Professional Documents
Culture Documents
Expe (5)
Expe (5)
ScienceDirect
w w w. j o u r n a l s . e l s e v i e r. c o m / b u s i n e s s - h o r i z o n s
a
University of North Carolina at Chapel Hill, Chapel Hill, NC, U.S.A.
b
California Polytechnic State University, San Luis Obispo, CA, U.S.A.
c
The University of Sydney Business School, Sydney, Australia
KEYWORDS Abstract Emerging artificial intelligence (AI) capabilities will likely pervade
Artificial intelligence; nearly all organizational contours and activities, including knowledge management
Deep learning; (KM). This article aims to uncover opportunities associated with the implementa-
Knowledge tion of emerging systems empowered by AI for KM. In doing so, we explicate the po-
management; tential role of AI in supporting fundamental dimensions of KM: creation, storage and
Knowledge scientists; retrieval, sharing, and application of knowledge. We then propose practical ways to
AI champions build the partnership between humans and AI in supporting organizational KM activ-
ities and provide several implications for the development and management of AI
systems based on the components of people, infrastructures, and processes.
ª 2022 Kelley School of Business, Indiana University. Published by Elsevier Inc. This
is an open access article under the CC BY license (http://creativecommons.org/
licenses/by/4.0/).
https://doi.org/10.1016/j.bushor.2022.03.002
0007-6813/ª 2022 Kelley School of Business, Indiana University. Published by Elsevier Inc. This is an open access article under the CC
BY license (http://creativecommons.org/licenses/by/4.0/).
88 M.H. Jarrahi et al.
learning. The most effective deep-learning algo- broach possible ways through which a synergistic
rithms often take a supervised approach, in which partnership between humans and artificial agents
large amounts of labeled data are employed to in organizational KM can be achieved.
train the connection strengths between nodes in a
massive, layered computational network, such 2.1. Knowledge creation
that patterns in the training data are used to make
accurate predictions on future unseen data The process of knowledge creation may occasion-
(Brynjolfsson & Mitchell, 2017). This approach ally involve developing ideas and solutions from
departs from traditional KM systems, such as scratch, but it more frequently amounts to a
expert systems, which leveraged a symbolic logic reconfiguration and recombination of already
in which rules were articulated and provided by existing background knowledge. This enables or-
humans to the system (Pushpa, 2019). ganizations to adapt to new situations (Bhatt,
As both AI and KM are inexorably bound up with the 2001; Davenport & Prusak, 1998). Additionally,
natures of knowledge and of learning, recent ad- firms may acquire knowledge from external sour-
vances in AI can provide new foundations for trans- ces (Alavi & Denford, 2011). As a result, knowledge
forming KM in organizations (Sanzogni et al., 2017). creation may take the form of knowledge acqui-
There are two complementary techno-organizational sition through searching for or sourcing
orientations in this space: (1) KM, which is directly information.
concerned with managing knowledge in organiza- The potential of deep-learning AI for knowledge
tions, and (2) AI, seen as a branch of computing, which creation lies in its predictive power in situations
is primarily focused on developing systems that can such as forecasting sales probabilities (Eitle &
mimic human knowledge and learning activities. Buxmann, 2019). Agrawal et al. (2017, p. 23)
Firms should examine the potential roles for new described this facet as “the ability to take infor-
AI systems in supporting organizational KM activities mation you have and generate information you
because of the intuitive connection between the didn’t previously have.” As a result of this capacity
two. There are lessons to be learned from traditional to derive their own rules based on similar patterns
rule-based KM; however, given the characteristics of in the data, deep-learning AI can discover patterns
contemporary deep learning as a rather different in the available data sets previously unknown to
way of computing, we need fresh insight into the the organization. For instance, AI can help raise
potential relationship between AI and KM. Thus, in new questions and develop new sets of declarative
this article, we envisage and identify such poten- knowledge in specific domains by connecting var-
tials. In doing so, we adopt a balanced perspective iables in new ways. This is only possible through
that does not put humans and AI in conflict but em- self-learning analytical capacities and features for
phasizes their partnership. Hence, this perspective pattern recognition, which allow organizations to
focuses on the partnership between humans and AI harness big data in unprecedented ways (Faraj
and on the unique capabilities of each for KM. In this et al., 2018).
