This emerging literature has been limited by the challenge of incorporating market effects.

As the scope of machine learning broadens, we observe a recurring theme of algorithmic monoculture: the same systems, or systems that share components (e.g.

Diversity in training data, architecture, and providers is assumed to mitigate homogeneity in LLMs. However, we lack empirical evidence on whether different LLMs differ meaningfully. We conduct a large-scale empirical evaluation on over 350 LLMs overall, using two popular leaderboards and a resume-screening task. We find substantial correlation in model errors -- on one leaderboard dataset, models agree 60% of the time when both models err. We identify factors driving model correlation, including shared architectures and providers. Crucially, however, larger and more accurate models have highly correlated errors, even with distinct architectures and providers. Finally, we show the effects of correlation in two downstream tasks: LLM-as-judge evaluation and hiring -- the latter reflecting theoretical predictions regarding algorithmic monoculture.

Dr. Steven Pinker, a Harvard psychologist and prolific author, has often been described as a cheerleader for science, reason, and humanism. He is often maligned by his critics as a defender of the status quo. Much of his research focuses on slow and steady incremental improvements that have defined rapid human development, both in the United States and globally, over the past century. His 2018 book, "Enlightenment Now" was famously cited by Bill Gates as "his new favorite book," and became a focal point for global policymakers. He is a fierce defender of liberalism, democracy, and market economies, and believes a variety of forces are conspiring against them: populism of both the right and left, religious fundamentalism, and political correctness, among others. He also has emerged as a champion of reasoned, civil debate on college campuses, pushing back against cancel culture, and what he views as a'political monoculture' in academia.

Over the past decade, AI research has focused heavily on building ever-larger deep learning models. This approach has simultaneously unlocked incredible achievements in science and technology, and hindered AI from overcoming long-standing limitations with respect to explainability, ethical harms, and environmental efficiency. Drawing on qualitative interviews and computational analyses, our three-part history of AI research traces the creation of this "epistemic monoculture" back to a radical reconceptualization of scientific progress that began in the late 1980s. In the first era of AI research (1950s-late 1980s), researchers and patrons approached AI as a "basic" science that would advance through autonomous exploration and organic assessments of progress (e.g., peer-review, theoretical consensus). The failure of this approach led to a retrenchment of funding in the 1980s. Amid this "AI Winter," an intervention by the U.S. government reoriented the field towards measurable progress on tasks of military and commercial interest. A new evaluation system called "benchmarking" provided an objective way to quantify progress on tasks by focusing exclusively on increasing predictive accuracy on example datasets. Distilling science down to verifiable metrics clarified the roles of scientists, allowed the field to rapidly integrate talent, and provided clear signals of significance and progress. But history has also revealed a tradeoff to this streamlined approach to science: the consolidation around external interests and inherent conservatism of benchmarking has disincentivized exploration beyond scaling monoculture. In the discussion, we explain how AI's monoculture offers a compelling challenge to the belief that basic, exploration-driven research is needed for scientific progress. Implications for the spread of AI monoculture to other sciences in the era of generative AI are also discussed.

As the scope of machine learning broadens, we observe a recurring theme of algorithmic monoculture: the same systems, or systems that share components (e.g. training data), are deployed by multiple decision-makers. While sharing offers clear advantages (e.g. amortizing costs), does it bear risks? We introduce and formalize one such risk, outcome homogenization: the extent to which particular individuals or groups experience negative outcomes from all decision-makers. If the same individuals or groups exclusively experience undesirable outcomes, this may institutionalize systemic exclusion and reinscribe social hierarchy. To relate algorithmic monoculture and outcome homogenization, we propose the component-sharing hypothesis: if decision-makers share components like training data or specific models, then they will produce more homogeneous outcomes. We test this hypothesis on algorithmic fairness benchmarks, demonstrating that sharing training data reliably exacerbates homogenization, with individual-level effects generally exceeding group-level effects. Further, given the dominant paradigm in AI of foundation models, i.e. models that can be adapted for myriad downstream tasks, we test whether model sharing homogenizes outcomes across tasks. We observe mixed results: we find that for both vision and language settings, the specific methods for adapting a foundation model significantly influence the degree of outcome homogenization. We conclude with philosophical analyses of and societal challenges for outcome homogenization, with an eye towards implications for deployed machine learning systems.

This story originally appeared on Undark and is part of the Climate Desk collaboration. Across Midwestern farms, if Girish Chowdhary has his way, farmers will someday release beagle-sized robots into their fields like a pack of hounds flushing pheasant. The robots, he says, will scurry in the cool shade beneath a wide diversity of plants, pulling weeds, planting cover crops, diagnosing plant infections, and gathering data to help farmers optimize their farms. Chowdhary, a researcher at the University of Illinois, works surrounded by corn, one of the most productive monocultures in the world. In the United States, the corn industry was valued at $82.6 billion in 2021, but it--like almost every other segment of the agricultural economy--faces daunting problems, including changing weather patterns, environmental degradation, severe labor shortages, and the rising cost of key inputs: herbicides, pesticides, and seed.

The Earth is losing forests at an alarming rate. The United Nations Food and Agriculture Organization estimates that 420 million hectares of forest have been lost to agricultural use (largely cattle ranching, soya bean and oil palm farming) since 1990. Between 2015 and 2020, some 10 million hectares were destroyed each year. The Amazon rainforest, for example, lost an area the size of Yellowstone (3,769 square miles) in 2019, and saw deforestation rates spike 30 percent to their highest point in a decade. What's more, Climate change-induced wildfires, as we've seen recently in Australia and in California, have been especially destructive.

The banana has been the subject of Andy Warhol's cover art for the Velvet Underground's debut album, can arguably be the most devastating item in the Mario Kart video game franchise and is one of the world's most consumed fruits. And humanity's love of bananas may still be on the rise, according to data from the Food and Agriculture Organization of the United Nations. On average, says Chris Barrett, a professor of agriculture at Cornell University, citing that U.N. data, every person on earth chows down on 130 bananas a year, at a rate of nearly three a week. But the banana as we know it may also be on the verge of extinction. The situation led Colombia--where the economy relies heavily on the crop, as it does in several other countries including Ecuador, Costa Rica and Guatemala--to declare a national state of emergency in August.