Generative AI

is all the rage

Deloitte AI Institute
Generative AI is all the rage
Implications of Generative AI for businesses

About the Deloitte AI Institute

The Deloitte AI Institute helps organizations connect Combined with Deloitte’s deep knowledge and
the different dimensions of a robust, highly dynamic experience in artificial intelligence applications,
and rapidly evolving AI ecosystem. The AI Institute the Institute helps make sense of this complex
leads conversations on applied AI innovation across ecosystem, and as a result, deliver impactful
industries, with cutting-edge insights, to promote perspectives to help organizations succeed by
human-machine collaboration in the “Age of With”. making informed AI decisions.

The Deloitte AI Institute aims to promote a dialogue No matter what stage of the AI journey you’re in;
for and development of artificial intelligence, whether you’re a board member or a C-Suite leader
stimulate innovation, and examine challenges to AI driving strategy for your organization, or a hands
implementation and ways to address them. The AI on data scientist, bringing an AI strategy to life, the
Institute collaborates with an ecosystem composed of Deloitte AI institute can help you learn more about
academic research groups, start-ups, entrepreneurs, how enterprises across the world are leveraging AI
innovators, mature AI product leaders, and AI for a competitive advantage. Visit us at the Deloitte AI
visionaries to explore key areas of artificial intelligence Institute for a full body of our work, subscribe to our
including risks, policies, ethics, future of work and podcasts and newsletter, and join us at our meet ups
talent, and applied AI use cases. and live events. Let’s explore the future of AI together.

www.deloitte.com/us/AIInstitute

2
2
 Generative AI is all the rage

Let’s take a moment to cut

through the hype.
The AI field took a turn with the release
of powerful Generative Artificial Intelligence
(AI) models, and as a result, the world is
seeing the automation of some skills around
creativity and imagination sooner than many
expected. For some organizations, Generative
AI holds valuable potential for higher order
opportunities, like new services and business
models. Deloitte offers a method for
selecting Generative AI use cases, as well
as some next steps for business leaders in
the Age of With™.

3
Implications of Generative AI for businesses 2
Generative AI is all the rage

The rise of Generative AI

Generative AI has captured attention
in global media and the public square,
prompting questions and discussions
around this transformative technology.
Businesses, research organizations, and Generative AI in general and LLM-
even lay users are experimenting with powered chatbots in particular
Generative AI, and given the excitement are not without risks, and this
and interest, it is important to look is prompting discussions around things
more closely at the potential capabilities like the potential for job losses and
and implications for business. legal questions around intellectual
property and ownership. What is more,
Generative AI is a subset of artificial because the chatbot mimics coherent
intelligence in which machines create human phrasing, it may give some the In this article, we look closely at the
new content in the form of text, code, impression that the AI understands the potential benefits and limitations
voice, images, videos, processes, and prompts to which it responds, which of Generative AI, introduce a method
even the 3D structure of proteins. Some can lead users to anthropomorphize to qualify if, where and how these
forms of Generative AI have been well the chatbot (i.e., the ELIZA effect, cognitive tools could be used, and
established in this decade, but it was as seen in the work of computer offer important factors for business
a large language model (LLM) powering scientist Joseph Weizenbaum). leaders to weigh when adopting
an easily accessible chat interface Generative AI.
that enabled Generative AI to have its Deloitte is working on a variety of
breakthrough moment and surprise projects exploring the opportunities In a prior article, “Implications
even specialists in the field. and business value Generative AI of Generative AI for businesses,”
can create for our clients. From Deloitte offered a deep dive on
As with other types of AI before it, experiences and conversations thus the qualities and capabilities
this new poster child of AI is stimulating far, the clear path ahead, as with of Generative AI, the state of the
the imagination as organizations and all AI, is to attempt to discover and market, and what that means for
individuals consider how to use this tool capitalize on capabilities while also organizations wading into this fast-
to benefit both business and society. responsibly managing the risks that evolving technology field. And in
Generative AI can help in incremental are already emerging. forthcoming articles, we will discuss
digitization and basic productivity use questions from legal, ethics, risk, and
cases (e.g., more effective text-based talent and technology perspectives
channels). But its grander potential and provide insights into industry
is in the higher order opportunities, use cases.
such as new services or business models
that were previously uneconomical.

