1. 1 | Realizing the real business impact of generative AI
Realizing the real business
impact of generative AI
2. 2 | Realizing the real business impact of generative AI
Table of contents
Introduction to generative AI 3
What is it exactly? 4
What is it made up of? 4
How does it work? 5
What are the key challenges? 8
How can generative AI be used by businesses? 10
Which industries could best adopt it? 12
Getting started 14
Use cases 15
Conclusion 17
2 | Realizing the real business impact of generative AI
3. 3 | Realizing the real business impact of generative AI
Introduction
Generative AI is a significant leap forward in the way artificial intelligence (AI)
can be used, with the potential to disrupt nearly every industry and economic
sector around the globe. However, it is an evolution not a revolution, offering
businesses a chance to build on their use of sophisticated algorithms and
machine learning to achieve new levels of competitive advantage.
This novel form of AI can automate many more
tasks, augment how we work, and accelerate
change within the workplace. According to
Goldman Sachs, 300 million jobs across the
globe could be affected by generative AI.
This technology has already been democratized
globally. When OpenAI released its chatbot late
last year 100 million people used it in the first
few months. This was the fastest adoption of a
consumer technology in history. Now a wave of
new generative AI tools are being deployed by
businesses around the globe from ChatGPT to
Bard, DALL-E to AWS’ Bedrock.
Yet many in the corporate world are rightly
sceptical about its use in real-world situations,
there is still a lot of hype. How businesses can
practically leverage generative AI’s true
capability is in its infancy. This white paper
lays out the facts.
“
Generative AI is neither a
fad, nor an apocalypse,
but a potentially useful tool
in its infancy that could
fundamentally change how
businesses operate.”
Philip Basford,
AWS Data AI Lead, Inawisdom,
a Cognizant company
3 | Realizing the real business impact of generative AI
4. 4 | Realizing the real business impact of generative AI
1. What is it exactly?
Generative AI refers to algorithms that generate
a particular output such as text, photos, code,
videos or 3D renderings from the data that
they’re trained on. This form of AI can create
brand new content based on training, instead
of just categorizing or identifying data, like other
types of machine learning. Generative AI can
answer queries and generate text responses
based on conversational language providing
readable answers. Natural language
processing is now used for both input and
output, this makes it highly accessible and
responsive to human interactions.
2. What is it made up of?
Data sets
This can be the vast swathes of data found on
the Internet, which the likes of ChatGPT have
trawled, or business specific data that enterprises
hold on their servers, then there are other
tranches of industry specific data. Generative
AI can be pre-trained on the voluminous,
unstructured data of the worldwide web. They
can also be fine-tuned on a business’ own unique
dataset, allowing this form of AI to offer content
specific to a corporation’s particular needs.
Foundational models - Large language, visual
and audio models
The underlying framework that enables
generative AI to work is called a foundation
model. This is what gives the AI its intelligence.
Transformers are a key component of these
models and are a type of artificial neural network
that is trained using deep learning. This is a
term that refers to the many layers within neural
networks. For instance, the GPT in ChatGPT, one
of the most famous generative AIs, stands for
generative Pre-trained Transformer.
So-called large-language models or LLMs refer to
a type of algorithm that deploys deep learning
techniques that are trained on large language-
focused data sets. The aim is to understand,
summarize, generate and predict new
content. It can work on both varied and
unstructured information from disparate
sources. From a text prompt it can then offer
up a human-like response.
Generative AI is also being used on images and
video, as well as 3D models, even code. It is also
being trained on music and voice. It means
that both its input and output can be incredibly
diverse and useful to those in a variety of
economic sectors.
User Interface and prompts
The interface between generative AI and the
person using it is often not considered enough.
The success of ChatGPT was dependent, not just
on its foundational model, but also on its user
interface or UI. The way in which people interact
with an easy web front-end, or through an API,
matters. Where consumers chat and responses
are delivered in a decidedly human way has
been generative AI’s unique selling point. So is
the value of the user prompt or query – which is
what guides the AI model’s output influencing
its tone, style and quality. Prompt engineering is
now crucial, optimizing textual input to effectively
communicate with models.
“
We are at the iPhone moment
for AI. ”
Jensen Huang,
Chief Executive Officer, Nvidia
5. 5 | Realizing the real business impact of generative AI
“
Generative AI is inherently personalized, as the prompts that drive the
content creation process are provided by the user. This means that
any content produced using generative AI is tailored to their specific
interests, which can have a number of benefits for businesses. ”
Duncan Roberts,
Technology Research Leader, Cognizant
3. How does it work?
Generative AI uses deep, neural networks, which
mimic how the human brain functions, to identify
structures and patterns within data. This type
of AI leverages different learning approaches,
including undersupervized or semi-supervized
learning when it comes to training. It means
generative AI can handle more complex
patterns than traditional machine learning. It
has the ability to proficiently collect and distil
large amounts of data. These models do not
always require human intervention to distinguish
differences in the training data.