context, deep-learning AI augments and improves As an example of the application of AI in this
people’s lived experience, rather than claiming to context, a deep-learning approach was recently
replace or surpass human intellect. employed to compute analogies between various
We aim to articulate the role of AI in supporting terms in 3.3 million abstracts of materials science
KM; in doing so, we build on the typology of KM research articles published over 96 years
processes presented by Alavi and her colleagues (Tshitoyan et al., 2019). Not only could the algo-
(Alavi & Denford, 2011; Alavi & Tiwana, 2003). rithm independently capture complex materials
According to this framework, KM in organizations science concepts (e.g., the structure of the peri-
helps the delivery of products and services by odic table), but it also identified unnoticed cor-
creating, storing and retrieving, sharing, and relations of materials with promising functional
applying knowledge. These processes are essential applications (e.g., compounds with similar ther-
for an organization to learn, reflect, and develop moelectric properties). Similar techniques can
core competencies to sustain its competitive edge contribute to organizational KM as many organi-
in the knowledge-based economy (e.g., Davenport zations collect and store much more knowledge
& Prusak, 1998). than they can use. One report estimates that 60%e
70% of data collected by enterprises go unused
2. Potential applications of AI in KM (Gualtieri, 2016). AI capabilities can similarly help
organizations find unknown and unexpected con-
In what follows, we discuss some potential impli- nections and insights from data sets containing
cations of AI systems for KM (Table 1) and then interactions with customers (e.g., call
Artificial intelligence and knowledge management 89
transcripts), email exchanges, enterprise chat data that were previously deemed unwieldy and
systems, discussions on social network sites, and difficult to parse (Paschen, Wilson, & Ferreira,
records on customer relationship management 2020). For example, AI can search for, organize,
(CRM) databases (Lavenda, 2019; O’Dell & and summarize previous legal precedents relevant
Davenport, 2019). to a new case (Bhattacharya et al., 2019).
Furthermore, AI can analyze multiple channels of
2.2. Knowledge storage and retrieval content and communication, generate summaries,
ascertain relevant topics (to emerging problems),
Another key function of KM is to create and isolate proprietary and confidential knowledge,
maintain an organizational memory that tracks and present reusable insights that are immediately
generated and acquired knowledge resources. applicable to new situations (O’Dell & Davenport,
Effective storage and retrieval strategies are one 2019). For instance, Google’s Gmail algorithms
of the primary mechanisms for preserving organi- recognize implicit social groups and recommend
zational memory (Alavi & Leidner, 2001). This potential email recipients when creating an email.
process is specifically focused on extracting In this way, deep-learning AI learns from in-
knowledge, making it explicit, and recording it dividuals’ or teams’ recurrent KM or communica-
systematically for future uses, often through the tion practices (e.g., history of emails) and offers
implementation of knowledge repositories. individualized solutions. For example, the system
The utility of AI in KM is most evident in can learn over time which documents or messages
enhancing the storage and retrieval of explicit to retain or to bring to the attention of the
knowledge. Since deep-learning AI holds a consti- knowledge worker, as well as what customer pur-
tutive relationship with big data (Brynjolfsson & chase history and supply-chain issues need to be
McAfee, 2017), these data-driven, self-learning retrieved for specific troubleshooting team meet-
algorithms open new possibilities for harvesting, ings. This output can increase productivity for
classifying, organizing, storing, and retrieving big many knowledge workers, who spend nearly 32
data being generated in organizations, including working days a year retrieving bits and pieces from
90 M.H. Jarrahi et al.
various documents and spreadsheets or switching may be able to more accurately and equitably
between folders in daily work practices (Lavenda, reward sources of expertise and knowledge
2019). contribution.
Distributing knowledge throughout the organiza- Knowledge application refers to putting knowledge
tion is a prerequisite to applying it effectively in into practice after it has been retrieved or shared.
problem-solving and decision-making. But knowl- It often involves repackaging available knowledge
edge sharing is often plagued by temporal, spatial, resources (e.g., a set of best practices) into
and, more evidently, functional barriers. As a applicable solutions or delivering new products
result, in many organizations, knowledge sharing and services to a new context (Bhatt, 2001).
tends to be local and fragmentary. AI can Knowledge application in many cases hinges upon
contribute to breaking down these organizational a reinvention process, which “is not merely about
silos in two ways: (1) by bringing together people tweaking the ideas of others. [But] It requires the
who are working on the same issues but are skillful selection, analysis, and assimilation of the
separated by different boundaries or geographies, right external knowledge.[matched] to the right
and (2) by creating more connected systems of local needs at the right time” (Cranefield &
coordination, which give managers a better sense Prusak, 2016, p. 100). IT can facilitate the pro-
of knowledge constraints. cess of knowledge application by providing faster,
One of the key challenges of knowledge sharing more effective access to knowledge resources, as
is overcoming silos to connect disparate people well as by codifying and automating routines that
with the practices and knowledge that they need help workers apply and integrate specialized
(Jarrahi, 2019). AI approaches can redress this by knowledge (Alavi & Leidner, 2001).
