Subscriber exclusive

Generative AI and the

future of work
The rise of generative AI is accelerating access to artificial intelligence across
industries, bringing with it risk—but also promise of great reward.

Produced by MIT Technology Review Insights, the custom content division of MIT Technology Review.
2  MIT Technology Review

What to know:

G
Generative AI and LLMs are increasing access enerative AI has enormous potential to
to AI for companies of all sizes and sparking the revolutionize business operations, but
beginnings of truly enterprise-wide AI. Powered by how companies decide to employ it will
the potential of newly emerging use cases, AI is finally make all the difference. Its full business
moving from pilot projects and “islands of excellence” value will only be achieved when it is used
to a generalized capability integrated into the fabric of thoughtfully to blend with human empathy and ingenuity.
organizational workflows, even in smaller companies.
Although the technology is still nascent, many
Automation anxiety should not be ignored, but generative AI use cases are starting to emerge. In
dystopian forecasts are overblown. Generative sales and marketing, generative AI can assist with
AI tools can already complete complex and varied creating targeted ad content, identifying leads,
workloads, but leaders and academics do not expect upselling, cross-selling, and providing real-time sales
large-scale automation threats. Instead, they believe analytics. When used for internal functions like IT,
the broader workforce will be liberated from time- HR, and finance, generative AI can improve help-desk
consuming work to focus on higher value areas of services, simplify recruitment processes, generate
insight, strategy, and business value. job descriptions, assist with onboarding and exit
processes, and even write code.
Unified and consistent governance are the rails on
which AI can speed forward. Generative AI brings One of AI’s great benefits for employees is its ability
commercial and societal risks, including protecting to take over mundane, rote, and time-consuming
commercially sensitive IP, copyright infringement, tasks. “Anything that’s repetitive and low-level can
unreliable or unexplainable results, and toxic content. be offloaded to AI,” says Ramandeep Randhawa,
When deploying generative AI in the enterprise, professor of data sciences and operations at USC
leaders must address its unique governance Marshall School of Business. This can improve
challenges, investing in technology, processes, and employee satisfaction, he says, since people are less
institutional structures. tied down by busywork.
3  MIT Technology Review

The emergence of consumer-facing generative AI tools Pervasive across industries

in late 2022 and early 2023 radically shifted public Thanks to its flexibility, range, and user-friendly, natural
conversation around the power and potential of AI. language–based interface, generative AI is showing
Although generative AI had been making waves among its mettle everywhere from copywriting to coding. Its
experts since the introduction of GPT-2 in 2019, its power and potential to revolutionize how work is done
revolutionary opportunities have now become clear to across industries and business functions suggests it
enterprises. will have a reverberating impact like that of the personal
computer, the internet, or the smartphone, launching
The impact of generative AI on economies and entirely new business models, birthing new industry
enterprise will be revolutionary. McKinsey Global leaders, and making it impossible to remember how we
Institute estimates that generative AI will add between worked before its spread (see Figure 1).
$2.6 trillion and $4.4 trillion in annual value to the
global economy, increasing the economic impact of AI AI is also venturing into creative applications, once
as a whole by 15% to 40%.1 The consultancy projects considered a uniquely human endeavor. Adobe’s Firefly,
that AI will automate half of all work between 2040 a family of creative generative AI models can act as
and 2060, with generative AI pushing that window a a copilot to creative and design workflows. Cynthia
decade earlier than previous estimates. Stoddard, the company’s senior vice president and
chief information officer says, “Generative AI evolves
Goldman Sachs predicts a 7%—or nearly $7 trillion— the possibility and promise of AI exponentially. You
increase in global GDP attributable to generative AI, can transform the conversation between the creator
and the firm expects that two-thirds of US occupations and the computer.” She continues, “But I don’t think it
will be affected by AI-powered automation.2 will ever replace humans. Instead, it’s going to be an
assistant to humans.”

Figure 1: Enterprise applications and use cases for generative AI

These are just a few of the business functions compellingly addressed by generative AI.

