Latest / 1 More Paper / The promise and peril of generative AI for organizational selection and socialization — Chu & Srivastava (2026)
Transcript
- You’re listening to “The promise and peril of generative AI for organizational selection and socialization,” by J.Y. Chu and S.B. Srivastava. Published in 2026. Abstract. Organizational survival and success depend on having members who have a shared understanding about the enterprise’s purpose and strategy. Organizations therefore invest heavily in the selection and socialization of new members. Since the public release of generative artificial intelligence based on large language models (GAI) in 2022, organizational leaders have been grappling with foundational questions about how this new technology will reshape these core activities. Although it is difficult to make precise predictions amid ongoing technological ferment, here we offer informed guesses about the trajectory of GAI-driven change in organizational selection and socialization. To organize our predictions, we draw on three key conceptual distinctions. First, we distinguish between the ability of GAI to select and socialize individuals who are internally committed to organizationally desirable values, versus individuals who only perform these values. Our second distinction pertains to the cross-pressures of fitting in versus standing out within organizations. Third, we distinguish between how GAI is adopted initially, and responses to these configurations by strategic actors, which we refer to as “second order effects”. Based on these distinctions, we array our predictions across three phases, with each new phase a response to the tensions and dissatisfactions of a preceding one. Introduction. Organizational survival and success depend on having members who have a shared understanding about the enterprise’s purpose and strategy, as well as the ways in which its members need to coordinate their activities. For this reason, most organizations make hefty investments in selection—that is, preferentially sourcing and hiring individuals who already hold or are expected to be able to rapidly learn organizationally sanctioned values, norms, and beliefs—and socialization—that is, helping new members acquire the social knowledge and skills necessary to act in ways that conform to prevailing values, beliefs, and norms. Together, selection and socialization ensure that an organization can build and maintain a shared culture. Although the production and maintenance of culture has been an integral part of organizations, how this is achieved has changed significantly over time. For example, the Fordist factory, with its standard mass assembly lines, also entailed the selection of individuals for various roles (through the division of labor) and socialization to these work roles. Yet some of the tactics of that time—for example, “human engineering” that sought to monitor whether workers were living virtuous lives outside the workplace— might be regarded as bizarre and unacceptable in a modern enterprise. While the reasons for such shifts in how organizations select and socialize workers are too legion to enu-merate here, a recurring driving force has been technological change. Consider, for example, how online job postings and two-sided platforms such as LinkedIn have fundamentally transformed selection or how digital onboarding tools and learning management systems have altered socialization. Since the public release of ChatGPT and other instances of generative artificial intelligence based on large language models (GAI) in 2022, organizational leaders have been grappling with foundational questions about how this new technology will reshape their enterprises’ core activities and functions. We focus here on the capacity of GAI to produce fine-grained representations of human culture. It is “generative” because it can be trained on symbols produced by humans (e.g., texts, images) and produce legible texts and images in return. Many technological advances have allowed employers, as well as current or prospective employees, to screen for what they desire and present curated selves. One primary way that GAI is distinct from prior technologies, however, is in how intensely and dynamically actors can use it to screen for and stage-manage cultural identities. Given these capabilities, this technology is likely to alter costly selection or socialization activities currently performed by humans. Although it is difficult to make precise predictions amid an era of ferment, here we offer informed speculation about the trajectory of GAI-specific change in the domains of organizational selection and socialization. Across our predictions, we draw on three key conceptual distinctions. First, we distinguish between the ability of GAI to select and socialize individuals who are internally committed to organizationally desirable values, versus individuals who only perform these values. The kind of things a new organizational member learns during onboarding—basic rules and regulations, safety protocols, desired and unaccep-table behaviors, and collaboration norms—reflect an organization’s espoused values but do not require any internal commitment, only the willingness and ability to perform in normatively compliant ways. Internal commitments to particular values or beliefs may also change over time, but this is hardly guaranteed. As we will discuss, one of our overarching expectations is that GAI will intensify and expedite organizational selection and socialization for performances of organizational culture, but often at the expense of socialization processes that change internal or “authentic” selves. Our second distinction pertains to the cross-pressures of fitting in versus standing out within organizations. Although these pressures wax and wane over an individual’s tenure in an organization, people