Once the machines came and took the menial jobs. Now they’re taking over the executive suite. Artificial intelligence has come of age, says Matthew Partridge.
Until recently, artificial intelligence (AI) – machines that can think for themselves – was the technology that was long promised but never quite delivered. Even when the Deep Blue computer defeated chess champion Garry Kasparov in a match in 1997, or when a similar machine solved draughts a decade later (ie, could follow a provably optimal strategy), these victories were written off as a triumph of brute computing power rather than of human-like thinking.
Most experts remained convinced that any further developments would be confined to slow incremental progress in specific areas, such as image and voice recognition. However, in the past five years barriers that were previously considered insurmountable have tumbled at a fast rate.
In 2011, “Watson”, a computer designed by IBM, defeated two champions of the hit US TV quiz show Jeopardy!. Three years later a computer managed to convince a panel of judges that it was really a 13-year-old Ukrainian boy, leading overexcited reporters to claim that a machine had for the first time passed the AI test laid out by legendary computer scientist Alan Turing (it hadn’t). However, even these achievements have now been eclipsed by the recent victory of a computer (AlphaGo) over Lee Sedol, the Korean champion of the board game Go.
This is important because Go is a much trickier game for a computer to crack than either draughts or chess. Because there are so many possible moves, you can’t win the game simply with calculating power. Winning demands intuitive and strategic thinking.
In fact, just a few months ago one expert predicted it would take as long as a decade for computers to have a chance of matching the best human players. Sedol admits that he expected to win easily, his only concern being that he might suffer the “humiliation” of losing one game in a multi-game match. In the event, he lost the best-of-five series 4-1, his sole victory coming after he had been swept in the first three games.
This is not to say that AlphaGo got beyond brute calculation entirely and now thinks like a human, but it does seem to some like a step in that direction. Even optimists are cautious. Futurologist Ray Kurzweil, for example, thinks that we won’t have conscious computers that are truly able to pass the Turing test – ie, convince a human that they too are human – until 2029, nor reach the “singularity” until 2045 (this is the point where machines are capable of doing things and developing themselves in ways we aren’t capable of predicting or even understanding). He thinks man/machine hybrids will be with us sooner – perhaps within 15 years.
That’s all for the future, possibly. But using the term in a narrower, more task-specific sense, AI is already transforming key sectors of the economy. While all sectors of the economy are likely to be affected, most of the big opportunities will be in medicine, business operations, internet search and robotics.
Artificial intelligence: the algorithm will see you now
One area that is particularly vulnerable to disruption is medical diagnosis. A rapidly ageing population and the growing global middle class is sending demand for health care through the roof. The problem is that it still takes a lot of time and money to train a doctor – one estimate puts the cost at £1m when you include postgraduate training and insurance. Another problem is that even the best medics make mistakes. Researchers estimate that at least 98,000 preventable deaths or permanent injuries occur in America each year as a result of diagnostic error, or poor treatment decisions.
Computers might be a cheaper, even a better, option. Computer systems are being developed that can diagnose some conditions more accurately than even the best physicians. Last year, San Francisco-based diagnostics company Enlitic ran a test that compared the ability of one of their programs correctly to diagnose early-stage lung cancer with that of a panel of leading doctors.
It claims that the computer was far less likely to come up with false positives (cases where it incorrectly believed the patient had cancer) and gave no false negatives (giving the all-clear to a patient who actually has cancer). By contrast, the human doctors, who were experts in their field, missed around 7% of cases, potentially putting lives at risk.
As well as better diagnosis, computers could also prescribe pills. After conquering TV quiz shows, IBM partnered with the prestigious Memorial Sloan Kettering Cancer Center in New York to apply its “Watson” AI system to medicine. The program can now scan through a huge database of textbooks and the latest medical journals to recommend the most effective therapies based on the patient’s symptoms. This has been so successful that it has been extended to 14 leading cancer centres in America and Canada as well as a chain of hospitals in India.
IBM is also supporting Modernizing Medicine, an electronic records management system that makes it much easier for doctors to discover effective drugs for a wide range of conditions, even those that are relatively obscure. Already, 5,000 hospitals and practices have signed up, including a third of dermatologists.
Early versions just used prescriptions by other doctors to generate recommendations (a bit like Amazon’s system for tipping books you might like). But the most recent updates now include a version that can come up with its own ideas (important if few doctors have encountered that particular problem, or a new drug has just been released).
Computers take over the board
Doctors aren’t the only middle-class professionals who should be worried. Until now advances in computing have tended to take over jobs at the bottom of the economic ladder, especially those that are administrative, secretarial or clerical. But as computers become smarter, managers and even some executives are now becoming increasingly vulnerable to automation.
Take Hong Kong’s Metro system, for example, which carries more than five million passengers every day. It has successfully transferred responsibility for deciding maintenance schedules and prioritising engineering tasks to AI. That’s made many managers redundant, making the whole system less bureaucratic and costly, and has also boosted response times. The system has been so successful that MTR Corporation, the company that runs the Hong Kong system, is rolling it out to other cities, including Beijing and the London Overground train network.
