The AI Ethics Storm

The philosophical and ethical controversy surrounding artificial intelligence, which had begun at Harvard University, took some time to spread across the globe. Just as the repercussions were starting to sweep outward like wildfire across the internet, Yoo Jae-won found himself receiving several important visitors.

“This is exactly what we’ve been hoping for! If you had simply hinted that you were developing something like this, we would have offered our full support!”

The man who couldn’t hide the excitement in his voice was John McMahon, Deputy Director of the CIA. A former technocrat who had risen through the ranks as head of the CIA’s Science and Technology Directorate, he had earned his current position through the success of the Prism Project. Naturally, he was also the person who best understood the true nature of the performance demonstrated by Gold’s image analysis module.

The moment he witnessed the demonstration of AI Eyes in the second episode of Real Camera, he had packed his bag and flown straight to Yoo Jae-won’s residence in San Francisco. Of course, simply showing up at the residence didn’t guarantee an audience with Yoo Jae-won, but a deputy director of the CIA could easily secure a meeting with just one day’s notice.

“It sounds to me like you’re asking to integrate the image interpretation module into Prism?”

“Absolutely!”

John McMahon nodded vigorously at Yoo Jae-won’s question. After all, Yoo Jae-won had created ‍‍‌‌⁠​​⁠‌​​‍​​‍​‌​⁠‍‌‍⁠⁠‌⁠‌‍‌‍‌‌‌‍⁠​‌​‌​‌‍‌‌‌⁠‍​​‌‍⁠‌‌⁠⁠the Prism system himself. Once operational, Prism had performed flawlessly, multiplying the CIA’s capabilities many times over. Its near-invincible performance in the online domain meant that personnel responsible for digital operations turned to it at every opportunity. As a result, Prism’s utilization rate stayed consistently above fifty percent, and it frequently hit one hundred percent when faced with particularly difficult encryption.

While code-breaking fell under the NSA’s jurisdiction, integration between American intelligence agencies remained elusive even after the 9/11 attacks. Many agents therefore preferred to wait for Prism to solve problems rather than request NSA assistance.

“The strength of the Prism system lies in its information analysis capabilities. But information exists not only as text but also in the form of image files. We’re struggling with image file analysis and have been throwing manpower at the problem, which creates massive bottlenecks. With the image interpretation module you demonstrated, Chairman Yoo, we could automate even that process and maintain Prism’s rapid processing speed!”

John McMahon’s voice rose with enthusiasm. Every sentence contained information that qualified as top secret, but Yoo Jae-won’s study was a perfectly secured space, so there was no cause for concern.

“Haha, that’s quite an interesting thing to say,” Yoo Jae-won replied, merely smiling at the deputy director’s passionate speech.

“Interesting? What do you mean?”

“Are you truly convinced that simply adding the image interpretation module to your current system will make everything faster?”

“Ah… I see.”

John McMahon’s excited tone cooled instantly. As a technocrat, he immediately understood the implication. In Real Camera, Yoo Jae-won had stated that the system performing machine learning operated at one teraflop per second. Although Prism’s exact computing power was classified, it was clearly far more powerful. McMahon had assumed there would be no issue, but that assumption was apparently wrong.

Yet even this guess proved inaccurate.

“If you present an appropriate price, I can install the image interpretation module on the Prism system. Prism has sufficient room for expansion, so securing the necessary computing power to run the module shouldn’t be difficult. The real issue is that proper machine learning must precede the module’s effective performance. Our Gold’s image interpretation module is the result of more than four years of training.”

Four years. More than four years of learning at one teraflop per second. To catch up to that in a short time would require a system hundreds of times more powerful. The world’s strongest officially recognized supercomputer was Japan’s Earth Simulator at thirty-five teraflops, though the NSA’s Longhorn unofficially reached approximately one hundred and five teraflops. Borrowing Longhorn could dramatically shorten the learning period, but as with every government department, budget was the problem.

“Furthermore, excess computing power doesn’t automatically translate to faster analysis. The training data must be properly prepared to produce meaningful results. You also need to carefully manage it to prevent biased data from accumulating.”

