The Artificial Intelligence Bandwagon
Opposite the London office of Teneo, next to the Shipwrights Arms pub, there is a taxi stand. There are usually a few of the city’s famous black cabs waiting.
Take a ride, and the cabbies invariably grumble about having to watch potential customers climb into Uber ride-hailing cars that collect passengers across the street.
The London taxi trade is an unlikely battleground for the future of Artificial Intelligence (AI). But economists Ajay Agrawal, Joshua Gans and Avi Goldfarb, based at the University of Toronto, argue that the taxi industry is just one of the sectors facing an AI upheaval that might be paralyzing.
In their book Prediction Machines, Agrawal, Gans and Goldfarb summarize the disruption to the London taxi trade by writing: “Cabbies who invested three years of studying to ‘The Knowledge’ did not know they would someday be competing with prediction machines. Over the years, they uploaded maps into their memory, tested routes, and filled in the blanks with their common sense. Now, navigation apps have access to the same map data and are able, through a combination of algorithms and predictive training, to find the best route whenever requested, using real-time data about traffic that the taxi driver cannot hope to know.”
Artificial intelligence has spawned a whole industry of ride-hailing services around the world. Automotive companies are rushing to join the bandwagon, with every leading manufacturer launching mobility services or forging partnerships with the likes of Uber or Lyft. Even the owner of the London Electric Vehicle Company, the manufacturer of London black cabs, has joined the AI arms race. Zhejiang Geely Holding Group, China’s largest privately-owed automotive group which acquired the former London Taxi Company in 2008, is also the founder investor of Cao Cao, the operator of more than 20,000 electric ride-hailing cars across 26 cities in China.
Li Shufu, Chairman of Geely Holding, believes such mobility services will revolutionize road transport when smart algorithms are combined with autonomous vehicle technologies. “Automated Driving is going to shake the foundations of the automotive industry,” according to Li. “It’s an environment that will be able to liberate the driver, lower the number of accidents, improve the efficiency of roads and improve fuel economy. Automated driving is the future of the automotive industry’s development and will bring greater value to society.”
The research firm IDC predicts that the market for the Internet of Things (IoT) – with AI as its beating heart – will grow to $1.7 trillion dollars by 2020, a near threefold increase over the past half-decade. Over that period, IDC expects the number of so-called ‘IoT’ endpoints such as automated cars, smart phones and robotic systems will reach almost 30 million.
Rival analysts and researchers at McKinsey believe that AI will create significant value across multiple industries, with the markets in retail, transport, logistics, travel and healthcare systems among those expected to see a major transformation. In its 2018 report Notes from the AI frontier, McKinsey says: “We estimate that the AI techniques together have the potential to create between $3.5 trillion and $5.8 trillion in value annually across nine business functions in 19 industries.” This value opportunity is creating hope and striking fear into large parts of the corporate world in seemingly equal measure.
Of course, AI is not new. Different forms of such technology have been transforming business since the Industrial Revolution. From autopilot in aircraft to cruise control in cars – and latterly to smartphones – consumers have mostly adapted and taken each new innovation for granted. What is now changing is the pace, the sophistication and complexities created by the latest generation of AI systems. According to McKinsey, the key transformation has been the ability of machine learning systems to model the way that neurons interact in the brain. This, in turn, is creating ‘deep learning’ techniques in what AI enthusiasts described as “neural networks” that have the power to transform data and to apply predictive processing power into completely new ways of doing business.
AI specialists talk of three categories of neural networks that are changing the way we all interact with machines and with each other. These categories are ‘feel forward, recurrent and convolutional’ neural networks. At this point, the language of AI frequently descends into levels of complexity and jargon that frazzles the minds of most ordinary businessmen and women. But in industries from agriculture to mining, or from consumer-packaged goods to telecommunications, the benefits boil down to a simple phenomenon: competitive advantage. Companies with smarter and relevant AI tend to be more competitive. They tend to have a lower cost of capital, a higher return on capital employed, more pricing power, more loyal customers and better-quality products and services. That, in turn, feeds through to higher margins, growing cashflow and share-value creation.
For industrial groups and manufacturers, one example of competitive advantage from AI lies in predictive maintenance. This is the power of machine learning to detect anomalies in assembly lines – and thereby reduce downtime and increase productivity. In logistics, meanwhile, AI systems can optimize delivering routings to cut costs and speed-up traffic management. And AI promises to revolutionize customer service and product recommendations.
