turn a room, as Ward told her more than once when they were at Stanford. Tonight, she was wearing dark blue slacks and a long-sleeved red and blue print silk blouse.
They had been close their last two years at Stanford, sharing classes, parties and weekend outings with friends. After she graduated, Zhang returned to China. Still unmarried, she worked for the Ministry of State Security, but never told Ward what she did.
Whatever her position was, Zhang was clearly well-connected. Although she had been helpful in opening government and business doors for PaprW8 on several occasions, Zhang always made clear her job and how she made things happen were not topics to be discussed. When she visited the States, she looked Ward up if she was in northern California. Ward did the same when she was in Beijing or Shanghai. This trip was a little different, as Zhang had texted Ward a couple of weeks ago asking when she planned to be in China next and suggesting they get together.
Spotting Ward, Zhang waved to her across the lobby and walked over to greet her. The women shook hands and hugged. Looking Ward over, Zhang said, “Hey, great dress! Is it a Yi-ming?”
In a rare exception to her conservative China travel wardrobe, Ward had on a silk print qipao dress hemmed several inches above the knee. The dress was by Yi-ming, a young Chinese fashion designer. It was a modern take on the classic close-fitting cheongsam sheath that originated in 1920s Shanghai after the founding of the Republic of China in 1912. Ward said, “It is! I bought it here on my last trip. I love her new design of a traditional style. I like to think it’s my way of showing respect for China’s history and future.”
Zhang said, “Nice marketing story. Besides that, it makes you look awesome.” She winked.
They took a cab to Zhang’s favorite restaurant in the old French Concession. After a round of Tsingtao beer, they shared soup dumplings, braised eggplant, cumin ribs and sweet candied bananas. They talked about the fun they’d had in college, as they always did. Zhang probed Ward about how her work was going, while remaining guarded about her own personal life and career. Ward was surprised to hear her ask whether PaprW8 was interested in attracting more social media advertising and marketing business from political customers. Ward said PaprW8 thought the political sector was a valuable revenue opportunity. She was even more surprised when Zhang mentioned David Bernbach’s investment in PaprW8 and asked Ward what she thought of him. Ward simply said he was bright, and she was enjoying working with him.
After dinner, they had a nightcap at a local bar. The conversation turned to how Ward’s trip was going. She mentioned she had walked the Bund promenade.
Zhang said, “You go to the Bund every time you come over here! Don’t you get tired of all the people?”
Ward said, “I don’t go every time, but close. You know how I love China—the history, the scale, the amazing pace of change. Tourist attraction or not, for me the Bund symbolizes China’s past and future and its place in the new world order. Nothing says more to me about where China has been and where it’s going.”
Zhang said, “Wow! We need to get you a job with the Shanghai visitors bureau!”
Ward laughed and said, “Probably not. Speaking of tourist spots, I did finally get to the West Lake World Heritage Site in Hangzhou earlier today. It’s quite beautiful.”
Ignoring her comment on the World Heritage Site, Zhang said, “Yes, Hangzhou has become a remarkable place for startups. You know Alibaba is headquartered there. They have had a big impact on the creation of new technology companies in Hangzhou.”
They took a cab back to Ward’s hotel. As they stood in the motor court hugging briefly and telling each other how much fun it was to be together, Zhang took Ward’s hand and said, “Your investor David Bernbach has important friends here who share his political goals in America. I know you were at CnEyeco Tech while you were in Hangzhou today. Their technology and access to data would be valuable for your company. If I can help, let me know. We should stay in touch. If you need to reach me, use this.” She slipped a small piece of paper into Ward’s hand and hopped into her cab, leaving Ward standing in the motor court.
◆◆◆
By Friday afternoon Shanghai time, Ward was on her way back to San Francisco.
Chapter 16
The report on the drone was not what Dan had expected. After he took the aircraft to JPAC’s office on Sunday, two of his engineers studied the components, made a few calls and did some testing and research. On Wednesday, they met with Dan and his IT Director to review their findings and discuss next steps.
The drone weighed 60 pounds with one rotor blade gone, which was over the FAA limit. It was a Chinese design, which was not a surprise, but the product wasn’t sold in the U.S., which raised questions about how it got here. The manufacturer was a state-owned company with close ties to the Chinese military.
The real shocker was the instrument package, which included two high res cameras, two other wide-angle cameras possibly used for navigation, four directional microphones and several radio frequency receivers his team was still assessing. Navigation commands and data went by satellite. Video and other data could be written to on-board solid-state storage or transmitted real-time via satellite.