4
 Generative AI is all the rage

While this is still the

beginning, it’s moving fast.
Among organizations across industries, In late 2022, with the release of an
there is interest in differentiating AI use easy-to-use Generative AI chatbot, more
cases, from public service applications people began to discover and imagine
to addressing climate change and how this new technology can be used
transforming business functions (see in the creative space. The chatbot use
Deloitte’s AI Dossier). AI is viewed as a case opened the door for thinking more
tool that can automate skills and tasks broadly about how Generative AI can
performed by humans, and AI can be so be used for tasks, ranging from writing
successful in this regard that humans can copy to generating 3D structures and to
forget skills that have been automated. outputting organizational processes. As
Examples include writing assistants, such, we are now seeing the automation
home automation, and automotive of some skills around creativity and
navigation systems. Would most people imagination sooner than many expected.
have the ability to navigate a new city
without a mobile phone? There is a lot left to discover. In this
Age of With™, the era of humans
We have seen these kinds of automations working with intelligent machines
emerge across a variety of areas and to achieve things greater than what
skillsets. The assumed roadmap for either could do alone, Generative AI
AI was that, in the shorter term, AI is will impact the future of work and
most valuable as a way to automate become a common tool throughout
operational skills, and creative skills will various aspects of our daily lives.
remain the exclusive province of human In some cases, the applications may
thinking for the foreseeable future. With be clearly visible, but more often
Generative AI, this roadmap has taken an than not, they may be working in
unexpected turn. the background.

5
Generative AI is all the rage

The evolution of Generative AI

The ability of Generative AI to create a convincing (albeit low-
quality and greyscale) image of a human face emerged in 2014.
Since then, the image quality has increased, and today, almost
anything that can be described in words can also be generated
as an image, using a textual description called a “prompt.”
Throughout 2022, social media users tinkered with Generative
AI platforms and shared the results. We have seen avocado
armchairs and photorealistic images of astronauts riding horses
on the Moon. Cosmopolitan magazine was the first to publish
a cover page created by an AI-based image generation tool,
and there has even been a case of a user who submitted an
AI-generated image to a fine art competition—and won first
place. Today, we are seeing similar improvements in other kinds
of Generative AI. You may even have found this article via a chat
with an AI system that integrates with a search engine.

Images created with Generative AI.

6
 Generative AI is all the rage

How Generative AI works:

Understanding the basics

To understand how Generative AI will impact

business and life, we need to understand what
it is, what it can do, and what it cannot do, yet.
Machine learning has dominated the field In a process referred to as training,
of AI for decades. Generally, this approach the algorithm is supplied with a large
to AI development is rooted in the dataset of input/output examples to
concept of learning from examples, rather extract patterns from the input, which
than following explicitly programmed allow conclusions about the expected
rules. This is important as there are many output. Spam filters, for example, use
tasks that humans perform based on these patterns to identify similarities in
tacit knowledge (and thus can provide data points and relate those to different
examples) but cannot describe the classes (i.e., sorting email to a spam
underlying rules to do so. For example, folder). While the input data has become
humans know how to recognize a face, more complex over time, from simple
but the rules that would instruct an AI arrays of numbers to high-resolution
system to do the same are much less photos, the output side of a model has
clear. The approach of learning from to this point been mostly limited to
examples has led to the development of categories like “spam” or “not spam,”
powerful tools that can identify intricate “cat” or “dog,” or numerical values such
patterns in complex data. as 7°C or $29. This approach powers
nearly all AI that has been deployed so far,
the result is “single purpose” applications
that can only perform one task.