generally seek to balance these competing imperatives. Organizations, too, strive to strike a balance between cultural homogeneity that can support effective task coordination and cultural diversity that can promote creativity and innovation. A second overarching prediction pertains to how GAI will upset this equilibrium for both individuals and organizations. Finally, and most importantly, we distinguish between how GAI is adopted initially, and responses to these con-figurations by strategic actors, which we call “second order effects”. The difficulty in making precise predictions about GAI’s impact on these processes stems, in part, from these second-order effects. GAI is now accessible to most organizations, and they must attend not only to concerns of efficiency but also strategic differentiation. An organization that adopts GAI in a manner that is nearly identical to its competitors is unlikely to derive much competitive advantage. Prospective or recently hired employees who are the targets of GAI-driven selection and socialization are also likely to adjust their behavior in response to the technology. Beyond gaming, which can occur whenever agents encoun-ter an evaluative system that they believe they can reverse-engineer to achieve better outcomes, GAI may trigger human reactance—overt and subtle acts of resistance to the percep-tion of being under the control of technology. These second-order effects will create or reveal new problems and dissatisfactions, which will further spur the development of new uses and technologies. Hence, we delineate our predictions across three phases, each one a response to the tensions and dissatisfactions highlighted by a preceding phase. Phase 1: intensifying and expediting current processes of selection and socialization Intensifying selection In the first phase, GAI will intensify cultural matching of new members to organizations. In what is already unfold-ing, generative AI will increase the scale by which hiring managers select individuals based on cultural attributes. We are already seeing the increased adoption of so-called helper or co-pilot tools that aggregate data from applicant resumes, social media presences, and other sources to produce recom-mendations about an individual’s cultural fit with the organization. A key motivation for the adoption of these practices are perceptions that current methods for identifying an individual’s cultural match with organizationally central values or styles are limited in scale and exacerbate various forms of bias. For example, managers may rely on interviews or inferences drawn from activities on a resume to assess the fit of new applicants, but these evaluations are often unrelated to job performance and can serve to reproduce various forms of inequality. To be sure, some organizations have responded to criticisms of hiring for cultural fit by selecting candidates that are perceived as offering a cultural “value-add” rather than mere matches with the organization’s culture. Nevertheless, we anticipate that the net effect of GAI will be to produce an employee base that is more, rather than less, homogeneous.1 We further predict that organizations will use GAI to advertise more granular information about their cultures. For instance, Netflix famously posted its culture slide deck (later turned into a memo, reflecting a shift in its internal commu-nications culture) to simultaneously encourage individuals with matching values to apply and discourage applications from those with misaligned values. A likely next step would be for firms to develop bots that interact with humans in ways that their typical (or ideal) worker would. Interactions with a single bot could convey signals about the content of a firm’s culture, whereas interactions with a range of bots in team-based scenarios could send signals about how cultural values and beliefs are distributed across the organization. Such tools would assist prospective job seekers in experiencing what it might be like to work in the organization and better inform their choice of whether to seek to join. Expediting socialization By producing representations of organizational policies, pro-cedures, norms, values, and other symbolic content, GAI will significantly expedite the onboarding process. Employee handbooks will remain an important legal formality. Yet the days of “walking through the handbook” in an onboarding program are already long gone. Prior to their first day, new organizational members will instead be asked to engage with individualized AI bots and AI-generated videos that intro-duce their specific roles. These bots will be able to answer questions tailored to specific roles (e.g., on formal policies but also informal norms about taking leave). We anticipate that these bots will become increasingly integrated as tools for ongoing socialization. Employees will receive tailored, on-demand suggestions on their cor-respondence regarding tone or style that would be in line with prevailing norms. Some organizations will adopt the Kafkaesque extreme of evaluating whether employees are writing or even speaking with colleagues according to the organization’s expected tone or wording. Over time, GAI will be fine-tuned on a wide variety of historical or subunit data to develop bots and other tools that help new entrants understand the various subcultures that exist within the organization. This will help them learn not only how to fit into their own group’s subculture, but also operate effec-tively across subcultural boundaries. Second‐order effects Early attempts at using GAI for selection and socialization are likely to suffer from inaccuracies or biases that will arise in identifying what constitutes a “typical” employee or “nor-mal” forms of behavior. Over time, with access to a broader range of data, we anticipate that GAI can overcome some of