Generating forecasts for Automating and personalizing

complex scenarios customer service

Analyzing massive volumes of Learning from previous and

complex and unstructured data unresolved support tickets

Aggregating key metrics Providing intelligent scripts for

across production systems agent interactions

Automating creation of text, Developing personalized

charts, and graphs for reports Generative AI marketing

Analyzing operational concerns, Increasing precision and

such as inventory and staffing effectivenes of ad targeting

Optimizing pricing strategies Detecting security threats and

account takeover attempts
Auto-generating and adapting contracts,
purchase orders, and invoices Analyzing communications for phishing
attempts and social engineering
Understanding user preferences,
behaviors, and contextual cues

Source: Compiled by MIT Technology Review Insights, based on data from “Retail in the Age of Generative AI,” 3 “The Great Unlock: Large Language Models in
Manufacturing,”4 “Generative AI Is Everything Everywhere, All at Once,”5 and “Large Language Models in Media & Entertainment,”6 Databricks, April – June 2023.
4  MIT Technology Review

“Anything that’s repetitive and low-level can be offloaded to AI.”

Ramandeep Randhawa, Professor of Data Sciences and Operations, USC Marshall
School of Business

The energy and chemical industries are applying AI in Increasing access to AI

domains that had previously been inaccessible. Multi- Just as affordable enterprise technology, especially
industrials giant DuPont, for instance, had worked on cloud computing and software-as-a-service (SaaS)
chatbots for both employee and consumer interfaces solutions made possible the startup boom of the
previously, but found their inaccuracy too frustrating. last decade, generative AI is the next such wave of
enterprise-enabling technology. It allows even smaller
“Large language models (LLMs) now have the capability companies to build new products and scale them with
to achieve the necessary accuracy, and at a faster pace,” ease. An AI feature that would have once taken a team
says Andrew Blyton, vice president and chief information of engineers years to build can now be developed
officer at DuPont Water & Protection. The company is now in a few weeks, thanks to application programming
using AI in production scheduling, predictive reliability and interfaces (APIs) provided by LLMs.
maintenance, and sales price optimization applications.

High hopes for health care

Health and medicine, a deeply human-centered Schaefer is also excited about the power of AI
field, has also been a productive testing ground to alleviate health-care employee fatigue while
for AI. Generative AI is now showing its promise building their trust, via small, practical operational
as a powerful assistant to front-line staff. Natural improvements. The VA, for example, is currently
language processing tools, for instance, can working on an AI-driven project to reduce
transcribe and summarize medical notes, while unnecessary alarms and alerts in its medical
chatbots might be trained to answer consumer centers. Machine learning models can help by
medical questions. defining dynamic and patient-centered thresholds
for provider interventions, rather than relying on the
Richard Spencer Schaefer, chief health informatics alarm systems’ traditional and static guardrails and
officer at the Kansas City VA Medical Center, thresholds. “For this to work,” says Schaefer, “we
which manages health care for American military have to build trust and help health-care workers
veterans, and AI solutions architect at the VA’s understand, hey, this is what AI can do.”
National Artificial Intelligence Institute, is optimistic
that automation and predictive analytics could help
the sector. The VA validated a model that identified
the 24-hour risk of a patient admitted to hospital
subsequently being transferred to a higher level
of care. “We found a very significant improvement
in the accuracy and predictability of using that AI
model, to a point where there could be significant
reduction in mortality,” he reports.
5  MIT Technology Review

And smaller companies are seizing this opportunity.

