Erik Brynjolfsson, Xiang Hui and Meng Liu
After half a century of hype and false starts, artificial intelligence may finally be starting to transform the US economy. An example is machine translation, as we found when analysing eBay’s deployment in 2014 of an AI-based tool that learned to translate by digesting millions of lines of eBay data and data from the Web.
The aim is to allow eBay sellers and buyers in different countries to more easily connect with one another. The tool detects the location of an eBay user’s Internet Protocol address in, say, a Spanish-speaking country and automatically translates the English title of the eBay offering.
After eBay unveiled its English-Spanish translator for search queries and item titles, exports on eBay from the United States to Latin America increased by more than 17 per cent. Other language pairs produced similarly significant gains. But the machine-learning tool is imperfect — it does not translate the entire description of an eBay offering. Refinements would almost drive even larger increases.
The eBay machine-translation results show how two barriers to productivity improvement can be overcome. That is a reason for optimism, but it also warrants a renewed effort to ensure that the economic gains from artificial intelligence are widely shared.
First, the reason for optimism: Language differences that have hindered trade are steadily going to become irrelevant as machine translation shrinks the world.
Historically, countries separated by languages or by geographic distance engaged in less trade than countries without those hurdles. Our research found that the increased trade enabled by machine translation was the equivalent of cutting the distance between countries by 37 per cent.
In addition to lowering the language barrier to trade, machine learning will overcome the historic barrier known as Polanyi’s Paradox: “We know more than we can tell.”
Bilingual people switch almost seamlessly between languages, translating almost as a reflex, but they’ve been unable to effectively explain to computers how to do it.
Unlike previous information technologies that required humans to explicitly codify tasks for computers, machine learning is designed essentially to teach itself by automatically studying millions of examples, such as pairs of corresponding English and Spanish sentences.
The machine-learning approach has proven remarkably powerful, not only for language translation but also for speech recognition (Apple’s Siri), facial recognition (Facebook’s photo tagging), product recommendations (Amazon) and even cancer diagnoses.
Machine translation is of course just part of a broader technological revolution. Surging productivity and the general rise in incomes it brings would be welcome, of course, but that isn’t sufficient. The same questions being raised about the advance of robotics in the workplace apply to machine learning.
While new jobs would be created, many existing jobs are susceptible to displacement. No economic law guarantees that productivity growth benefits everyone equally. Unless we manage the transition, some people, even a majority, are vulnerable to being left behind even as others reap billions.
Entrepreneurs need to invent new business models, workers need access to new skills, and policymakers need to be urged by voters to invest in research that will redesign approaches to human learning for an era of machine learning.
At the Massachusetts Institute of Technology, the Inclusive Innovation Challenge was inaugurated three years ago to help speed the transition to a high-growth and high-opportunity digital economy.
More than $1 million (Sh100 million) is awarded annually to recognise and reward people and organisations that are working toward the goal of widely shared participation in the digital economy.
Award recipients include Laboratoria, a six-month coding boot camp that trains low-income women in both technical skills and soft skills such as teamwork and collaboration.
After graduation, 80 per cent of Laboratoria’s students find jobs that pay three times what they earned before the program. Another recipient, Apli, uses an AI-enabled chatbot to “interview” students, single mothers, shift workers and other jobseekers, then uses machine learning to match them with employment opportunities. This approach connects them with jobs within 24 hours rather than the 52-day average recruiting cycle.
Artificial intelligence is beginning to transform the economy. Human intelligence is needed to make sure it benefits the many, not just the few. —This is an abridged version that first appeared on Washington Post.