Intelligence is man’s most potent weapon. If not for it, we would have remained just another species. The unique kind of intelligence we possess gives us dominion over everything else. Granted, when Mother Nature shrugs or sighs with an earthquake or a hurricane, man’s vulnerability is starkly exposed, but humans negotiate the reality of existence better than other species by learning from experience, solving complex problems, and uncovering patterns that lie deep in nature’s womb.
From the discovery of fire to the invention of the microprocessor, the spark of human ingenuity has always served to make life safe, productive and convenient. Equally impressive has been the remarkable rate at which technology has advanced. More things changed in the last 10 years than they did in the 50 years before that. Rapid acceleration in technological progress is hurrying us towards a new frontier where human intelligence finally gives birth to an artificial one that is just as formidable. The way we harness it will determine our future well-being.
Some experts believe that like all innovations of the past, Artificial Intelligence (AI) will make life better for us. There are others who believe that AI will cause massive job losses and increase income inequality. At the extreme end of sceptical opinion are some futurists who believe that AI will relegate humans to a perpetual underclass. As always, reality presents a mixed picture. To understand how AI affects our future, we should first understand what AI is and why we are embracing and fearing it in equal measure.
What is Artificial Intelligence
AI is a suitcase word — it is stuffed with many meanings. Hollywood’s take on AI, with its preference for robot heroes and gravity-defying action, muddled its meaning in the minds of the general public. To simplify, AI refers to a system that is capable of thinking and acting like humans. Whether it is a robot or a software programme does not matter. Machines are good at taking instructions, but they are not good at thinking for themselves — like we can. For true AI to be possible, machines should be able to learn and think autonomously by finding patterns, which calls for a tonne of data and huge computing power. Even then, a machine brain would be a poor match for the human brain.
Any promising new technology attracts a lot of hype. The same was true of AI too, which had a spirited start in the late 1950s. In the 60s, leading AI researchers declared that machines will be as smart as human beings in a decade. It turned out that it was not so easy and by the mid-70s, the mood turned pessimistic as some AI projects failed to deliver on their advertised promise. Soon, funding was cut and interest in the area waned leading to a period that’s referred to as the ‘AI winter’. By 1993, things started looking up again. When IBM’s chess programme, Deep Blue, beat Gary Kasparov in 1997, it caught headlines and the AI spring began.
In 1965, Gordon Moore, co-founder of Intel, made a prediction that computer chips — essentially the brains of computers — double in power every two years. Called the Moore’s law, this prediction turned out to be incredibly accurate. This improvement is made possible by advances in chip design that allow more transistors to be crammed into a square inch of an integrated circuit. Today, the availability of large-scale computing power is driving a new wave of AI research, as are advances in machine learning techniques and the availability of immense data troves. Think of all our online activity and the data streamed by sensors that are everywhere. AI systems have started getting better. They are also moving out of research labs and into our daily lives with greater speed.
Narrow AI vs Full AI
Whether we know it or not, AI is a regular feature of our daily lives. When we book a cab via Uber, the AI-enabled algorithm determines what price to show you based on the demand for rides in that area. Facebook’s algorithm will rank and sort your feed to surface stuff it thinks you may find interesting. Video suggestions on YouTube and shopping recommendations on Amazon depend on intelligent algorithms. These are examples of what is called as ‘Narrow AI’. These algorithms do a specific task better than humans, but they do not have general intelligence in the sense that we do. Most of us embrace Narrow AI without hesitation because its presence improves the experience of using a product or service.
Some ambitious researchers are working on building machines with the general intelligence of the kind that humans possess. They are trying to develop artificial general intelligence (AGI) – also called ‘Full AI’ – which would allow machines to perform all the intellectual tasks that humans can, like abstract thinking, natural language communication and continuous learning. One of the ways researchers are seeking to achieve Full AI is by mimicking the structure of our brains. Deep learning – a subset of machine learning – is an example of it. Still, for all the breakthroughs, we are nowhere close to replicating the intelligence of a 5-year-old. Given the exponential increase in computing power though, experts believe that we are not too far from that day.
As things stand currently, computers are good at doing things that we find difficult, but they are miserable at things that are a breeze for us. Computers can multiply two six-digit numbers in a fraction of a second. But throw a ball at a robot to catch and it will stumble and freeze. Since the rules of chess and the combination of moves can be programmed, it is easy to create an algorithm that can beat a grandmaster. That’s Narrow AI. On the other hand, it is exceedingly difficult to create a programme that can engage in friendly small talk or detect sarcasm. There is good reason to believe that Full AI has a long uphill road to tread. Since technological advance represents just one factor in the development of AI, more computing power does not always mean better AI.