A 2023 survey conducted by MIT Technology Review
“Much of this technology can
Insights on 1,000 executives about their generative AI be within the reach of many
deployment found that small companies (those with
annual revenue less than $500 million) were three
more organizations. It’s not
times more likely than mid-sized firms ($500 million to just the OpenAIs, Googles,
$1 billion) to have already deployed a generative AI use
case (13% versus 4%). In fact, these small companies and Microsofts of the world,
had deployment and experimentation rates similar but more average-size
to those of the very largest companies (those with
revenue greater than $10 billion) (see Figure 2). businesses, even startups.”
“Much of this technology can be within the reach Michael Carbin, Associate Professor, MIT,
of many more organizations,” says Michael Carbin,
and Founding Advisor, MosaicML
associate professor at MIT and founding advisor at
MosaicML. “It’s not just the OpenAIs and the Googles
and the Microsofts of the world, but more average-size on certain tasks.7 Open source’s greater transparency
businesses, even startups.” also means researchers and users can more easily
identify biases and flaws in these models.
Smaller open-source models, like Meta’s LLaMA,
are enabling this accessibility while also offering This democratization of access to technical tools could be
capabilities that could rival the performance of large society-wide. “This is going to lower the barrier for people
models and allow practitioners to innovate, share, and accessing technology and programming or manipulating
collaborate. One team built an LLM using the weights software. Now you have this natural-language interface
from LLaMA at a cost of less than $600, compared for design software allowing you to navigate what was
to the $100 million involved in training GPT-4. The otherwise an extremely difficult tool used only by the
model, called Alpaca, performs as well as the original best creative professionals,” says Carbin.

Figure 2. Generative AI deployment plans by company size (annual revenue)

Survey question: When will your organization begin deploying generative AI technologies
(apart from isolated experiments)?

We are already formally experimenting We have already deployed at least one use case

Annual revenue
More than $10 billion

$1 billion to $10 billion

$500 million to $1 billion

Less than $500 million

0% 10% 20% 30%

Source: MIT Technology Review Insights’ survey on 1,000 executives globally about generative AI deployment in the enterprise, summer 2023.
6  MIT Technology Review

Automation anxiety
From the telephone to the desktop computer, each
leap in everyday technology sparks worries about
unemployment and threats to human craft and skill. But
The AI skills challenge
generative AI appears to mark a new phase given the
unprecedented range of tasks that can now feasibly be
on the factory floor
automated.
Many economists and technology futurists, voicing
concerns about AI’s impact on jobs, emphasize the
In the MIT Technology Review Insights poll on
importance of re-skilling factory-floor workers as AI
generative AI deployment in the enterprise, one-
changes roles. The focus of this argument is often on
quarter of respondents expect generative AI’s primary
training employees who lack advanced technology skills
effect to be a reduction in their workforce8. But many
to use AI models. MIT’s Ben Armstrong believes these
of the leaders and academics we have spoken to do
calls are off target.
not expect large-scale automation threats. Instead,
they believe the broader workforce will be liberated
In the future, says Armstrong, AI factory workers will
from time-consuming work to focus on higher value
need more domain-specific skills. “The type of flexible
areas of insight, strategy, and business value.
LLM-based tools that are emerging now do not require
a lot of skills to use,” he says. “You offer a query, and it
In fact, AI’s contributions may be welcomed in
gives you a response. What will really be needed is the
sectors that are critically short of workers, such as
skill to tell whether the response is valid for the job at
manufacturing. “There are 700,000 job openings out
hand. For that, a lot of domain expertise is needed.”
there [in the US economy],” observes Julie Shah, who
leads the Interactive Robotics Group of the Computer
These are high-stakes scenarios for people who
Science and Artificial Intelligence Laboratory at
work in manufacturing, says Armstrong. “And those
MIT. Far from replacing workers, generative AI could
scenarios require skills and knowledge that the
invigorate industrial sectors and help smaller firms
worker will have but not the LLM in all situations.” In
compete in the labor market.
this context, the challenge is not so much in reskilling
workers but in ensuring their core skills and domain
And technology-driven upskilling can help humans
expertise are maintained as AI becomes a bigger
do their work more effectively—and even more safely
presence on the factory floor.
in some cases. Alex Alonso, chief knowledge officer
at the Society for Human Resource Management,
cites crane operators who, thanks to automation and
drones, are stepping away from physically dangerous
roles and shifting to analytical and creative areas, such
as working with AI to design work schedules around
weather patterns.