discovering and strengthening weak ties, thus Emerging tools powered by AI, such as intelli-
facilitating community-based learning. For gent assistants, can enhance knowledge retrieval
example, intelligent KM systems used by MITREda and representation needs for situated knowledge
federally funded research organization in the application (Maedche et al., 2019). This addresses
U.S.daggregates knowledge concerning where one of the nontrivial challenges of knowledge re-
workers are present, when they are active, and positories concerning knowledge application:
what they are doing to proactively connect people “getting the right knowledge to front line workers
working on similar projects or technical problems in real-time” (Davenport, 2019). Traditional
(O’Dell & Davenport, 2019). In this context, the knowledge repositories are typically challenging to
most fundamental use of AI for knowledge sharing search through when a quick answer is needed. By
pertains to its contribution toward collaborative contrast, intelligent assistants can make what is
intelligencedsomething not attainable by tradi- already known readily available for processing
tional database systems. AI can help animate through AI-enabled storage/retrieval strategies,
creative thinking, can create a shared memory like automatic content cleaning, classification,
among various team members, and can facilitate and tagging. Repsol, a global energy and utility
feedback and peer review. These smart features company, used AI to whittle 5 million crude oil
are increasingly embedded into enterprise or per- production scenariosd100 million data points a
sonal communication systems such as Slack or dayddown to a manageable group that its engi-
Yammer, providing more than just a communica- neers could then evaluate. The accessibility of this
tion channel for connecting workers. timely, contextualized knowledge to its engineers
AI systems can likewise generate dynamic social helped Repsol achieve a 40%e50% reduction in
graphs that capture interconnections between nonproductive time across 30 drill sites
people and teams to provide a comprehensive (Ransbotham, Khodabandeh, Kiron, et al., 2020).
perspective about knowledge sources and bottle- Situated knowledge offered by AI can be particu-
necks in the organization (Jarrahi, 2019). For larly valuable in customer service, where a
example, Campaign 360, offered by People.ai, knowledge worker needs immediate access to
enables businesses to gain real-time insight into previous cases and a summary of the issue at hand.
selling opportunities created by various marketing As an example, Talla, an AI-based service knowl-
campaigns, enabling more cohesive coordination edge automation platform, cleans up, searches
and sharing between marketing channels and sales through, and classifies manuals and online docu-
pipelines (Latinovic & Chatterjee, 2019). In addi- mentation, finding questioneanswer pairs to
tion, via enterprise social graphs, organizations manage service knowledge (Davenport, 2019).
Artificial intelligence and knowledge management 91
AI systems can also provide more interactive 3.1. Personal intelligent assistants and
and intuitive interfaces that may converse with or personal KM
debate the knowledge worker using everyday lan-
guage. For example, natural language processing is KM for organizations embodies an important
a type of AI that enables machines like digital as- dimension of personalization in which the focus is
sistants or chatbots to understand and simulate on individual workers and their “personal enquiry”
human conversations (O’Dell & Davenport, 2019; (Pauleen & Gorman, 2016). For that reason, con-
Raghavan, 2019). These more natural and human- ceptualizations of knowledge often recognize the
oid ways of interacting with the retrieval mecha- centrality of each worker in defining the concept
nisms can facilitate knowledge application. of knowledge. For example, Davenport and Prusak
Specifically, chatbots can provide knowledge and (1998, p. 5) define knowledge as “a fluid mix of
aid in its application through a natural conversa- framed experience, values, contextual informa-
tion with humans. This can reduce social barriers tion, and expert insight that provides a framework
that impede equal access to organizational for evaluating and incorporating new experiences
knowledge. A compelling example here comes and information.”
from the use of chatbots in higher education to An emerging genre of AI systems called personal
help students access organizational knowledge and intelligent assistants can occupy a unique position
feel more supported by the university (Agrawal, in personal KM. Information overload is one of the
2021). The chatbot allows students to ask ques- key challenges of the information environment for
tions in a casual tone, as if texting with a friend, knowledge workers (Pauleen & Gorman, 2016).
and to receive relevant answers. Because students Personal intelligent assistants can help broaden
were interacting with a machine, rather than a the cognitive bandwidth of knowledge workers and
human, they could ask sensitive questions about change the way they digest relevant knowledge by
finances and mental health without fear of po- providing more effective capabilities for process-
tential social costs. ing, filtering, sorting, and navigating information
resources (Maedche et al., 2019).
