If Lawyers Became Algorithms
LONDON — “We cannot afford to let our bright young minds become appendages of software. Software companies need to offer fulfilling and generous opportunities, or else they need to pay reparations for the way they’ve transformed our economy.”
The words were Bernard Willston’s, the newly elected Conservative Party leader, whose ascent was roiling British markets. Breaking with his party’s traditional pro-business posture, Willston was demanding that software giants provide monetary compensation for at least 20 percent of the millions of jobs their technologies had replaced. The one-time payment would amount to a total of 9 billion pounds ($11 billion).
Willston’s rhetorical crusade had begun shortly after mass automation, long predicted for manufacturing industries, had struck white-collar office jobs. The shift had come far quicker than expected. The resulting job dismissals and wage and hours cuts had put the Conservatives under heavy pressure. Willston was the result, the latest in a series of party outsiders promising radical economic remedies founded mostly on making tech companies pay.
Simon Bossell was part of Willston’s target audience, a 29-year-old aspiring arbitration lawyer at a well-established London firm. Pressed against other commuters on the Uber-pool bus, Bossell turned off the video of Willston’s speech playing on his phone. The reparations, Bossell thought, were probably a bluff — the courts would probably block it.
But he couldn’t help but agree that Willston had a point. Besides the pay, he’d trained as a lawyer for the intellectual challenge, the competition and the prestige. But like most white-collar workers, in 2025 Bossell found himself in a job made largely, and increasingly, redundant by software, overqualified and with few career prospects.
When Bossell graduated from University College London with a degree in law, working as a paralegal for one of London’s big firms was no longer a steppingstone into the profession. Instead, for the past three years Bossell had been working as a database assistant, entering data from cases and precedents into a spreadsheet. Pretrial research, report drafting and briefing were all done by Symantec’s latest Glacier software, which was able to scan and write faster than Bossell could think. And it never got sick, took a break or went on leave.
Back in 2011, Symantec created its Clearwell software, which was based on language analysis that allowed it to identify general concepts in documents; it proved capable of analyzing and sorting more than 570,000 documents in two days. The turning point came in 2018, when Symantec released a tool that could not only scan big data but also synthesized information in writing at a level unprecedented for software applications. Its competitor, Wordsmith, was already used by Yahoo and USA Today for simple sports reporting, and by The Associated Press for corporate earnings reports and minor league baseball. Glacier’s edge was its capacity to write more descriptive and analytical material.
Initially, the fear was that Glacier would become a way for students to cheat on writing their essays. But by 2021, the top three British law firms had replaced their paralegal teams with Glacier. In the following months, Symantec released sector-specific versions of Glacier, slowly replacing accountants, researchers and, increasingly, public policy bureaucrats. 2024 was the first year that the Daily Mail used reporters and software for its investigative writing — some readers found Glacier to be less biased and the language easier to read.
Law had been especially affected; even judges were threatened. The first “AI judge” had appeared in late 2016. The machine, invented by researchers from the University of Sheffield and University College London, had analyzed hundreds of cases from the European Court of Human Rights and was able to predict verdicts with 79 percent accuracy. It was not that cases were being decided entirely by machines, as some doomsayers had predicted; instead, judges were taking into account verdict predictions produced by them. An early pilot effort in the U.S. — to have machines rule on minor criminal cases — had embarrassingly derailed after an algorithm, fed tens of thousands of cases, began ruling against defendants based on their addresses and names. That briefly undermined public confidence, but overall efficiencies had been inarguable. Now 80 percent of minor criminal cases had machine-suggested verdicts, considered and then largely rubber-stamped by judges.
All these kinds of changes had transformed the daily experience of work for the likes of junior white-collar workers like Bossell. He rarely interacted with lawyers or clients and had little exchange with his co-workers from the Database Maintenance Team. He spent most of his day silently inputting information and spot-checking Glacier’s work, making minor adjustments where necessary. Despite his work, he was not a specialist. Beyond Glacier, he had little training in software maintenance; his university law program had introduced basic coding classes shortly after his graduation, but for only a few hours a week.
Bossell’s office was a microcosm of the “fourth industrial revolution” economy. Advances in computer science, machine learning and data architecture gave algorithms and software a higher level of adaptive intelligence, accelerating the pace at which computers advanced their learning and competence. While the need for human interaction and strategic decisions preserved some corporate jobs, many once-human tasks were replicated by code.
In past years, software was often used to check the work of humans. Now humans were checking the work of software. Everything could be done faster, and increasingly cheaper, by computers.
Nothing lost
People’s greatest technology fear has always been job loss. When William Lee sought patent protection for his stocking frame knitting machine in 1589, Queen Elizabeth I replied, “Consider thou what the invention could do to my poor subjects, Master Lee. It would assuredly bring to them ruin by depriving them of employment, thus making them beggars.” Before World War II, U.S. President Franklin D. Roosevelt said inventions were destroying jobs faster than they were creating new ones. In 2015, a Bank of England report predicted 15 million jobs were at risk of being replaced by software, mostly low-skilled jobs in customer service and skilled-trade occupations.
But jobs had not simply disappeared. Though unemployment had risen, material engineering, software development and data architecture had created thousands of new positions for young university graduates. What had happened, rather, was a mismatch; a generation overqualified for software maintenance tasks and underqualified for the potential innovative “jobs of the future.”
It was an age of robotic feudalism, where a few companies owned the software that eliminated so many middle-class, high-productivity jobs. The spectrum of office jobs had narrowed considerably, with some people getting one of the few managerial posts or stuck in a routine job merely aiding the software’s efficiency. Half of Bossell’s friends were now suffering work-related depression, while the other half were seeing physiotherapists to alleviate finger cramps from endless typing. Friends who once dreamed of becoming teachers, accountants or policy experts were stuck in low-paying positions, sustaining pieces of software.
Some were considering become freelancers, if Willston succeeded in passing his proposal for a universal basic income for all British citizens. The U.K. was already a relative latecomer to the idea — the Netherlands had introduced it in 2023.
As Bossell’s bus approached his office, he looked around. The once buzzing city had changed — there were fewer pubs and cafés. The walkways were quieter after office hours. There was less need for people to work late, but they also had less cash to enjoy during their free time.
The bus stopped, and Bossell looked glumly at his fellow commuters as they began to disembark. Still, he thought, “It could be worse. I could be a banker.”
