.

,F SESSION

ucing
nerar ive Al World

Do you know what is Generative AI ? To introduce it, just ldel

know that it is the creative
form of AI. Let us now notice how creative can Generative AI be,
through the following Sol
activity.

,4ctiviry t Can you spot the real ys Algenerated image ?

Corefully look ot the {ollowing two imoges

A B

ldentifu which of these images is real and which one is created

thraugh generatiae AI ?

Solution. Real Image is : A Generatiae AI image is : B

,4ctivity 2 Can you spot the real rrs Al-generated image ?

Corefully look at the {ollowing iwo imoges

c D

Identifu which af these images is real and which one is created through
generatiae AI ?

Solution. Real lmage is : C Generatioe Al image is : D

 ession i : INTRODUCING GENERATIVE Al WORLD 243
fl.ctivit1 3 Can you $polthe real vs Algenerated image ?
%il, l'
Core{ully look ot thre following two imoges.

aa

E F

Identify which of these im*ges is resl snd whiclz one is crested through generatiue Al ?

Sblution. Real Image is : E Generatiae AI image is : F

Isn't it surprising how generative AI can create entirely new images that took real ?

After looking at these images. you must have figured out that Generative N is a form of
Artificial Intelligence that can create or generate some real tooking images. yes" you are
right, but generative AI is way more than that. It can just not only create images, it can
also create articies and text just tike a human can, or it can even generate music or some
real looking videos.

The fotlowing session, Introduction to Generative N would talk about what it is, how it
works, its advantages, challenges, and so forth.

E
 J

l\ Whot is Generotive Al ?
lnlroduclion Io l\ Generoiive Al vs Conventionol Al

Eeneralive AI l\ How Generotive Al Works ?

/l Generotive Al Tools
/t Elhicol Considerotions of Using Generoiive
AI
l3
2.1 INTRoDUCTIoN

You all are aware that AI, Artificiat Intelligence,

has taken the wortd by storm. Everyone
aware of it and everyone is using and is
benefitting from it directly or indirectly. There
different types of AI that work in different are
*uy, fo, different purposes.
AI belongs to one of these three main types
:
L
(i) Narrow AI' This AI can only perform specific,
narrow tasks /ike tmage recognition/
creation or language/text processing.
virtual assistants like sirz or Alexaare forms
of narrow AI focused on understanding voice
commands.
(ii) Reactive Machines' These AI systems
can only react to current situations
and Tabte
stimuli' without learning or building memories.
iramples include spam filters and
recammendation systems that suggest
content based on your current behavior. 5,ffc
(izi) limited Memory AI. This A] can 1,
store knowledge, learn from it, and then
that learning to perform new tasks and make predi.,lonr. use
It's used in applications
ltke chatbots' self-driving cars etc. This 2
is one of the most common types of AI
today.
Generative AI is a type of narrow AI.In this session,
we shail discuss
what Generative AI is,
its types' tools' benefits and limitations of generative
AI, and ethical considerations of 4,
using generative AI.
(
Generative Ai is a type of narrow AI,
that can only perform
specific, narrow tasks.
2.2 WHAT IS GENE RATIVE AI } -
Generative AI is a type of artificial intelligence that can
create new content of different types,
such as images,
mwsic, text, or even vtd.eo.s. Imagine you
have a robot that
can paint pictures, write stories, or
compose songs. That,s 2.3.1
what Generative AI can do! It does this
by rearning patterns
from a large amount of existing data usini -
special
atgorithrns and then create something
new that is similar. @ onu.r z

244
 Session 2: INTRODUCTION TO GENERATTVE At 245
Another example use of generative AI could be this. For
example, if you train a generative*f on millions of pictures
of cats, it can then generate brand new images of cats that
ve Ai
is a tvpe
that
,il
itl
look real but have never existed before. It essentially learns of

what a cat looks tike by studying the training data. liil

iil

are bas:c :r ==:.=i.iir e -{L

2.3 GENERATIVE AI VS CONVENTIONAL AI

Regular AI or conventionalAl is good at classifying and predicting based on existing data.

But it can't create brand new data from scratch. However, generative Ai is creative and can
create new things. It can take what it has learned and use that to make something totally
new and originat, much like an artist paints a new picture.
Conventional Al follows rules to perform tasks. For example, a calculator uses math ruies to
solve problems. In contrast, Generative AI creates new things, €.g., DALL-E creating an
image. Generative AI doesn't just follow rules; it iearns patterns from data and uses these
patterns to generate new content.