Citing examples from the retail and content industries, Michael Chui and his fellow researchers at McKinsey say: “Combining customer demographic and past transaction data, with social media monitoring can help generate individualized product recommendations. ‘Next product-to-buy’ recommendations that target individual customers – as companies such as Amazon and Netflix have successfully been doing – can lead to a twofold increase in the rate of sales conversions.”
Clearly the transformative nature of AI applications could be enormous. But they come at a price. Greater automation will change the way current business processes happen, with an obvious knock-on effect on jobs. Some commentators fear that AI will make assembly-line workers, logistics staff and retail employees metaphorically and literally redundant. Those who earn the least – non-skilled or semi-skilled factory workers or warehouse employees – and who can least afford to lose their jobs may be displaced by growing numbers of robots and algorithmic systems.
Martin Wolf, the chief economics commentator of the Financial Times, paints an apocalyptic picture of the social consequences of machines replacing workers. He fears grave social consequences if smart algorithms make knowledge-acquisition just a past-time and dispense with the need to work. In a recent column, Wolf questioned the future of society “in a world in which few people can do anything that is obviously economically productive. The world might become techno-feudal, with an owning elite hiring great numbers of human servants not for their value, but for the pleasure of domination.” The alternative, he suggested, could lead to a society on which: “People might instead share the abundance more equally, with all enjoying civilized leisure that was once the province of the very few. Ours is the first civilization to view work as the highest calling. Maybe that strange prejudice will need to be discarded.”
The reality is that no-one quite knows how the AI revolution will evolve, or what the broader societal impact will be. We are just at the frontier of what might be possible in terms of automation. But it is a safe bet that it will, at the very least, transform the world of employment.
Andrew N. Liveris, Chairman Emeritus of the Dow Chemical Company, is among those concerned about this risk-benefit equation. “The digitization of supply chains is going to accelerate with the rise of artificial intelligence, along with the automation of factories as first-generation robotics are upgraded with the addition of Big Data,” he says. “If companies fully adopt artificial intelligence it might increase their operating efficiency but lead to significant declines in employment. Millions of jobs could be affected by the digitization of production lines.”
And it will not just be manufacturing or blue-collar jobs that could become dispensable. Even sectors as oriented to personal choice as fashion and design could be affected. “Clothing design is only the leading edge of the way algorithms are transforming the fashion and retail industries,” wrote Naom Scheiber in the New York Times in July. “Companies now routinely use artificial intelligence to decide which clothes to stock and what to recommend to customers.”
Return on Investment Hurdles
One of the basic business laws that could hold back this tide, and give companies and their workers time to adapt, is the need to guarantee a return on investment.
Some consultants regard the cost and complexity of deploying AI as simply not worthwhile compared with the value generated. Few carmakers, for example, have yet explained the business case for AI. Tesla, one of the early adopters and main advocates of automated driving, has never turned a profit. Most car brands with electric vehicles make much more money and higher margins from gasoline and diesel models than their zero-emission counterparts, even though they have all vowed to combine electrification with automation in years to come.
AI also comes with huge potential legal and regulatory liabilities. An automated car that hits a pedestrian, as happened in Arizona earlier this year, can pose a nightmare for insurers and could inflate the potential legal costs of ride-hailing services. The same is true in medicine where zero failure rates would be required to minimize malpractice claims for AI diagnostics and healthcare testing. To this mix should be added the growing public backlash over the misuse and reselling of private data, which has undermined trust in companies seeking to commercialize individuals’ personal information.
Rising consumer, legal and regulatory concerns about AI have therefore reduced trust in its benefits. This, in turn, has forced AI advocates to be open about its limitations.
Even the most ardent AI enthusiasts admit that the one thing machine intelligence lacks is judgment. Essentially, AI relies not on the ability to make judgments, but on predictive learning and the constant refinement of data that – when combined together – helps reduce uncertainty in many forms of commercial activity and business services.
Take the retail industry. As AI is adopted more widely, there will be more predictive shopping services, with anticipatory product deliveries before customers even order them. Predictability will be enhanced by smarter use of data and machine-understanding of consumer preferences. And constant evolution of AI machines based on data feedbacks and scenario-mapping will require less human control for everyday tasks from parking a car to school-testing or book-keeping.
So, what may yet save us in terms of human-machine interaction is personal judgment. It requires human intervention and ingenuity to decide how to apply AI and, critically, to recognize its limitations. This is particularly true when it comes to safety-critical machinery.
