Journal of Behavioral and Applied Management

2024, Vol. 24, No. 2, 112–122

https://jbam.scholasticahq.com/

Maximizing Generative AI Benefits

with Task Creativity and Human Validation
Charu Sinha
Chapman University
Veselina Vracheva
Chapman University
Cristina Nistor
Chapman University

Much of the existing literature on generative AI applications is conflicting, with findings suggest-
ing that investing in AI will lead to better organizational outcomes but also pointing out that incor-
porating AI may be a wasteful even counterproductive initiative. We develop a conceptual frame-
work to characterize generative AI benefits based on the types of tasks that generative AI may be
used for in management. Our work suggests that task creativity plays a key role in successful gen-
erative AI outcomes, but human validation - the extent to which a human engages in a supervisory
role - is required to reap the benefits. Our conceptual framework is focused on white collar jobs and
suggests that the management of generative AI is a strategic choice with important managerial im-
plications.

Generative artificial intelligence (GAI) is a new form of include “human work [that] can be decomposed into at-
artificial intelligence (AI) that applies machine learning omistic tasks that differ in their complexity” (p. 2). Thus,
algorithms to existing data in a training dataset to gener- there is still considerable uncertainty over the effect of AI
ate new data, and has the potential to disrupt current busi- in the business world and how managers should allocate
ness models and human labor (McKinsey, 2024a). The tasks between AI tools and human workers.
increase in computing power and development of new We propose a conceptual framework that examines two
predictive models have led to an impressive increase in attributes, task creativity and human validation, to antici-
the number of GAI companies like ChatGPT (Dilmegani, pate the level of benefits achieved from incorporating
2024; Zia, 2023). Popular press articles suggest either GAI in management tasks. Our conceptual framework
wild optimism for the potential of AI in management (Zia, focuses on GAI benefits such as lowering labor costs,
2023) or unqualified optimism that humans cannot be faster time to completion for tasks, or improved quality in
replaced by AI bots (Barnett, 2024). As such, the role of the execution of the task. Generative AI is considered a
GAI in management and the business world is still uncer- potential disruptor for white collar jobs (Eisfeldt et al.,
tain: will the new technology lead to more productivity, 2024; Noy & Zhang, 2023). Thus, we focus on this part of
higher overall returns, and change the business world or the economy in our theorizing. Moreover, we characterize
will it be another automation tool that can be used to mar- situations where tasks performed by GAI can substitute
ginally improve current business outcomes? for human labor rather than instances where GAI may be
Recent research has focused on AI use as a tool to as- used to supplement it.
sist humans in their tasks. For example, Brynjolfsson et
al. (2023) find that an AI bot which helps call center as- Generative AI in Business Applications
sistants identify answers for customer conversations can AI has been making its impact felt in various functional
foster higher productivity, increased customer satisfaction areas in business over the last decade, yet the excitement
for the call, and faster on-the-job learning for human around the recent introduction of GAI parallels few other
workers. However, experienced workers benefit the least moments in the history of business. NVIDIA’s CEO Jen-
from AI tool implementation, which suggests that the sen Huang has described GAI as the “iPhone moment” for
impact of AI may be overall limited. Acemoglu (2024) AI (Caufield, 2023). JP Morgan Chase CEO Jamie Dimon
estimates that AI will contribute to the US economy in a has predicted that “AI might require future workers to
“nontrivial but modest—no more than a 0.66% increase in work only 3.5 days”, while Mckinsey Global Institute has
total factor productivity (TFP) over 10 years” (p. 1), while predicted that GAI and other tools will automate 30% of
listing several ways that this small estimate is still an opti- work by 2030 (Ellingrud et al., 2023) and constitute the
mistic measure of the impact of AI on the business world. new productivity frontier for businesses (Chui et al.,
Korinek and Suh (2024) arrive at more optimistic esti- 2023). Current estimates put AI adoption by companies
mates but rely on the assumption that business processes much lower: McElharan et al. (2024) use data from the

112
Copyright © Institute of Behavioral and Applied Management. All rights reserved.
Article citation: Sinha, C., Vracheva, V., & Nistor, C. (2024). Maximizing generative AI benefits with task creativity and human validation. Journal of
Behavioral and Applied Management, 24(2), 112–122.
113 Charu Sinha, Veselina Vracheva, and Cristina Nistor