The senior hardware engineer said, “It’s a heck of a drone. Way beyond any hobbyist or recreational flyer. Almost certainly military grade.”
The communications engineer said, “We’re still trying to understand the electronics. We’ve identified a lot of commercial semiconductors and other components that give us some idea of what the systems are designed to do. The onboard storage uses two 256 gig cards for each camera, enough to hold more than 20 hours of 4K video or 60 hours of full HD video, plus gigs of other data. That’s a huge amount.”
Dan asked, “Have you been able to tell whether any video has been stored in memory?”
The communications engineer said, “No. We can power on the cameras and some of the other sensors and navigation electronics, but we haven’t been able to boot up any CPU, memory or data upload or download devices, which appear to need a password. We want to be careful about trying credentials. We don’t want to accidentally trigger some auto-erase function or have it phone home. If we really want to understand this thing, we’re going to need some military or governmen
t resources.”
Dan said, “What about electronics for network mapping or intrusion?”
The engineer said, “From the components, we think it can identify and map Wi-Fi and Bluetooth signals, including SSID network names for Wi-Fi. We don’t know whether it can probe security levels, try to connect with unsecured networks or send passwords to try to connect with secured networks, but those things wouldn’t be hard to do with the components onboard.”
The IT Director said, “Based on our review of the firewall and router logs for Dan’s home networks, something was going on. The logs showed unusual spikes in attempted logins to his home Wi-Fi networks about the time the drone was hovering over the pool. There were similar penetration attempts on two days earlier in the week, which could be consistent with what Eva had said about hearing a drone. Each spike only lasted for 15 or 20 minutes. None of the attempts appeared to have been successful. There were no increases in login attempts coming in over the wired internet during these periods. If this were all the logs showed, we’d just suspect the drone. But the logs also showed a continuous level of increased penetration attempts for a period of roughly 40 hours, beginning about the time the party started on Saturday and ending on the following Monday morning. Again, we don’t think any of these attempts was successful.
Dan said, “I’m trying to wrap my head around the idea someone tried to map or hack into our home network through our Wi-Fi. You expect this kind of effort at a commercial office, but not at home. The idea they’d try to do this with a drone sounds like it belongs in a movie rather than our backyard. If the drone was involved, what about the 40 hours of other penetration attempts that started when our party started?”
The IT Director said, “We don’t know and probably won’t. Could have been another device, maybe even a package planted by the drone.”
Dan said, “Thanks for the good work, gentlemen. I’d hoped we could keep the drone confidential, both to protect the privacy of our party guests and to avoid publicity that could be embarrassing for JPAC. But the FBI’s questions about the home invasion are making it hard to leave the drone out of the conversation. I don’t want to get charged with lying to an FBI agent. I think we need to turn the drone over to the FBI. They have the resources to dissect it far better and faster than we could ever do. I’ll call the agent in charge this afternoon.
◆◆◆
The FBI picked up the drone from Dan’s offices the following morning.
Once they had seen the drone, the FBI’s interest in the investigation changed overnight. Suddenly, Dan and Eva were dealing with experts from the FBI’s Counterintelligence and Cyber Divisions as well as the San Antonio field office. They barraged Dan and Eva for network logs, gate access and security camera videos and other data, even a list of the people who had attended their pool party or had been in their home within a few weeks before. What they didn’t do was provide any answers.
Chapter 17
Sitting in a conference room at JPAC’s offices, Rakesh and Dan were reviewing the details behind Dan’s status email earlier in the week.
“Overall, Crystal is going well,” Dan said as he reached the end of his PowerPoint deck. “I don’t think anyone can do what we’re doing. The latest AI tools we’ve added are increasing our ability to predict and influence electoral reaction to specific social media content. All our indexes are tracking well.”
“Don’t we need more data for everything we’re doing?” asked Rakesh?
“Correct,” said Dan. “The AIs need data to learn and they learn faster and more accurately with more data.
“As you know, we’re doing a combination of deterministic and probabilistic identity resolution. With deterministic identity resolution, we’re zooming in, looking for instances where we can tag information to a person. For example, if a prospective voter logs into a candidate’s website using an email address, we can match that email address with the website cookie the site dropped for that session. If the email address is also connected to a mobile ad ID, we can link that to the email address and cookie that identify the prospective voter. As we rinse, wash and repeat with additional data, we get a better and better mosaic of the voter. It’s a slow, granular process that needs multiple data sources with clear attribution to a specific voter. This is where logging into your apps through Facebook or Goggle gives them a treasure trove of data about you that they can link together for their advertisers to use.
