Additional pre-registered experiments showed that subjects’ querying behavior was no more optimal – or less similaritydriven – in our active learning task than a traditional semantic search task , and no more optimal or less similarity driven when directly told to query more optimally by querying dissimilar items , suggesting that memory based active learning is at the mercy of extremely stubborn memory constraints, which are difficult to alleviate by task instructions. A final experiment showed that subjects can distinguish between the more and less optimal query sets, suggesting that subjects understand what optimality entails, but that memory constraints make the spontaneous generation of optimal queries from memory difficult. Our results stand in stark contrast with the large body of work that finds optimal search in active learning. The theory that people acquire information optimally has been very successful in explaining human inquiry in several domains. However, most prior studies use fairly simple, artificial stimuli, and do not require subjects to generate queries from memory. We thus suggest that the scope of the optimality hypothesis in explaining human active learning may be more limited than previously thought. Indeed, we suspect that any setting in which subjects must formulate sequences of queries in natural language will probably be constrained by memory processes, particularly the similarity-driven associative memory search. Although associative memory processes curtail optimal active learning, that does not mean that people’s memory processes are inherently flawed. Rather, memory serves multiple cognitive functions and the associative biases documented in this paper may reflect optimal trade offs between diverging task demands. Indeed, many researchers have argued that association or similarity-driven memory search is part of an optimal system for semantic memory retrieval . Related work has shown that associative memory processes implicated in judgment and decision biases are adaptive in that they often lead to accurate inference and generalization with minimal cognitive cost . Regulating these processes in active learning tasks may be too effortful,blueberries in pots and people may be optimally trading off performance with the cognitive cost required to succeed in our task .

This theory predicts that even though we were unable to reduce semantic congruence and increase optimal search through coaching, performance may improve with higher incentives or practice. Testing these predictions is an important topic for future work. Other future directions include the refinement of our memory and learning models. For example, subjects in our study learned about novel target properties. Yet they came into the experiments with idiosyncratic knowledge about food items or animals. Thus, it is likely they held different prior belief about the novel target properties. Since prior belief is not the focus of this paper, we assumed all subjects held the same prior belief in the experiments. In future work, the shape of prior belief can be set as free parameters and the same framework can be used to derive the prior representation of target properties in a given domain. Individual differences in this regard can be revealed. The Bayesian learning model also assumes that subjects maintain a distribution of belief over multiple hypotheses . However, other research suggests that in a closely related – and not even as complex – active category learning setting, subjects maintain a single hypothesis at a time . Previous research also reveals other simple heuristics, such as the split-half heuristic and the likelihood difference heuristic , in active learning tasks. It is possible that such heuristics play a role in the query search in our active learning tasks and, therefore, can be considered in the modeling of algorithmic processes in future research. Our work contributes to the emerging body of research that offers researchers a naturalistic search domain to study active learning. Additionally, our computational models integrate insights from several fields, and are able to jointly describe both algorithmic memory search processes as well as the optimality or suboptimality of these search processes for active learning. In this way, our paper presents a powerful new research paradigm for naturalistic active learning. There has been an increasing interest in porting computational cognitive models beyond abstract lab stimuli, to attempt to describe everyday cognition. This has been driven by the availability of new machine learning models that offer quantitative representations for natural entities , as well as the growing demand from policy makers and practitioners for theory-driven behavioral and cognitive insights.

Our research is part of this trend, and we look forward to future work that applies established algorithmic and rational theories of cognition to rich stimuli sets to better understand human cognition and behavior in the wild. The slowdown in unauthorized Mexico–U.S. migration has set off a race in U.S. agriculture between rising imports, more machines, and foreign guest workers. Trade policy, including North American Free Trade Agreement re-negotiations, and immigration policy, including more enforcement and new or revised guest worker programs, will determine the winner. Fewer and larger farms that depend on hired workers produce most U.S. fruits, vegetables, and horticultural crops such as nursery plants. The number of farms in the United States is stable at about 2 million, but the largest 10% of all farms account for three fourths of U.S. farm sales. In fresh vegetables, the largest 10 producers account for more than half of the lettuce, broccoli and carrots produced. Americans do not dream of growing up to be farm workers. About 70% of the hired workers on U.S. crop farms were born in Mexico, and 70% of these Mexican born workers are unauthorized, so half of crop workers are working illegally. California has a higher share of unauthorized workers because more of its workers were born in Mexico, 90% versus less than 70% in other states. Crop workers are aging and settling. Most have families that include children born in the United States, and few are migrants who follow the crop harvests from south to north. Unauthorized newcomers, who are primarily Mexican-born workers in the United States less than a year, have been the flexible fresh blood of the farm workforce, willing to move to fill vacant jobs. Their share of crop workers peaked at a quarter in 2000, but today such newcomers represent just 1% of crop workers. Farmers are responding to the end of large-scale Mexico–United States migration and California’s rising minimum wage with four strategies: satisfy current workers to retain them, stretch them with mechanical aids that increase their productivity, substitute machines for workers, and supplement current workers with H-2A guest workers. Seasonal farm work is generally a decade-long job rather than a lifetime career. Training first-level supervisors to reduce favoritism and harassment, paying bonuses to workers who stay through the season, and offering other benefits helps to satisfy current workers and keep them in farm work longer.