Following table lists some differences in Generative AI and Conventionat AI

Tabte 2.1 Generative AI vs. Conventianal AI

S.lfo. Conventional AI Generative AI

7. It is good at classifying and predicting based on it is good at creating new content of different types.
existing data.
2. It cannot create brand new data from scratch It can create brand new data from scratch
3. It fotLows rules to perform tasks. It doesn't just follow ruLes; it Learns patterns from
data and uses these patterns to new content.
4. it is l"imited to doing atready ptanned jobs and It is more creative and ftexibte and can create new
tas ks. things.
5. It js suitabte task-specific responses, e.g., It can create innovative and creative content, and it
reat-time response in automation, or vehjctes etc. can create a va of thin
6" It requires clear rules and standards. It can create content without preset clear rules.

create

2.3.1 How Generotive Al Works ?

Generative AI works by learning from data. Generative modeis work by first being trained
on a large dataset, like millions of images or billions of words of text data. During training,
the model learns the patterns and characteristics of the data. For images, it might learn
features like shapes, textures, colours etc.
I
 246 ARTTFtciAL INTELLTGENCE-rx

lt Then when you provide the trained model with a text prompt or seed data, it uses what it
learned to generate brand nevr content that matches the patterns in a realistic way. It's
like the model has absorbed and understood the rules or statistics behind what a
image or text shouid look like. So its generated output matches those learned patterns.
Thus, we can summarise that generative AI works by following these steps :

(i) Training with Data. The AI looks at lcts of examples (e.g., pictures of cats)
(iz) Learning Patterns. It iearns the patterns in these examples (e.g.,
fur, eyes,
whiskers).
(fi4 Generating. It uses these patterns to create new images (e.g,, of cats that
realistic).

2.4 TYPES oF GENERATIVE AI

Generative AI comes in different types. Few main types of generative AI models are being
Iisted below :

l. Generotive Adversoriol Networks (GANs)

GANs are like a game between two AI systems (two com peting A] -based neural networks
P Generator and the Discriminator. The Generator creates new c0n ten t, and the
A tries to guess if the content is real or fake. They leam from each other until the
R becomes very good at creating realistic content.
T
1
2. Voriotionol Autoencoders (VAEs)
VAEs learnto compress data into smaller bits and then reconstruct it. This hetps
understand the data's underlying patterns and generate new, similar content.
3. Tronsformers
Transformers are used in language models tike ChatGPT. They tearn from lots of text
and can generate new text that looks like a human,s work.
4. Diffusion Models
Diffusion models learn data distributions to generate things like photorealistic imag
from text descriptions.

2.5 EXAMpLES oF GENERATTvE Al

Generative AI is used these days in man5r fields. Some real-life examples of how Gene
AI is used are :
(i) Image Generation. AI can create reaiistic images of people who don't exist. This
useful in movies and video games.
(ii)Music Composition. AI can compose new music pieces by iearning from existing
(iu) Text Generation. AI can write stcries, articies, and even chat with you like a human.
(iv) Video Creation. AI can generate new videos or enhance existinE ones by
details"
r'

t
 Session 2: INTRODUCTION TO GENERATIVE Al 247

it Following figure 2.1 shows some common use cases of generative AI. !

['s
ffir&
r&1
* .EI ffi{
ffi
lmage Video 3D shape
Visual content
en hancern e^: prediction generation

qd.
* ,/
qp,) -
A
la
ok Audio generatron lrluslc Text-to-speech Speechto-speech
composrng generator conversron

+ /+
</ i': '. ;lM

H-
Go

Text Creative Chatbots Translation

generation writing

trtrn
:ir{ }:iili + E#
Code Code Code
generation

Figure 2.1 Generative AI use cases

2.6 BENEFITS AND LIMITATIONS OF USING GENERATIVE A!

Generative AI has made fast inroads ln our daily work and activities. There are many
benefits of using generative AI models, hut there are cedain limitations too.
For example, generative AI offers these benefits :

+) Greativity. It
can aid creativity and automate content creation for things like
marketing materials, artwork, code generation and more.
o) Efficiency. It can save time by automating routine tasks like generating product
descriptions or design mockups.
*) Innovation. Generative AI can explore new ideas and concepts that humans may not
think of.

M
+ Data Generation. It allows companies to quickly create Synthetic data like images to
train other AI systems.

;is generatine neri

ls.
Iiowever, Generative AI also has some limitations :

*) Quality. The output can be nonsensical, biased, or contain mistakes since the model
mg
just iearns patterns.
 248 ARTTFTcTAL TNTELLTcENCE rx

r) Bias. Models can unintentiona[y learn societal biases present in their training data, J.

which will impact their outcomes (which witl be biased outcomes). fpl, -

r) Ethics. There are potential ethical issues around things like deepfakes or impersonating
real people /data. +

+) Quatity. Al-generated content may not always be high quality or factual. So users
mustcheckfactsandensurethecorrectnessof databeforeusingit, e.g.,ifyougotan
article related lo policy implementation related to SDG tn your area, then you must
verify the produced content's facts and correctness.

Following table lists the benefits and limitations of Generative AI I

Table 2.2 The Benefits and Limitations of Generative AI

2.7 I
S./Vo. Limitations
G
7. It can help artists and musicians create new works. Sometjmes the generated content isn't perfect and
d
needs human improvement.
P I
2. It can come up with ideas that humans might not If the training data has biases, the AI wjtl" atso be
A think of. biased.
R
3. It can generate content quickly, saving time and There are ethical concerns about using AI to create
T effort. fake content.

