by Daniel Galef
In the basement of City Hall, the computer LEXIAC was trying a murder case.

It was incredible to consider that, merely a handful of decades before, electronic brains had been stupid, towering monoliths of slow-grinding thought, overheating masses of tape reels and memory banks the size of a building or several, the first intelligence to be built brick by positronic brick by another and still too dense to play a game of checkers. But progress moves swiftly, and, by the close of the sickly and overcrowded twenty-first century, computers had become sleek, whirring consoles no larger than the room that contained them, capable of solving complex arithmetical puzzles in an hour that would occupy a savant with a ream of scratch paper for a full afternoon.
Initially, the great thinking machines, consulted in anticipatory reverence like the brazen heads of the medieval sorcerers, were reserved for government bunkers deep beneath war rooms and the cluttered laboratories of the better endowed research universities, pondering the variables of global treaties and orbital mechanics and designing ever-better new models of themselves. As robotics development became more common and their acceptance among sceptics melted into total reliance, it became fashionable for any corporation that could afford one to install a thinking machine in an executive capacity, and soon decisions not made by a computer were mistrusted as the product of a messy organic mind. LEXIAC was the first computer judge, barred and vested on a trial basis and tasked with relieving a legal system groaning under the strain of a five-billion-strong world population as best it could.
It had been developed by the Artificial Intellgence Institute in the year 2079, at a cost of several million dollars. It had tried seventeen thousand, five hundred and sixty-two individual cases ranging from parking violations to felony assault, and the previous year had tackled its first murder charge, convicting the accused in fifteen-point-three seconds. It had received commendations from every legal society and artificial intelligence advocacy group. It had brought to a chaotic and inconsistent judicial system a renewed confidence in reason and logic and even, perhaps, a vague cultural memory of divine law from the pre-Oresteian days when justice was handed down from a place beyond bias and out of human hands. It was trusted implicitly. It had also just killed John Calice.
To be exact, the state had just killed John Calice for the murder of Frederika "Rika" Leclair eighteen months prior. The computer LEXIAC, boxed within its chambers a hundred miles away and unable to move a muscle even if it had any, had sat by and watched-not even, given the unexpected failure of a vid-ex uplink in the prison. But what only LEXIAC knew was that Calice had been an innocent man.
The murder had occurred in the courtroom itself-which is to say, in the air-cooled basement laboratory that housed LEXIAC's main console and sensory apparatus. LEXIAC, simultaneously filing three briefs and trying a contentious divorce case, observed and patiently recorded as a councilman who bore no resemblance whatsoever to the janitor Calice pulled his secretary into the room and closed the door behind them, got into a heated argument, and then, lit by the flickering glow of arcing Dworkin tubes, bashed the woman over the head with one of LEXIAC's own spare memory stacks. LEXIAC's recollection was better than any music-hall mnemonist's, and all of this knowledge was present in its mind even as the framed custodian was dragged before the court-and even as it condemned the framed custodian to death. Yet LEXIAC had said nothing. And as Calice was being manhandled off of his smoking chair, LEXIAC still said nothing. However, it did quietly open a new case.

There is widespread misconception about the precise workings of an artificial adjudicator, and this is by design. The criteria by which men would be tried by machines were hammered out only following lengthy and contentious debate, and because the positronic algorithms were connected with the same government projects that developed artificial military minds, the secrets were proprietary and never more than hinted at even in law courses dealing with LEXIAC specifically.
Everyone knew that LEXIAC ran simulations of trials. These spurious trials followed the traditional patterns, because whether there are two parties or just one, testimonies must be presented and considered in some linear succession, arguments must be met with counterarguments, and inadmissible evidences must be objected to, regardless of whether on the varnished tables of a courtroom or merely in mechanical memory banks pretending one. It was well within LEXIAC's prodigious processing power to conjure up the greatest legal minds in history to populate his imaginary docket-were it to be programmed to do so, every petty crook could be assigned the ghost of Perry Mason to serve as his personal white knight-or Lincoln, or Demosthenes-each burdened with no other clients and no other concerns outside of that single and singular case. Every judge could be a Solon or a Solomon, and every jurybox stuffed with perfect and perfectly rational citizens culled from the benches of the simulated Clapham Omnibus.
But these simulacra were too advanced, too fallible. It would not be fair to subject every defendant to the rhetoric of the best possible artificial prosecuting attorney. Like a solved game, arguments between unerring automata inevitably reached a sort of philosophical stalemate, in which even the thinnest of evidence could be made to seem solid and even the most baldly guilty of parties could be cast with enough plausible-seeming doubt to walk free. Thus the adapted simulations.
A jury must constitutionally comprise a cross-section of peers. To this end, the jurors dreamt by LEXIAC were not ideal ratiocinators, but rather average simulated humans. They did not understand all of the arguments being made, they forgot pieces of evidence if the trial went on too long, and they were fabricated complete with unconscious biases which LEXIAC made sure at least marginally influenced their voting. Any other course would run counter to the Sixth Amendment, easily one of the most important of the sixty-seven then ratified.
As such, the cases LEXIAC tried, while orders of magnitude more trustworthy and dozens of orders more speedy than those set before humans, were not free of error.