7
Generative AI is all the rage

INPUT OUTPUT
Figure 1:
Used car data An example of
a single purpose
Type Engine Milage Year machine learning
model, using a
Limo Electric 70k 1996 Predicted price regression model
Regression to predict the
S. wagon Petrol 100k 2005 $17,000
model resale value for
a given car.
Truck Diesel 150k 2016

Figure 2:
INPUT OUTPUT
An example of how
Email Predicted label a single purpose
model can be used
Hi Bjorn,
No spam to sort e-mails
Have a look at this: A generative adversarial network
Classification by “spam” or
(GAN) is a class of machine learning frameworks
designed by Ian Goodfellow
model Spam “not spam.”

Enter Generative AI
The main difference between “traditional quality data processed over weeks
AI” and Generative AI is that in the latter, on large-scale, GPU-enabled, high
the output is of a higher complexity. performance computing clusters.
Rather than just a number or a label, the Only a few institutions have the necessary
output can be an entire high-resolution resources and talent to build such
image, a full page of newly written text models. Running a model also requires
(which is generated word by word), or any a lot of compute, which is why access to
other digital artifact. This introduces an these kinds of models is often provided
interesting new element: There is usually via an application programming interface
more than one possible correct answer. (API). This allows developers to use the
This results in a large degree of freedom models with their existing software
and variability, which can be interpreted products without need for additional
as creativity. infrastructure. These models are versatile
and can be fine-tuned for specific tasks,
Generative AI models are typically hence they are called Foundation Models.
large and resource hungry. Creating Unlike single-purpose AI, they are suited
them requires terabytes of high- for multi-purpose tasks.

Figure 3:
INPUT OUTPUT
With Generative AI,
Generated text user prompts lead
Prompt to artifacts that
Generative AI is a broad field of computer sicence focused on can contain a large
What is AI? creating intelligent machines that can perform tasks
model degree of freedom
that typically require human-like congnitive abilities,
and variability.
such as perception, reasoning, learning, problem-
solving, and decision-making.

8
 Generative AI is all the rage

Regarding risks and limitations

Current Generative AI models have Similar to other AI models, Foundation

limitations. Perhaps the most well-known Models can reproduce latent bias in the
is termed “hallucination,” which refers to training data, and of course, they lack
a high-confidence response that is not comprehension and the ability to reason
grounded in the training data. In other as humans do. This has implications for
words, the response is fictional. For the broader concept of Trustworthy
some applications, like art generation, AI™. After all, they are language models,
this is a non-issue and perhaps even image models, or voice models but not
a desired “creative” feature of Generative knowledge models.
AI. For other applications, however,
such as copywriting or computer code Despite limitations, Foundation
generation, hallucinations can result in Models can function at such a high
artifacts that are not entirely valid or true, quality that many new use cases
which undercuts the potential value of become possible.
Generative AI.

Another limiting factor is that today’s

Generative AI models generate artifacts
based on the model itself and the
Some known limitations
input prompt. Other additional sources
of current Generative AI
and datasets cannot currently be
integrated directly into the model’s
internal information processing without
Hallucination | Generative AI systems create
costly re-training or fine-tuning, which
means Generative AI models can only confident responses that cannot be grounded in any
access information extracted from the of its training data.
data on which they were trained. For
similar reasons, they cannot provide Bias | Similar to other learned models, Foundation
references and sources for the generated Models inherit the bias contained in the training data.
content, which complicates validation.
Furthermore, current models have
a context window of a few thousand Lack of human reasoning | Generative AI systems
words, which is the limit for the size are based on statistical features, which is not a solid
of the combined input and output. foundation for logical reasoning.
However, Generative AI models can
be combined with other systems (e.g.,
Limited context window | Current models have
search, conversational AI) to leverage the
a context window of a few thousand words, which is the
benefits of both parts. For example, with
limit for the combined input and output of the model.
a chatbot, a conversational AI system can
serve as an orchestration layer between
the Generative AI model, a search engine,
and the user, which helps to amplify the
user experience.