L; 4. Generative AI can create synthetic data jn butk, AI-generated c0ntent may not always be high
which proves usefu[ in situations where buLk data is qual,ity or factua|'.
required. such as training AI systems.
,-;r.
@*
Table 2,r

7 Use Al to Generate Real Looking lmage using GAN Paint

1.

GAN Paint, a tool that lets you modify images with AI. You can ad.d trees,
buildings, and more to a picture iust by clicking! Use URL : ganpaint.io
1. Go to ganpaint.io in a web browser.
?
2" It will show a list of images in the lower side and some predefined paint tools,/brushes
on the left of image such as to paint grass, trees, clouds etc. (see below)

I
Pick brush
type

Pick 2. Texl
action
Pick

l lmage
 Sesslon 2 : INTRODLCi:.. -: _-: .:: _
Z4g
3.
Explore -,.,;th -,-arious tools,
actions and images.
Follon'ing imaqes show the
before and after usingcroud
brush. rutt

Before After fl

2.7 GENERATIVE AI TOOLS

Generative AI tools leverage
machine learning models
diffusion models to create like GAlfs, vAEs, transformers, ili
nu* ryn*,etic data outpritJ Like and.
These toots have a wide images, text, audio,and
range Jf potential applications. more.

Various Generative AI tools

are given beLow :

I lmoge Genercrlion Tools

These toots provide features
(realistic digitar brushes)
i*t'l::HT;Tl,'.T #ili[l5 il;":;#l;
or use prompts to generate
Generative ; ;;;. s eneratio n
fable 2.3 Generative AI Image Generation Took
Tool
7. irii'i F=ir:! Poi nt ts
GA a too th a t It is usefuI for artists/ An a rch itect
a 0 s use S t m0d fy to quickty proto_
destgners ca n u5e GAN
m se5 bv dd ns 0r type changes in images Pa n tn u ze h o
re mov'i n a d di ng
trees might c hange
s o bjects ti ke trees, without manua[ty editing.
bui or cto u ds Io 0 k of a b u tdi n g s
Lh e
2 il***Ai-i surroundi n
DeepArt uses AI to turn
photos into artworks by
it hetps artists / hobbvists A user can transform
create unique art pieces yacation photo into a
apptying stytes of famous
lrom ordinary photographs. an
artists. artwork that tooks like it
was painted by a g-eaie,
a rtlst.
Some other Generative
AI image generation tools are : DALL_E
Diffusion, Jasper Art, prompt
Hunt, NightCafe, AutaDraw,
, MidJourney, Stable
Designs.ai, and many rnore.
2. Texf Generotion Tools
Generative text generation
tools can write human-like
of Generative AI text generation text on almost any topic. Examples
toors are being risted in the
tabre below.
 250 ARTTFTcTAL TNTELLTGENCE-rx

Table 2.4 Generative AI Text Generation TooLs

.l
Tool Description Application
1. !.-:itr !: : These are examptes of con- These are usedfor creating Businesses use them to
ii:.,-s.a,. versationaI AI that content, customer service provide 24f7 customer
1i i ir i l''j i, generates human-[jke text chatbots, and even for support through chatbots,
i,::-:',1!*: jit based on the input (prompt) educationa[ purposes. answering common queries
'it receives. from customers.
2. ali:;:-::r::t+r: AI Dungeon js a text-based It provides an interactive Players can create and
game that uses AI to gene- and engaging storyteLLing exptore their own unique
rate interactive stories experience for entertain- adventures, with the AI
based on the player's ment and education. generating new scenarios
choices. based on their decisions.

Some other Generative AI text generation tools are : DuetAI, Bard, AnyT,lford, and many more.

3. Music Generotion Tools

Generative music generation tools can create real music of different genres. Examples
Generative AI music generation toois are being listed in the table below.

Tabte 2.5 Generative AI Music Generation Taak

Tool Description Application Example

P
1". Amper Music is a p[atform It is ideaI for video A YouTuber can generate
A that allows users to create creators, marketers, and background music for thei.
R custom music tracks using musicians to produce videos without worrying
AI. royatty-free music quick[V. about copvriqht issues.
T
2. r::,": ra.,i.i,-:: AIVA cornposes originaL It is used in creatinE sound- A fiLmmaker can use AIVA
:::;a i:iri:r:::,r music by Learning from tracks for fiLms, video games, to generate an originaL
1.i1 : t.. :::i :i ii i:-ii : existin g compositions. & other multimedia projects. scoi'e for a movie scene.

Some other Generative AI rnusic generation tools are : Baamy, S?ANDRAW, Soundful.
Loudly, Musicfy, FM, and many more.

4. Video Generotion Tools

I
Generative video generation tools can generate entirely new video clips from teil
tools are beins listed in the tabte betow'
,",," ,.: :;;:": ::;::::,:,:::,:;neration I
Tool Description Appltcaticn Example
1" l.:,;t,: i:.:1,.: Deepfake technoLogy uses It can be used for entertain- In movjes, deepfake
i;l:rlr-:i:ril',r AI to create reaListic videos ment. eciucationaL content, technology can be used to
where people appear to do and even in political satire bring historica[ figures to
or say things they never (though it has significant ljfe or to de-age actors.
did. ethjcaI considerations).
I Synthesia aL[ows users to It is usefuI for creating A company can use
create AI-generated videos training vjdeos, marketing Synthesia to create a series
with cirstomizabte avatars. content, and personaLized of training videos featuring a
messages. consistent virtuaI presenter.