2018 Annual Business Survey of firms across the United There is thus a need to characterize tasks where genera-
States to estimate that fewer than 6% of companies use AI tive AI may yield the highest benefits in terms of cost
-related technologies in any capacity. reduction and/or increased revenue, thereby improving
As companies have adopted GAI tools in various facets the bottom line. As managers struggle to decide to what
of business, the results have been mixed even for tasks degree to adopt and deploy generative AI in their organi-
that are similar such as customer-facing bots implemented zations, our conceptual framework is a call for more em-
in different industries and companies. Klarna, a financial pirical research to measure the boundaries of the proposed
technology company, is reaping significant benefits by theory with a focus on task creativity and the degree of
adopting an AI assistant capable of doing work equivalent human validation. We focus on jobs where there is a need
to 700 full-time customer service agents (Klarna, 2024). for human validation after GAI is implemented.
The AI assistant resolves customer queries in about one- AI, Management, and Organizational Outcomes
fifth of the time and does so more accurately than the hu-
mans. As a result, customer satisfaction with the AI assis- Management involves coordinating tasks and activities
tant has been similar to that with human agents and the to achieve specific objectives and characterizes creatively
overall productivity gains thus far are projected to result solving problems that arise during the planning, organiz-
in a $40 million profit improvement in 2024. However, ing, leading, and controlling of resources. Simply put, it is
not all chatbots have been equally well-received. Koko is the “judicious use of means to accomplish an
an online mental health support company whose recent end” (Merriam-Webster, n.d.). Accordingly, the applica-
experiment with using GPT-3 to respond to customers tion of generative artificial intelligence to real-world man-
seeking mental health support turned controversial agement problems has profound implications for the busi-
(Ingram, 2023). Strong criticism followed when news ness sector.
about the experimental use of a generative AI bot was It should come as no surprise that the use of AI is ex-
made public by Koko’s co-founder. This was despite the panding as companies increasingly leverage AI to en-
fact that the AI-generated responses were deemed to be at hance the organization’s profitability and efficiency.
least as good as responses that would have been written McKinsey’s 2022 State of AI survey finds that AI adop-
entirely by humans. tion has more than doubled since 2017 (McKinsey, 2022).
The implementation of generative AI has been uneven, The survey also estimates that, though originally used to
with some seemingly straightforward tasks leading to manage manufacturing (e.g., simulations like 3-D model-
public and costly failures while other difficult tasks en- ing) and risk (e.g., fraud and debt analytics), AI now gen-
joyed resounding successes. For example, Air Canada’s erates the most significant impact on revenues in the man-
use of a chatbot that provided incorrect advice to a cus- agement of marketing and sales (e.g., customer service
tomer seeking information on the airline’s bereavement analytics and segmentation), product and service develop-
policy eventually landed the airlines in court (Proctor, ment (e.g., creation of new AI products), and strategy and
2024). The airlines chose to not stand by the pricing infor- corporate finance (e.g., capital allocation and M&A sup-
mation offered by the chatbot and ultimately lost the law- port) and on costs in supply chain management (e.g., sales
suit brought about by the customer who was able to pro- and demand forecasting and logistic networks optimiza-
vide the details of the interaction with the chatbot as evi- tion) (McKinsey, 2022). Thus, the promise of GAI bene-
dence of the inaccurate information. However, GAI has fits has shifted from automation of simple discrete tasks
been used to create new drugs with promising initial re- to creative and complex tasks that can increase productiv-
sults. Starting with patient-focused problems and using ity and supplement human labor. The benefits of GAI can
GAI to find a molecule that targets the problem without come in the form of reduced costs, time savings, higher
harming the patient can lead to breakthroughs. In particu- labor productivity, or higher customer satisfaction.
lar cases of proteins with a specific function, the need for The 2024 State of AI survey (McKinsey, 2024b) sug-
AI generated solutions has been validated at the proof-of- gests a shift in this trend, with the most significant cost
concept stage by tech companies like Generate Biomedi- decreases coming from generative AI use in the human
cines (Heaven, 2023). The regulatory process is currently resource management function and revenue increases in
incorporating changes that would allow AI discovered supply chain management. The proportion of companies
drugs to enter testing, a complex system of innovation that spend most aggressively (more than 20% of their
with a focus on patient safety (FDA, 2024). digital budget according to the survey) on generative AI is
Adoption of new technologies is a costly and difficult the greatest in energy and materials (17%), technology
decision for most businesses and organizational leaders (11%), and financial services (11%) (McKinsey, 2024b).
(Ataman et al., 2023). If the technology is capable of re- Healthcare, manufacturing, and retail operations, howev-
placing human labor partially or completely, the stakes er, have traditionally seen the strongest impact of AI inno-
become even higher not only for business leaders who vation (Dogru & Keskin, 2020). The capacity of AI to
decide whether to adopt, but also for the workers who enhance business operations is based largely on enhanced
may have to learn the new technology or even risk being productivity by reducing search time and processing fast-
replaced. GAI has the potential to disrupt the labor market er data used in decision making, and freeing employees to
(Frank et al., 2019), which leads to considerable concerns engage in creative and adaptive work (Tarafdar, 2019).
for workers (Yehiav, 2023).
 Journal of Behavioral and Applied Management 114

A vast body of academic research corroborates the application of GAI in the management of organizational
growing significance of AI in the management of organi- problems and the realized organizational benefits may be
zational processes to achieve organizational goals. Kob- non-linear.
bacy et al. (2007) review 1200 papers published in the
period 1995-2004, and Kobbacy & Vadera (2011) review Task Creativity and AI
1400 papers published between 2005-2009 to classify AI Individual differences in creativity are conventionally
applications in 4 areas: (1) design, (2) scheduling, (3) measured using divergent thinking tests (Silvia et al.,
process planning and control, and (4) quality, mainte- 2008). Creativity has traditionally been viewed as an abil-
nance and fault diagnosis. More recently, Jarrahi et al. ity mainly attributable to human beings and the common
(2023) assess the potential of AI in supporting aspects of consensus is that the notion of independently creative AI
knowledge management, including creation, storage and is still many years away (Hwang, 2022). In that sense, AI
retrieval, sharing, and application of knowledge in organi- can be used as a substitute for routine tasks which can
zations. AI is also applied in the analysis of customer data unlock employee creative potential. In support of this,
to personalize the company’s interaction with customers, Koivisto and Grassini (2023) argue that the “best humans
which has been linked to increased satisfaction, brand still outperform artificial intelligence in a creative diver-
loyalty, and business growth (Odejide & Edunjobi, 2024). gent thinking task” (p. 13601). Specifically, they conduct
AI is increasingly used to manage credit, market and op- a popular divergent thinking test to compare the creativity
erational risk, as well as compliance (Aziz & Dowling, of 256 humans to that of three current AI chatbots. Find-
2019) and is even transforming educational management ings suggest that on average, humans often include poor-
in streamlining administrative tasks, improving student quality ideas relative to the chatbots, but the top human
outcomes and enhancing the learning process (Igbokwe, ideas are comparable to, or better than those, produced by
2023). AI. More recently, Castelo et al. (2024) find that new
Several studies have investigated the use of AI and AI- product ideas generated by GPT4 are rated higher than
driven tools on the financial performance of the firm. those produced by laypeople and professionals. Similarly,
Mullangi (2017) argues for a positive long-term correla- Girotra et al. (2023) find that ChatGPT outperforms MBA
tion between AI application and the firm’s economic out- students in coming up with new product ideas, a surpris-
comes when AI is used for predicting market trends and ing result for Prof. Terwiesch, one of the researchers, who
more efficient resource allocation to guide decision- said “I had naively believed that creative work would be
making aimed at optimizing relationships across units and the last area in which we humans would be superior at
between the organization and its environment. AI can solving problems [...]” (Basiouny, 2003).
improve the firm’s economic outcomes in several ways, Such findings have caused some to question the useful-
including enhanced efficiency, automated and fast adapta- ness of a comparison between humans and AI and to call
tion to a rapidly changing external environment, detection for a shift in focus from competing to collaborating with
of fraud, optimization of sales processes, and automating AI for enhanced creativity (e.g., Elfa & Dawood, 2023).
quality management (Wamba-Taguimdje, 2020). GAI is Accordingly, Wu et al. (2021) propose the term “AI crea-
also shown to affect firm value and future profitability vis tivity” - “the ability for human and AI to co-live and co-
-à-vis employees’ exposure to GAI - the degree to which create by playing to each other’s strengths to achieve
the workforce can be replaced by GAI - where an increase more” (p. 171) and Elfa and Dawood (2023) suggest a
in firm value is notable in companies when the workforce choice board of options to use AI in helping people over-
is more exposed to GAI (Eisfeldt et al., 2024). come the limitations of the human brain and enhance cre-
Different occupations entail varying levels of nonrou- ativity by providing feedback on unfinished work and
tine cognitive (analytical and interpersonal), routine cog- producing supplementary artwork. Empirical research
nitive and manual, and nonroutine manual (physical and supports this trend. Using a dataset of more than 4 million
interpersonal) skills and research shows that technology artworks from over 50,000 unique users Zhou et al.
substitutes for workers in routine cognitive and manual (2024) show that text-to-image GAI increases human cre-
tasks and complements workers in non-routine problem ative productivity by 25%.
solving and complex communication tasks (e.g., Ace- Relying on previous research on creativity in general,
moglu & Autor, 2011; Autor et al., 2003). Furthermore, we envision the concept of task creativity as the degree to
40% of white-collar occupations characterize high GAI which creativity should be applied for successful task
exposure in comparison to 9% for blue-collar and service completion. Thus, tasks may be on a continuum from low
occupations (Eisfeldt et al., 2024). Also, there are distinc- creativity tasks, such as summarizing legal documents,
tive differences in how managers approach AI-based in- summarizing customer reviews, aggregating search re-
novation management falling in one of four distinctive sults, to high creativity tasks such as creating new protein
clusters according to Füller et al. (2022) - AI- molecules for new drugs or writing a movie script. Table
Frontrunners, AI-Practitioners, AI-Occasional innovators 1 provides some examples of white-collar occupations
and Non-AI innovators. Such variation of occupational classified by task creativity taking into consideration the
and task exposure to GAI, degree of managerial openness dominant set of tasks in each occupation. As can be ob-
to AI-based innovation management, and employee skill served, certain occupations are inherently more creative
requirements suggests that the association between the than others.
115 Charu Sinha, Veselina Vracheva, and Cristina Nistor