Personal intelligent assistants are self-learning,
3. HumaneAI symbiosis in KM interactive systems that can learn and refine their
interactions with the user over time. In doing so,
As we outlined above, AI systems will likely they can learn each worker’s dynamic knowledge
percolate throughout the KM foundation of orga- needs, workflows, preferences, and feedback to
nizations, but it is important to bear in mind that provide individualized solutions. For example, at
knowledge production and management are one financial institution, personal intelligent as-
inherently human-centered (Davenport & Prusak, sistants are deployed alongside traders to assist in
1998; Sanzogni et al., 2017). Therefore, the most managing the information overload and to broaden
effective roles assigned to AI in KM will mostly workers’ cognitive bandwidth (Ransbotham,
augment humans rather than replace them, Khodabandeh, Fehling, et al., 2020). After
thereby achieving collaborative intelligence, in observing patterns in the traders’ data-monitoring
which AI and humans enhance each other’s com- practices, decision-making, and results of their
plementary strengths (Paschen, Wilson, & trades, these personal intelligent assistants
Ferreira, 2020; Wilson & Daugherty, 2018). This learned to support traders by bringing relevant
relationship can be captured through the concept information to their attention in real time. This
of humaneAI symbiosis. In the rest of this article, timely information enables better decisions on the
we outline some potential scenarios and practical part of the traders and represents a continually
ways for building this partnership in KM (Figure 1). improving partnership, as the personal intelligent
In Figure 1, the application of personal intelligent assistant receives additional feedback both each
assistants in KM mostly concerns storing and time a trader dismisses or acts on the information
retrieving knowledge; the discussion of specialized it provides and from the outcome of that trade.
or general intelligence focuses on knowledge- Humans are still central in shaping the effec-
creation processes; codification or collaboration tiveness of personal intelligent assistants, as these
of knowledge has to do with knowledge-sharing systems thrive on the worker’s feedback and
processes; and the discussion of know-how, interaction with the information they provide. In
know-what, and know-why is primarily about other words, personal assistants must undergo
knowledge application. continual training by humans to be able to display
92 M.H. Jarrahi et al.
complex and subtle human traits (Wilson & elsewhere is largely task-centered and cannot
Daugherty, 2018). After all, these tools are assis- easily transfer across different contexts.
tants that are designed and trained to augment General intelligence remains a human-centered
individual workers, and augmentation involves a characteristic, as “we remain very far from artifi-
mutual learning process through which both cial general intelligence. Machines cannot do the
humans and AI learn from each other and hence full range of tasks that humans can do”
coevolve (Raisch & Krakowski, 2020). A heedless (Brynjolfsson & Mitchell, 2017 p. 1530). Specifically,
approach to the use of personal intelligent assis- the application of knowledge for strategic-level
tants can result in what researchers call cognitive thinking and decision-making requires elements of
complacency, whereby workers put too much trust general intelligencedwhich tend to be more holi-
in their artificial partners without mindfully sticdand builds from the uniquely human pre-
analyzing the outcomes (Jarrahi, 2019). rogatives such as foresight, social and emotional
intelligence, self-development, imagination, and
3.2. Specialized intelligence versus general curiosity. Strategic-level decisions can be data-
intelligence driven but are rarely repetitive and mundane.
Furthermore, unlike data manipulation and in-
AI presents specialized intelligence that enables formation management, KM entails a crucial
sensing the environment, learning from experi- judgment component (Davenport & Prusak, 1998).
ence, and creating possibilities for action in rela- Judgment is a profoundly human-centered capa-
tion to specific task contexts. For example, it can bility; humans can judge a new situation, consider
store, retrieve, and analyze explicit forms of what is known, and decide the best course of ac-
knowledge (e.g., images, text data) and sensibly tion by holistically weighing the social and orga-
respond to customer inquiries about the compati- nizational ramifications of various options.
bility of their heating system with smart thermo- Whereas AI systems provide a powerful lever for
stats (e.g., Google Nest). But AI’s role in KM and storing, retrieving, and managing knowledge
Artificial intelligence and knowledge management 93
content, humans are well-positioned to navigate transferred through social interactions (Hansen
and discern knowledge contextdthat is, the spe- et al., 1999). Malone (2018) summed up the po-
cific circumstances that surround the knowledge tential roles of AI in connecting people in new
problem. Knowledge application activities require ways: “the most valuable contributions of com-
an understanding of the peculiarities of the new puters will not just be their AI but also their ability
context in which knowledge is put into use, its to provide hyperconnectivitydconnecting people
relationship with the contexts of discovery, and to other people in rich new ways.” AI technologies
what can be transferred between the two con- provide great capabilities for generating know-
texts. Such a contextualization process typically who (i.e., sources of expertise) within and across
requires a human’s general intelligence. organizational boundaries and for extending and
Finally, a central attribute of general intelli- augmenting social networks that serve as conduits
gence is self-reflection. Self-reflection perpetu- of knowledge. But transferring tacit knowledge
ates the need for self-awareness about one’s remains a highly human-centered practice,
knowledge regarding areas of strength and weak- gleaned through person-to-person social in-
ness. This undergirds the capacities to learn and to teractions and informal relationships such as ap-
reinvent, which are germane to knowledge- prenticeships, enabling workers to engage in
application processes (Jarrahi et al., 2019). practice-centered knowledge encounters. For this
Compared with humans, AI systems lack self- reason, attempts to turn inherently tacit knowl-
awareness and self-reflection. Human decision edge into explicit knowledge and to facilitate its
makers can sustain KM systems by emphasizing transfer through technological mechanisms have
their unique capabilities in observing and reflect- failed in the past (Orlikowski, 2002).