Management consultants could soon become a thing of the past. In the last few years, firms have become increasingly interested in “big data”, the ways in which large amounts of information can be harvested (from customers, for example) and analysed to improve corporate performance. The problem has been that, while it’s never been easier (or cheaper) to capture information, it still needs to be analysed and there are a limited number of analysts skilled enough to do it. Indeed, McKinsey Global Institute predicts that, unless something radical happens, there will be an acute skill shortage in this area within two years.
Firms such as IBM and Microsoft are trying to fill this gap by developing software that can do much of the work of a human analyst. In the past, such automated data mining was only possible using highly structured data that could be easily categorised. But the technology now is becoming much better at dealing with unstructured data, such as photos, sales records and even emails. Such technology is better than humans are at making unexpected connections and spotting meaningful patterns.
Would you trust a robot to do such analysis for you? Well, why not: Japanese venture-capital firm Deep Knowledge is so confident in the ability of its AI program to make investment decisions that it’s decided to put it on its board of directors. The program, Validating Investment Tool for Advancing Life Sciences, analyses research data and makes recommendations about the viability of treatments for age-related conditions, which the company invests in. Its analysis and recommendations have already persuaded Deep Knowledge to invest in two life-sciences companies.
Robots to sort out your inbox
AI could make the experience of contacting your bank or phoning a call centre more, well, human. Faced with increasing regulation on the fees they can charge for financial advice, advisers plan to shift low-end work to “robo-advisors”, who can deal with more routine problems. In March, Royal Bank of Scotland said it would replace around 200 staff with a computer program named Luvo, designed to mimic human empathy when answering customers’ questions. Last year, US brokerage Charles Schwab launched Schwab Intelligent Portfolios, to create automatic investment portfolios for its customers. This now has $4.1bn of assets under management.
And how about a robot to help you tackle your inbox? Five years ago, Apple’s introduction of the virtual personal assistant Siri on its smartphones proved a major factor in the success of the iPhone 4. As a result, search engine and smartphone companies have invested heavily in similar technologies. One of the most interesting is an application from Google that can scan the content of emails and generate a list of potential responses. This isn’t quite an automatic reply, but it does promise to make it easier for people to deal with the ever-larger number of emails they receive each day.
Finally, AI is providing the “brains” for a new generation of serfs. One area of growth is robots to do the chores. The International Federation of Robotics estimates that in 2014, 4.7 million personal and domestic robots were sold, an increase of over a quarter in one year. Sales of industrial robots also hit a record level of 229,000 in the same timeframe. Market researcher Tractica predicts the global robotics market, both industrial and non-industrial, will be worth $150bn by the end of the decade.
Six ways to back the rise of the machines
IBM (NYSE: IBM) is trying to move away from focusing on more traditional IT services towards business analytics. Key to this is a huge investment in artificial intelligence, notably its Watson system (see above). IBM has also acquired one of the largest portfolios of AI-related patents. Already, analytics accounts for just under a quarter of revenue and is growing at more than 15% a year. Despite this, IBM is priced conservatively, at only ten times 2017 earnings.
As the leading manufacture of computer chips, Intel (Nasdaq: INTC) has played a key part in the exponential growth of computer processing power. It’s now attempting to build AI capabilities into its products. As a result, it has been aggressively acquiring firms specialising in AI. These include Indisys, Xtremeinsights and chipmaker Altera. Last October it bought Saffron Technology, which has developed AI technologies with large potential in the field of cybersecurity. Intel shares trade at twelve times 2017 earnings.
NVidia (Nasdaq: NVDA) makes high-powered chips for use in graphics cards.Modern AI applications use a lot of power and NVidia has the technology to help them cope. The firm is buying up machine-learning firms in an attempt to make its approach to processing standard throughout the industry. Its shares trade on 20 times 2017 earnings.
The value of Chinese search engine Baidu (Nasdaq: BIDU) has been hit by fears that China’s economy is dramatically slowing down (though the stock has rallied a bit in the past few weeks). To maintain growth rates of around 30% a year, it has been investing in AI. The primary focus is on improving search results and providing related services, such as language recognition software for smartphones. The shares trade at just under 20 times 2017 earnings.
iRobot (Nasdaq: IRBT) makes domestic robots, most notably its range of cleaners and floor washers. All these systems rely on AI to guide themselves around rooms without human input. The company is spending large sums upgrading the AI systems of its machines in an attempt to keep ahead of its competitors, and is developing robots for the military and the police. The shares currently trade on a price/earnings ratio of 22, but this should fall to below 16 times by 2018.
Beyond individual stock-picking, ROBO Global Robotics and Automation (Nasdaq: ROBO) is an exchange-traded fund that tracks robotics-related firms. It’s a niche product, which allows it to get away with charging a relatively high fee of 0.95%, but it remains the best way to get broad exposure to the robotics sectors. With US firms accounting for only a third of the ETF’s holdings, it is also well diversified geographically.