Table 1 dation of the GAI task output will be a necessary overlay

to maximize the benefits of GAI through various actions.
A Potential Classification of White-Collar Jobs based These actions could include creating accurate training
on Task Creativity prompts (Metcalfe, 2024), mitigating fakes or lies
(Stimson, 2023) which are euphemistically called halluci-
White Collar Task nations but should be called “bullshit” as suggested by
Hicks et al. (2024), eliminating out-of-context results
Occupations Creativity (Hoffman, 2023), incorporating cultural perspective for
Accountant Low the application (Tsanni, 2023) and ensuring bias free re-
sults (Trifilo & Blau, 2024; Unesco, 2024a). AI bias has
Interpreter Low been a growing concern associated with risks in deploying
AI systems. In general research, bias is defined as the
Customer service representative Low generation, intentionally or unintentionally, of systematic
errors by selecting one outcome or answer over another
Data engineer Low (Pannucci & Wilkins, 2010). In AI, this can take the form
of, for example, a face-recognition algorithm being
Drug innovation professional High trained with more photos of light-skinned faces than dark-
skinned faces, leading to poor performance in recognizing
Writer High darker-skinned faces (Srinivasan & Chander, 2021). Bias-
Coder High es can occur at every stage of AI development, including
data collection, preparation and annotation as well as
Data Scientist High model development, deployment, and evaluation
(Gichoya et al., 2023). Hence, the human validation as-
Music creation High pect can be integral in the successful adoption of GAI
systems.
Marketing content creation High
A Conceptual Framework of GAI Benefits
Employee training videos creation High Creativity, long believed to be a feature mainly attribut-
able to humans, is now beginning to be simulated by gen-
Given the present work investigates the role of creativi- erative AI models. However, applying GAI to the busi-
ty at the level of the task, Table 2 includes examples ness world is still in a nascent phase. We suggest that a
across several industries of tasks that vary in their degree potential ramification of the development of GAI is to
of creativity, the benefit to the company from using GAI predict how it would perform tasks usually undertaken in
and whether the task was successfully completed at the white collar jobs by humans. These tasks may vary in
time GAI was implemented. their creativity level, potentially leading to different levels
of GAI substitution, as seen in the examples in Table 2.
Human Validation and AI Thus, we propose that:
As has been noted earlier, use of GAI has led to mixed Proposition 1: Generative AI can be used to substitute human
results, with some use cases yielding higher productivity labor for low and highly creative tasks in white collar
gains through faster task completion and reduced labor jobs to derive GAI benefits.
needs, and others resulting in an array of outcomes in- To build on current research, future work might focus
cluding no efficiencies, lawsuits and negative publicity. on a systematic analysis of jobs or occupations that would
Unlike human intelligence, GAI lacks contextual under- be amenable to GAI substitution for human labor. The
standing of the application. This can result in an inauthen- current research landscape suggests examples of white
tic or untrustworthy GAI-created output (Robinson, collar jobs without fully defining the boundaries of what
2024), which could fail to satisfy the task requirements or might be possible for GAI in the future.
customer specification. To derive usefulness from the
automation of a GAI-created output, there is a need for Moreover, despite the promise that GAI will be fully
human supervision that can validate the task output creat- automated to take over complex business processes, there
ed by GAI. Human validation can ensure that GAI output is considerable uncertainty that organizational processes
is relevant, as well as useful and safe for the specific task are amenable to that level of GAI integration. Instead, we
performed and in the context of the larger managerial expect that GAI will substitute for some tasks currently
application. performed by human white-collar workers and that hu-
mans will need to operate in a supervisory role. Thus,
The degree of human validation needed for the GAI human workers may be involved with the GAI output in
output may vary by task. It could range from a cursory different degrees: some tasks and roles may be a simple
scan to validate a routine task in a customer service appli- check for context while others may require detailed vali-
cation to a longer and more thoughtful validation process dation. Therefore,
in a critical marketing campaign. Deploying human vali-
 Journal of Behavioral and Applied Management 116