ing on past experiences and in developing correc- One example of an AI system that harnesses the
tive strategies, as well as in formulating problems unique contributions of both humans and machines
that are worthy of attention (von Krogh, 2018). is Swarm AI. Modeled after the kinds of swarm
intelligence found in biological systems, this plat-
3.3. Codification versus collaboration form enables groups of humans to collectively
converge on a decision in real time (Metcalf et al.,
Previous work (e.g., Alavi & Tiwana, 2003; Hansen 2019). The AI algorithm trains on human behaviors
et al., 1999) highlights two dimensions of KM that made through the platform to implicitly determine
can be supported by IT uses: (1) codification of the confidence each individual has for a decision.
knowledge and (2) human collaboration (person- The dynamic group preference is then visualized in
alization). While codification aims to deal with real time, allowing participants to apply their tacit
explicit aspects of KM, collaboration concerns tacit knowledge and to engage in reflection as they
elements of functional KM. Here, we rethink the update their preferences. Forecasts made on this
utility of new waves of AI in relation to these platform have been shown to outperform ML,
dimensions. prediction markets, and online crowds.
learning come with high levels of complexity and on the three complements of people, infrastruc-
low levels of interpretability, making it difficult, if ture, and process (Figure 2).
not impossible, to explain the inferences gener-
ated by AI (Maedche et al., 2019). The reasons for 4.1. People
an AI algorithm providing a particular recommen-
dation exist in an unknowable black box. The po- 4.1.1. Elevate humans in KM
tential for spurious relationships to hide within Metaphors such as “human-in-the-loop” AI or “the
opaque AI outputs accentuates the need for last mile problem” in AI-empowered automation
testing the inferences to ensure their validity. But fall short of capturing the key role workers can
an AI algorithm cannot evaluate its own inferences play. These formulations tend to present humans
because it cannot contextualize and ground them as an afterthought, as an appendage to the tech-
in meaning (Mitchell, 2019). The role of humans as nological AI system, or as a roadblock to full
explainers in their partnership with intelligent automation. Instead, we argue for a symbiotic
machines is even more salient in evidence-based relationship, one that both recognizes the irre-
industriesdsuch as medical and legal, financial, placeable contributions of humans to knowledge
or public administration contexts with more strict work and that seeks ways to reinvent and elevate
compliance requirementsdwherein there are their role. No one is more central to AI success
clear obligations for explaining how AI systems than the workers who work at the front line of
may weigh inputs they receive into certain rec- organizations and who integrate AI into their work
ommendations (Wilson & Daugherty, 2018). As practices.
such, the role of humans is indispensable in One way in which AI can elevate knowledge
formulating know-why for AI-based inferences; workers is through reskilling and upskilling. As
know-why is essential for alleviating the black box human skills and activities become more visible to
of AI, justifying decisions, training budding human AI systems, organizations can harness this knowl-
experts, and garnering organizational support. edge to provide targeted training and develop-
ment. For example, EdCast uses AI to personalize
training programs for individuals by comparing
4. Practical implications their current skills with future labor market needs
(Caine & Firth-Butterfield, 2020). Realizing the
A recent survey indicates that a vast majority of AI promise of AI in organizations will likely force
investments bring about minimal or no impact knowledge workers, including knowledge scientists
(Ransbotham, Khodabandeh, Fehling, et al., 2020). and AI champions, to adopt new roles in the or-
Years of research indicate that for an IT deploy- ganization. We explain knowledge scientists and AI
ment to be successful, there need to be accompa- champions in the following sections.
nying organizational changes. These are referred to
as organizational complements (Brynjolfsson et al., 4.1.2. Train knowledge scientists
2017). So the value of AI for KM lies not only in Recent developments in explainable AI (XAI) or
technology but in new infrastructures, trained neural-symbolic learning advocate for new roles,
people, and redesigned processes. We discuss the such as knowledge scientists and data scientists,
practical implications of AI and its role in KM based who collect and prepare training data sets for ML
generate a comprehensive milieu for AI- can draw from its understanding of the relationships
empowered analysis. between objects to explain the reasons behind their
recommendations.
