It’s the age of anxAIety

Artificial intelligence, until recently, could not perform jobs that required human judgment and creativity. Not so today. Is any profession safe from the encroachment of AI?

Alexandre Le Bouthillier, an operational research and parallel computing expert and entrepreneur, lost his father a year ago to cancer. “It took a whole year to know what he was suffering from. Had we detected the signs of cancer earlier, we probably could have saved him,” Le Bouthillier says.

It coincided with his founding of Imagia in Montreal, a startup he founded in 2015 that develops AI software platforms for the early detection and treatment of cancers. Existing diagnostic systems already perform quite well. On nascent lung tumours of five millimetres in diameter, systems accurately discover the tumours 83% of the time. Imagia’s system, not yet commercially available, delivers an accuracy level of 95%, says Le Bouthillier.

AI SPRAWL

Radiology is not the only discipline where AI systems are beating their human counterparts. In major companies, AI is performing better than the very best sales forecasters and is more successful than hiring managers are at recruiting candidates who stay longer on the job. At Volkswagen, for example, machine learning algorithms (a popular branch of AI) developed to predict sales performance reach accuracy levels up to 90%, “a stunning, surprising success,” Volkswagen’s CIO, Martin Hofmann, told web magazine ZDNet. Human forecasters reach an average of 60%, he said.

In tax matters, AI is making significant inroads, for example, with the Tax Foresight system by Blue J Legal in Toronto, currently implemented at a number of Canadian accounting firms.

“The system compares the client’s situation to other cases, gives you a predictive outcome, generates an explanation of that predicted outcome, reports confidence in the predicted outcome as a percentage, and identifies the most similar past cases that have led to court and to court decisions,” says Benjamin Alarie, the startup’s cofounder and CEO, who concluded his term as associate dean of the University of Toronto’s faculty of law before launching Blue J Legal. (He remains Osler Chair in Business Law at U of T.)

Tax Foresight is very impressive, says Marlene Cepparo, partner-in-charge of the National Tax Center at KPMG, which participated in the algorithm’s pilot implementation in 2016. Results that could require up to 40 hours of research by a junior come up within two minutes, she says.

AI is also making surprising strides in finance. In 2016, Hong Kong’s Aidyia launched a hedge fund that is totally run by AI — “no human intervention required,” reports Wired. The machine selects stocks and trades them on its own.

Do top financial algorithms perform better than their human counterparts? Luc Bégnoche, founder of Montreal-based Aidvisors, which is developing AI investing systems, claims that some financial AI systems are doing a lot better than humans. For instance, since its launch in 1988 by James Simons, Renaissance Technologies’ Medallion fund, a massive user of AI, has amassed yearly returns in the area of 35%, peaking above 70% in 2000 and in 2008 when markets were at their worst.

MONAI

AI systems are pushing ahead like never before. After going through what’s been called the “AI winter” in the 1980s and 1990s, the discipline’s pace of development picked up in the early 2000s with the invention of deep learning neuronal networks, large matrices of processing nodes many layers deep that seek to imitate the structure of the human brain.

Everything is unfolding faster than the practitioners of AI expected. In 2015 Nick Bostrom, a leading light in the field, author of Superintelligence and director of the Future of Humanity Institute at the University of Oxford, thought that it would take at least 10 or 20 years before an AI system could beat a champion of Go, a game considered more complex than chess. “It happened the following year, in 2016,” says Alain Tapp, professor at the Montreal Institute for Learning Algorithms at the Université de Montréal, when Google’s AlphaGo system beat Lee Sedol, the world’s uncontested Go champion, four games out of five.

Now, money is pouring into the discipline, notably in Canada, which has carved out a spot for itself at the leading edge. At the Université de Montréal’s IVADO project (a collaboration with HEC Montréal and Polytechnique Montréal), commitments total $245 million, and at the University of Toronto’s Vector Institute for Artificial Intelligence, $150 million.