Some other Generative AI video generation tools are : Colossyan, Hour 1ne, D-ID, Elai,
HeyGen, Runway, Pictory, Deepbrain AI, InVidea, and many more.
 257
Benefits of Generotive Al Tools
Generative AI tools have become very popular I
as they offer the forlowing
benefits :
t" herp artists, writers, and musicians
::fT:;?;T::f:1'il:,I* by proviriins

.(l') Efficiency' They can produce high-quality

content quickly, saving time and
t"" complex content creation
effort.
accessibre to peopre without
ittffitl#i1,,[1t'make
(iv) Personalization' They allow for the creation
individual preferences and needs. of personalized content tailored
to

2.8 Fr'-lftnl lqryllqq$4rloNs oF u SING GENERA TIVE AI

As generative AI becomes more
advanced and widespread,
we need to think about the
Ar toois orrer numerous benents,
,*['it#']ln::11';fff}:X::i.'ative they arso raise
*) Misinformation' Al-generated content can be
used to create fake news or
information' For example, during misleading
elections, many fake news or
misleading content,
Generative Ar, were circurated
which some peopl. ,oura not
:lTi:1,:l:rr'sh suspect
t) Bias' If the AI is trained on
biased data, it can produce
generative AI ad for hiring biased contenr, ror instance,
people disptayed high-pa;rg a
than to women' because oitt',. positions to maLes more often
iata it *u, tran.la witi. wiricrr had
biases against women.
t") Privacy' using personal
data to generate content can
example' you must have read raise privacy concerns. ror
about or seen some fake videos
famous peopre. peopre's personai circulating of some
data such us infor*rtion, pictures
misused by generative AI etc. can be
to create fake content, which is a
*) copyright viotations' Al-generated big threat to privacy.
content might unintentionaly
existing copyrights' For example, infringe on
a generative AI tool created a
inspired by a novel, which script for a movie
is copyrighted. This i, .opvriglrt
infringement.
generative Ar tools, one
l[il;llill!-:;:l,1il must folrow and ensure the fouowins

(i) The users must ensure that

the AI training data is fair and
unbiased.
,', protect individuar privacy
l|,u,#:ff.frX': and prevent il;; spreadins
(iii)
users must consider challenges
iike automated generation of malicious
expLicit content and must code or
hJve ways to handle it.
(iv) users must honour intellectuat
property and ricensing. They
Al-generated content is not must ensure that the
,riotrtirg il,ir.
To hetp guide ethical AI
development, groups tike the
IECD (\rganisation
far Economic
';:':::;y;!r,{,!r",,ii'{#;X?r:lHl-posed principres around transparencv, ,obu,tnuss,

!
 62 ARTTFTcTAL tNTELLIcENCE-rx

'l Technology companies depioying generative AI witt need 1.4. A

clear policies and practices to
identify and mitigate ethicai- risks. Humans should always have G
oversight over these
powerful AI systems. G
15. W
ar":"*m 16.
(n

Ge
MuLttpLe Clw'tce @estio?rs (MCqd (a
i. What is Generative AI ? (c
A type of AI that follows rules to perform tasks
(a) 17. Vill
(&) An AI used for data analysis (I
(c) A type of AI that creates new conten t (d) A robot that can paint pictures G
2. Generative AI can create: 18. An
(a) Only text (b) Only images (t
(c) Only music (d) Images, text, music, and videos 19. Ar
-',. Which of these is an example of narrow AI ? (a
(a) A chess engine (D) siri or Arexa (c) A spam filter (c
(d) Both (a) and (&)
4" which of the follou.ing is a tool that turns photos into artworks 20. wI
?
(a) cAN Iraint (&) DeepArt (c) ChatGpr (o:
P @) ArDungeon
5' what are Generative Adversariar Networks (GANs) composed
of ? 21. Wl
A
(a) Generato, and Discriminator (b) Encoder and necoder (d
R
T
(c) Painter and Viewer (d) player and Opponent (cl
6' which Generative AI toor is used for generating human-rike text ? 22. Cn
(a) DeepArt (b) Amper Music (c) ChatGPT (d,
(d) Synthesia
7. What does VAE stand for in Generative AI ? (c)
(a) Variational Adversarial Encoder (b) Virtual Artificial Environment
(c) Variational Autoencoder
(d) Visual AI Encoder
8' tool allows users to create custom music tracks using AI
\Atrhich
? 1.
(a) AIVA (&) DeepArt (c) Amper Music
9. Deepfake technology is mainly used for :
(d) GAN paint 2. G
3. G.:
(a) Creating realistic videos (b) Composing music 4. Tr;
(c) Writing stories
@) Analyzing data 5.
10. \Atrhich of the following is a key benefit of Generative
AI ? 6. Cei
(a) Reduces creativity
(b) produces low-quality content 7. Tht
(c) Enhances efficiency (d) Decreases accessibilitli 8. Ge:
11' A tool that creates Al-generated videos with customizableavatars is called: 9. To,--
' (a) ChatGPT synthesia
(b) (c) AI Dungeon (d) GAN paint 10. The
72. Ethical concerns of Generative AI include : 11. Tle
creativity
(a) Increasing
(b) Speeding up processes 12.
(c) Misinformation
(fl Enhancing accessibility 13.
13. izVhich type of AI learns from text data to generate new text ? 1,4. c,{
(a) GAN (b) VAE (c) Transformer (d) Deepfake 15. VA
 Session2: l.,irl----, -' --: :r - . 253
i4. A comm.rrl '.1sc tri AIVA is :