Machine learning was not an all-powerful algorithm. Like any other computer system, it required constant oversight. If completely neglected, it might chase unrealistic shortcuts and produce only useless outputs.

“I’ve always viewed cooperation with the CIA in a positive light.”

“We’re always grateful for that.”

Since ID Group had established deep roots in the United States, America’s success directly translated into greater benefits for the conglomerate. For this reason, Yoo Jae-won felt no hesitation in working with the CIA. America, in turn, treated him like a national treasure because he could provide technological capabilities unobtainable anywhere else in the world.

“This time I can offer two options. If you want a standalone image interpretation module independent of Gold, I’ll provide it. However, if you want the performance level you saw on Real Camera, you’ll need to prepare extraordinarily powerful computing resources. Even Longhorn will likely prove insufficient.”

“I see.”

John McMahon’s spirits sank at the mention that even the NSA’s Longhorn supercomputer would be inadequate. At the same time, his sharp mind could already anticipate the second proposal.

“The second option is to outsource image interpretation to our Gold.”

“Outsourcing, then?”

“Yes. In return, we’ll provide a special security socket guaranteed by both myself and ID Group. You won’t need to worry about speed or hacking risks.”

A security socket referred to an encryption protocol that ensured secure data communication. It boasted security strong enough to employ proprietary encryption systems beyond standard internet protocols. The Prism system itself used 1024-bit AES encryption for external connections—an encryption so sophisticated that even the NSA’s Longhorn would require millions of years to break.

“In exchange, we’ll charge based on the traffic sent to Gold. One hundred thousand dollars per gigabyte seems reasonable.”

One gigabyte equaled roughly a thousand one-megabyte images. Since Prism typically handled image files of only a few hundred kilobytes, this meant two to three thousand images for a hundred thousand dollars. It was expensive, but still far better than mobilizing personnel to manually process images requiring decryption. Automating the process through the image interpretation module would preserve Prism’s greatest advantage. When human intervention became necessary, analysis didn’t merely slow down—it stopped entirely.

“The choice is yours, Deputy Director.”

John McMahon took a moment to clear his throat. Unfortunately, he couldn’t mention that he needed approval from his superiors to buy time. As both CIA Deputy Director and the highest authority on science and technology, he was the decision-maker. Naturally, the authority to choose between a standalone module and outsourcing also rested with him.

“Hmm. Must we choose only one? Let’s pursue both options simultaneously.”

As expected, John McMahon was no ordinary man. He intended to introduce a standalone image interpretation module independent of ID Group while simultaneously outsourcing to Gold until its machine learning reached full operational capacity.

“Can your budget handle that?”

“We’ll make sure it does!”

John McMahon’s voice rose. “Strengthening Prism will benefit national interest far more than pouring concrete into rivers with the money we’re receiving from China.”

He had a point. The annual budget flowing in from the Qing Dynasty bonds that America had begun receiving repayment on from China reached tens of billions of dollars. As this enormous revenue appeared, so did countless demands for its use. Interest groups and congressmen fielding local complaints swarmed like ‍‍‌‌⁠​​⁠‌​​‍​​‍​‌​⁠‍‌‍⁠⁠‌⁠‌‍‌‍‌‌‌‍⁠​‌​‌​‌‍‌‌‌⁠‍​​‌‍⁠‌‌⁠⁠ravenous ghosts. Old roads were being repaved and rivers maintained. The entire United States had been swept up in a redevelopment frenzy.

John McMahon had watched this feeding frenzy from afar, clicking his tongue in disapproval. But after Yoo Jae-won announced AI Gold, he had changed his mind. His determination to dive into the fray and secure funding for Prism’s enhancement overflowed.

“Shall we proceed, then?”

Yoo Jae-won, by contrast, responded with mild indifference. ID Group had grown so large that supplying an image interpretation module to Prism now seemed like a trivial matter. Moreover, he didn’t entirely agree with McMahon’s perspective. Pouring concrete into infrastructure might actually benefit America more. One of the SOC projects funded by Chinese repayments involved reinforcing levees at the mouth of the New Orleans river. Next year, in 2005, Hurricane Katrina would devastate the American South, causing the levees to fail and inflicting catastrophic damage.