This type of AI deploys transformers, which
are neural networks that learn context and
understanding through sequential data analysis.
They are used primarily in the field of natural
language processing where context is important.
Generative AI platforms are then able to create
rules, and then make judgements
and predictions.
5 | Realizing the real business impact of generative AI
6. 6 | Realizing the real business impact of generative AI
“
Generative AI represents a breakthrough in what computers can
accomplish for users. Previous generations of advancements have
excelled at providing precise answers to narrowly defined queries.
However, altering the input question would require overhauling the
underlying algorithm.
In contrast, generative AI overcomes this constraint and facilitates
conversational interfaces with minimal friction. The user can engage
in natural dialog while the system remains focused on delivering the
desired information.
By comprehending broader context and meaning, generative models
push the frontiers of AI beyond predefined responses. This technology
promises to dramatically enhance how humans and computers interact
and collaborate. ”
Naveen Sharma
Global AIA Practice Head, Cognizant
6 | Realizing the real business impact of generative AI
7. 7 | Realizing the real business impact of generative AI
“
AWS has played key role in democratizing ML and making it accessible
to anyone who wants to use it, including supporting more than
100,000 customers of all sizes and industries. We are taking the same
approach to generative AI. Customers want to be able to start using
generative AI to transform their applications and businesses as quickly
as possible, and we’re here to help them do that.
Recently, we have announced 1/ Amazon Bedrock, a new service that
makes foundation models (FMs) from AI21 Labs, Anthropic, Stability AI,
and Amazon accessible via an API, 2/ AWS Trainium AWS Inferentia
are Amazon custom silicon, cost-effective, high-performance GPU
instances and purpose-built chips, 3/ Amazon CodeWhisperer is a
coding companion that enables developers to build applications on an
average ~57% faster and more securely. ”
RK Kuppurao
Technology Leader, AWS
US $350.4 billion Size of the AI market by 2028 - IMARC GROUP
7 | Realizing the real business impact of generative AI
8. 8 | Realizing the real business impact of generative AI
4. What are the key challenges around generative AI?
Generative AI platforms are being trained
on billions of parameters, processing huge
volumes of text and images, audio and video,
this process comes with risks. Before businesses
can embrace generative AI, they need to
understand the challenges:
•
Copyright issues – Think intellectual property
infringement, concerning which content is used
for AI creations. Legal questions are still being
resolved and whether patent, trademark and
copyright violations apply to both the input and
output of generative AI.
•
Security concerns – The nature of these
complex algorithms make it difficult for
developers to identify security flaws. Generative
AI could also be used for easier malicious code
generation.
•
Privacy issues – Training data is often stored for
training models. Businesses can easily expose
sensitive data interacting with AI. Protecting
personally identifiable information will be
crucial. The recent EU AI Act aims to address
such privacy and ethical concerns.
•
Lack of transparency – It’s still perceived
as a black box technology and difficult for
businesses to understand how these complex
models work and make decisions. This can
create biases in output and erode acceptance.
This needs to be addressed.
•
Hallucinations – Generative AI is prone to
making mistakes and fabricate information.
At the moment models are 80-90% accurate.
Factual inaccuracies, social biases from
training data and misinformation from AI
systems could undermine their reliability.
•
Cost and time - Training a custom model offers
greater flexibility, but it comes with a high price.
It is estimated to cost US$1.6 million to train a
1.5-billion-parameter model. It can also take a
long time to train and deploy.
•
Size of data pool - Training AI models from
scratch is often not possible because it requires
an enormous amount of data. This is overcome
by using pre-trained models and ‘transfer
learning’, where businesses fine-tune the model
using a small number of labelled documents.
“
Just as generative AI is
working on the ability to be
multimodal and comprehend
and create diverse types of
content, the measures to
keep it in check and avoid
the worst-case scenario
must also be diverse and
comprehensive.”
Duncan Roberts,
Technology Research Leader, Cognizant
9. 9 | Realizing the real business impact of generative AI
Responsible use of generative AI
Who owns the input data you’re fine-tuning generative AI with matters.