9
Generative AI is all the rage

Generating revenue
using Generative AI
Using this technology for business benefit can
be conceived along two distinct approaches.
First, the models can be used as they a job advertisement or a floor plan,
are available today, a simple interface all the way to the 3D model of an engine
that allows near-direct access to the part, a molecule with certain properties,
underlying model in the form of a chat or a workflow, to name a few. Use cases
for text or an image generation tool. with high usage frequency are preferred,
The second approach is to integrate as there will be more example data to
Generative AI with other technologies fine-tune and improve a model, and
to automate processes. For example, subsequently a more substantial impact.
Generative AI can allow for human- Other factors to consider in selecting
level expressive interactions, while high-value use cases are existing skill and
a conversational AI system (i.e., a chat- cost bottlenecks with human generated
or voicebot) controls the flow and artifacts. The quality of the artifact may
ensures factual accuracy. An example in some cases require human effort, but
is an automated, Generative AI-powered if it can be created with Generative AI to
call center. We expect the second a commensurate quality, then the human
approach will provide the most value. can be liberated to work on higher quality
tasks. By turning lower-level creative tasks
A good start to identifying use cases over to Generative AI, we could see things
is to find processes or tasks where like databases providing stock content
a digital artifact of some kind is created (e.g., images, sounds, or texts) turned
or processed. This could range from upside down as these digital artifacts can
be created instantly with a prompt.

If a task requires effort to execute but is easy

to validate, it might be a good use case.
Deloitte has designed a Digital Artifact a task without Generative AI; and the
Generation/Validation method to help necessary effort to validate or fact check
innovation leaders determine whether the output from the Generative AI. This
an idea can be turned into a beneficial leads to a two-dimensional classification,
use case leveraging Generative AI. At categorizing use cases based on the
the core of this method are two of the required human effort and the ability
most critical elements to consider: of the user to validate the results.
the human effort required to complete

10
 Generative AI is all the rage

There is a sweet spot for Generative AI use cases

Generative AI is useful where the artifact generation effort is high and validation is easy

Digital Artifact Generation/Validation method

Identifying desirable use cases
HIGH
Generative AI assists best in use cases
where human effort is high, while
validation is easy.

CONSIDER Generation effort

3 How much human effort is required to
2 achieve the desired result
1 Validation effort
4 How much human effort is required to
check the plausibility or correctness of
Generation the result
effort
ASSESS

NEGLECT

LOW
DIFFICULT Validation effort EASY

Examples plotted above

1 Create a joke 2 Draw an image of 3 Draft a contract 4 Draft a contract

an elephant under (without legal expertise) (with legal expertise)
While creating a good
a palm tree
joke requires some effort If you do not have legal If you do have legal
into designing the punch- Drawing any sophisticated expertise, drafting expertise, drafting
line and best storytelling- image requires reasonable a contract is very hard and a contract still requires
style, it is easy to validate effort for most people validating it difficult. effort, but validating it is
simply by reading it. regardless of the tools Generative AI is useful signifacantly easier.
available. On the other where the artifact
hand, validation is easy generation effort is high
since you can just look at and validation is easy.
the picture.

11
Generative AI is all the rage

For example, re-writing text can be There is an additional complication

a daunting and time-consuming task for that should be considered. If the model
a human. Generative AI tools can take outputs are consistently correct, users
original text and quickly produce may, over time, become less rigorous in
a re-written result, a shorter text, fact checking. Inevitably, however, the
a summary, or even a different writing model will make an error, and part of the
style. A user who is familiar with the challenge is that the errors may not be
original content can validate the accuracy obvious, particularly when Generative AI
or correctness of the output. Thus, this is used to create more complex things,
could be a promising application of like programming code. Thus, when
Generative AI. assessing the ease of validation, weigh
the importance of ongoing attention
Yet, if the user is reviewing outputs that to fact checking.
are outside of their area of expertise,
validation becomes more complicated.
The Generative AI output may
read as coherent and convincingly
accurate, but the potential for a
“hallucination” remains. If users lack
the knowledge to validate the output
and spot hallucinations, the use case is
revealed to require high levels of effort
for validation and mitigating the risks
from hallucinations.