If generative AI adoption follows a similar trajectory to

robotics, it may both increase labor demand and make
the companies that use it more attractive places to
work. Ben Armstrong, executive director and research
scientist in the Industrial Performance Center at MIT,
cites evidence from several countries that when
manufacturers adopt robots, on average, they also end
up hiring more people. And Shah’s research has found
that smaller companies use robotics not to replace
workers but to attract fresh talent. “A young person
would rather be working with the latest technology in a
cleaner setting,” Shah observes.
7  MIT Technology Review

For the broader workforce, next-generation AI could Risks and responsibilities

alleviate burnout by automating repetitive, rote tasks, Ranging from bias to copyright infringement to privacy
and boost engagement by enabling creative work. “The and security breaches, the adoption of AI in the
impact on knowledge work will be more qualitative enterprise comes with significant risk. Understanding
than quantitative,” says data scientist Ben Lorica who these risks is the most common challenge to
hosts The Data Exchange podcast. “Work will be implementing AI in the enterprise, according to 59%
transformed rather than eliminated.” of respondents in the MIT Technology Review Insights
survey on generative AI deployment in the enterprise
Deploying it well will be imperative, however. If poorly (see Figure 3).
executed, automation can diminish people’s value and
skills as they turn into machine appendages. Shah The governance challenges that come with generative
points out that automation can increase cognitive load AI in particular may exceed the capabilities of existing
by requiring them to oversee, monitor, and fix its errors. data governance frameworks. When working with
In the worst case, it’s harder for humans to find and fix generative models that absorb and regurgitate all
an AI’s errors than to do the work themselves. the data they are exposed to, without regard for its
sensitivity, organizations must attend to security and
In any case, tomorrow’s worker is more likely to privacy in a new way. Enterprises must also now
become “AI-competent” than replaced, and AI manage exponentially growing data sources, and
could give entry-level or less traditionally educated data that is machine-generated or of questionable
employees the tools to perform more expert work. provenance, requiring a unified and consistent
“The social norms will change, and it will be an governance approach. Lawmakers and regulators
expectation that you have competence in these tools have grown conscious of generative AI’s risks as well,
like you do in Excel or Word,” predicts AI expert and leading to legal cases, usage restrictions, and new
adviser Henry Ajder. regulations, such as the European Union’s AI Act, which
came into effect in August 2024.

Figure 3. Generative AI implementation challenges

Survey question: What are the primary challenges your organization faces in successfully implementing generative
AI? (Rank three.)

Understanding the risks

Better understanding our data

Developing use cases

Engaging with regulators or

uncertainty around regulation

Hiring talent or upskilling employees

Determining how it could

help our bottom line

Creating an overall AI strategy

Developing AI governance policy

Gaining support from

executive team and board
Partnering with suppliers
or customers

Source: MIT Technology Review Insights’ survey on 1,000 executives globally about generative AI deployment in the enterprise, summer 2023.
8  MIT Technology Review

“The social norms will change, and it will be an expectation

that you have competence in AI tools like you do in Excel or
Word.”
Henry Ajder, AI Expert and Adviser

One concern is protecting privacy at a time when With these precautions in mind, the most forward-
reams of new data are becoming visible and usable. looking CIOs are moving decisively into this AI era.
“Because the technology is at an early stage, there is a “People who went through the computer and internet
greater need for large data sets for training, validation, revolution talk about when computers first came online,”
verification, and drift analysis,” observes Schaefer. At says Blyton. “If you were one of those people who
the Kansas City VA Medical Center, “that opens up a learned how to work with computers, you had a very
Pandora’s box in terms of ensuring protected patient good career. This is a similar turning point: as long as
health information is not exposed. We have invested you embrace the technology, you will benefit from it.”
heavily in federally governed and secured high-
compute cloud resources.”

Commercial privacy and IP protection is a related

concern. “If your entire business model is based on
the IP you own, protection is everything,” says Blyton
of DuPont. “There are many bad actors who want to
get their hands on our internal documentation, and the
creation of new avenues for the loss of IP is always a
concern.”

As generative AI becomes “agentized” into, say,

autonomous customer bots, there is also a risk that
companies could lose control of communications
channels, says Ajder. “If you’ve baked a model into
one of your customer-facing products, and suddenly
it generates something defamatory or misleading and
leads to actions which cause harm to individuals, or
companies, you’re going to be in real trouble.” Cyber
risks are mounting too. Hackers can use deepfake or
synthetic media to deploy more authentic phishing
campaigns, mimicking the communication style or
voice of a known and trusted contact, whether a CFO
or a supply chain partner.