In the world of equity funding, US$14.9 billion has been injected into 2,250 deals since 2012, according to CB Insights, numbers practically doubling each year. And that’s not counting the billions invested by the four leaders of AI development: Amazon, Google, Microsoft and IBM, vying for an AI market that should grow from US$8 billion in 2016 to US$47 billion in 2020, according to market research firm IDC.

The flow of money responds to the irrepressible economic impetus of AI. “The cost of a professional job offshored represents one-third [of the cost in North America],” says Ramy Sedra, partner and data analytics consulting leader at PwC in Montreal. “With AI,” he says, “the cost is one-ninth.”

ME, MY ROBOT, MY JOB

Insecurity is in the air, particularly among professionals, with many wondering, Are robots after my job? Technology has certainly not waited for AI to destroy and create millions of jobs. But decimation has happened mostly in manual and repetitive jobs, not “thinking” jobs. And many technologies, notably computers and the Internet, have already shifted things around and are eating away at the fringe of many professions, as Richard and Daniel Susskind chronicle in their 2015 opus, The Future of the Professions — How Technology Will Transform the Work of Human Experts.

The authors note that in the legal sphere, for example, tasks such as document review in litigation and due diligence “are now being outsourced, offshored, passed along to paralegals, subcontracted and sold to clients on a fixed-price basis. Some leading firms are setting up their own low-cost service facilities.” In management consulting, “the availability of basic analytical tools, and more sophisticated systems as well, enables people outside traditional consultancies also to process data and to tease out the sort of insights over which consultants once held a monopoly. ... As a result, traditional strategy consulting firms now offshore a great deal of their routine research.”

Now, the fear is that AI wants to snatch away almost everybody’s job, a fear that a number of studies have vigorously stoked. In 2013, Oxford University’s Carl Benedikt Frey and Michael Osborne came to the issue with a flamethrower in their report The Future of Employment: How Susceptible Are Jobs to Computerisation?

“According to our estimates,” they wrote, “around 47% of total US employment is in the high-risk category, i.e. jobs we expect could be automated relatively soon, perhaps over the next decade or two.” Of course, many of those jobs sit in middle-income manufacturing and low-income service occupations, but many are at the professional level.

More recent studies significantly tone down the fear-mongering rhetoric. A representative report from the McKinsey Global Institute (A Future that Works: Automation, Employment and Productivity, January 2017) shifts the focus from “jobs” to “tasks.” Jobs are comprised of many tasks, and while tasks can be automated, jobs will mostly hold steady. “While less than 5% of all occupations can be automated entirely using demonstrated technologies,” writes McKinsey, “about 60% of all occupations have at least 30% of constituent activities that could be automated. More occupations will change than will be automated away.”

MORE FOR EVERYBODY

For now, AI will not bump professionals; it will support them. “It’s hard to say that you’re replacing a person,” says Cepparo in reference to Tax Foresight. “It’s essentially a research tool we use along with others. It doesn’t take the place of judgment and decision.” Le Bouthillier shares that view: “Our system doesn’t herald the disappearance of radiologists, not even the reduction of their numbers.”

Judgment, decision, intuition, consciousness: four ramparts that will hold back the flood of AI, according to some. But many specialists believe those ramparts are fragile. Algorithms already make countless decisions, as in investment systems. The question is, how much decision-making power are humans ready to hand over to machines? Present-day airliners have a pilot always standing by, but the planes still fly on their own, a process that requires a lot of judgment and decision-making. And even without consciousness, AlphaGo nevertheless beat Sedol and some believe that the system exhibited creativity, even intuition, according to Sedra.

A certain effect of AI will be a democratization of professional services, thanks to a dramatic increase in performance speed and a reduction in costs, a frequent logic behind technology’s thrust. Bégnoche definitely leans on the side of democratization: “I want to give tools to everyone, not just to the wealthy, that are on a level with the best that’s done today.”

