The system had only limited capabilities for independent action without input from the crew. It had no real initiative, although when the assemblers were deprived of human input the system reviewed previous experience related to human actions, weighing most heavily the more recent human actions, and the system would consider similar directed action, depending on the current status of the risk-gradient and the outside environment as well as the status of the mission and the seriousness of the threat to that mission. Deprivation of input from the modular neural networks would at first lower neural flux, then begin increasing it to compensate for the lack of processing raw material, much the way a human deprived of vision and touch and hearing would hallucinate in laboratory experiments.
Out of concern for the unknown quality of this electronic increase in function during sensory deprivation, the ship was wired with internal microphones. This was a part of the system not revealed in the technical manuals or in the training courses. One of the modular neural networks was an analyzer of the human speech inside the vessel as received on microphones placed in each room of the command-module compartment. These voice recorders were at first justified based on previous ship designs in which speech records were part of accident investigations, much as airliners had black-box voice-recorders to analyze the last moments of an airplane that had crashed. Valuable tactical data could be obtained even from a ship that lost a battle if the entire event could be reconstructed from recovered voice records. Not all tactical conversations occurred in the control room. Many came about in the captain’s stateroom, in the doorway to the first officer’s stateroom, over breakfast in the officers’ mess, in presleep ruminations by the lower ranking officers. In short, the designers had made the decision to wire the entire ship for sound without the awareness of the crew for three purposes. One was to avoid the sensory deprivation that would cause the system to fall out of alignment. The second was the same as the black-box designs. The third was related to the Second Captain. The system’s neural artificial-intelligence assembler modules, the frontal lobes, needed data for the estimate of the success of the ship’s mission and to continue the fight if the crew died or was disabled. The system’s main source for the crew’s estimate of the tactical situation was the input from the speech detectors in the ship’s forward compartment. It was imperfect, and the value of the eavesdropping system would be in effect only if the entire crew was lost, but in that unlikely event the Second Captain would carry on based on what the crew had been doing just prior to their loss, the eavesdropping a sort of programming.
If the Yokogawa designers had been able to wire the captain’s brain to detect his thoughts, they might have done that too.
The system description would lead some engineers outside the Yokogawa enclave to conclude that the Second Captain had an intelligence very closely resembling a human’s, but that needed refinement. If the Second Captain’s artificial intelligence were to be rigorously compared to a human’s it might come closest to being a five-year-old human … a five-year-old capable of advanced thought patterns, highly developed learning abilities, successful applications of experience-based initiative, extraordinary adaptability to new situations — in short, a five-year-old human possessed of considerable brain power. But a five-year-old was not the person who should be driving a car or flying an airliner, or directing the actions of the world’s most advanced attack submarine. Unless, of course, the system concluded that the crew was gone.
In the eastern Atlantic, during the hour after the Mark 50 torpedo explosion, the Hegira continued west along the track inserted into it by Lt. At Ishak. During that hour the sonar system detected the sounds of the recovery of the 688 class submarine that had launched the hostile torpedoes.
The Second Captain recognized the ship as the one the crew had attempted to kill earlier, and a large part of the assemblers’ internal neural flux dialogue was devoted to the discussion of whether the recovering submarine should be fired on and destroyed.
The ship’s mission, as the Second Captain understood it, was to proceed along the track to the Labrador Sea off Greenland, where it would fire the high-altitude radar-evading supersonic cruise missiles fitted with their new Scorpion warheads toward Washington, D.C. The 688 submarine related to that mission only as far as it threatened the passage of the Hegira. The plasma-bubble memory modules contained numerous references of the crew — before they perished, the system thought with something much like grief — to the fact that the 688 could counterdetect the Hegira and fire back or fire first. Firing a Nagasaki torpedo was one valid course of action, but remaining undetected by the 688 was equally valid. There was also the fact that the crew had thought along similar lines as the voice memories showed — their motivations in firing at the ship in the first place had been grounded in the fact that it guarded the opening at Gibraltar. There was significant risk to the ship and the mission involved in a hostile torpedo shot at the 688, perhaps less risk in attempting to sneak by the other submarine.
So the risk that was involved at first induced a hesitation in the Second Captain, the system initially deciding to collect more data on the probability of hostile intent by the 688.
The initial estimates showed that the 688 was not masking its own noise signature and was, in fact, generating the loudest series of noises in the Second Captain’s plasma-bubblememory’s history. Therefore it was probably not acting along a hostile-threat curve but was concerned with its own survivability estimates. The noises grew quieter as the two vessels drew closer, causing further hesitation in the Second Captain, which now devoted processing time to the question of avoiding an encounter by steering clear of the 688.
There were also valid reasons not to do that, including delaying the mission and an uncertainty of the 688’s course and mission intentions. Finally, the two ships’ tracks converged, bringing the Hegira within a few hundred meters of the 688, a closeness that the crew would probably not allow to happen, but since it had, the Second Captain — now much deeper into a negative-value risk-gradient — estimated that increasing speed to take the ship away would cause a louder own-ship noise emission that would make them more easily detected by the 688.