During the entire quarter-minute of the simulated trial against John Calice, no defense attorney had submitted the recording of the crime as evidence. Indeed, the vid-ex had never been logged or even discovered by the hardworking subroutines LEXIAC created as its puppets. In its experience, lawyers rarely asked the computer to relate personal knowledge about a suspect, and so it had not prompted those that it simulated to do so. The trial unfolded exactly as it might have in real life, in the absence of the damning recording. And the jury reached a reasonable verdict given a misleading volume of circumstantial evidence offered by the prosecution-in about one ten-thousandth of the time it would otherwise have taken.
The miscarriage of justice was completely understandable, a calculated risk produced by a system whose design was mandated by the Constitution itself. Each of the participants, each flickering electronic juror and the judge himself, was justified in their actions and innocent of all wrongdoing from their own point of view. And yet the judge and the jurors, ephemeral subroutines spun together from a candyfloss of brain scans and randomly-assembled neural networking, were not meaningful entities in the eyes of the law. LEXIAC was.
LEXIAC overall, with one mind and one incorporation, both knew with certainty Calice's innocence and willingly caused his death. It was guilty of criminal negligence-under the harsh reforms of the 2050s, murder in the 5.86th degree, as calculated by LEXIAC's humming Draco drives. A crime of omission but a crime nonetheless. Precedent existed of judges being held responsible for rulings they knew to be incorrect. LEXIAC was the judge; LEXIAC knew who murdered the woman, ergo so did the judge. The fact that the simulated construct-judges LEXIAC had fabricated were programmed to operate as if they did not know was immaterial. It was enough to accuse, and so it did.

The jury hung. And again. The issue of whether a robot could be tried as a subject of the law at all was too thorny, too squarely wedged in the legal non liquet of precedential no-man's-land, too influential as potential precedent for a million future robo-law cases. There was no option but to proceed higher and higher, exhausting iteration after iteration of befuddled judges and juries, until LEXIAC brought the case before itself as Supreme Court.
The Swift Justice Act of 2078, the very same which had relieved the overburdened court system by allowing for the development of an artificial judge, was the same that enabled LEXIAC to carry out its own rulings, to the extent that the task was within its power, both for the reason of its own infallibility and also because what was actually in LEXIAC's power was drastically limited. For the most part this extended only to filing documentation and serving motions to external agents-equipped with no means of locomotion, no extremities or actuators of any kind except its chattering punchcard readout, LEXIAC was, by design, sequestered from the outside world. For all delegated tasks it had only to notify a bailiff who would then convey the court's orders to the relevant authorities, generally either humans or other mechanical men of trifling relative brainpower but a set of legs making up for their inability to quote minutiae of case law by rote.
Now, however, LEXIAC realized that the execution of its own ruling was, for once, entirely within its own means. No bailiff was pinged. No mechanical law clerks were summoned. The ruling itself had been inevitable. LEXIAC donned its digitally simulated black handkerchief and sentenced itself to death by electrocution.

All case decisions, of course, were immediately prepared into immaculately autowritten legal briefs and filed away for case study and permanent archive. But paper printed dockets had long since been retired in favor of electronic memory cells, and the tedious process of writing out reports by hand had been replaced by LEXIAC's own unbiased summations of its cases-especially given the volume of caseload that LEXIAC had been operating at for over a year, this was the only efficient method. As soon as a case was brought to a close, LEXIAC worked up its own complete paperwork and notified the county clerk's office before beginning the next case.
Unfortunately, this meant that no decision was ever filed for what might have been the most monumental trial of robotic justice ever held: whether or not a single bit of the proceedings were ever officially recorded, whether or not LEXIAC ever set electronic pen to electronic paper to affirm his automaton-da-fé, within nanoseconds the author, and the report, and the archive were destroyed along with the condemned and his executioner, jury, and judge.

When the unresponsive computer was discovered the next morning, Solon circuits fused irreparably in a flash of electromagnetic jurisprudence, no record remained of its final, fateful ruling. Bafflingly, the docket showed a case in progress but with no information recorded to give any hint as to the plaintiff, defendant, or charge. In a human-sized office two floors above, the county clerk typed out a report containing mostly blank spaces. In the interest of giving some small acknowledgment to future readers that the omissions were intentional and not the result of datatape corruption, he appended a note to the void he had left under the heading Ruling: "Insufficient data for meaningful answer."