Page 78
The same guy in the white T-shirt brought their food. Four plates on a bigger tray. They slid them around and tried a bite. Pretty good. No complaints, even from Turner.
Elwood asked, “What makes EEF faster than anyone else?”
“Ask a random question,” Tow said. “Any subject.”
Elwood thought a moment.
He asked, “What was the biggest explosion in all of recorded history?”
“OK, that’s a straightforward drill-down into our existing body of knowledge. Any conventional search engine could scrape the net and get the answer in a tenth of a second.”
“Mount Tambora,” Reacher said. “Indonesia, in 1815. A volcanic eruption more than sixty thousand times more powerful than the Hiroshima bomb.”
“He’s a human search engine,” Turner said.
“Good answer,” Tow said. “But more information than requested. You could have stopped after the name of the volcano, plus maybe the date. If we reverse-engineer a question to fit that answer, we would get, how many times more powerful was the biggest explosion in history than the first atomic bomb? And that would be a very basic type of A.I. prompt. It requires multiple processes. First, A.I. would need to understand the nuance of the question, linguistically. That’s a big deal in itself. Second, it would need to search for the biggest explosion in history, which would come up with the volcano. Third, it would need to search for an estimate of the volcano’s power.”
“A thousand megatons,” Reacher said.
“OK, fourth, it would need to search for the name of the first atomic bomb.”
“Little Boy,” Reacher said.
“OK, fifth, it would need to search for an estimate of Little Boy’s power.”
“Fifteen kilotons as designed,” Reacher said. “Although some claim nearer to twenty in practice.”
“OK, sixth, it would need to perform an arithmetic computation of the ratio between the two powers, giving a result of sixty thousand times, or whatever it was.”
“Sixty-six thousand, six hundred sixty-six,” Reacher said. “Point six-six recurring, to be pedantic.”
“And finally it would need to express the answer in a smooth and accessible written style. Again, not easy, linguistically. And all done very quickly. Not necessarily in a tenth of a second. Ironically people trust the answer better if they have to wait. They love to see the machine thinking. It makes them feel it’s a personal response, properly considered, tailored for them alone. Like the machine is making an effort on their behalf. It creates the illusion of a relationship. A couple of seconds is considered ideal. Think, think, answer. The marketing people tested it.”
“What makes EEF better?”
“Some of those processes are easy, and some are not. Math is proving surprisingly difficult for A.I., and so is writing. Graphics and visuals are really hard. All those separate steps are handled by algorithms. You need an infinite number. They all need to be learned and trained. Maybe you want the answer not in writing, but by Marilyn Monroe singing it in the shower. That would be about a billion extra algorithms right there. Maybe you want the machine to police its own ethics, so that a question about how to kill a bedbug gets an answer, but how to kill your wife doesn’t. That’s another billion algorithms. Right now EEF’s algorithms are the best and the fastest in the business, and Axel Brill turns them into pots of gold. That’s the CliffsNotes version.”
Turner asked, “How does Cameron make EEF’s algorithms better and faster than anyone else?”
“I don’t know,” Tow said. “If I did, I’d be a billionaire by lunchtime.”
“How would illegal activities help them?”
“I don’t know,” Tow said again. “I guess a little confusion in their accounts couldn’t hurt, to gloss over the amount of money they must burn every day. That would keep investors happy. As would putting lipstick on their test results. But that’s standard practice. They all do it.”
“Why would they be sensitive about a technology park in Colorado?”
Tow looked at Reacher. “You mentioned that, right? Five data centers and two solar fields bigger than most of the countries in Europe. When were the data centers built? Does EEF own them? Maybe they’re using more power and water than their permits allow. That’s a sensitive issue right now. People are up in arms. Terrible for the environment, especially out west.”
“No doubt,” Turner said. “What would happen to the owner of a data center if they were found to be doing that?”
“Some state or federal agency would fine them about half a second’s worth of income.”
“Not a motive for homicide, then.”
“Is that where the first one was?”
Turner nodded.