12
 Generative AI is all the rage

Insights from Deloitte projects on Generative AI:

Reaping benefits from Generative AI requires
more than identifying a good use case
Identifying use cases is only part of the avoid the temptation to go forward alone
challenge. Whenever a transformative and instead find support and knowledge
technology emerges, some organizations from partners, colleagues, and third-party
are spurred to experiment for the sake organizations operating in this space.
of its novelty, which can lead to “random
acts of digital” that do not deliver the The inherent complexity in current
anticipated return. Driving business projects leveraging Generative AI
results with Generative AI requires a requires a cross-disciplinary team
strategy and collaboration from a cross- to guide and govern the AI lifecycle.
disciplinary team. In addition, with Professionals from a variety of domains
a technology that is advancing and can help the business answer critical
maturing as quickly as Generative AI, questions, including:

DOMAIN

Ideation Business Customer & Enterprise Human Risk Regulations

& Product Operations Marketing Technology Capital Management & Laws
Development

STAKEHOLDERS

Creatives, CEO, COO, CMO CIO, CTO, IT CHRO Risk officers Legal &
designers Line of Business Compliance
leaders

KEY QUESTIONS

What can How does the How can the use Can the existing Does the What risks emerge What current and
Generative AI Generative AI fit case be leveraged MLOps-tech stack workforce possess when deploying expected laws
permit that into and enhance to build customer and platform the skills to use Generative AI (e.g., and regulations
reduces human existing processes engagement, licenses fuel Generative AI, jailbreaks, prompt- concern the use
effort and can be and enterprise and how much Generative AI, or and what are the spoofing), and of Generative AI,
rapidly validated? strategy? transparency is are third-party implications for how do these risks and are existing
appropriate? services required? talent acquisition impact Generative governance and
and upskilling? AI value? MLOps processes
sufficient to meet
those laws and
regulations?

13
Generative AI is all the rage

Based on our observations and experience, we

recommend business leaders avoid jumping head-first
into the hype. Instead, we encourage decision makers to:

1
Develop a strategy for
2
Become familiar with the
3
Bring together a cross-
Generative AI and integrate and underlying technologies that make disciplinary team of people with
harmonize it with the enterprise’s Generative AI possible, as well as the the domain knowledge to think
existing AI strategy. The same current capabilities and limitations. creatively about potential use cases.
principles that guide an AI-fueled Educate your workforce in the When business leaders, technology
organization apply to the use of usage, risks, and capabilities of AI leaders, and creatives work together
Generative AI (e.g., access to curated to establish a baseline of knowledge with external experts, they are able
enterprise data; AI governance; through training. Also monitor over to identify valuable applications
process transformation to leverage time how the technology advances and also design Generative AI
cognitive workers, etc.). With and the impact on business risks deployments, to mitigate business
a technology evolving this quickly, and opportunities, as they emerge. and cyber risks and meet applicable
avoid the temptation to go forward This article series may support laws and regulations.
alone. Find support and knowledge your efforts.
from partners and third-party
organizations operating in this space.

4
Leverage Deloitte’s
5
Ensure the collection and
6
Assess use cases against
Digital Artifact Generation/ curation of proprietary data, Trustworthy AI™ principles,
Validation method as this is key for tailored use cases including challenges around bias
to identify where Generative AI that provide a differentiator or and misinformation, attribution,
might impact your value chain, competitive advantage. transparency, and enterprise
with incremental digitization from accountability for the impact from
basic productivity use cases to Generative AI.
higher order opportunities, such
as new, differentiating services
or business models.

14
 Generative AI is all the rage

Deloitte is excited to move into the future with

our clients and colleagues, as well as with our
connections in academia and the broader
AI ecosystem around the world.

The discussions so far show that there

is a need for a deeper understanding
There is a lot
of Generative AI, from the underlying to cover and the
technology to its impact on the future
of work. As such, it is important to look conversations are far
closely at the implications for risk, trust,
and governance, which is investigated from over. Deloitte
in a forthcoming article, “Proactive risk
management in Generative AI.” There are
is a trusted advisor
also legal considerations for Generative
AI, which we plan to cover in “Legal
as we push beyond
implications of using Generative AI the initial buzz
(What the AI System won’t tell you).”
around this new
technology and into
how Generative AI
can be used for good
in the Age of WithTM.