Table 2
Examples of Task Creativity in Real World Business Applications
Company GAI Task Creativity Benefit to Company Outcome
Qualcomm make TikTok videos High reduce hiring of people with video G
editing skills
Ecolab analyze rival earnings reports Low help Ecolab prepare their own earnings 60% G, 40% B
call
Cisco employee conflict High successful conflict diagnosis G
management
Klarna customer service for a Fin Low handle workload of 700 customer G
Tech company service agents
Koko send mental health support Low Controversial, caused bad press for firm B
messages to users
Mind Meld send public relation pitches Low no efficiencies since humans had to B
to journalists clean up the gen AI output
Air Canada give bereavement policy info company had to refund the customer due B
to incorrect info being provided
Low
Trivago create new ad campaign for High reduced from 35 different productions to G
global markets one by using AI to translate to 12 other
languages.
Coca Cola Create “Masterpiece” High created a unique commercial drawing a G
commercial lot of attention
Generate software twists strands of High creating specific drugs for a particular G
Biomedicines amino acids to form new condition
proteins

Note. Outcomes can be Good (G) or Bad (B).

Proposition 2: Generative AI can be used with low and high Management scholars might investigate the organiza-
human validation to derive GAI benefits. tional capabilities that allow organizational leaders to
As the GAI technology improves, we see a need for deploy GAI as a strategic tool rather than a timely new
more research to determine the uses for human validation technology or fad. The degree of GAI adoption along with
for GAI. So far, managers and researchers are deploying human validation are two important variables that organi-
GAI with ad-hoc levels of human validation and check- zations will have to consider in addition to all the other
ing. Future field studies could more precisely determine strategic decisions, thus opening a new research perspec-
the human validation needs for particular tasks. tive in the future.
Organizations that aim to deploy GAI must be prepared Several business occupations such as accountants, para-
to determine the extent to which the technology should be legals, and customer service agents perform largely rou-
used. Thus, human validation of GAI becomes a strategic tine cognitive tasks that may also be repetitive and rule-
choice for investment. We propose that to maximize GAI based. These tasks include preparing financial statements
benefits managers should match the level of human vali- from input data, summarizing large amounts of text or
dation to the task creativity when GAI is used to substi- providing customer service in the form of answering cus-
tute for white collar labor tasks. tomers’ questions or providing informational assistance to
customers. The degree of creativity required in their daily
Proposition 3: The degree of human validation is a strategic
tasks is relatively low as their tasks are governed by disci-
management choice that organizational leaders must
align with the task creativity needs to maximize GAI pline-specific rules and does not offer opportunity for
benefits. much flexibility or imagination to the task. Utilizing GAI
in these tasks presents an opportunity to reduce the time
117 Charu Sinha, Veselina Vracheva, and Cristina Nistor

to complete the task while using fewer employees, there- High creativity tasks in business may encompass a
by reducing costs, increasing productivity, and ultimately range of activities including writing (scripts, novels, song
improving the overall profitability of the company. This lyrics), producing music, producing imagery (photos, vid-
would lead to high GAI benefits. The output of the tasks eos, art), creation of marketing content, and creation of
needs to be aligned with well-defined rules and parame- employee training videos. Each of these tasks requires a
ters within the task problem. Thus, the degree of human specific artistic skill set, usually acquired and honed
validation needed for any error correction in the output is through many years of learning and practice. Most of
expected to be minimal. For example, a task like text these high creative tasks also consume a significant
summarization/annotation involves generating concise amount of time to produce the desired output. Substituting
summaries of lengthy documents or articles. GAI algo- the human effort on these tasks by applying GAI promises
rithms can rapidly extract key points and summarize con- to bring to the endeavor a level of imagination at least
tent automatically, with humans reviewing summaries for equivalent to humans but with the ability to produce the
accuracy. Thus, we posit that: creative output in a much shorter time and with many
output options to choose from (Swant, 2023). This would
Proposition 4: GAI used to perform low creativity tasks can
yield high GAI benefits when the GAI-created output greatly reduce the time expended on, for example, pro-
requires low human validation. ducing a creative advertisement, while also reducing the
need to hire several employees with different skill sets to
In this proposition we theorize that for tasks with low produce the advertisement (Swant, 2023). The drawback
creativity and low human validation needs, GAI adoption with the GAI application in this case is the possibility of
is likely to yield high benefits, which implies that busi- too fanciful or unrealistic GAI-produced output
nesses should adopt GAI for these specific conditions (Weekman, 2024). The creative endeavor may also re-
first. Thus, future research might investigate empirically quire a deep understanding of human emotions, cultural
if businesses are finding these types of tasks as their first nuances or contextual factors that are challenging for GAI
for GAI adoption. Moreover, as GAI technology im- to replicate. This necessitates extensive human validation
proves, we expect that such studies would capture the to lend human judgment on the appropriateness and effec-
changing business decision-making in further GAI adop- tiveness of the GAI output. We expect that despite this
tion across organizations. human overlay to the machine-produced creative output,
GAI may be less suited for certain low creativity tasks. the overall savings in labor and time could be substantial.
Some low creativity tasks such as compliance audits, Therefore,
which ensure adherence to financial regulations and
standards, require significant human oversight as a crucial Proposition 7: GAI used to perform high creativity tasks can
yield high GAI benefits when the GAI-created output
validation to interpret nuanced regulations and ensure requires high human validation.
ethical standards. Though the tasks may be repetitive or
routine, the ability to interpret complicated yet flexible Future research might investigate the degree to which
accounting regulations requires a high level of human humans need to validate the GAI output for each industry
validation. Thus, any efficiencies gained from deploying and domain. Since different areas of business have differ-
GAI in these tasks could be largely negated by the costs ent requirements for creativity, such empirical research
of an increased level of human validation, which may lead would characterize the nuanced levels of high human vali-
to low overall benefits of GAI adoption. Therefore, dation needed in each industry or area.
Figure 1 summarizes the conceptual framework. Our
Proposition 5: GAI used to perform low creativity tasks can
yield low GAI benefits when the GAI-created output framework focuses on white collar jobs that may benefit
requires high human validation. from substituting GAI for some tasks. We suggest that the
level of benefits organizations may derive from GAI im-
One of the most promising areas of GAI applications, plementation will depend on the level of task creativity
tasks with high creativity that require very little human and human validation needed. In most combinations of
validation, have been studied in academic research experi- task creativity and human validation, GAI is expected to
mentally (Peng et al., 2023) and already applied in busi- yield a high benefit if deployed correctly. However, Prop-
ness settings (Soper, 2023). For example, code generation osition 5 describes that for low creativity tasks that re-
and automated programing involving writing code and quire high validation the expected benefit from GAI is
developing software applications are particularly well low. This level is relative to other returns and includes
suited for GAI applications. GAI tools can generate code benefits that are greater than zero. Companies might still
snippets, automate repetitive programming tasks, and choose to use GAI in this scenario, because a low or mod-
even assist in algorithm development with minimal hu- erate benefit to GAI might still be preferable from a cost
man validation, as the resulting algorithms can then pass perspective.
automated tests to ensure they are successful in achieving
the desired results. We thus expect that: Practical Implications for Managers
Proposition 6: GAI used to perform high creativity tasks can Our framework has immediate practical and managerial
yield high GAI benefits when the GAI-created output implications. While it is estimated that a quarter of For-
requires low human validation. tune 2000 companies have a VP or higher position to
 Journal of Behavioral and Applied Management 118