The tendency of LLMs to hallucinate facts or deliver

falsehoods is also a significant obstacle, especially in
highly regulated or sensitive sectors like science or law.
Generative AI skeptics like Gary Marcus, a cognitive
neuroscientist, suggest that hallucinations may be an
inescapable feature of generative AI models, since they
are probabilistic engines rather than agents capable of
true reasoning.9
9  MIT Technology Review

Generative AI and the future of work is an exclusive report by MIT Technology Review. It draws from four reports. One,
Generative AI deployment: Strategies for smooth scaling, in association with Adobe, EY and Owkin, was based on a
summer 2023 survey and interviews with academics and data and AI experts. The second, The great acceleration: CIO
perspectives on generative AI, in association with Databricks, was based on desk research and in-depth interviews with
senior executives and experts, conducted in April and May 2023. The third, Taking AI to the next level in manufacturing,
in association with Microsoft, is based on a survey of senior executives in manufacturing organizations and interviews
with experts on the use of AI in manufacturing, conducted in December 2023 and January 2024. The fourth, Humans
at the heart of generative AI, in association with Teleperformance, was based on desk research and in-depth interviews
with senior executives and experts, conducted in March and April of 2023.

MIT Technology Review has collected and reported on all findings contained in this paper independently. Francesca
Fanshawe and Laurel Ruma were the editors of the report, and Nicola Crepaldi was the publisher.

About MIT Technology Review

Founded at the Massachusetts Institute of Technology in 1899, MIT Technology Review is a world-renowned,
independent media company whose insight, analysis, reviews, interviews and live events explain the newest
technologies and their commercial, social, and political impacts.

MIT Technology Review derives authority from its relationship to the world’s foremost technology institution and
from its editors’ deep technical knowledge, capacity to see technologies in their broadest context, and
unequaled access to leading innovators and researchers.

We are the destination for those seeking to understand how technology is shaping our world. To dive deeper into
our award-winning journalism and get ahead of future trends, download the MIT Technology Review mobile app.

Footnotes
1. “The economic potential of generative AI,” McKinsey & Company, June 14, 2023, https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-
potential-of-generative-ai-the-next-productivity-frontier#/.
2. “Generative AI could raise global GDP by 7%,” Goldman Sachs, April 5, 2023, April 5, 2023, https://www.goldmansachs.com/intelligence/pages/generative-ai-could-raise-
global-gdp-by-7-percent.html.
3. “Retail in the Age of Generative AI,” Databricks, April 13, 2023, https://www.databricks.com/blog/2023/04/13/retail-age-generative-ai.html.
4. “The Great Unlock: Large Language Models in Manufacturing,” Databricks, May 30, 2023, https://www.databricks.com/blog/great-unlock-large-language-models-
manufacturing.
5. “Generative AI Is Everything Everywhere, All at Once,” Databricks, June 7, 2023, https://www.databricks.com/blog/generative-ai-everything-everywhere-all-once.
6. “Large Language Models in Media & Entertainment,” Databricks, June 6, 2023, https://www.databricks.com/blog/large-language-models-media-entertainment.
7. “Just how good can China get at generative AI?” The Economist, May 9, 2023, https://www.economist.com/business/2023/05/09/just-how-good-can-china-get-at-
generative-ai.
8. “Generative AI deployment: Strategies for smooth scaling,” MIT Technology Review Insights, 2023, https://www.technologyreview.com/2023/10/10/1081117/generative-ai-
deployment-strategies-for-smooth-scaling/.
9. Gary Marcus, “What If Generative AI Turned Out to Be a Dud?” Marcus on AI, August 20, 2023, https://garymarcus.substack.com/p/what-if-generative-ai-turned-out.

Illustrations
Illustrations assembled by Tim Huxford with elements from Shutterstock and Adobe Stock.

While every effort has been taken to verify the accuracy of this information, MIT Technology Review cannot accept any responsibility or liability for reliance by any person on this
report or any of the information, opinions, or conclusions set out in this report.

© Copyright MIT Technology Review, 2024. All rights reserved.

www.technologyreview.com