Recent research co-authored by MIT economists demonstrates how enhanced translation technology can significantly accelerate global online commerce, providing compelling evidence of machine learning economic impact in real-world scenarios.
The study reveals that following eBay's implementation of an advanced automatic translation system in 2014, trade volume surged by 10.9% between country pairs utilizing the new AI translation international trade technology.
"That's a remarkable figure. To observe such a substantial effect in such a brief timeframe truly speaks volumes about the transformative power of this technology," notes Erik Brynjolfsson, MIT economist and co-author of the research paper detailing these findings.
To contextualize these results, Brynjolfsson explains that physical distance represents a significant barrier to global commerce. The 10.9% increase generated by eBay's enhanced translation software effectively boosts trade by an amount equivalent to "making the world 26 percent smaller, in terms of its impact on the goods we studied," he states.
The paper, titled "Does Machine Translation Affect International Trade? Evidence from a Large Digital Platform," appears in the December issue of Management Science. The authors include Brynjolfsson, who serves as the Schussel Family Professor of Management Science at the MIT Sloan School of Management, along with Xiang Hui and Meng Liu, both assistant professors at Washington University in St. Louis's Olin Business School.
Research Methodology
For their investigation, the researchers analyzed the aftermath of eBay's 2014 introduction of its proprietary eBay Machine Translation (eMT) system—an artificial intelligence commerce growth platform that, by multiple objective metrics, substantially improved translation quality across eBay's marketplace. The system initially focused on English-Spanish translations to facilitate trade between the United States and Latin America.
Previously, eBay had utilized Bing Translator to render product titles. According to the Human Acceptance Rate (HAR)—where three experts evaluate translation quality—the eMT system increased acceptable Spanish-language item titles on eBay from 82% to 90%.
Using eBay's administrative data, the researchers then examined trade volume changes following the eMT system's implementation. With other factors held constant, the study demonstrated that the new translation system significantly impacted sales, with trade increasing by 1.06% for each additional word in product titles.
This represents a substantial change for a commerce platform where items typically feature extensive, descriptive titles such as "Diamond-Cut Stackable Thin Wedding Ring New .925 Sterling Silver Band Sizes 4-12," or "Alpine Swiss Keira Women's Trench Coast Double Breasted Wool Jacket Belted." In these instances, improved translation helps potential buyers precisely understand what they might purchase.
Given the study's precise methodology, Brynjolfsson describes it as "a truly fortunate natural experiment, with a clear before-and-after distinction that sharply illustrates what happens with and without machine translation."
The study's structure, he adds, has enabled researchers to confidently assert that eBay's new program—not external factors—directly caused the trade volume increase among affected countries.
"In economics, establishing causation rather than mere correlation can be challenging," Brynjolfsson explains. "However, in this case, I'm completely comfortable stating that improvements in machine translation directly caused the increase in international trade."
Broader Implications: The Productivity Puzzle
The research originates from ongoing questions about new technology and economic productivity. While numerous artificial intelligence applications have emerged over the past two decades, the economic impact of AI—including machine translation global business systems—hasn't been readily apparent in economic statistics.
"There's certainly been remarkable progress in core technologies, including natural language processing and translation," Brynjolfsson observes. "However, what's been missing is clear evidence of economic or business impact. That's been somewhat puzzling."
When seeking measurable economic impacts from various AI forms, Brynjolfsson, Hui, and Liu determined machine translation "made sense, because it represents a relatively straightforward implementation," Brynjolfsson adds. That is, improved translations could influence economic activity on eBay without requiring additional technological changes.
In this context, the findings align with a broader hypothesis Brynjolfsson has developed in recent years—that AI technology adoption produces a "J-curve" in productivity. As Brynjolfsson has previously written, comprehensive AI technologies "require significant complementary investments, including business process redesign, co-invention of new products and business models, and investments in human capital" to generate substantial economic impact.
Consequently, when AI technologies are introduced, productivity may initially appear to decline, and when complementary technologies are developed, productivity may surge—creating the characteristic "J-curve" pattern.
While Brynjolfsson believes this study's results are unequivocal, he cautions against overgeneralizing about machine learning's and other AI forms' economic impacts based solely on these findings. Each case differs, and AI won't always produce such notable changes independently.
"This represented a scenario where minimal additional changes were needed for the technology to benefit the company," Brynjolfsson concludes. "However, in many other cases, more complex, complementary changes are necessary. That's why, in most machine learning applications, delivering benefits takes considerably longer."