4.2.2. Facilitate interpretability and
accountability 4.3. Processes
AI technology itself also needs to be adapted for
organizational rollout. Many of the AI capabilities A major complementary investment for successful
developed in academic and corporate labs may not AI deployment has to do with aligning ways of
be ready to be deployed in knowledge practices working with AI. A process-centered approach re-
owing to the issues of interpretability and quires shifting from a mindset of using AI to auto-
accountability. The most prevalent machine- mate processes to seeing it as an opportunity to
learning models in businesses today are still su- augment or reinvent processes.
pervised learning models, in part because the la-
beling of their training data enhances the 4.3.1. Pursue mutual learning
interpretability and accountability of the model. Process redesign must both draw on and facilitate
One significant cost that businesses must incur in mutual learning between humans and AI. Designing
producing these models is the labor-intensive for mutual learning recognizes the limits of the AI
cleansing and preparation required to implement system in managing knowledge and precipitates
a minimum viable AI product using proprietary the need for constant auditing and involvement of
data (Davenport & Seseri, 2020). The granular human supervisors. Datafication and digitalization,
manner in which data must be tagged to be coupled with the use of AI, can give the wrong
effective for supervised learning models takes a impression of authentic knowledge, but over-
unified, cross-functional effort. estimating or trusting too much in the learning
capacities of AI may thwart the real value of these
4.2.3. Develop knowledge graphs systems for KM. For example, AI systems can carry
One of the key challenges of leveraging knowledge or even amplify implicit biases embedded in
has to do with the fact that data are dynamically existing organizational systems. The knowledge
being generated in real time, and most such data that these systems generate, share, or propagate
are unstructured. This presents barriers to timely throughout the organization can provide the
and meaningful AI-empowered analysis. Knowledge impression of objectivity or cogency even when
graphs are an emerging way in which organizations the system is as biased as the human thinking it
can harness this data. Knowledge graphs manage replaces. Hence, organizations need to undertake
the key concepts, terms, and entities and their continual AI audits (Brown et al., 2021).
relationships in the business. In doing so, a knowl-
edge graph produces an understanding of the re- 4.3.2. Form cross-functional teams
lationships between data points, allowing for In addition to redesigning processes to foster
complex integration, discovery, and analysis. For collaboration between AI and knowledge workers,
example, a knowledge graph might be constructed AI capabilities themselves can also be used to re-
to understand the complex relationships between engineer existing organizational processes
regions, weather, cost, and manufacturing, allow- (Daugherty & Wilson, 2018). As noted, algorithmic
ing the business to learn how inclement weather systems thrive on comprehensive data that span
will impact the cost of a part produced in Texas. multiple units. Providing this data for AI systems
Many businesses already use knowledge graphs for requires organization-wide initiatives. Interdisci-
AI-enabled KM. AstraZeneca uses knowledge graphs plinary and cross-functional teams could facilitate
to inform drug discovery through exploring a network these initiatives and connect business domain
of known relationships between chemical com- proficiency with algorithmic capabilities. These
pounds, molecules, diseases, and biological entities teams provide a mix of skills and perspectives that
(Dhuri, 2020). AI systems build on knowledge graphs, are critical to the implementation of AI systems.
but at the same time, their role in creating knowl- The teams involve AI and analytics experts working
edge graphs is essential as they can dynamically closely with domain experts, frontline workers,
assign meaning to the relationships between various and operational people.
data points. In other words, knowledge graphs can The redesign of workflows (e.g., digitalization,
overcome the limitations of dynamic, unstructured, datafication, and quantification) and the identifica-
and varied sources of data typically present in orga- tion of ways in which algorithms’ recommendations
nizations (Elnagar & Weistroffer, 2019). Knowledge can augment various knowledge activities require a
graphs are particularly useful for KM because they continual conversation and negotiation between
Artificial intelligence and knowledge management 97
technology and domain experts. Such a process partners for KM while avoiding such pitfalls of
cannot be performed top-down; AI will touch upon automation as cognitive complacency or algo-
many processes and people (Paschen, Pitt, & rithmic aversion. Such preparations by organiza-
Kietzmann, 2020). Knowledge workers have a tions help put into practice the unique capabilities
unique perspective about their core activities and of AI in KM, which are only utilized and realized
must therefore feel empowered to make decisions, through an effective symbiotic partnership be-
to decide on how to integrate algorithms, and sub- tween knowledge workers and intelligent systems.
sequently to develop trust in their applications.