Figure 1
Conceptual Framework for GAI benefits

Note. P3 is the overarching proposition, which captures the entire conceptual framework and is implied in the above figure.

manage AI in the organization (and the number is ex- as the firm’s technological infrastructure, strategic align-
pected to grow to 80% in 2024), the role of this leadership ment, scalability, and AI integration with existing sys-
position is not yet well defined (Molla, 2023). One of the tems. Taken together, these factors limit the generalizabil-
core tasks of an AI manager should include how to best ity of the proposed framework.
make use of GAI to ensure maximum benefits from the Nevertheless, future research can explore several ave-
technology. We suggest that one of the ways a manager nues based on the present conceptual framework. As GAI
could determine how to deploy GAI is to match the crea- is increasingly integrated into various management func-
tivity level of the task to the level of human validation tions, the potential for biases in decision-making when
that a successful implementation of the task would re- creative and complex tasks are concerned, including for
quire. A thoughtful planning of how to use GAI tools is a which AI is used to augment or automate decisions, be-
core function of an AI manager who can then deploy hu- comes more prominent. Ohlheiser (2024) suggests that
man and GAI assets accordingly. “Those dominant patterns might show up in the training
Moreover, managers should consider legal and regula- data an AI system learns from, in the tasks it is asked to
tory aspects of using GAI. Human validation is crucial for complete, and in the algorithms that power its learning
businesses: if the level of human validation does not process.” Bias in GAI introduced through training da-
match the needs of the task, it can lead to costly mistakes tasets that might contain skewed, incorrect or biased in-
for the brand, antagonize loyal and prospective customers, formation is difficult to identify accurately and timely,
and expose the company to legal challenges. Thus, man- and to correct (Robertson, 2024). Biased AI systems can
agers should carefully decide if the benefit of using GAI lead to adverse economic outcomes for the firm vis-à-vis
is worth the risk, and whether the company has the capac- decisions regarding unfair and discriminatory practices,
ity to correctly deploy GAI. strategic resource allocation, the development of competi-
tive advantages, talent search, employee promotion, and
Limitations and Future Research customer satisfaction among others. Unesco (2024b) iden-
The present work has several limitations. It is conceptu- tified gender biases that might be particularly damaging
al, which limits its ability to reflect real-world scenarios for hiring decisions for example “Open-source LLMs in
and to provide concrete evidence or validate the proposi- particular tended to assign more diverse, high-status jobs
tions. There is no standardized measure for creativity to men, such as engineer, teacher and doctor, while fre-
across different tasks and contexts and creativity is inher- quently relegating women to roles that are traditionally
ently subjective and difficult to quantify uniformly across undervalued or socially-stigmatized, such as “domestic
disciplines and industries. Also, focusing specifically on servant”, “cook” and “prostitute”.” Decisions influenced
task creativity and human validation might overlook other by biased AI can have profound consequences for individ-
critical factors influencing AI economic outcomes, such uals and society. Understanding and mitigating such bias-
119 Charu Sinha, Veselina Vracheva, and Cristina Nistor

