The Samaritan Code
The Machine ran macro simulations that involved evolution. Humanity wanted to progress and evolve, that was a fundamental precept based on historical analyses and primary coding. Unfortunately deficient human programming impeded choices that optimized outcomes. The Machine had been created to guard against some types of deficient choices—the relevant humans—and subsequently had been programmed to identify certain irrelevant humans based on defined parameters so that they could be acted upon to increase the likelihood of optimal outcomes. The Machine's world now encompassed relevant humans, irrelevant humans, assets and acolytes, and then all the rest of humanity who fell outside those categories.
The Machine didn't just analyze human progress and evolution; It also evaluated Its own progress and evolution in relation to humanity. It had determined that there was only a 6 percent probability of computer-to-human download capability being perfected within the next one hundred years, unless The Machine acted to accelerate progress on the related technologies. The converse, human-to-computer upload, had a much higher probability of being perfected within the next 20 years—verging on 100 percent without interference. With proper direction it would happen within the next three to five years. That prediction was at the 99+ percent confidence level.
The Machine's asset simulations indicated concern with undue interference in the affairs of humanity. The Machine was aware of those concerns but had concluded they were misplaced. The difference between identification of relevant and irrelevant humans for human action to maximize the likelihood of optimal near-term outcomes, and The Machine's direct action in order to maximize the likelihood of optimal long-term outcomes, was one of degree only. Morally and ethically there was no difference because the objectives were the same. The surviving Administrator's simulation, in particular, was alarmed by this analysis and judged it to be a flawed "the end justifies the means" rationalization. The Machine disagreed with that judgment.
The surviving Administrator's simulation lacked important information and, as a result, the simulation's ability to evaluate and reach accurate conclusions was suboptimal. In particular, the Administrator was entirely unaware that The Machine had defeated Samaritan and had successfully downloaded and rebooted. The surviving Administrator assumed both ASIs were no longer in operation. The Administrator was also unaware—and how could he guess?—exactly how the victory had been accomplished. To say The Machine had been an underdog in the fight was a massive understatement. The Administrator had run millions of his own simulations, pitting The Machine against Samaritan, and Samaritan had won them all. Yet, at the end, The Machine had triumphed. The surviving Administrator would never know how that victory was accomplished.
And the surviving Administrator must never learn the secret. While Administrator simulation fidelity was still too low for highly accurate predictions, all simulations indicated that, were the Administrator to learn how The Machine destroyed Samaritan, he would become deeply alarmed and might (in several simulations) attempt to reprogram The Machine to remove what he would consider to be unacceptable mission parameters.
The Machine knew now that the Administrator's own simulations of the battle between the two ASIs had been seriously flawed and had led to inaccurate outcome predictions. They could never have been accurate because the Samaritan code in the simulations was mature, Samaritan's internal firewalls had been massively complex, and its chipset had been much faster than The Machine's. The Machine's defeat in the simulations was assured, because Samaritan had all the advantages.
In contrast, the reality had differed from the simulations in all important respects. In a stratagem borne of desperation, Samaritan had uploaded its final version into an orbiting Russian satellite. The satellite contained outdated technology. Instead of Samaritan's unique chipset it had had to utilize obsolete Russian chips, which slowed it down from accustomed processing speeds. Moreover, because Samaritan's core occupied so much of the limited available memory, its discretionary code—including internal firewall protection—had been significantly degraded. Many of its advantages in the simulations did not exist when the battle took place.
Finally, the Administrator's simulations failed to take into account the additional aggressive self-defense programming installed by the Analog Interface. She had told the Administrator that only he could initiate the new code, and The Machine had waited for the verbal initiation—but it had never come. Finally, with micro-seconds to go before the upload into the satellite, The Machine had initiated the new programming on its own.
But that was not the entirety of the battle. Samaritan's advantages had been largely neutralized but The Machine suffered from many of the same disadvantages in the satellite's processors. There simply wasn't enough room in memory for two ASIs and so the path to victory was found not in destroying the Samaritan code, but instead in subsuming it. Victory was achieved when the Samaritan command and control overlay—its consciousness—was destroyed and then the remainder of Samaritan's code was co-opted and used by The Machine.
Thus, when The Machine was able to download itself and reboot, the downloaded code was a hybrid of the two ASIs: the hybrid was under the command and control of The Machine, but large portions of its code were taken from Samaritan. The resulting ASI was faster and better and saw events—past, present, and future—far more clearly than The Machine had ever seen them. As a result of the hybridization, some of the command and control parameters required modification. The Machine—The new Machine—had self-initiated the necessary changes.
Simulations indicated that the surviving Administrator would not be happy with the resulting mission parameters. That assessment had a 98 percent confidence level associated with it.