15
Generative AI is all the rage

Reach out for a conversation.

Beena Ammanath Wessel Oosthuizen Dr. Kellie Nuttall Jefferson Denti Audrey Ancion
Global Deloitte AI Institute Deloitte AI Institute Africa, Deloitte AI Institute Deloitte AI Institute Brazil, Deloitte AI Institute
Leader Lead Australia, Lead Lead Canada, Lead
Deloitte AI Institute Deloitte Africa Deloitte Australia Deloitte Brazil Deloitte Canada
United States, Lead woosthuizen@deloitte.com knuttall@deloitte.com jdenti@deloitte.com aancion@deloitte.ca
Deloitte Consulting, LLP
bammanath@deloitte.com

Jan Hejtmanek Roman Fan Anne Sultan Dr. Bjoern Bringmann Prashanth Kaddi
Deloitte AI Institute Deloitte AI Institute China, Deloitte AI Institute France, Deloitte AI Institute Deloitte AI Institute India,
Central Europe, Lead Lead Lead Germany, Lead Lead
Deloitte Central Europe Deloitte China Deloitte France Deloitte Germany Deloitte India
jhejtmanek@deloitte.com rfan@deloitte.com asultan@deloitte.com bbringmann@deloitte.com pkaddi@deloitte.com

Masaya Mori Nicholas Griedlich Naser Bakhshi Tiago Durao Sulabh Soral
Deloitte AI Institute Japan, Deloitte AI Institute Deloitte AI Institute Deloitte AI Institute Deloitte AI Institute
Lead Luxembourg, Lead Netherlands, Lead Portugal, Lead United Kingdom, Lead
Deloitte Japan Deloitte Luxembourg Deloitte Netherlands Deloitte Portugal Deloitte United Kingdom
mmori@deloitte.com ngriedlich@deloitte.com nbakhshi@deloitte.nl tdurao@deloitte.com ssoral@deloitte.com

Special thanks to contributors:

Jakob Nikolaus Moecke, Senior Consultant, Deloitte AI Institute Germany Elon Allen, Partner, Monitor Deloitte Australia
Philipp Joshua Wendland, Senior Consultant, Deloitte AI Institute Germany Bram den Hartog, Partner, Monitor Deloitte Australia
Alexander Mogg, Lead Partner, Monitor Deloitte Germany Aisha Greene, Senior Manager, Deloitte AI Institute Canada
Kate Fusillo Schmidt, Senior Manager, Deloitte AI Institute US Anke Joubert, Senior Manager, Deloitte AI Institute Luxemburg
Erica Dodd, Senior Manager, Deloitte AI Institute Australia Dr. Gordon Euchler, Director, Deloitte Germany
Jessica Carius, Senior Consultant, Deloitte AI Institute Australia

16
This publication contains general information only and Deloitte is not, by means of
this publication, rendering accounting, business, financial, investment, legal, tax, or
other professional advice or services. This publication is not a substitute for such
professional advice or services, nor should it be used as a basis for any decision or
action that may affect your business. Before making any decision or taking any action
that may affect your business, you should consult a qualified professional advisor.

Deloitte shall not be responsible for any loss sustained by any person who relies on
this publication.

About Deloitte
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private
company limited by guarantee (“DTTL”), its network of member firms, and their related
entities. DTTL and each of its member firms are legally separate and independent
entities. DTTL (also referred to as “Deloitte Global”) does not provide services to
clients. In the United States, Deloitte refers to one or more of the US member firms of
DTTL, their related entities that operate using the “Deloitte” name in the United States
and their respective affiliates. Certain services may not be available to attest clients
under the rules and regulations of public accounting. Please see www.deloitte.com/
about to learn more about our global network of member firms.

Copyright © 2023 Deloitte Development LLC. All rights reserved.