these limitations. Legal and public critiques of such models as opaque and unfair will also arise. In response, GAI models will likely evolve to support greater interpretability (i.e., the ability for individuals to understand how the algorithm arrives at its judgments). Even so, GAI tools in this first phase will ultimately only render the surface-level performances of individuals and how they align with the firm’s desired or existing culture. In other words, they will intensify and expedite existing processes of selection and socialization for outward display of organizational culture. As is common with the introduction of any interpretable metric, once it becomes a target, employees will strategically adjust their behavior to game the system. Organizations might hire “culture optimization” consultants to ensure that the text, audio, and video data that are ingested in GAI bots represents elements that are desirable to employees. Meanwhile, prospective employees will learn how to curate the selves they put on display and ensure that data aggregators selectively ingest elements about themselves that match firms’ desired traits. Over time, employees will begin to bristle at the forced conformity in outward behavior, especially when it is at odds with their deeply held values and beliefs. Indeed, sepa-rate from gaming to strategically derive advantages, employees who feel that GAI is threatening their autonomy or other important values may engage in acts of direct or implicit resistance, such as producing data that are designed to interfere with machine learning models (e.g., “poisoning” data) or unionize to force employers from deploying certain GAI tools. The upshot is that GAI in this initial phase will make little progress in fostering the internalization of culture—for example, of the kind that Amway relies on for its success. Socialization that can transform individual identities is typically ritualistic, and human beings still require some degree of embodied interaction for these “deep socialization” efforts to work. Changes in, or crystallization of, central identities still rely on ritual smells and bells, repetition in community, and a sense of shared reality. Indeed, the adoption of GAI might even detract from attempts to socialize individuals to adopt organizational identities, as it will be perceived as an external, disembodied actor that individuals will find difficult to identify with. Employer interests in recruiting not only outward performances of culture, but also the internal commitment of employees, as well as employee desire for bringing more authentic selves to work, will likely prompt the development of the next phase of GAI use in selection and socialization. Phase 2: predicting internalization and balancing conformity with authentic self‐expression In the second stage, we anticipate that organizations will begin using GAI to measure the values and beliefs of individuals’ internal selves, select new hires on these traits, and help workers navigate when and how to express their authentic selves at work. Selection for internal commitment The dissatisfactions produced in the first phase will drive increasing attempts to develop GAI models that can predict deeply held, or authentic, values or beliefs. For instance, various machine learning techniques are already being used to identify the linguistic or social network signature of various desired values and beliefs—often in ways that cannot be easily gamed or masked by strategic action. Over time, these techniques will extend beyond their current application to internal communication data. With the increasing prevalence of online interaction—for example, collaboration on GitHub—and the increasing difficulty of masking one’s digital trace, we predict that GAI models will increasingly ingest these more difficult-to-mask data. If the technology can successfully identify authentic selves, we anticipate that its usage will expand further as organizations sort for individuals who hold cultural traits that are most desirable to them. That is, firms will screen and select individuals at scale, not only those with the capacity to outwardly perform in alignment with prevailing norms and beliefs, but also those whose internal commitments are aligned with the organization’s. These efforts constitute a deeper attempt at control, in response to acts of gaming or reactance faced in the prior phase. Tailored socialization Organizational control is arguably at its strongest when individuals feel that they are performing tasks that emanate from their authentic selves. We anticipate that organizations will leverage the ability for GAI to identify deeply held values and beliefs and tailor socialization practices. For instance, GAI will be able to provide better and more targeted guid-ance on how to fit into the organization while remaining true to oneself. In particular, GAI will suggest outlets for individuals to deviate from prevailing norms or rules that do not ultimately undermine the goals desired by organizational leaders. This might start with simple guidance about the situations under which it is especially important to hew to prevailing norms versus times when it is acceptable and perhaps even preferable for the person to deviate and stand out from the crowd. More sophisticated versions of the tool might go a step further and allow the person to practice enacting acceptable forms of deviance with AI agents that represent different internal or external stakeholders. GAI is designed to convince new members that they are performing their organizational role out of their own volition and authentic selves. For some time, such technology may feel liberating to employees. Yet, as we out-line below, such positive responses will likely prove fleeting. Second‐order effects The advent of more sophisticated GAI will spur a new cycle