Fine-tuning them on private data sets that have been vetted, pruned of
toxic data and copyrighted information is crucial. The output also has
to be reviewed by humans to ensure there are no hallucinations, that it
is has the right tone of voice, and the output – text or images – are not
contrary to your brand. The results of generative AI must be consistent
with your corporate policies and standards.
Philip Basford,
AWS Data AI Lead, Inawisdom, a Cognizant company
19% of the US workforce could have 50% of their tasks affected
by generative AI.
- Source: OpenAI
“
Rather than replace humans,
this technology will enhance and
augment human intelligence
and decision-making, making us
better at what we already do. ”
Duncan Roberts,
Technology Research Leader,
Cognizant
10. 10 | Realizing the real business impact of generative AI
5. How can generative AI be used by businesses?
Since generative AI creates new content it can
be used for a variety of business functions within
organizations including classifying, editing,
answering questions, summarizing, as well as
drafting new content. It produces outputs in the
same medium in which it is prompted. Here are
some examples:
•
Intelligent document processing – Allows
organizations to analyse and extract valuable
insights from unstructured documents including
forms, emails, contracts and spreadsheets.
•
Writing code – Generative AI can now help
coders work, in some cases, at twice their
previous speed. It automates repetitive tasks
and suggests code using tools like GitHub
Copilot. It can recommend code modifications
to boost performance.
•
Complimenting human work – Augmenting
the output and productivity of managers,
journalists, PR and marketing professionals or
healthcare workers, rather than replacing them.
•
Personal assistant - It can act as a smart
virtual assistant aiding knowledge workers to
expand their capacity and analyse their work
from transcribing meetings to developing
presentations.
•
Spot inaccuracies – Generative AI can spot
issues with text, images, code, in fact all forms
of content that it has been trained on. It can
also point out social biases and prejudices in
human generated content that workers may
not even be aware they have.
•
Human-focused output – AI tools use
advances in natural language processing,
this means that they are able to interact in a
more human way in terms of enquiry journeys
and interrogation, such tools could substantially
increase productivity.
By 2025, 30% of marketing content will be created
human-augmented, generative AI.
- Source: Gartner
10 | Realizing the real business impact of generative AI
11. 11 | Realizing the real business impact of generative AI
“
Generative AI technologies has
opened up an interesting arms
race between generative AI
software providers, businesses
and government regulators.
We believe that the number of
use cases across the enterprise
will explode in terms of volume
and variety. This in turn will
cause a tremendous pressure
on governments to continuously
legislate to ensure fairness
and transparency on how the
enterprise decisions are made
and how it affects all segments of
the population. ”
Sean Heshmat,
Head of Data AI,
Global Growth Markets, Cognizant
Myth busting
•
Most businesses don’t need to train large language models – They use them instead. Corporations
can deploy LLMs that have been pre-trained on publicly available data. These models generate
accurate content when given just a handful of labelled examples to work on.
•
No data footprint is created using generative AI - When you use a foundation model you are not
adding your enterprise data to it. If used in the right way, your interactions with generative AI do not
leave a footprint on the public model itself. However, you can use transfer learning and fine-tune the
model to your private data. If you are using a public model such as Chat GPT, without opting out,
Open AI will undoubtedly use your data to train generative AI.
•
It may not be a cheap solution - Creating your own private, large language model from scratch can
cost US$4 million in computing power and time alone. Data, skills and ongoing maintenance could
easily double this.
11 | Realizing the real business impact of generative AI
12. 12 | Realizing the real business impact of generative AI
Think about ESG
Creating large foundational models are not only hugely expensive but are also heavy on compute
power. To create also consumes both a lot of energy and materials. On ESG grounds, businesses need
to think whether they need to create one from scratch, since they are resource heavy. Enterprises have
to weigh up whether it will deliver on the results of that investment and whether fine-tuning an existing
model could work. Equally generative AI is being used in the ESG space to achieve Net Zero goals such
as optimize energy networks and tackle emissions.
Philip Basford,
AWS Data AI Lead, Inawisdom, a Cognizant company
6. Which industries could best adopt it
Established, incumbent companies in many
economic sectors are likely to deploy generative
AI, feeding their own proprietary data into AI
models and fine tuning the outputs. There is also
a move towards off-the-shelf industry-specific
models that account for the specific needs and
concerns of each sector.
•
Insurance – From assessing risk to detecting
fraud, reducing human error in the claims
process to personalized customer experiences.
Expect intelligent document processing and
generative AI augmenting policy enquiries, as
well as faster, more accurate policy processing.
•
Legal – Generative AI can summarize
documents in seconds with incredible accuracy,
slashing the time spent by legal counsels. It
can process invoice approvals or extract key
terms from lengthy legal documents, 44% of
legal tasks could be performed by AI, according
to Goldman Sachs. Expect improved legal
research and document review.