(a) \\-ritir-r,q articles (&) Generating realistic images

(c) Composing music (d) Creating video games
15. !\hich AI tool is used to modify images by adcling or removing objects ?

cAN Pai.t
(a) (b) DeepArt (c) chatGpr (d) Amper Music
16. Generative AI tools can help artists by :

slower
(a) Making content creation (D) Increasing the compiexity of artwork
(c) Providing new h'ays to create content (d) Reducing the need for creativity
lL7- which of the following is Nor a limitation of Generative AI ?
(a) Qualitv issues (&) Bias in data
(c) Creati'itv enha,cement (r/) Ethical concerns
lc$"An AI tool that allows users to plav text-based interactive stories is:
(;) Synthesia (b) GAN paint (c) AI Dungeon (d) DeepArt
19. A common ethical issue with deepfake technology is :

(a) trmproving video quality (b) Creating personalizecl content

with fake videos
(c) Misleading viewers (ri) composing music for movie_s
2ij. which grolrp proposed principles for ethical AI der.elopment ?
(a) OECD (&) OpenAI (c) Deep\{ind r,jr Ant}.,ro:rc
2l ' What type of AI is used in Runu.avlVll lor creatrng irnaees arLd r iric..s I
(a) Conventional AI l,r Generatire -\I
(c) Anah.tical Ai (ri ) Reactir-e -{l
??. Generative AI is different irom conr.entional AI becar_rse it :

rules
(a) Follows strict (b) Creates ne\^, content
(c) only a.alyzes data (d) Does not iearn from data

Fill in the Blanhs

1. is a type of artificial intelligence that creates new content.
2' Regular AI is good at classifying and predicting based on _ data.
-
3. Generative AI learns by studying patterns from a large _ of data.
4. Transformers are a type of model great at _ text.
5. -- are used to create photorealistic images and deepfakes.
6. Generative AI could potentially learn and amplify societal
7. The has proposed principles around ethical AI development.
8. Generative AI can explore ideas that humans may not _ of.
9. Tools like DALL-E and stable Diffusion use _ to generate images.
10. There are concerns around the _ property of Al-generated content.
11" These models leam the behincl what real data looks like.
72' is a tool that allows users to modify images by adding or removing objects.
i3. _ uses AI to turn photos into artworks.
14. -GANs consist of a Generator and a _ ,

15" VAE stands for _ Autoencoder.

 254 ARTTFTcIAL TNTELLTGENCE-lx

-ir. _ is a tooJ that creales Al-generated videos r,r''ith customizable avatars.

l'
17, One limitation of Generative AI is potential in data'
iil. ChatGPT is an erample of a model.

i. Generatjve AI can create entirelY ner'r'data like il-rages and tert' 3.

.L. GANs i-rat'e onh' one ner'lral netrr-ork'

;i Narro.,r- AI can onlr- perlorm speciiic, narrorr' tasks.
4. Limited memorv AI cannot store knol'r'ledge or learn'
5. Generative AI creates new content by foliowing fixed rules'
ij. VAEs are used for image generation.
:'. Generative AI can be used to enhance videos'
$" AIVA is used to generate realistic text.
9. GAN Paint can add or remove objects in images'
Ji!. Ethical concerns of Generative AI include misinformation.
ii Generative AI always produces perfect content'
'l
2. Generative AI onlv analyzes data r,\,ithout creatilrg new content.
ii. ChatGPT is a Transformer model used for tert generation'
P ii DeepArt is an AI tool for cor:rposing music'
A
i 5. One limitation of Generative AI is potential bias in data'
R
ir,r. Generative Al is different from conventional AI because it foliows strict rules.
T

ffi !. A company wants to create a series of training videos quickly without hiring actors' They r'r'ar':
customizable avatars that can deliver training content effectively.
(l) Which AI tool should theY use ?
(a) DeepArt (&) Synthesia (c) AI Dungeon (d) Amper l\4usic

(ll) Why'is Synthesia a good choice for this task ?

1,t) lt conrposes music

(h) It creates Al-generated r,ideos with customizable avatars ).
(c) It generates hr-rman-iike text (ri) It turns photos into artworks
(iii) If thev have to rlse a tool ot1'Ler t1-L.rr-r Svnthesia, u'hich of the foilowing toois can help ther:
create videos ?