AI and Employment
AI prompts an equal amount of optimism and fear currently. Some believe that it will have a benign effect by complementing human ability and making us more productive, while others feel that it will make a large mass of humanity dispensable. One issue that dominates mainstream discussion of AI is its impact on jobs. A popular 2013 study by Oxford economists, Carl Benedikt Frey and Michael Osborne, estimated that “47 per cent of total US employment is in the high risk category” because it is “potentially automatable in perhaps a decade or two”. A few books and academic papers in recent years reinforced the same message.
The fear of job losses is both overblown and real -overblown in the short-term and real in the long-term. AI is currently not at a level where it can replace human beings. At best, it complements human labour. In some cases, it certainly reduces the number of people needed to do a job — warehouses for example. This is qualitatively not very different from ATMs reducing the need for bank tellers or online booking engines eliminating the need for reservation clerks. We have adjusted to such shifts because the self-same advances created new jobs elsewhere. Why should we think things will be different this time?
Economists argue this time is different because, unlike in the past, the rate of change is quite rapid. The consensus view holds that AI will create new jobs over the long-term, but there will be a carnage in the short-term as thousands of jobs are lost and people struggle to adjust. The transition could be very painful for low-skilled people at the bottom of the income hierarchy, whose jobs are the most vulnerable to automation. Given that many white-collar workers also perform repetitive tasks, an AI algorithm could soon render many of them redundant. It could also make the problem of middle-class wage stagnation worse. The fact that all this is happening at a time when the economy has entered a period of slow growth further compounds the problem.
Some jobs are more vulnerable to AI than others. Among the 702 occupations included in the Frey and Osborne study, telemarketers and tax preparers face the highest risk of their jobs being computerised. Recreational therapists, surgeons and choreographers face the lowest risk. The more mechanical and repetitive a job is, the higher the chances of it being automated. If you have a job that mostly involves repetitive work, you should start acquiring new skills to fend off competition from the robots. Many jobs in the IT industry, like network administrators and support professionals, are also vulnerable to automation.
As for the long-term, it is true that a majority of the current jobs will become redundant. That’s how technological progress works. Driverless vehicles could eliminate driving as a profession, but it is reasonable to expect that new jobs will be created in maintaining the fleet of self-driving vehicles. If history is a reliable guide, we should certainly expect AI to create new jobs. A decade ago, few among us expected ‘SEO Consultant’ to be a full-time job, let alone a job in much demand. It is easy to say which jobs AI will automate away, but almost impossible to predict which ones will be created in their wake. It is wise not to get worked up by long-term predictions about disruptive technologies, because they may be wildly inaccurate and also because, as British economist John Maynard Keynes wrote ‘in the long run we are all dead’.
We would do well to remember that, along with intelligence, adaptability is one of humanity’s special traits. Until the mid-19th century, agriculture employed over 50% of the population in advanced economies like the US. Today, thanks to advanced farm machinery and agricultural science, less than 2% works in agriculture, but farm output continues to increase to meet the demands of a growing population. Similarly, AI boosts the efficiency of companies. If companies pass this benefit on to consumers, it will translate into lower prices and therefore improved living standards. It is in these areas that smart government policy is needed to ensure that AI does not increase income inequality.
To be sure, it is not possible to regulate technological advances, but it is possible to ensure that its outcomes are distributed fairly. AI has great potential to make the world a better place, provided we direct its power towards the right goals. Human intelligence created artificial intelligence. It is not only defeatist but also vain to think that our well-being or survival is threatened by it.
Impact of AI on jobs
If a job involves repetitive work, either manual or cognitive, it is at a high risk of being automated. Creative and social skills will continue to be in demand.
Jobs facing low-risk from automation
Doctors & healthcare workers
Creative artists & professionals
Engineers & scientists
Teachers & education administrators
Jobs facing high-risk from automation
Low-skilled manual workers
Source: The Future of Employment: How susceptible are jobs to computerisation?; Carl Benedikt Frey and Michael Osborne
Three common encounters with AI
Facebook News Feed
The News Feed that we see when we log on to Facebook is our personalised front page. The algorithm that determines what to show at the top of the feed and in what order is a classic example of Narrow AI. This algorithm ranks stories based on the time of the post, our engagement with the person who posted it, the overall performance of the post and a thousand other criteria
Uber’s Dynamic Pricing
If you hate Uber’s surge pricing at peak hours, go argue with the algorithm. When you book a ride, Uber’s shrewd algorithm calculates the fare based on the real-time demand for rides in that area. Surge pricing ensures that you get a cab when you need it if you are willing to pay more. It is not fair, but the algorithms are not known for such moral scruples.
Amazon knows how to make the most of our impulsive shopping tendencies. The shopping recommendations on Amazon are the result of analysing real-time data on our past purchases, browsing history, wish list items, and the buying behaviour of people who bought products similar to the one we bought or are planning to buy
(The author is co-founder of Lexys Labs, a Hyderabad-based IT services company)