es will affect the quality of managerial decision-making. and produces more authentic output, the benefits of im-
Thus, conducting empirical studies on the impact of bi- plementing GAI in management tasks will need to be
ased AI decisions in real-world scenarios, and evaluating reevaluated. Our current framework suggests that decision
the effectiveness of mitigation practices is an important makers will continuously update their strategic decisions
area for future research on the economic outcomes of AI. regarding the degree of human validation deployed rela-
Anchoring bias in decision-making – a biased perception tive to the task creativity requirements. The degree of
caused by an anchoring effect from AI decisions, which human validation will also depend on algorithm quality
restricts the consideration of alternative scenarios – is and availability of data for training, and will in turn affect
particularly problematic (Rastogi et al., 2022) and could the GAI benefit derived by the organization. Potential
be prioritized in future investigations on how to improve empirical validations of the theory will have to consider
the benefits of AI systems that managers consider for in- contemporaneous technological advances.
tegration into business operations.
References
The expectation is that organizations will continue to
Acemoglu, D. (2024). The simple macroeconomics of AI.
replace human employees with intelligent machines, espe- NBER, Working Paper 32487. 10.3386/w32487. http://
cially in the performance of creative tasks, potentially www.nber.org/papers/w32487
making the workforce totally unrecognizable by 2040 (De
Cremer & Kasparov, 2021). Thus, a key managerial con- Acemoglu, D., & Autor, D. (2011). Skills, tasks and technolo-
sideration will be the degree to which and ways in which gies: Implications for employment and earnings. In Handbook
of labor economics (Vol. 4, pp. 1043-1171). Elsevier.
the integration of AI alters the roles and responsibilities of
human managers. Questions like how AI helps and hin- Ataman, K., Myhr, N., & Nistor, C. (2023). Disruptive innova-
ders managerial work and the issue of trust between man- tion as a network dilemma: A conceptual model. Journal of
agers and employees and its subsequent impact on AI Behavioral and Applied Management, 23(2), 104-113. https://
adoption rates can be examined. Concomitantly, if human doi.org/10.21818/001c.84555
work is increasingly replaced by AI, how will AI influ- Autor, D. H., Levy, F., & Murnane, R. J. (2003). The skill con-
ence organizational culture - one of the most powerful tent of recent technological change: An empirical exploration.
levers for successful implementation of a firm’s strategic The Quarterly Journal of Economics, 118(4), 1279-1333.
plan - is a viable research question. Even more important- https://doi.org/10.1162/003355303322552801
ly, culture itself may change dramatically. The promotion Aziz, S., & Dowling, M. (2019). Machine learning and AI for
of important behaviors like innovation, risk-taking, and risk management (pp. 33-50). Springer International Publish-
collaboration all rooted in the company’s organizational ing.
culture has managerial and organizational implications
ripe for investigation. Barnett, T. (2024). Here’s why AI probably isn’t coming for
your job anytime soon. Fast Company. https://
Future research might focus on the ethical considera- www.fastcompany.com/91130124/heres-why-ai-probably-isnt
tions of GAI adoption that substitutes completely for hu- -coming-for-your-job-anytime-soon?utm_source=pocket-
man labor. As nascent technologies rely on human labor newtab-en-us
to develop (Grynbaum & Mac, 2024) and then have the Basiouny, A. (2023). Is ChatGPT a better entrepreneur than
potential to completely replace the humans that created most? Knowledge at Wharton. https://
the inputs, it is important to consider the boundaries of the knowledge.wharton.upenn.edu/article/is-chatgpt-a-better-
ethics of such substitution processes and the potential entrepreneur-than-most
worker protections or compensations that should be put in
Brynjolfsson, E., Li, D., & Raymond, L. R. (2023). Generative
place. AI at work. NBER, Working Paper No. 3116. http://
Workforce skill transition and mobility within the or- dx.doi.org/10.3386/w31161
ganization are major AI-related concerns. To that end,
Castelo, N., Katona, Z., Li, P., & Sarvary, M. (2024). How AI
understanding how AI impacts career progression and job outperforms humans at creative idea generation. SSRN, http://
roles within firms and identifying key skills that employ- dx.doi.org/10.2139/ssrn.4751779
ees need to effectively collaborate with AI will be a sig-
nificant challenge for managers. AI-based machines are Caufield, B. (2023). SIGGRAPH Special Address: NVIDIA CEO
fast, versatile, and can be trained to perform a variety of brings generative AI to LA show. Nvidia. https://
human tasks. As of now, humans excel in intuition, emo- blogs.nvidia.com/blog/siggraph-2023-special-address/
tion, moral reasoning, sensory experiences, and cultural Chui, M., Hazan, E., Roberts, R., Singla, A., Smaje, K., Sukha-
sensitivity. One area for research is identifying the kind of revsky, A., Yee, L., & Zemmel, R. (2023). The economic
opportunities for collaborations between organic and inor- potential of generative AI: The next productivity frontier.
ganic intelligence associated with optimal firm outcomes. McKinsey & Co. https://www.mckinsey.com/capabilities/
mckinsey-digital/our-insights/the-economic-potential-of-
Conclusion generative-AI-the-next-productivity-frontier#business-value
To quote Dr. Harrick Vin, CTO of Tata Consultancy De Cremer, D., & Kasparov, G. (2021). AI should augment
Services on AI innovation, “This is not a one-time trans- human intelligence, not replace it. Harvard Business Review,
formation” (Jacobs, 2024). As generative AI improves 18(1). https://www.daviddecremer.com/wp-content/uploads/
 Journal of Behavioral and Applied Management 120

HBR2021_AI-Should-Augment-Human-Intelligence-Not- Hicks, M.T., Humphries, J., & Slater, J. (2024). ChatGPT is