(a) Claude AI (b) Boomy (c) Runway (d) Gemini

; An artist wants to experiment with creating artwork in the stvle of famous painters using AI'
(l) Wrich tool should the artist use ?
(a) ChatGPT (Il) DeepArt (c) GAN Paint (d) Synthesia

(lf \\fnat benefit does DeepArt offer to the artist ?

(a) Generates text
(&)Turnsphotosintoartworksinthestyleoffamouspainters
(c) Modifies images by adding objects
(r/) ComPoses music
 (iii) U the artisr IS e\ploring using AI to generate 111'\tr. -:.-:-.-' - . rr.: i'I^'c::i.'' \1 \
"''''i'-''
approa.h .reates new images ?
(r;) Reiniorcement leaming (b) Ciassificatit-rL i:-' -=
(c1 Ger-rerative adversarial network (CAN)
(d) Supervised learning

3 A writer wants to include some conversations for their novel using AI.
(i) Which AI tool should the writer use ?
(a) Claude AI (b) Svntlresia (C) Cl1atGPT (d) An.iPgl \'1....':
(ii) The tool(s) that vou chose for the above question, what do they do?
(a) generate images (i;) create music tracks
(c) generate human-1ike text conversations
(d) create iots of text
(lli) What type of AI system would it be it also includes language translation?
(a) Narrorv AI (&) Reactive machine
(c) Limited memory AI (d) General AT

4. A company r,rrants to automatically generate product descriptions for e-commerce rt'ebsite.

(l) Which generative AI tool r'vould be most suitabie ?
(a)DALL-E(1l)ChatCPT(c)Boomv(il)Sor-rr-rdf..r1
(ll) \A{rat is a potential lirnitation of using gencratrr c AI like C}-utGPT iuri {cr'..i.rtir'.q Pr!rL'11-1.i

descriptions ?
(a) The olitpllt tert rr-111 airr-avs be faclr-ralh' actlliate'
(li) It cannot handle iong-form content.
(c) The descriptions may lack contextual understandiug'
(d) The AI will require too much computational power'
(lli) What approach could the cotnpany take to impror.e the quality of Al-generated piodr-rct
descriptions ?

(a) Use narrorr/er training dataset

(b) Combine the AI output with human post-editir-rg
(c) Train the model on competitor r,r,ebsite data (d) Reduce the model's complexitv

:. A biotech firm wants to accelerate drug discovery by generating new molecular structures.
(l) Which generative AI model shouid they use ?
t:n-I (a) A diffusion model (b) A variational autoencoder (vAE)
(c)GPT-3(d)Aconvoitrtionalneuralnetwork
(ii) what is a key chailenge in using generative AI for drug discovery ?
(a) Models can proPose molecules th.rt violate chemicai rules
(b) The compute costs are too high (c) Models cannot learn 3D molecular .'stftrcttlres
(d) Lack of large, high-quality training data
(lil) How. can firms address some of the iimitations in this application ?

(a) Use simpler machine learning models

(&) Combine generative AI u,ith other rnethods like virtual screening
(c) Train the model onh' on failed drug candidates
(d) Reduce the complexity of the model
 256 ARTIFIcIAL INTELLIGENcE-lx
*G
v
fusertion & Reasoning Qgestlotts
*A
+ siarerreni *i :==s:;':'. iE)"
I
*; ass*.-ti*r* iA) is i+ii*w** b:r
i|']::*i::.r.r.=: i.*=ri*e:, a siaiei:ieni *,1

R is the correct expianation

of A' )rolut
(a) Both A and R are true and of A'
R is not the correct explanation
(b) Both A and R are true but 1. Unr
partly true)'
(c) A is true but R is false (or
(,1)Aisfalse(orpartlytrue)butRistrue.(e)BothAandRarefalseornotfullytrue. mt
text, music, and videos.
as images,
AI can create new content such
I

], Assertiqrrr iA}' Generative rle(

Reason{R').GenerativeAIuSeSpattem,-t"u,,","afromexistingdatatogeneratenewcontent.
objects to an rmage' 2.
2. Assertioi-r {A)' GAN Paint can add text'
Reason iE)' GAN Paint
is used to create human-like
mtlslc'
is mainly used for composing
l. ,{sserticr 1A'i' Deepfake technology videos where p"opt" appear
to do or say things
creates
Reasirr-. rRi. Deepfake technology '"uirr,t. 3.H

'u**'' t
n
P il- I\
A
R
T ffigff41g*!f;#ffifi#::;1 ! f

D
' *:**t3J*Id:q;iliiiT:IJTJ#ff:'r
t'i"' it to classify as real or fake'
s"""'u'or 'fha'f
crea'1es ou'fpu'f'
"'' ^l
discriminatol tt''ui
- il:f ff :::*l:, .r unin'1ended'u'fpu'is' " *"' 6

l:i*:'a,'t*:"f "tj;*:;l;iled
careful human monitoring
ani controi lens'
l
require r^r.."^r.u e1
observed from an ethical
]
oirtcomes must be critically
l
g. Assertion (A). Generative AI s
malicious content generation'
Reason(R).TherearepotentialethicalissueswithgenerativeAllikeprivacyviolations,bias,]
*t#ttti^ip",;;;'t' and
deepfakes, l