Nowhere is that combination of human ingenuity better displayed than at Tytyri, a limestone mine about 50 kilometers west of Helsinki, the Finnish capital. This mine is the research and development hub of Kone, one of the world’s largest elevator manufacturers. Here, deep underground, the company uses AI to determine the safety of next generation elevators in shafts that drop hundreds of meters below the earth’s surface.
There are, in total, 11 elevator shafts with a combined length of 1.6 kilometers that test the speed, the braking power and the latest standards of elevators – one of the earliest forms of mass transit. We tend to take elevators for granted. But they are vital and complex components of all tall buildings, without which high-rise urbanization would grind to a halt.
At the Tytyri mine, Kone uses AI sensors and data systems to test critical safety on elevators which might be dropped 200 meters, at speeds of up to 26 meters per second, to see if its safety equipment can stop it successfully. Kone uses AI to detect that the car is speeding downward, and which then deploy a metal wedge into a channel in the elevator-shaft guide rails. As friction builds between the wedge and the rail, the elevator car comes to a stop at a comfortable rate. The company has teamed up with IBM to take predictive machine-learning to the next level. It is using the IBM Watson IoT platform to enable escalators and elevators to talk to the cloud intelligently and in real time. The data derived from the cloud allows building managers to monitor systems and to schedule upgrades or maintenance work before problems occur.
Paying for Al
In a recent article published on Kone’s website, Andy Stanford-Clark, Chief Technology Officer, UK and Ireland, Distinguished Engineer, and Master Inventor at IBM, said: “If we can predict maintenance problems, to tell when something is going to fail rather than service something once a year, or wait and then react when it breaks down, the savings in terms of cost and time can often pay for the investment in IoT and AI technologies on its own.” The Kone-IBM partnership illustrates another key element in the evolution of AI: few companies can develop it on their own. The AI sector is spawning a new era of partnerships. Legacy manufacturers, aware of huge development costs and uncertain returns from machine learning, are increasingly forging deals or buying in AI expertise to mitigate the investment risks.
This is particularly true in the auto industry, where some of the world’s largest brands are turning to outside experts for automation systems. Nissan, one of the most advanced users of self-driving systems, has, for example, teamed up with DeNA, the Japanese internet technology company, to develop a new ride-hailing service using automated vehicles. Their ‘Easy Ride’ service has begun tests with robo-vehicles in Nissan’s hometown of Yokohama. The goal is to allow customers to use a dedicated mobile app to complete the whole process from setting destinations and summoning vehicles, to paying the fare.
And companies such as Nissan are not just signing commercial ties-ups. They are hiring experts who can apply different sorts of judgment to next-generation autonomy. One of those is Melissa Cefkin, a cultural anthropologist who now leads the “Human Centered Systems” practice at the Nissan Research Center in Silicon Valley. Her job: to predict the unpredictable and see how a machine can apply it. The idea is to try to ensure that an automated car will, in future, be able to make a judgment call about an unforeseen road event, such as a child chasing a ball into the road and making a decision to swerve into oncoming traffic – something most automated systems would prevent – to avoid the worse of two accident-scenarios.
Working with other social scientists at Nissan, Cefkin explains: “We are expected to provide results that can be implemented into algorithms, resulting in a challenge to our social science perspective: How do we translate what are observably social practices into implementable algorithms when road use practices are so often contingent on the particulars of a situation, and these situations defy easy categorization and generalization?”
Critical Input: CEO Judgement
This is the crux of the AI frontier. Machines can be programmed to predict behaviors in a whole range of industries and services. But they have not yet learned – or been programmed – to apply judgment to make different types of decision. Whether in medicine or schooling, or from assembly lines to warehouse management, AI will clearly reduce waste, speed up systems, enable predictive ordering and product deliveries. It could transform cost-management and greatly enhance operating efficiency – partly at the cost of traditional jobs. Machine learning, combined with smart data management and predictive systems, will become vital for companies to compete in the new age of efficiency linked to AI. But it is the ability to apply and judge when and how to use such AI that will redefine and ultimately protect the role of the individual. As the authors of Prediction Machines admit: “If judgment could be well specified, then it could be programmed, and we wouldn’t need humans to provide it.” Happily, that is not yet the case. The applications and value of machine learning remain something that individuals in business, in public service, in education and health, can still determine how best to use and when to avoid – using human intuition and judgment. It is human unpredictability, the irrational behavior that for many years was seen as a management weakness that might – just might – become a management strength in the coming era of ubiquitous AI.