Stretching farm workers involves management changes and mechanical aids that increase productivity. Most fresh fruits and vegetables are over 90% water, and workers spend much of their time carrying harvested produce down ladders to bins or to the end of rows to receive credit for their work. Dwarf trees mean fewer ladders and faster picking, reducing the need to fill 50- to 60-pound bags of apples and oranges from tall ladders. Slow-moving conveyor belts that travel ahead of workers in the fields reduce the need to carry harvested produce, increasing worker productivity and making jobs more attractive to older workers and women. Substitution is replacing workers with machines. There are machines available to handle most tasks done by farm workers, but human hands are gentler than mechanical fingers on fragile fresh fruits and vegetables, so that a higher share of hand-harvested produce can be sent to consumers. Machines have other disadvantages as well. They are fixed costs, meaning that farmers must pay for, say, a $200,000 harvesting machine whether there are apples to pick or not, while workers are variable costs who are not paid if storms or disease destroy the apple crop. Nonetheless, rising minimum wages, fewer flexible newcomers,square plant pots and advances in mechanization have encouraged many farmers to experiment with machines, prompting manufacturers to develop and market labor-saving machines that are doing more planting and pruning and are improving rapidly to harvest blueberries, peaches and leaf lettuces. The fourth option is to recruit guest workers under the federal H-2A program, which admits an unlimited number of foreign farm workers to fill seasonal jobs. Receiving permission to hire H-2A guest workers requires farmers to try and fail to recruit U.S.-born workers, provide free housing, and pay an Adverse Effect Wage Rate , which is $13.18 an hour in California in 2018. The number of U.S. farm jobs certified to be filled by H-2A workers tripled over the past decade to 200,000 in fiscal year 2017 and may surpass the peak number of Braceros by 2025 . The number of jobs certified to be filled by H-2A workers in California tripled in 5 years, from 3,000 in 2012 to 15,000 in 2017, and appears poised to continue increasing. Half of the fresh fruit and a quarter of the fresh vegetables available to Americans are imported, and imports of everything from avocados to raspberries are rising. Mexico is the major source of fresh fruit and vegetable imports, supplying half of the imported fresh fruit and three-fourths of the imported fresh vegetables. Many of the fruits and vegetables imported from Mexico are produced on farms that involve partnerships between U.S. and Mexican growers and shippers, with U.S. partners providing capital and technology and marketing Mexican-grown produce. Satisfying and stretching current workers are shorter term strategies to increase the productivity of an aging farm workforce. Substituting machines, hiring guest workers, and increasing imports are longer term strategies to supply fresh fruits and vegetables to Americans.

Policy will help to determine the winner of the race in the fields between machines, migrants and imports. Technologies that could replace farm workers are improving rapidly and decreasing in cost, potentially putting agriculture on the cusp of another wave of labor-saving mechanization. Farmers have long sought new or revised guest worker programs that eliminate requirements to try to recruit U.S.-born workers, provide housing, and pay the super minimum AEWR wage. The House Judiciary Committee approved a bill in November 2017 that includes these farmer wishes, but it has drawn opposition from advocates for removing worker protections and from some farmers for capping the number of guest worker visas at 450,000 a year. If the new H-2C program included in the Agricultural Guest worker Act is enacted, the influx of farm guest workers would likely accelerate, which may reduce support for the engineers and scientists developing machines to replace farm workers. The United States has an overall agricultural trade surplus, but a deficit in agricultural trade with Mexico reflecting ever-more Mexican avocado, tomato and berry imports. The Trump Administration aims to reduce the trade deficit with Mexico in NAFTA renegotiations, perhaps by imposing tariffs or other restrictions on Mexican imports. This could slow the integration of the North American produce industry, which has evolved to provide year-round supplies of fresh fruits and vegetables to Americans. Agriculture has been at farm labor crossroads many times, asking who will pick the crops after the exclusion of the Chinese in the 1880s and the termination of the Bracero program in the 1960s. Today’s race in the fields will determine whether Americans will consume more imported produce or whether fruits and vegetables will continue to be grown in the United States and picked by machines or guest workers. Lowbush “wild” blueberries are considered a nutrient-rich healthy food, due in large part to their exceptional phenolic content and antioxidant activity. Lowbush blueberries are particularly rich in anthocyanins and the anthocyanin profile is complex compared with other fruits. They contain five of the six anthocyanidins commonly found in nature , which can have three different sugar moieties attached as well as acyl groups such as acetyl-, malonyl-, or coumaryl- also attached to the sugar moieties. Blueberries are also rich in proanthocyanidins, chlorogenic acid, and flavonols. Diets rich in blueberries or their polyphenolic-rich extracts have been associated with lower cardiovascular risk, weight gain and metabolic syndrome, and neurological diseases . In addition, studies involving blueberries have identified polyphenolic-derived phenolic acids that improve cell differentiation and proliferation of osteoblasts in vitro and promote bone growth and limit bone loss in rodents. These health-promoting effects are due to a myriad of mechanisms associated with blueberry polyphenolics, including prevention of oxidative stress and inflammation, and vaso- and lipid modulation.