{ *t
ru.---
us Rrvtse -, r:
*Nbelongstooneofthesethreemaintypes:NarrowN,ReacttveMachines,and.LimitedMemoryN. s.N
+ Generative N is a tYPe of Narrow N' that can create riew content
of dffirent types, such as
arttficial interigence
* Generative N is a we of
videos'
images, music, text' or even
on generative AI'
.+ Popular tools like ChatGPT and DALL-E are based" but can,t create new data or
predicting based on existing data,
:, Regular Al is good at classifuing and

things. However, genera'tive'U

is creative and can create new things' (iii) Generating.
wtth Data, (ii) Learning Pattems,
.. these steps : (i) Training
by
Generattve AI works following (GANs), Variational
N mod'els are: Generative Adversarial Networks
:;. Few main types of
generative
and Difusion Models'
Autoencoders $msi' Transformers'
 Session 2 : INTRODUCION TO GENERATM At 257
+ Generative AI is used fur many things, such as : Image Generation, ttusic Composition, Tart
Video Creation, etc.
Gmqation, \
* Although Generative N tools offer numerous benefits, they atso ratse
important ethicalissues, such as:
Misinformation, Bias, privacy, Coffight Violations, etc.
* Before using the generative AI tools, one must follow and ensure ethical cansiderations.

lJ
-\ olution Time
*@#

7. lMat is Generatiae Al? lMat type Al is it ?

Ans' GenerativeAl is a type of artificial intelligence that creates new content

such as images, tex!
music, and videos by leaming patterns from existing data.
Generative Ar is a type of narrow Ar that can only perform specific,
narrow tasks like image
recognitionL/ creation or language/text processing
2. IMat are the two main components of GANs ? rNhat are their rores ?
''Ans' The two main components of GANs are : the Generator and the
Discriminator.
The Generator creates new content, and the Discriminator tries
to guess if the content is real or fake.
These two work in tandem by helping each other and improving
the content being generated.
3. How do text generator toors tike Crqude, Ar, ChatGpr generate text
?
Ans' Text generator tools like Claudc, Al, ChatGPT generate text
"they by-identi$.irg
leaming from large amounts of
text data' From the bulk of data fed to them, starting pattems and their
interconnections. Then using the leamt patterns, they can create human-like
text.
4. Name a to.ol that uses AI to turn photos into artuarks.
Ans. DeepArt.
5" lNhat does VAE stand
for ? V{hnt daes it do ?
a Ans. VAE stands for Variational Autoencoder.
VAEs learn to compress data into smaller bits and then reconsfruct
it. This helps them understand
the data's underlying pattems and generate new, similar content.
:y 6. lNhat nre softrc ethical concerns related to Generatioe AI ?

Ans. The main ethical concerns related to Generative AI are :

x Misinfonnation. Ar-generated content can be used to seate fake news or misleading
information.
3/ x Bias. If the AI is trained on biased. data it can produce biased
content.
x Privacy. Using personal data to generate content can raise privacy
concems.
x Copytight violations. Al-generated conterrt might trnintentionally infringe
on existing copyrights.
7. l&at sre sorne benefits and limitations of Generatioe Al ?
Generative AI offers the follow,ing benefits and has the following,
Iimitations.
S.lfo.
Limitations
1.. It hetps artists and musicians create new works.
I
Sometjmes the generateci content jsn,t perfect and
needs human im nigv,ertrgrlk_
2. It-comes up with ideas that humans might not think
of. I If the training data has biases, the AI wil.t also be

It generates
-T-*
biased
3 content quickty, saving time and effort" There are ethicaL concerns about using AI to create
fake content.
4. It can ata in butk, which proves Al-gen erated content may not atways be high
usefuI bu[k data is required, such quatity or factual.
as AI

UNIT I\' : lNrRooucrtoN To Ce NrRarrvr Al

 258 ARTIFTctAL INTELLIGENCE-tx

8. How is generatiae AI dffirent from Conaentional AI ? 4.

Ans' The di:Eferences in Generative AI and convendonal AI can be summarised
as
S.ffo" AI AI
q
!o9d at classifying and predicting hased on existing
1.
Good at creating new content of differen,l types
data.
) Cannot create br.and new data from scratch. Can create brand new data frorn scratch
3 Fottows ruies to perform tasks. Learns patterns from data and uses these patterns to
new content" 6.t
,/^
Lirnited to doi d and tasks. creative and flexible and can create neisl
E
Suitabie for task-specific responses, e.g., reat-time Can create innovative and creative content and
in automation.
it
can a of 7. 1
6. Requires clear rules and standards. Can create content without preset c|ear rules.