•
Construction – AI can improve 3D modelling
and make the design process more efficient,
encapsulating drone or satellite images,
helping generate maps or augmented reality
experiences of a site. It can help automate
how architects visualise plans, from processing
to text to generating blueprints, as well as
reducing building waste due to design errors.
•
Financial Services – This form of AI can halve
the time taken to complete certain written tasks.
Improving risk management and detecting
fraud is also the realm of generative AI, but it
can also refine investment strategies and help
personalize customer services.
•
Life Sciences – Artificial intelligence can
accelerate the research and development
process for new drugs, since there can be a
better understanding of complex data. It can
help with the discovery of novel drugs, improve
medical writing and create synthetic data.
•
Education – Generative AI can create lesson
plans to enquiry journeys, as well as compile
worksheets in many different levels in different
languages. Then there are chat bots that
support teachers. Pupils can debate with bots
to improve their knowledge.
•
Sales - Trained AI models are able to suggest
upselling opportunities, not just based on static
customer data, but on real-time responses
and the current conversation, drawing on past
recordings, customer and market data, as
well as social media information. It can make
call centres 14% more productive, improving
customer service response times
and helpdesks.
•
Marketing – Generative AI can help personalize
and adapt marketing presentations, offer
corrections and suggestions. It can also help
generate marketing content, social media
messaging, email campaigns, blog posts and
product descriptions, which are more targeted.
13. 13 | Realizing the real business impact of generative AI
Two-thirds of jobs in the US and Europe are exposed to some degree of
AI automation.
- Source: Goldman Sachs
Not a threat to jobs – but supercharging employees
We don’t envisage it taking jobs away from people, but realigning
the workforce, freeing up employees to do different, value-added and
interesting tasks. Generative AI can augment human-work creating a
new dynamic in the workplace.
Philip Basford,
AWS Data AI Lead, Inawisdom, a Cognizant company
13 | Realizing the real business impact of generative AI
14. 14 | Realizing the real business impact of generative AI
7. Getting started
Businesses are testing out this technology to see
if it can replace certain tools they are currently
using or make their existing ones more effective.
However, those organizations that are ahead are
those looking at how generative AI can transform
their business and operating models.
1.
Identify golden use cases – Work out where
your company can have a true competitive
advantage deploying generative AI. You can
use our Discovery-as-a-Service methodology to
achieve this.
2.
Take stock – Assess whether your businesses
has the necessary technical expertise,
tech systems, data architecture and risk
management processes in place to account
for generative AI, especially in its responsible
and accountable use.
3.
Start small – Contained, focused foundation
models, cheap to fine-tune, fast to run, with
easily accessed data will win. Businesses can
now use a pre-trained model with additional
labelled data, where a general-purpose model
is adapted for a specific task.
4.
Build an MVP – A minimum viable AI product
can now be deployed in your organization.
This pilot will help businesses realise the value
of generative AI quickly. An MVP will also allow
you to gather valuable feedback for future
development.
5.
Think data – When using generative AI, you
should guarantee clear data ownership and
establish a review process to prevent incorrect
content being produced, as well as protect the
proprietary data of the organization.
It’s all about the value
proposition
We concentrate on the value
proposition when it comes to
generative AI - what’s the real value, the
opportunity and return on investment?
The focus should be on what problems
is AI trying to solve. Businesses should
do some ideation around this and
calculate the real business impact. For
some companies generative AI is not
always appropriate to develop at this
time, even though it may have huge
potential. It depends on the maturity
of the business and its digital and data
architecture, or the investment needed.
Philip Basford,
AWS Data AI Lead, Inawisdom,
a Cognizant company
15. 15 | Realizing the real business impact of generative AI
“
Using generative AI in an
innovative way matters. Asking
the question - how are you trying
to transform your business?
This is where client conversations,
use cases and ideation are vital. ”
Philip Basford,
AWS Data AI Lead, Inawisdom,
a Cognizant company
8. Use cases
Unlike previous AI and machine learning
technologies that have distinct use cases,
generative AI’s applications are so expansive that
they can be hard to nail down, (see boxout.) Yet
the lifeblood of generative AI involves access to
data, fine-tuned for a specific business problem.
Those tasks that have extensively relied on
repetitive and heavy, human processing are likely
to be the ones that can be easily automated with
generative AI. For instance, dealing with large
quantities of verbal or written communication
or image processing. In the past tech tools have
struggled to properly automate these processes.