of prospective job applicants and organizational newcomers attempting to game the system by exhibiting behaviors that match those the model infers as being signals of internalization. This will, in turn, lead organizations to up the ante by developing tools to detect and counteract such gaming. Despite this cat-and-mouse dynamic, we anticipate that the use of GAI in this phase will ultimately lead to still greater homogenization of culture within a given organization. Excepting significant legal change, we anticipate increasing difficulty for individuals to fully curate or control their own digital traces. Even accounting for individuals’ capacity to discover hacks and workarounds that allow them to drive a stronger wedge between their front- and backstages, we anticipate that human adaptation to GAI will tend to lag GAI’s ability to ferret out the internal commitments that ultimately guide action and expression. As a result, we anticipate massive expansion in the scope of organizational control. In what will make Ford’s attempts to control the behaviors of workers outside of the assembly line appear unambitious, this second phase of GAI development will enable organizations to select and socialize for individuals who believe their work reflects authentic selves. This expansion of organizational control over the authentic self will entail new problems. For individual firms, the adoption of customized GAIs that are fine-tuned to their own data will usher in a period of exceptional cultural homogeneity within organizations. This is a problem because tensions between one’s outward performance of culture and internal commitments can serve as latent sources of cultural innovation. Consider if top leaders at Boeing were to leverage GAI to select and socialize individuals that were fully committed to their desired culture of efficiency and cost-cutting. In such a world, the firm might become exceptionally efficient and cruise through multiple years of record profits, but its cost-cutting zeal might further under-mine competing values like product safety or innovation. With new revelations about the firm’s poor safety record, the capacity for new leadership to successfully pilot Boe-ing to prioritize safety or innovation depends, in part, on the presence of existing members who privately champion these values, even if their outward behavior has emphasized cost-cutting and efficiency. Indeed, cultural innovation and adaptation often requires having members who internally hold different values or beliefs from the organization. By collapsing across that divide, GAI enables extreme short-term efficiency at the expense of this ability for longer-term cultural adaptation. We anticipate that there will also be greater between-firm cultural homogeneity. This will be driven by the widespread adoption of off-the-shelf GAI tools that, despite being cali-brated with firm-specific data, will still be trained on shared cultural material. The adoption of such tools will lead firms to converge in more similar ways of selecting and social-izing individuals. Eventually, firms will begin to recognize that between-firm cultural homogeneity makes it difficult to differentiate oneself with any stakeholder, whether a cus-tomer, a supplier, an investor, or an employee, driving some to ultimately resist or even publicly reject the use of GAI in selection and socialization as a point of differentiation. To be sure, in a foreshadowing of the next phase, some organizations may begin pursuing an alternative tack, adver-tising greater leeway in the cultural identities they deem acceptable from employees to inject a modicum of noise into selection and socialization. This strategy may help these firms avoid excessive cultural homogeneity among employees and disarm human resistance to GAI. Indeed, we specu-late that some firms may advertise how little they use GAI in their selection and socialization to obtain a competitive advantage in hiring. For workers, we anticipate that the most talented or self-aware individuals will ultimately recognize GAI as being a form of cultural engineering that does less to liberate the “authentic” self than serve as a straitjacket to control the self. Even if the rise of GAI will help workers feel like they are able to bring their full selves to work, such tools will eventually create problems of burnout and attendant worker resistance to GAI. The small licenses that GAI will afford for displays of authenticity will be insuf-ficient. At a macroeconomic level, if GAI can truly identify internal beliefs or values, and if there is convergence in the kind of values that are desirable in firms due to increasing cultural homogeneity across-firms, it is likely that GAI will drive astronomical demand and compensation for some individuals at the expense of others, perhaps hastening the exit of individuals deemed poor fits. The inequalities in rewards produced by this shift, combined with the difficulty in learning or acquiring internal traits desired by firms, will have high potential to drive widespread frustration and resistance toward GAI tools. Phase 3: selecting for cultural innovation and promoting deep socialization In response to concerns about cultural homogeneity and worker resistance, in the third stage, organizations will use GAI to select for individuals whose presence along-side existing members would create conditions for cultural innovation. Additionally, rather than merely tai-loring socialization processes to offer small licenses for individuals to bring their authentic selves to work, there will also be attempts to adapt GAI to directly socialize individuals’ internal selves—a process that we call “deep” socialization, in contrast to the “shallow” socialization required for external performances of cultural values or scripts. Selection for cultural novelty Recognizing longer-term