article 1

The Economic and Technical Feasibility of AI Substitution of Harvesting Jobs in the United States During COVID-19 3 lowest average value of all industries, and a significantly low number as well for farming per se. Although the RLI is not given for specific occupations within agriculture, the job of a harvester is assumed to have a very low RLI (below .5), since the main object of a harvester’s work resides in a field, which is not at home. A measure of the “essentiality” of the job of a harvesting worker can be revealed by examining the fraction of six-digit North American Industry Classification System codes relating to harvesting/farming that are categorized as essential. If this fraction is greater than one- half, the occupation of a harvester/farmer in the large can be considered essential in further analysis. This fraction is 19/30, categorizing harvesting/farming as essential (Bennet). The third data point is the risk of infection for harvesters. O*NET, an online job and work focused database, gives a work context variable entitled “exposed to disease or infection,” which is a rating from 1 to 100 of an occupation’s rate of workplace exposure to disease or infection. A 1 rating denotes very little exposure and 100 denotes very frequent exposure. This number for the occupation of harvester is 1 (O*NET). Collectively then these three numbers indicate that the work a harvester accomplishes cannot be feasibly done online, that harvesters can be reasonably categorized as essential, and that harvesters are relatively safe from COVID-19 in the workplace. Since it is a reasonable assumption that harvesters will be safe while at work, a review of the economic feasibility of AI substitution must be a comparison of the cost of AI workers versus human laborers during COVID-19. Such a cost analysis would be enormously complex for a variety of reasons, including the reality that many harvesters are undocumented and due to the fact that they are hired on short-term and periodic contracts. Thus, an in-depth analysis has been deemed beyond the scope of this project. Instead, a simplified comparison can be reached by looking at what actions industries are actually engaging in during COVID-19 and thus used to gauge which is seemingly more cost-efficient. It is reasonable to conclude that some of the percentage decline of harvesting workers in the past decade has been attributed to the rise of harvesting robots in the agricultural sector. These robots have continued to meet some of the demand from the agri-food sector in the United States during the time of COVID-19. According to a market research report, the global crop harvesting robots market is expected to grow at compound annual growth rate of 27%, with the North American region accounting for 39% of this growth(Technavio).Thisgrowthsignalsthat theutilizationof automated harvesting robots increased during the pandemic regardless of the fact that workers could still work without much risk of infection, indicating that robots are a more cost- effective option. Since an increase in robotic substitution has been directly observed during COVID-19, it is reasonable to assume that robotic substitution is economically feasible. However, even though the development of crop harvesting robots has the potential to increase the productivity of many farms, it has significant consequences for the labor force and the unemployment rate in the harvesting sector that must be recognized and resolved through policy changes. It is important to understand that economic feasibility does not necessarily entail technical/robotic feasibility. Indeed, according to a 2013 study, the major occupation group of agriculture, fishing, and forestry is onewithahighprobability of computerization (Frey &Osborne, 37). Over the last couple of years, advancement in AI and machine learning has led to thedevelopments of cropharvesting robots that are capable of harvesting delicate crops and fruits such as strawberries with high precision and quality (Lewis). Despite these theoretical results, in order to examine the total feasibility of robotic replacement, one must also take a shop-floor approach to make sure robots can actually do the job of a harvester. According to O*NET, the main abilities of a harvester are multilimb coordination, static strength, manual dexterity, trunk strength, arm-hand steadiness, information ordering, finger dexterity, stamina, and near vision. To harvest a crop in automatic mode, several complex technological problems need to be solved. These problems include accurate and correct positioning at the collection point, synchronization of actions with other collectors, analysis and selection of the correct algorithms for use with different cultures, recognition of ripe fruits, disposal of damaged or nongerminated elements, accurate grip and cutting of ready-to-harvest crops without damage, checking the correctness of current actions, and minimizing damage at the collection site, etc. In addition, it is necessary to perform all of these tasks at high speeds under constantly changing environmental conditions. Tang et al. suggest that key factors inhibiting harvesting robots are the accuracy of 3D visual perception and ensuring stability in challenging conditions. This can make carrying out the O*NET-defined capability of near vision difficult. Fruit recognition and localization in certain circumstances such as occlusion and illumination can cause errors due to the complexity of the environment. As an example, iceberg lettuce is routinely harvested with a hand knife, which, as Birell et al. point out, presents several major challenges to automation. One, it is