Forum. Available at https://www.weforum.org/agenda/ Jarrahi, M. H., Philips, G., Sutherland, W., Sawyer, S., &
2020/10/ai-jobs/ Erickson, I. (2019). Personalization of knowledge, personal
Canhoto, A. I., & Clear, F. (2020). Artificial intelligence and knowledge ecology, and digital nomadism. Journal of the
machine learning as business tools: A framework for diag- Association for Information Science and Technology, 70(4),
nosing value destruction potential. Business Horizons, 63(2), 313e324.
183e193. Jarrahi, M. H., & Sutherland, W. (2019). Algorithmic manage-
Cranefield, J., & Prusak, L. (2016). Managing your own knowl- ment and algorithmic competencies: Understanding and
edge: A personal perspective. In D. J. Pauleen, & appropriating algorithms in gig work. Available at https://
G. E. Gorman (Eds.), Personal knowledge management (pp. www.researchgate.net/publication/329336053_
121e136). Abingdon, UK: Routledge. Algorithmic_Management_and_Algorithmic_Competencies_
Dan, W., Elliott, T., & Noga, M. (2017, May 31). Eight ways Understanding_and_Appropriating_Algorithms_in_Gig_work
machine learning is improving companies’ work processes. Joshi, A., & Wade, M. (2020, August 10). The building blocks of
Harvard Business Review. Available at https://hbr.org/ an AI strategy. MIT Sloan Management Review. Available at
2017/05/8-ways-machine-learning-is-improving-companies- https://sloanreview.mit.edu/article/the-building-blocks-
work-processes of-an-ai-strategy/
Daugherty, P., & Wilson, J. (2018). Processes reimagined. Kaplan, A., & Haenlein, M. (2019). Siri, Siri, in my hand: Who’s
Accenture. Available at https://www.accenture.com/_ the fairest in the land? On the interpretations, illustrations,
acnmedia/pdf-76/accenture-process-reimagined.pdf and implications of artificial intelligence. Business Horizons,
Davenport, T. (2019, October 10). Managing support knowledge 62(1), 15e25.
with AI: Talla helps Toast. Forbes. Available at https:// Latinovic, Z., & Chatterjee, S. C. (2019, September 24). How AI
www.forbes.com/sites/tomdavenport/2019/10/10/ is helping companies break silos. MIT Sloan Management
managing-support-knowledge-with-ai-talla-helps-toast/ Review. Available at https://sloanreview.mit.edu/article/
#4d88ade77267 how-ai-is-helping-companies-break-silos/
Davenport, T., & Mahidhar, V. (2018). What’s your cognitive Lavenda, D. (2019, August 28). Machine learning should make
strategy? MIT Sloan Management Review, 59(4), 19e23. tech work for usdNot the other way around. TechTalks.
Davenport, T., & Prusak, L. (1998). Working knowledge: How Available at https://bdtechtalks.com/2019/08/28/
organizations manage what they know. Boston, MA: Harvard machine-learning-data-management/
Business Press. Maedche, A., Legner, C., Benlian, A., Berger, B., Gimpel, H.,
Davenport, T., & Seseri, R. (2020, December 15). What is a Hess, T., Hinz, O., Morana, S., & Sollner, M. (2019). AI-based
minimum viable AI product? MIT Sloan Management Review. digital assistants: Opportunities, threats, and research per-
Available at https://sloanreview.mit.edu/article/what-is-a- spectives. Business and Information Systems Engineering,
minimum-viable-ai-product/ 61(4), 535e544.
Dhuri, S. (2020, July 29). How AstraZeneca is using Netflix like Malone, T. W. (2018). How human-computer ‘superminds’ are
knowledge graph to discover new drugs. Medium. Available at redefining the future of work. MIT Sloan Management Re-
https://medium.com/@siddhesh.s.dhuri/how-astrazeneca- view, 59(4), 34e41.
is-using-netflix-like-knowledge-graph-to-discover-new-drugs- Metcalf, L., Askay, D. A., & Rosenberg, L. B. (2019). Keeping
a6cd187e5b10 humans in the loop: Pooling knowledge through artificial
Doran, D., Schulz, S., & Besold, T. R. (2017). What does explainable swarm intelligence to improve business decision making.
AI really mean? A new conceptualization of perspectives. California Management Review, 61(4), 84e109.