benefits of cultural adaptability, and hence the need for organizational members who hold differing values and perspectives, organizational leaders will increasingly demand GAI with capacity to select for individuals who, when brought together under the right conditions, are most likely to surface novel and beneficial cultural values. GAI will assess existing organizational members, identify what kinds of individuals who could spur new organizational cultures, and then determine how best to find and attract them to the organization. In essence, organizations may rely on GAI not only to identify those who fit in, but also those who stand out. Deep socialization Whereas in Phase 2, GAI will mostly take a person’s authentic self as given and tailor socialization to persuade them that they are able to bring this authentic self to work, the next step will be to use GAI to engineer authentic selves that are desirable to the organization. As noted above, it is unclear if GAI will have the capacity to transform internal selves if it is perceived as a disembodied algorithmic tool. At minimum, we anticipate that GAI will assist organizational leaders’ efforts to craft desired employees by matching trainees to mentors that have the highest probability of changing their internal commitments, or by matching trainees to task groups with strong peer effects. But, by Phase 3, we also anticipate chang-ing perceptions about GAI from being an algorithmic tool (serving the intentions of human beings) to having the capacity for moral authority (and thus able to guide what human beings should consider desirable). This change will likely occur as religious and political authorities adopt GAI as surrogates to amplify their messaging. Indeed, if primary sources of solidarity and information continue to shift into digital worlds, it is not impossible to imagine the emergence of deeply held political or personal identities that are coordinated by GAI. In Phase 2, workers will eventually come to see through GAI as a source of cultural engineering and control. In Phase 3, this resistance will at first be muted as organizations command greater control over employee selves—both performed and internal. Second‐order effects While there is a world where GAI can perfectly identify and manipulate individual beliefs and values in ways that people find desirable and authentic, we suspect the technology will not be up to this task. Indeed, the alternative world is one in which increasing attempts at deep socialization drive people to feel like they are being exploited, which will eventually lead them to more forcefully resist organizations’ GAI-fueled attempts at social engineering. Similarly, while it is possible that GAI will allow organizational leaders to select a perfect balance of individuals who fit in with organizational culture, along with individuals who stand out enough to enable organizational adaptability and change, we are skeptical that any tool can identify, much less produce, this perfect balance. In fact, the noisiness of current hiring practices for identifying cultural fit may do just as well as GAI in identifying individuals who stand out. After all, if the goal of identifying cultural novelty is merely to have variance in beliefs and values, then an imprecise selection tool could accom-plish this end. Of course, these limitations will not stop organizations from trying to pursue deep socialization or selection on cultural novelty. But these limitations suggest the possibility of serious resistance—for example, collec-tive bargaining agreements that limit the use of GAI, mass strikes, or perhaps even the creation of new organizational forms that, ironically, look more like what organizations used to look like before AI hit the scene. Concluding thoughts We present here one plausible trajectory of how GAI will transform the ways in which organizations select and socialize new members, which we also summarize in Fig. 1. Given the many uncertainties that exist about how the technology will evolve, how governments might regulate it, what legal challenges might arise to the use of employee or job can-didate data for model training, and how employees might themselves resist such efforts individually or collectively, what we offer is, at best, informed speculation rather than a hard prediction. We acknowledge that many other trajecto-ries, including those in which humans and GAI co-exist in relative harmony, are still possible given that the technology is still at a nascent stage of development. We do, however, feel more confident in asserting that, as organizations deploy GAI for selection and socialization, there are likely to be unintended consequences stemming from the tensions of targeting the performance of organizationally desired values versus internal alignment and of selecting and socializing for cultural homogeneity versus differentiation. Organizational leaders will tend to view employee reactions to these tensions—the “second-order” effects described above—as nuisances to be designed out in the next generation of tools. Yet only the wisest of organizational leaders will recognize that these “second-order” effects are also expressions of humanity and reflections of people’s deeply rooted need for both individuation and group affiliation. GAI may ultimately have a constructive role to play in supporting these personal desires in ways that also promote broader organizational goals and economic impera-tives. We anticipate, however, that the road to this destina-tion is likely to be potholed, bumpy, and long. eration and the writing of this piece. Data availability statement A data availability statement is not appli-cable as this a point-of-view/perspective piece that does not draw on any data. Declaration Competing interests The authors hereby declare that they have no competing interests.