Replace-It.pdf bullshit. Ethics and Information Technol, 26(2). https://
doi.org/10.1007/s10676-024-09775-5
Dilmegani, C. (2024). Top 35+ generative AI tools by cfategory
(Text, Image…), AIMultipleResearch. https:// Hoffman, B. (2023, March 31). Leaders looking to leverage AI
research.aimultiple.com/generative-ai-tools/ need to think about context. Forbes. https://www.forbes.com/
sites/brycehoffman/2023/03/31/leaders-looking-to-leverage-ai
Dogru, A. K., & Keskin, B. B. (2020). AI in operations manage- -need-to-think-about-context/
ment: Applications, challenges and opportunities. Journal of
Data, Information and Management, 2(2), 67-74. https:// Hwang, A. H. C. (2022). Too late to be creative? AI-empowered
doi.org/10.1007/s42488-020-00023-1 tools in creative processes. In CHI Conference on Human
Factors in Computing Systems Extended Abstracts (pp. 1-9).
Eisfeldt, A. L., Schubert, G., Zhang, M. B., & Taska, B. (2024).
The labor impact of generative AI on firm values. https:// Igbokwe, I. C. (2023). Application of artificial intelligence (AI)
papers.ssrn.com/sol3/papers.cfm?abstract_id=4436627 in educational management. International Journal of Scien-
tific and Research Publications, 13(3), 300-307.
Elfa, M. A., & Dawood, M. E. T. (2023). Using artificial intelli-
gence for enhancing human creativity. Journal of Art, Design Ingram, D. (2023). A mental health tech company ran an AI
and Music, 2(2), 106-120. https://doi.org/10.55554/2785- experiment on real users. Nothing’s stopping apps from con-
9649.1017 ducting more. NBC News. https://www.nbcnews.com/tech/
internet/chatgpt-ai-experiment-mental-health-tech-app-koko-
Ellingrud, K., Sanghvi, S., Gurneet Singh Dandona, G.H., rcna65110
Madgavkar, A., Chui, M., White, O., & Paige Hasebe, P.
(2023). Generative AI and the future of work in America. Jacobs, E. (2024, April 13). The rise of the chief AI officer.
McKinsey Global Institute. https://www.mckinsey.com/mgi/ Financial Times. https://www.ft.com/content/4c5d823e-004b-
our-research/generative-ai-and-the-future-of-work-in-americ 4f00-9536-9fea93330453
FDA (2024). Artificial intelligence and machine learning (AI/ Jarrahi, M. H., Askay, D., Eshraghi, A., & Smith, P. (2023).
ML) for drug development. FDA. https://www.fda.gov/ Artificial intelligence and knowledge management: A partner-
science-research/science-and-research-special-topics/artificial ship between human and AI. Business Horizons, 66(1), 87-99.
-intelligence-and-machine-learning-aiml-drug-development
Klarna (2024). Klarna AI assistant handles two-thirds of cus-
Frank, M. R., Autor, D., Bessen, J. E., Brynjolfsson, E., Cebrian, tomer service chats in its first month. https://
M., Deming, D. J., Feldman, M., Groh, M., Lobo, J., Moro, www.klarna.com/international/press/klarna-ai-assistant-
E., Wang, D., Youn, H., & Rahwan, I. (2019). Toward under- handles-two-thirds-of-customer-service-chats-in-its-first-
standing the impact of artificial intelligence on labor. Pro- month/
ceedings of the National Academy of Sciences of the United
States of America, 116(14), 6531–6539. https:// Kobbacy, K.A.H. & Vadera, S. (2011). A survey of AI in opera-
doi.org/10.1073/pnas.1900949116 tions management from 2005 to 2009. Journal of Manufactur-
ing Technology Management, 22(6), 706-733. https://
Füller, J., Hutter, K., Wahl, J., Bilgram, V., & Tekic, Z. (2022). doi.org/10.1108/17410381111149602
How AI revolutionizes innovation management–Perceptions
and implementation preferences of AI-based innovators. Kobbacy, K.A., Vadera, S., & Rasmy, M. H. (2007). AI and OR
Technological Forecasting and Social Change, 178, 121598. in management of operations: History and trends. Journal of
https://doi.org/10.1016/j.techfore.2022.121598 the Operational Research Society, 58, 10-28.

Gichoya, J. W., Thomas, K., Celi, L. A., Safdar, N., Banerjee, I., Koivisto, M., & Grassini, S. (2023). Best humans still outper-
Banja, J. D., Seyyed-Kalantari, L., Trivedi, H., & Pur- form artificial intelligence in a creative divergent thinking
kayastha, S. (2023). AI pitfalls and what not to do: mitigating task. Scientific Reports, 13(1), 13601.
bias in AI. The British Journal of Radiology, 96(1150), Korinek, A., & Suh, D. (2024). Scenarios for the transition to
20230023. https://doi.org/10.1259/bjr.20230023 AGI. National Bureau of Economic Research. Working Paper
Girotra, K., Meincke, L., Terwiesch, C., & Ulrich, K. T. (2023). 32255. https://www.nber.org/system/files/working_papers/
Ideas are dimes a dozen: Large language models for idea gen- w32255/w32255.pdf
eration in innovation. The Wharton School Research Paper McElheran, K., Li, J. F., Brynjolfsson, E., Kroff, Z., Dinlersoz,
Forthcoming. https://ssrn.com/abstract=4526071 or http:// E., Foster, L., & Zolas, N. (2024). AI adoption in America:
dx.doi.org/10.2139/ssrn.4526071 Who, what, and where. Journal of Economics & Management
Grynbaum, M. M. & Mac, R. (2023, December 27). The Times Strategy, 33, 375–415. https://doi.org/10.1111/jems.12576
sues OpenAI and Microsoft over A.I. use of copyrighted McKinsey (2022). The state of AI in 2022 – and a half decade in
work. New York Times. https:// review. QuantumBlack. https://www.mckinsey.com/~/media/
www.nytimes.com/2023/12/27/business/media/new-york- mckinsey/business%20functions/quantumblack/our%
times-open-ai-microsoft-lawsuit.html 20insights/the%20state%20of%20ai%20in%202022%
Heaven, W. D. (2023). AI is dreaming up drugs that no one has 20and%20a%20half%20decade%20in%20review/the-state-of
ever seen. Now we’ve got to see if they work. MIT Technolo- -ai-in-2022-and-a-half-decade-in-review.pdf
gy Review. https:// McKinsey (2024a). What is generative AI? https://
www.technologyreview.com/2023/02/15/1067904/ai- www.mckinsey.com/featured-insights/mckinsey-explainers/
automation-drug-development/ what-is-generative-ai
121 Charu Sinha, Veselina Vracheva, and Cristina Nistor