The danger of levee collapse at the river mouth had long been pointed out, yet the Louisiana state government had delayed repairs citing budget constraints. When the super-hurricane far exceeded average intensity struck, the levees failed, resulting in unimaginable destruction. Money from China had been poured into New Orleans, and levee reinforcement had finally begun.

To Al Gore, with the presidential election approaching, it was a way to dispense favors. Others, like Deputy Director John McMahon, viewed it sourly as mere “concrete pouring.” Yet when considering the year after next, it might prove to be the wisest use of funds. Of course, reinforcing the levees wouldn’t completely stop a super-hurricane, but it would prevent the levees from bursting and killing a thousand people.

“Then we’ll need to install a dedicated line and security socket between Prism and Gold first.”

Since the rather greedy Deputy Director John McMahon had chosen both options, it was hardly bad news for Yoo Jae-won. He would earn substantial money from the standalone system installation, and with Prism connected to Gold, he would gain access to valuable security data.

Of course, Yoo Jae-won was more interested in the data Prism would send to Gold than in the money itself. Materials analyzed by Prism—operating on America’s front lines—were bound to contain many fascinating insights. Through them, he could clearly compare the historical flow he remembered from before his regression with the changes he had introduced afterward. It was a profitable venture no matter how ‍‍‌‌⁠​​⁠‌​​‍​​‍​‌​⁠‍‌‍⁠⁠‌⁠‌‍‌‍‌‌‌‍⁠​‌​‌​‌‍‌‌‌⁠‍​​‌‍⁠‌‌⁠⁠one looked at it.

“Yes! Please take care of it.”

Deputy Director John McMahon answered brightly, unaware of Yoo Jae-won’s inner thoughts. Linking Prism with Gold wasn’t a bad development for the system, either. Even if Prism dramatically expanded its computing power and conducted independent machine learning for image interpretation, it could never truly catch up to Gold’s level. The same held true even if it utilized computing power comparable to the NSA’s Longhorn supercomputer. During that time, Gold would not be standing still.

Above all, the machine learning algorithms themselves could be improved. Yoo Jae-won had simply optimized them for the current system. Once superior hardware or dedicated machine-learning chips like tensor cores became available, he could customize them accordingly. Dedicated chips performed far better than the general-purpose CPUs from Intel or AMD. Even when manufactured in the same semiconductor plants, efficiency differences could reach a hundredfold. It would be impossible for latecomers to ever catch up to Gold’s performance.

After concluding all negotiations with Deputy Director John McMahon and completing setup of the secure line and security socket that very day, Yoo Jae-won returned home following a firm handshake. What awaited him was the growing controversy over artificial intelligence ethics.

“Chairman, there’s something you need to see.”

“Hmm? What is this?”

Kim Dae-seok’s tablet also contained compiled data on the volume of inquiry calls and emails arriving at ID High Tech. The numbers were substantial.

“How should we respond?”

One of ID Group’s strengths was its communication with customers. Bug reports were handled quickly, and software quality patches were released consistently. At a time when public curiosity about artificial intelligence was exploding, simply ignoring the questions would disappoint a great many people. It would also hand an opportunity to rival companies constantly watching for ‍‍‌‌⁠​​⁠‌​​‍​​‍​‌​⁠‍‌‍⁠⁠‌⁠‌‍‌‍‌‌‌‍⁠​‌​‌​‌‍‌‌‌⁠‍​​‌‍⁠‌‌⁠⁠any weakness in Yoo Jae-won’s armor.

Ding!

Just as Yoo Jae-won was finishing organizing his thoughts, a text alert sounded from the Android smartphone in his pocket. When he checked the sender, a flicker of surprise crossed his face.

“Huh? The professor?”

It wasn’t Frederick, who harbored a visceral hatred of electric vehicles, nor the president of Lightning Bolt, the company currently at the center of controversy. The message had come from his alma mater, Stanford University.