SHIFTING GROUND

Even if the AI forces at work are not as threatening as Frey and Osborne predict, they still create an unstable landscape. Undoubtedly, the jobs of auditors, actuaries and materials engineers are not at risk, but as tasks become automated, other jobs will probably be restructured, downgraded or deleted.

“What has been happening to a lot of jobs is that the qualifications required to carry [them out] have been pushed down [because of new technology that automates part, but not all, of their work],” said Albert Wenger, a managing partner at Union Square Ventures, during a panel discussion in April. Workers requiring less skill because machines have more, he said, will likely command cheaper wages, which will slowly be pushed down below the cost of any technology that could replace them.

While many jobs are still safe, some are definitely in the firing trajectory of AI, notably financial advisers and, arguably, portfolio managers, as developments at Aidyia and Aidvisors suggest.

And let’s not forget all the pressure that AI innovations are bringing to bear on large corporations, notably big accounting and consulting firms. “It changes equations for big players,” says Sedra. “Small companies with lower investment capital can still produce solutions and services as efficient, if not more efficient, than those of rivals a hundred times bigger. It’s a huge challenge for them to remain competitive.” It also puts enormous pressure on many high-wage and high-skilled workers in these corporations.

So, considering only the partial push of AI in the area of tasks, many jobs can still be sliced, diced, reshaped, repriced, downgraded and potentially eliminated.

But jobs will also be created.

JOB-CREATION ENGINES

While most studies to date have examined the loss of jobs due to AI, surprisingly few have looked at the numbers of jobs created. AI will infiltrate just about every sphere of activity, not only work. In fact, most AI developments happen outside formal jobs, for example in the countless applications that identify photos on iPhones, analyze massive amounts of data collected on Facebook and that will eventually drive an increasing number of interfaces in cars, websites and appliances. These AI interfaces and assistants will not be performing anyone’s present jobs, but their implementation will command the creation of countless jobs in programming, interface design, language understanding, interactivity management, etc. For this story, we haven’t found a single study that tries to put a number on those new jobs.

However, in its study “How companies are reimagining business processes with IT,” Accenture suggests the creation of three totally new categories where jobs could multiply, although it didn’t extrapolate numbers. Trainers will teach AI systems how they should perform, explainers will bridge the gap between technologists and business leaders, and sustainers will ensure that AI systems are operating as designed and that unintended consequences are addressed with the appropriate urgency.

And that’s probably only the beginning. Optimists are confident that new jobs will emerge. “The robot economy will invent work we can’t even dream of today, much as the Internet gave birth to unforeseen careers,” says Kevin Maney, columnist and author of Play Bigger, in a 2016 Newsweek article. “Nobody’s grandmother was a search engine optimization specialist. Today, that job pays pretty well.”

However, the logic of technology doesn’t run in a straight line. The three work categories outlined by the Accenture study will not necessarily require high levels of training. “Empathy trainers, for example, may not need a college degree,” notes Accenture. So accountants whose jobs are undermined by AI will not necessarily find equivalent jobs as explainers, unless they train accordingly; and whether they will climb back to the same salary level is an open question.

A recent study by the Information Technology and Innovation Foundation (ITIF), titled False Alarmism: Technological Disruption and the US Labor Market, 1850-2015, confirms that technology creates jobs, but “normally not as many as it eliminates.” Oops.

Most new jobs “will not be created in the new machinery firms,” writes ITIF. “Rather, they will be created across the economy from the new demand that higher productivity enables.” The 1960s give an eloquent example. While the TV industry was rising and automation was increasing productivity and cutting down employment in manufacturing plants, job creation happened elsewhere: janitors, labourers, freight-stock-material movers, maids and house cleaners, secretaries and administrative assistants accounted for 56% of all net job creation in that decade.

Fine, AI will create lots of jobs at all levels. Which ones? One thing is quite sure: many jobs will revolve around that most human of qualities, empathy, predicts Pierre Jean-Claude Allard, an economist, lawyer and author of Crisis and Beyond. “The other two human traits where jobs will concentrate,” he says, “are decision-making and creativity.” But it is not clear how that will translate for professionals and the highly skilled. It could be very positive, but potentially not without a great deal of disruption.