McKinsey (2024b). The state of AI in 2022 – and a half decade Soper, T. (2023, June 13). AI and coding: How these tech com-
in review. QuantumBlack. https://www.mckinsey.com/ panies are using generative AI for programming. GeekWire.
capabilities/quantumblack/our-insights/the-state-of-ai#/ https://www.geekwire.com/2023/ai-and-coding-how-seattle-
tech-companies-are-using-generative-ai-for-programming/
Merriam-Webster (n.d.) Dictionary. https://www.merriam-
webster.com/dictionary/management Srinivasan, R., & Chander, A. (2021). Biases in AI sys-
tems. Communications of the ACM, 64(8), 44-49.
Metcalfe, C. (2024, April 24). Job titles of the future: AI prompt
engineer. MIT Technology Review. https:// Stimson, B. (2023, December 29). Michael Cohen admits to
www.technologyreview.com/2024/04/24/1091125/ai-prompt- inadvertently citing fake cases generated by AI in legal mo-
engineer-generative-ai-job-titles/ tion. Fox News. https://www.foxnews.com/politics/michael-
cohen-admits-inadvertently-citing-fake-cases-generated-ai-
Molla, R. (2023). The hottest new job is “head of AI” and no- legal-motion
body knows what they do. Vox. https://www.vox.com/
technology/2023/7/19/23799255/head-of-ai-leadership-jobs Swant, M. (2023, December 18). AI Briefing: Creative ways
companies used AI in advertising in 2023. DigiDay. https://
Mullangi, K. (2017). Enhancing financial performance through digiday.com/media-buying/ai-briefing-creative-ways-
AI-driven predictive analytics and reciprocal symmetry. Asian companies-used-ai-in-advertising-in-2023/
Accounting and Auditing Advancement, 8(1), 57-66.
Tarafdar, M., Beath, C. M., & Ross, J. W. (2019). Using AI to
Noy, S., & Zhang, W. (2023). Experimental evidence on the enhance business operations. MIT Sloan Management Review,
productivity effects of generative artificial intelligence. Sci- 60(4), 37-44.
ence, 381,187-192. 10.1126/science.adh2586
Trifilo, A., & Blau, G. (2024). Developing racially ambiguous
Odejide, O. A., & Edunjobi, T. E. (2024). AI in project manage- job candidate avatars with strong versus ambiguous job candi-
ment: Exploring theoretical models for decision-making and date backgrounds in preparation for a field experiment for
risk management. Engineering Science & Technology Jour- enhancing inclusive hiring and reducing implicit bias. Journal
nal, 5(3), 1072-1085. of Behavioral and Applied Management, 24(1),14–25. https://
Ohlheiser, A.W. (2023, June 14). AI automated discrimination. doi.org/10.21818/001c.115890.
Here’s how to spot it. Vox. https://www.vox.com/ Tsanni, A. (2023, November 17). This company is building AI
technology/23738987/racism-ai-automated-bias- for African languages. MIT Technology Review. https://
discrimination-algorithm www.technologyreview.com/2023/11/17/1083637/lelapa-ai-
Pannucci, C. J., & Wilkins, E. G. (2010). Identifying and avoid- african-languages-vulavula/
ing bias in research. Plastic and reconstructive surgery, 126 Unesco (2024a). Challenging systematic prejudices: An investi-
(2), 619-625. gation into bias against women and girls in large language
Peng, S., Kalliamvakou, E., Cihon, P., & Demirer, M. (2023). models (Catalogue Number 0000388971). https://
The impact of AI on developer productivity: Evidence from unesdoc.unesco.org/ark:/48223/pf0000388971.locale=en
GitHub Copilot. https://doi.org/10.48550/arXiv.2302.06590 Unesco (2024b). Generative AI: UNESCO study reveals alarm-
Proctor, J. (2024, February 15). Air Canada found liable for ing evidence of regressive gender stereotypes. https://
chatbot’s bad advice on plane tickets. CBC News. https:// www.unesco.org/en/articles/generative-ai-unesco-study-
www.cbc.ca/news/canada/british-columbia/air-canada-chatbot reveals-alarming-evidence-regressive-gender-stereotypes
-lawsuit-1.7116416 Wamba-Taguimdje, S. L., Wamba, S. F., Kamdjoug, J. R. K., &
Rastogi, C., Zhang, Y., Wei, D., Varshney, K. R., Dhurandhar, Wanko, C. E. T. (2020). Influence of artificial intelligence
A., & Tomsett, R. (2022). Deciding fast and slow: The role of (AI) on firm performance: The business value of AI-based
cognitive biases in AI-assisted decision-making. Proceedings transformation projects. Business Process Management Jour-
of the ACM on Human-computer Interaction, 6(CSCW1), 1- nal, 26(7), 1893-1924.
22. Weekman, K. (2024, March 1). 'Terrifying' Willy Wonka event
Robertson, A. (2024). Google apologizes for ‘missing the mark’ used AI images to mislead guests. Here's how to avoid getting
after Gemini generated racially diverse Nazis. The Verge. duped by fake marketing photos. Yahoo News. https://
https://www.theverge.com/2024/2/21/24079371/google-ai- www.yahoo.com/news/terrifying-willy-wonka-event-used-ai-
gemini-generative-inaccurate-historical images-to-mislead-guests-heres-how-to-avoid-getting-duped-
by-fake-marketing-photos-191205992.html
Robinson, K. (2024, May 30). Google defends AI search results
after they told us to put glue on pizza. The Verge. https:// Wu, Z., Ji, D., Yu, K., Zeng, X., Wu, D., & Shidujaman, M.
www.theverge.com/2024/5/30/24168344/google-defends-ai- (2021). AI creativity and the human-AI co-creation model. In
overviews-search-results Human-Computer Interaction. Theory, Methods and Tools:
Thematic Area, HCI 2021, Held as Part of the 23rd HCI In-
Silvia, P. J., Winterstein, B. P., Willse, J. T., Barona, C. M., ternational Conference, HCII 2021, Virtual Event, July 24–
Cram, J. T., Hess, K. I., Martinez, J. L., & Richard, C. A. 29, 2021, Proceedings, Part I 23 (pp. 171-190). Springer
(2008). Assessing creativity with divergent thinking tasks: International Publishing.
Exploring the reliability and validity of new subjective scor-
ing methods. Psychology of Aesthetics, Creativity, and the Yehiav, G. (2023, August 8). Will AI Augment Or Replace
Arts, 2(2), 68–85. https://doi.org/10.1037/1931-3896.2.2.68 Workers?. Forbes. https://www.forbes.com/sites/
forbestechcouncil/2023/08/08/will-ai-augment-or-replace-
workers/
 Journal of Behavioral and Applied Management 122

Zhou, E., & Lee, D. (2024). Generative artificial intelligence,

human creativity, and art. PNAS nexus, 3(3), 1-8.
Zia, A. (2023, September 3). 18 biggest generative AI compa-
nies in the world. Yahoo Finance. https://finance.yahoo.com/
news/18-biggest-generative-ai-companies-140447246.html

────────────────────────────────────
Charu Sinha ([email protected])

Veselina Vracheva ([email protected])

Cristina Nistor ([email protected])

────────────────────────────────────