Available at https://arxiv.org/pdf/1710.00794.pdf Mitchell, M. (2019). Artificial intelligence hits the barrier of
Duffy, R. (2019, January 31). AI and robotics made 10 years of meaning. Information: An International Interdisciplinary
steady gains. Morning Brew. Available at https://www. Journal, 10(2), 51.
morningbrew.com/emerging-tech/stories/2019/12/20/ai- O’Dell, C., & Davenport, T. (2019). Application of AI for
robotics-make-10-years-steady-gains?fbclidZIwAR0HpX-IxXr knowledge management. CIOReview. Available at https://
xhZId8DcESvOOkEXcwUjmBdBxaHFtQi1WYjh98HL0Tpnw1OY knowledgemanagement.cioreview.com/cxoinsight/
Eitle, V., & Buxmann, P. (2019). Business analytics for sales application-of-ai-for-knowledge-management-nid-30328-
pipeline management in the software industry: A machine cid-132.html
learning perspective. In Proceedings of the 52nd Hawaii in- Orlikowski, W. J. (2002). Knowing in practice: Enacting a col-
ternational conference on system sciences. Available at lective capability in distributed organizing. Organization
https://aisel.aisnet.org/hicss-52/ Science, 13(3), 249e273.
Elnagar, S., & Weistroffer, H. R. (2019). Introducing knowledge Paschen, U., Pitt, C., & Kietzmann, J. (2020). Artificial intelli-
graphs to decision support systems design. In Information gence: Building blocks and an innovation typology. Business
systems: Research, development, applications (pp. 3e11). Horizons, 63(2), 147e155.
Cham, Switzerland: Springer. Paschen, J., Wilson, M., & Ferreira, J. J. (2020). Collaborative
Faraj, S., Pachidi, S., & Sayegh, K. (2018). Working and orga- intelligence: How human and AI create value along the B2B
nizing in the age of the learning algorithm. Information and sales funnel. Business Horizons, 63(3), 403e414.
Organization, 28(1), 62e70. Pauleen, D. J., & Gorman, G. E. (2016). The nature and value of
Gualtieri, M. (2016, January 21). Hadoop is data’s darling for a personal knowledge management. In Personal knowledge
reason. Forrester. Available at https://go.forrester.com/ management (pp. 23e38). Abingdon, UK: Routledge.
blogs/hadoop-is-datas-darling-for-a-reason/ Pushpa, R. (2019, June 16). Artificial intelligence and knowledge
Hansen, M. T., Nohria, N., & Tierney, T. (1999). What’s your managementeUnderstanding how they are linked. LinkedIn
strategy for managing knowledge? The Knowledge Manage- Blogs.Availableathttps://www.linkedin.com/pulse/artificial-
ment Yearbook 2000e2001, 77(2), 106e116. intelligence-knowledge-management-how-linked-pushpa/
Jarrahi, M. H. (2019). In the age of the smart artificial intelli- Raghavan, S. (2019). 2020 AI predictions from IBM Research. IBM
gence: AI’s dual capacities for automating and informating Research. Available at https://www.ibm.com/blogs/
work. Business Information Review, 36(4), 178e187. research/2019/12/2020-ai-predictions/
Artificial intelligence and knowledge management 99
Raisch, S., & Krakowski, S. (2020). Artificial intelligence and Tshitoyan, V., Dagdelen, J., Weston, L., Dunn, A., Rong, Z.,
management: The automation-augmentation paradox. Kononova, O., Persson, K. A., Ceder, G., & Jain, A. (2019).
Academy of Management Review, 46(1), 192e210. Unsupervised word embeddings capture latent knowledge
Ransbotham, S., Khodabandeh, S., Fehling, R., LaFountain, B., from materials science literature. Nature, 571(7763),
& Kiron, D. (2020, October 15). Winning with AI. MIT Sloan 95e98.
Management Review. Available at https://sloanreview.mit. Tsui, E. (2005). The role of IT in KM: Where are we now and
edu/projects/winning-with-ai/ where are we heading? Journal of Knowledge Management,
Ransbotham, S., Khodabandeh, S., Kiron, D., Candelon, F., 9(1), 3e6.
Chu, M., & LaFountain, B. (2020, October 20). Expanding AI’s von Krogh, G. (2018). Artificial intelligence in organiza-
impact with organizational learning. MIT Sloan Management tions: New opportunities for phenomenon-based theo-
Review. Available at https://sloanreview.mit.edu/projects/ rizing. Academy of Management Discoveries, 4(4),
expanding-ais-impact-with-organizational-learning/ 404e409.
Sanzogni, L., Guzman, G., & Busch, P. (2017). Artificial intelli- Wilson, H. J., & Daugherty, P. R. (2018). Collaborative intelli-
gence and knowledge management: Questioning the tacit gence: Humans and AI are joining forces. Harvard Business
dimension. Prometheus, 35(1), 37e56. Review, 96(4), 114e123.