THE BIGGEST EVENT IN HISTORY

Something lurks inside AI that most job creation projections can’t account for. Granted, as AI infiltrates activities and occupations, it will almost certainly create jobs and job categories. But here’s the rub: it is also likely that AI will itself be able to rapidly take up those new jobs.

“It takes only a few weeks to turn a game-playing system into a champion,” says Richard Zemel, professor of computer science and research director at Toronto’s Vector Institute for Artificial Intelligence.

But we are not yet at the point where AI systems will just gobble up whole jobs as fast as they pop up. “For that, we will need another AI revolution,” says Le Bouthillier.

That revolution mostly involves the capacity to instil AI with common sense. Current AI systems, outside their narrow field of expertise, are totally dumb. Let a stone fall off the Go board and AlphaGo would be paralyzed because it doesn’t have a clue what it means for an object to fall. “But we’re getting there,” says Zemel. “In the next five to 10 years we could get machines to do more of what people call common sense.”

Of course, scientists are very bad at extrapolating, says Tapp. The revolution might very well never happen, though that is unlikely. It could be in 10 years, he says. Or it could be 20 years. But it could be next year.

“In the long term, all jobs could be in jeopardy, even that of researcher, though we don’t really know,” says Tapp. Algorithms could potentially step in quickly to fill the jobs that AI creates, says Stuart Armstrong, a research fellow at the Future of Humanity Institute.

The AI community believes that injecting algorithms with common sense could eventually lead to the development of human-level AI, an event researchers and futurologists set in an uncertain future: 20 years from now, or 200. Who knows? But we must remember that things are moving faster than anyone anticipated.

When that happens, it could be “the biggest event in human history,” says Stuart Russell, a professor of computer science and founder of the Center for Intelligent Systems at the University of California, Berkeley. “And if that’s true, then we need to put a lot more thought into what the precise shape of that event might be.”

After human-level AI, the next biggest event in history would probably happen quite quickly: the rise of a superhuman level of intelligence, what Bostrom calls superintelligence. Systems so efficient, lucid and powerful we would have no clue what they could be up to. Bostrom, Armstrong, and a few others think one possibility is that they would plan the extinction of humanity.

And here we enter the nebulous world of futurology, though not fantasy. Many serious people prompt us to really think about the societal impact of AI, among them Russell, Microsoft’s Bill Gates, physicist Stephen Hawking, entrepreneur extraordinaire Elon Musk and thousands of others who have signed an open letter warning humanity that we have to think very carefully about what we want AI systems to do.

Questions abound. Here’s a small sample. Will AI usher in an age where work will become obsolete? Is that desirable? Would it bring us back to ancient times, when the Roman aristocracy discoursed, philosophized and competed in public squares while slaves toiled and the lower classes spent their lives in idleness and misery? (In this new social organization, slaves and labourers would be robots and AI agents.) Should we consider a universal guaranteed income for large categories of citizens, distributing the riches of the zero marginal cost economy that many see coming? And should we not make sure that AI and its robotic extensions remain aligned with human goals and values, and not come up with objectives of their own that are potentially at odds with ours? And if humans are challenged by AI, would they be compelled to “upgrade” into a posthuman cyborg state? Ultimately, the foremost question will emerge: what is it really, truly, essentially, to be human? Do we want to preserve that?

Such questions seem like science fiction. Wrong. They are preoccupations that lie consistently at the fringe of the field of AI and chances are their urgency will increase.

When asked about the future of the legal profession and if he would advise his two young daughters to pursue a career in law as their father did, Alarie was cautiously optimistic. “I would,” he answers, “but I would also tell them that being a lawyer in the future will not be the same as today. They would need to be nimble and flexible and to constantly revisit their practice.”