FORGET Skynet. Hypothetical world-ending artificial intelligence makes headlines, but the hype ignores what’s happening right under our noses. Cheap, fast AI is already taking our jobs, we just haven’t noticed.
This isn’t dumb automation that can rapidly repeat identical tasks. It’s software that can learn about and adapt to its environment, allowing it to do work that used to be the exclusive domain of humans, from customer services to answering legal queries.
These systems don’t threaten to enslave humanity, but they do pose a challenge: if software that does the work of humans exists, what work will we do?
In the last three years, UK telecoms firm O2 has replaced 150 workers with a single piece of software. A large portion of O2’s customer service is now automatic, says Wayne Butterfield, who works on improving O2’s operations. “Sim swaps, porting mobile numbers, migrating from prepaid onto a contract, unlocking a phone from O2” – all are now automated, he says.
Humans used to manually move data between the relevant systems to complete these tasks, copying a phone number from one database to another, for instance. The user still has to call up and speak to a human, but now an AI does the actual work.
To train the AI, it watches and learns while humans do simple, repetitive database tasks. With enough training data, the AIs can then go to work on their own. “They navigate a virtual environment,” says Jason Kingdon, chairman of Blue Prism, the start-up which developed O2’s artificial workers. “They mimic a human. They do exactly what a human does. If you watch one of these things working it looks a bit mad. You see it typing. Screens pop-up, you see it cutting and pasting.”
One of the world’s largest banks, Barclays, has also dipped a toe into this specialised AI. It used Blue Prism to deal with the torrent of demands that poured in from its customers after UK regulators demanded that it pay back billions of pounds of mis-sold insurance. It would have been expensive to rely entirely on human labour to field the sudden flood of requests. Having software agents that could take some of the simpler claims meant Barclays could employ fewer people.
The back office work that Blue Prism automates is undeniably dull, but it’s not the limit for AI’s foray into office space. In January, Canadian start-up ROSS started using IBM’s Watson supercomputer to automate a whole chunk of the legal research normally carried out by entry-level paralegals.
Legal research tools already exist, but they don’t offer much more than keyword searches. This returns a list of documents that may or may not be relevant. Combing through these for the argument a lawyer needs to make a case can take days.
ROSS returns precise answers to specific legal questions, along with a citation, just like a human researcher would. It also includes its level of confidence in its answer. For now, it is focused on questions about Canadian law, but CEO Andrew Arruda says he plans for ROSS to digest the law around the world.
Since its artificial intelligence is focused narrowly on the law, ROSS’s answers can be a little dry. Asked whether it’s OK for 20 per cent of the directors present at a directors’ meeting to be Canadian, it responds that no, that’s not enough. Under Canadian law, no directors’ meeting may go ahead with less than 25 per cent of the directors present being Canadian. ROSS’s source? The Canada Business Corporations Act, which it scanned and understood in an instant to find the answer.
By eliminating legal drudge work, Arruda says that ROSS’s automation will open up the market for lawyers, reducing the time they need to spend on each case. People who need a lawyer but cannot afford one would suddenly find legal help within their means.
ROSS’s searches are faster and broader than any human’s. Arruda says this means it doesn’t just get answers that a human would have had difficulty finding, it can search in places no human would have thought to look. “Lawyers can start crafting very insightful arguments that wouldn’t have been achievable before,” he says. Eventually, ROSS may become so good at answering specific kinds of legal question that it could handle simple cases on its own.
Where Blue Prism learns and adapts to the various software interfaces designed for humans working within large corporations, ROSS learns and adapts to the legal language that human lawyers use in courts and firms. It repurposes the natural language-processing abilities of IBM’s Watson supercomputer to do this, scanning and analysing 10,000 pages of text every second before pulling out its best answers, ranked by confidence.
Lawyers are giving it feedback too, says Jimoh Ovbiagele, ROSS’s chief technology officer. “ROSS is learning through experience.”
Massachusetts-based Nuance Communications is building AIs that solve some of the same language problems as ROSS, but in a different part of the economy: medicine. In the US, after doctors and nurses type up case notes, another person uses those notes to try to match the description with one of thousands of billing codes for insurance purposes.
Nuance’s language-focused AIs can now understand the typed notes, and figure out which billing code is a match. The system is already in use in a handful of US hospitals.
Kingdon doesn’t shy away from the implications of his work: “This is aimed at being a replacement for a human, an automated person who knows how to do a task in much the same way that a colleague would.”
But what will the world be like as we increasingly find ourselves working alongside AIs? David Autor, an economist at the Massachusetts Institute of Technology, says automation has tended to reduce drudgery in the past, and allowed people to do more interesting work.
“Old assembly line jobs were things like screwing caps on bottles,” Autor says. “A lot of that stuff has been eliminated and that’s good. Our working lives are safer and more interesting than they used to be.”
The potential problem with new kinds of automation like Blue Prism and ROSS is that they are starting to perform the kinds of jobs which can be the first rung on the corporate ladders, which could result in deepening inequality.
Autor remains optimistic about humanity’s role in the future it is creating, but cautions that there’s nothing to stop us engineering our own obsolescence, or that of a large swathe of workers that further splits rich from poor. “We’ve not seen widespread technological unemployment, but this time could be different,” he says. “There’s nothing that says it can’t happen.”
Kingdon says the changes are just beginning. “How far and fast? My prediction would be that in the next few years everyone will be familiar with this. It will be in every single office.”
Once it reaches that scale, narrow, specialised AIs may start to offer something more, as their computation roots allow them to call upon more knowledge than human intelligence could.
“Right now ROSS has a year of experience,” says Ovbiagele. “If 10,000 lawyers use ROSS for a year, that’s 10,000 years of experience.”