Vast document summarization, query, data
extraction and smart assistants are the low-
hanging fruit for generative AI.
Case studies include:
•
Octopus Energy – The company rolled out
generative AI in seven weeks. It now answers
44% of customer emails, in part, doing the work
of several FTE’s . “Emails written by AI delivered
80% customer satisfaction,” says Greg Jackson,
CEO of the company. Staff supervize the
answers generative AI provides, it can also draft
personalised response that a human team
member then review before sending on.
•
Duolingo - Engineering teams are now 25%
more productive since adopting GitHub
Co-pilot, a generative AI tool, it is poised to
support developers helping them write code
faster, not displace them. Duolingo is also using
this form of AI to help language students get in-
depth explanations of why their answers to test
questions were correct or incorrect.
•
Khan Academy – The global education provider
has launched a pilot of Khanmigo, its virtual
guide that uses GPT-4, supporting teachers
and pupils. This chatbot acts as a tutor guiding
students through maths and science problems.
The not for profit has given the AI model
additional training and crafted custom prompts
for specific learning situations.
16. 16 | Realizing the real business impact of generative AI
What is generative AI good for?
•
Content creation: Generative AI models can be used to create new and
original content, such as images, videos, music, and text. This can be useful
for content creators, marketers, and advertisers.
•
Product design: It can be used to create new designs for products, such as
clothing, furniture even buildings. It can create digital twins for new product
development or to simulate running systems.
•
Personalization: Models can create unique experiences for individual
users, based on their preferences and behaviour. It also has applications in
personalized chat-based applications, expect more targeted marketing and
truly customized brand engagement.
•
Data augmentation: This type of AI can generate new data samples that
can be used to augment existing data sets. This can help improve the
performance of machine learning models.
•
Simulation: Generative AI can be used to create simulations of real-world
scenarios, such as weather or traffic patterns, even human behaviour. This
can help predict complex systems. It can also be used in the metaverse
generating more accurate digital twins.
16 | Realizing the real business impact of generative AI
17. 17 | Realizing the real business impact of generative AI
“
Maturity in terms of being able to use generative AI, or data issues,
should not be a barrier to adoption. Any company can use AI in a bite-
sized, innovative way. ”
Philip Basford,
AWS Data AI Lead, Inawisdom, a Cognizant company
“
Generative AI, which can not only interpret data but present its findings
in clear, conversational language, raises the prospect of AI not just
assisting the decision-making process, but participating in it.”
Duncan Roberts,
Technology Research Leader, Cognizant
9. Conclusion
Artificial intelligence is already being used
extensively by many businesses across huge
swathes of the global economy. Generative AI is
another step forward in this process, moving the
technology on. Could it create new markets and
business models that no one can imagine today?
Yes, definitely.
If you are already using AI and machine
learning in your business right now, you can use
generative AI. Teaming up with partners makes
it easy to deploy, yet its responsible use requires
skill and understanding. This is where specialist AI
partners and leaders in cloud technology such as
AWS can help make the most of this new tool.
The likes of Cognizant and AWS are at the
cutting edge of generative AI creating services
and solutions to help businesses make the most if
it. They have done the hard work, now it’s time for
companies to take advantage of this technology.
If it is developed responsibly, and with solid
parameters in place, the use cases and potential
for generative AI could be limitless. Legislation
and regulation will help, but organizations
should look to specialists and partners who have
experience in rolling out generative AI to ensure
security and safeguards are put in place.
Generative AI has already been democratized,
which means its deployment will be table stakes
for businesses wanting to innovate and stay
ahead of competitors. Its utilization could also
reduce large enterprises’ manpower advantage
and elevate smaller companies’ capabilities,
since this form of AI can automate many more
human processes.
In fact, the benefits of generative AI could be
exponential. Technology can often move the
needle for a process or operation, a single point
percentage, but with AI, and now generative AI,
outcomes are being improved by double, even
triple digits.
18. 18 | Realizing the real business impact of generative AI
It’s the future
“
We have just started the journey to fully understand how generative AI
will completely re-wire how business is done in 5 years. Enterprise
generative AI requirements that cannot be fulfilled today when viewed
in the prism of cost / benefits (e.g., should I build my own LLM) will
have a completely different answer within the next couple of years. It
is therefore important for companies to keep track of this exciting and
fast-moving space and continuously engage their partner ecosystem. ”
Sean Heshmat,
Head of Data AI, Global Growth Markets, Cognizant
Generative AI could raise annual global GDP by 7% over a
10-year period.
- Source: Goldman Sachs
18 | Realizing the real business impact of generative AI