In 2011, Krista Jones discovered a lump in her breast. She immediately visited her family doctor for a mammogram. While the lump turned out to be a common, noncancerous growth, the mammogram showed two other “unknown” spots in Jones’s breast. These small spots initiated a nightmarish four year journey in which doctors’ offices and surgery rooms upstaged her “real” life as a mother of three and engineer at a startup incubator.
Despite a constant barrage of ultrasounds, biopsies, and surgeries, Jones’s doctors could not correctly diagnose or eliminate the tumors that kept appearing. Finally, doctors proposed a double mastectomy as the last hope. Jones was terrified that the cancer might spread beyond her breasts to her brain or her lungs.
While she was contemplating undergoing radical surgery, she met a medical physicist who wanted to try something new: artificial intelligence (AI). The new technology diagnosed Jones’s tumors as a rare form of cancer and steered doctors toward a radiation treatment. This treatment regime stopped the growth of the tumors and allowed Jones to enter remission. Ultimately, Jones believes, AI saved her life.
Jones’s story, as she recounts in Quartz, illustrates the incredible potential of AI to improve the well-being of individuals around the world. Yet, no technology is without its drawbacks and AI is no different. Concerns about bias and job loss exist and have some merit. Even so, the amazing benefits that AI offers far outweigh the risks.
AI is such an exciting technology, in part, due to its versatility. It can be trained for a vast array of tasks, making it useful for most industries. Many experts believe that AI will be so widely adopted and influential that they have deemed the advent of AI as the “Fourth Industrial Revolution.” Similar to the switch to mechanization in the First Industrial Revolution, using AI to make sense of modernity’s data-rich world will bring productivity gains and generate new solutions to today’s problems.
AI already pervades the modern economy. Many individuals likely interact with some form of AI on a daily basis even now. It pilots our planes, enhances the safety of our cars, and helps us find our life partners on dating apps. AI powers many of the platforms that form the rapidly expanding sharing economy like Uber and Lyft that have become indispensable to both consumers and producers. These services bring immense value to our lives and allow our modern economy to function.
As AI technology is becoming more powerful and versatile, professionals in a wider array of industries are adapting it to their needs.
One of the most high-profile applications of AI is autonomous driving. Several companies, including Waymo, Uber, and most of the major car manufacturers, are racing to be the first to deploy this futuristic technology. Despite the recent death of a pedestrian hit by an autonomous vehicle, this technology will almost assuredly save lives by reducing the approximately thirty thousand annual car accident fatalities.
AI has similarly been disrupting finance over the last several years. Last year, the Wall Street Journal published a series of reports on the rise of algorithmic trading at hedge funds and other Wall Street institutions. So far, AI has shown promise, beating the annual returns of their human counterparts. Online personal investing services that rely on AI to help their customers with investment advice are also gaining market share and may disrupt the traditional Wall Street players. Personal investment firm Betterment has been growing at a three hundred percent rate and has gained over 250,000 customers since it launched a decade ago.
AI could arguably most significantly impact healthcare, as Jones’s story demonstrates. Hoping to reduce healthcare spending and save lives by boosting early detection of disease, Stanford researchers designed an algorithm that can be used in a smartphone application to detect skin cancer. The algorithm matched the performance of twenty one board-certified dermatologists when asked to diagnose skin lesions in pictures of the most common forms of skin cancer. The researchers hoped that a readily available smartphone app could overcome the cost and access issues that often keep patients from seeking early detection and treatment. Catching melanoma in its earliest stages correlates with a five-year survival rate of around 97 percent, while waiting means a survival rate of around 14 percent. With over five million new cases of skin cancer each year, early detection using an AI-powered smartphone app could mean hundreds of thousands of live saved or extended.
Disease detection is not the only useful application of AI in healthcare. A Scottish company called Exscientia is employing the technology to improve small molecule drug discovery. Chemists estimate there are trillions and trillions of molecules that may be useful for medicine; Exscientia’s AI helps pharmaceutical researchers make sense out of this vast universe of molecules. Scientists can start with an approved drug with known properties and use the technology to quickly identify other, closely-related molecules worth further exploration. This combination of human and machine can lead to safer drugs and monetary savings due to fewer failed clinical trials.
Analysts from consulting firm Frost & Sullivan expect the application of AI in healthcare to grow rapidly in the coming years. By 2021, revenue in the AI health market will be over $6.5 billion, compared with just $600 million in 2014.
Yes, There Is a Down Side
Every technology has its limitations and problem areas, and AI is no different. Unfortunately, objective debate over the risks and benefits have been somewhat hindered by hyperbolic comments made by popular figures like Elon Musk, who called the technology a “fundamental threat” to human civilizations and said it was more dangerous than nuclear weapons.
At this point, AI systems are simply not good enough to merit such fear-mongering. The technology is limited to narrow, repetitive applications for which it has been trained on an enormous amount of data. It is especially bad at adapting to new situations, something humans learn at a young age. The creators of AlphaGo, an AI capable of dominating games of Chinese Go, noted that the system would have been worthless if even a small aspect of the game, like the size of the board, had been changed.
Musk’s prognosticating also falls into the trap of technological determinism. Human researchers control the progress of AI, so better and better systems are not a foregone conclusion. History demonstrates that humans are adept at avoiding apocalypse while cultivating a technology’s benefits, even a technology as destructive as nuclear power.
Ignoring history and AI’s limitations incites fear and panic, and inhibits rational policy-making. Acting on worst-case fears will inevitably obviate the considerable benefits that AI can bring.
Instead, decision-makers ought to focus on mitigating more legitimate and realistic concerns like algorithmic bias. Researchers recently opened the AI Now Institute at New York University to study the social implications of AI. Algorithmic bias, one of the Institute’s focus areas, occurs when AI systems “reflect and amplify existing cultural prejudices and inequalities.” This bias can occur when the data sets used to train the algorithms are flawed or limited. One example of such prejudice is Google’s automated photo tagger labeling pictures of African Americans as gorillas. As AI is used in increasingly more serious and complex decisions like academic admissions, credit scores, and employment, uncovering and correcting such bias will be all the more important.
Another important concern that has received much attention and ought to be considered is selective job loss. Two researchers from the University of Oxford estimated in a 2013 paper that almost half of the jobs in the U.S. are at risk of being automated in the next twenty years, thanks to both AI and robotics. Especially at risk are routine, repetitive professions like those of taxi drivers or paralegals. Yet historically, automation has usually increased the net number of available jobs by increasing demand for human workers in related jobs that cannot be automated. Nevertheless, a net increase in jobs will not ease the pain of job loss for those workers who are displaced. Programs for retraining these individuals ought to be explored. Some scholars have suggested a small tax on AI-assisted industries that could be applied to such retraining. This must also be combined with relocation assistance of some form since post-automation jobs generally do not exist in the same pre-automation job centers.
A Net Benefit
Even considering these legitimate concerns, the U.S. and other major countries ought to embrace AI. The technology offers the potential to enhance human well-being by making us richer, safer, and healthier. It will also likely spur new innovation and create whole new, unforeseen industries. These benefits are too compelling not to pursue. But such pursuit need not, and ought not, be reckless. Both industry-led and government-led policies can be implemented to mitigate AI’s downsides, thus maximizing the benefits.
AI has the potential to change and save lives, as it did for Krista Jones and her family. Not pursuing such benefits would be a grave mistake.
Originally written for an essay contest at The Economist. The essay was long-listed.