“With the probabilistic approach, we’re zooming way out, looking at large amounts of data at once to figure out patterns and context based on probabilities of what identifiers are likely to match together. It’s like the big data analytics and Hadoop processing you use in other industry sectors. Although we need a large number of data points and individuals, there’s more opportunity to use data that’s anonymized or associated by group rather than identified by specific individual. This offers potential advantages as the data privacy rules become more restrictive. But probabilistic identity resolution works effectively only when we have some way to tie the big data patterns back to specific individuals or groups.”
“I understand. But either way, the more data and data sources we can tie to an individual or a group, the better the accuracy of our results.”
“Definitely,” said Dan. “This is where purchased or shared data can be very valuable, even public data from government sources. In the commercial sector, brands often combine data they’ve developed through multiple internal methodologies with data they’ve obtained from third-party sources.”
“Overall, do we have access to the data we need?” asked Rakesh.
Dan said, “The validation accuracy of most of our results would suggest we do, at least for now. But more data sources and more data points would be valuable across the board. They’d accelerate learning by our AIs, allow better cross-checking and improve our accuracy and margin of error.
“Data availability is not only important to our business, it’s important to America’s ability to compete with China on AI. As you know, the U.S. and Europe are moving toward giving users much more control over their personal data. That potentially reduces the amount of data available to train AIs. In contrast, the Chinese expect personal data held by social media companies operating in China to be made available to the government. China has the data and they’re prepared to use it, even if some of it is stolen from American companies and users. It’s like Kai-Fu Lee, the Chinese venture capitalist, says, ‘If data is the new oil, China is the new Saudi Arabia.’”
Rakesh said, “That’s troubling. I hadn’t thought about the global implications of access to personal data. We need to consider the data sources available to JPAC’s clients and to us, and what we can do to help increase them.”
“Agreed,” said Dan. “We also need to remember the privacy issues. If we end up with tainted data, data where the required usage consents haven’t been obtained, it could create a nightmare scenario for us or a client. We don’t want to be the next Cambridge Analytica.”
Rakesh said, “Before we talk about deployment, could you summarize the enhancements you’ve developed for Crystal?
Dan said, “Sure. First, new tools that expand the psychological, demographic, behavioral and other characteristics we can use for AI-enhanced microtargeting. These tools also improve our ability to combine deterministic data with information extracted from anonymized probabilistic data.
“Second, the AI tools we use in determining, crafting, applying and disseminating the desired messages to the microtargeted audience. It’s what the sociologist Zeynep Tufekci calls ‘persuasion architectures at scale’. Think automatically tailoring propaganda to the individual level across small and large groups of people. The improvements we’ve achieved here are huge and work for social stereotypes as well as candidates and issues.
“Third, more automated deployment of internet bots to spread these tailored messages. Our latest developments give us a substantial advantage in speed, cost and survival against the bot
defenses social media companies employ.
“Fourth, enhanced ability to use our algorithms and automation to test and improve propaganda campaigns in advance of deployment. We can do virtual A/B testing with high accuracy. This capability accelerates learning by our AIs and improves our automated crafting of messages to microtargeted audiences.
“Finally, the unique modular architecture of our systems, which enables us to quickly add, test and validate further measurement factors as they become available. Facial recognition is a good example. We’re working on additional psychological and emotional measurements, both standalone and linked to DNA profiles.
“That’s probably enough for a high-level summary,” Dan said.
Rakesh said, “Very impressive. You and your team have worked hard and accomplished a lot.”
Dan’s assistant brought two salads and a couple bottles of water into the conference room. Rakesh took the opportunity to shift the discussion to deployment. They ate as they talked.
“How long would it take us to release something for beta use?” Rakesh asked.
“For client beta use, 60 to 90 days, assuming we use the software-as-a-service model we use for similar marketing and analytic tools today,” Dan said. “Cloud-based data and processing, 24-7 access and support, automatic updates. For your use only, maybe as little as 30 days. We could probably launch your application within a week or ten days.”
Rakesh said, “Let’s start with me as the sole beta customer.”
“What do you want to do? Support issues? Candidates?” asked Dan.
“Almost certainly issues to start, but maybe not the usual issues we see in the PAC ads. I’ve been talking with Valerie and some business colleagues about pulling together a coalition to reunite America.” He explained what he wanted his project to do.
Intentional Consequences Page 10