Practical A"tir,ty
'1" AI 8. I
can generate human'lit(e speech anil oocals, imitating ooices. Try interacting
with an AI assistant that caA
hazte natuml canoersations- {you ffiay use a
fun chatbit like simsiini - url simsimi.comt or oy otin ol youl

Solution-
& s B: 9-t
1. Open URL simsimi.com Sim5imi stBY,@ SE* A8@S
P
2. Choose preferred language nE;mltd
A
and start Make sure
R "lrutti*g.
to object if you do not like any 10.
T resPonse.
Y!:!r

11. {

@
6*s*Ry 12.
Generative 4f
a type of ar"cificiai intelligence that can create new content of different
types.
canve*tionai A/ a type of afiificial intelligence that follows ruies to perfcrm
tasks, such as clas:iiylng and
oredicting based on existing data. t3.

rt
,?f
"*t'=f.sstgnrneni
14. t

1- Generative AI can create new data like :

(a) Images (&) Text (c) Audio (d) AX of these
15- (
2. Which is NOT a type of generative AI model ?
(a) Generative Adversarial Networks (GANs)
(b) Variational Autoencoders (VAEs)
1,6.
(c) Convokrtional Nzural Networks (CNNs)
(d) Transfarmers
3- How does gerrerative AI leam to create new content ?
(a) By studying a dataset of edsting data and leaming
pattems 17.
(b) T?uough manual programming
(c) It doesn't achrally leam (d) None of tlrese
 lit
- /ht!
LJ} LI

,i. "Ihe prlcess r-ri trair-,ing Generatil,e AI invoives :

(,;) G'lt rules

I,-,i1or,;ing iixed (b) Leamir,: -- - h
(c) igroring data patterns (d) Cenerarir'.i ----ri:-' - '' '-

.i. What r,. the CAN in GAN Paint ?

(a) Gcneratil,,e Adversarial Network (h) Graphics i\nii:-,ati,r:- ..

(c) Generat Art Netr,vork (ri) I'Jone oI the abo- e

TlS tO
:;. Cenerative Al iras potential benefits ia ',r"'hicir alea ?

(rz)Autcrnating cor-rtent creation ii;) Aiding scientific drscover,'

(c) Creating synthetic training data (d) ''dii of these
-r il
?. \M-rat is a kev ethical risk ct' generative Al ?

(a) it cogid aqtomate too many jobs (b) It requires toc'r much coinputing:'r, =

(c) it cor"rld preser\/e societai biases (d) None of the above

l':. trVhich statement about generatir.e Al is (FAL;qE) ?

(a) it
leams patterns from data to create nel'v outpttts
(b) Current s-l/stems harre no iirnitations
(c) It can generate images, text, aueiio etc" (d) lt raises intellectual lircpertlr concerns
t. Ceneratirre AI modeis neeC to be tra:ined on :
(n) A small dataset of examples {li) FianC'written ftr1es and iogic
(c) A large corprls of ctata (d) Notl-ring, thev create clata spontaneouslY
j:,. What is a diffiision moci,ei good at ge;'eI'atir.r-g ?

+
(a) Deepfakes (&) Compr"rter code ic) Photorealistic images (d) Al1 of the above
' i.x
e
r I Generative Ai lliffers froni conventional Ai in tirat it catr :

t (a) Generate entireiy new

(c) itun on very limited
data
hardi.t'are
(&) Sirnpiy clas,eify

{d) Bot}r (&) nnd (ci

or p:redict on existing data

.i,jr. A ianguage moCei ljke GPT-3 is an exar:rple of y,rhat tvpe of generatir":e AI ?

(c) GAN {i;}) VAE (r) Transformer (d) Diflusion model

and j i Wirat does the abbreviation CAIrJ stand for?
(a) Cenerative Adversarial Network (b) Generative Art Net"'r'oik
(c) Cer,"metric Ar-rimation Netw'ork {d) l{cne of tlrt' above
, ,. Generatir.e Ai cannot be useC- for lr'hich o{ these ?

(c) Generating new fasliion designs (&) Automating 'website coding

(c) Providing meclical ciiagnoses (d) Creating marketing videos
:,= Generative AI helps in irrnovation bY :

(c) Sticking to traditional methods (b) Generatin,g ne\^/ idea-o and ccn-ieni

(c) Siorting dorvn the creative process (d) Reciucing ihe numtrer of creative G1-itp1,its

,,: lVhicl-r of the following is an advantage of using Generative AI?

(ri) Al-wa;,s prodr-rces perfect content (l') Can be biased
(c) Saves time ancl effort (di Ethicai concerns
,
- IAt'rat does generatir.e AI do?
(a) Classifies data (&) Creates ner'r' data

(c) Analyzes existing data (d) All of these

lV: lNrRooucTloN To C:\=-=:- .'- I

 260 A,RTrFrcrAL TNTELLtcENCE tx

L8. Generative AI learns patterns from: L.r

(a) A small amount of data . (b) No data at all

(c) A large amount of existing data (d1 Human instructions !i-

19. What does a GAN consist of ?

(a) one neural network (b) Two competing neural networks

(c) Three neural networks (d) No neural networks

20. What is a potential benefit of generative AI ?

(a) Increased cybersecurity risks (b) ]ob displacement

(c) Aiding creativity and automating content creation
(d) More energy consumPtion
21. \A/hat is a limitation of generative AI ?
(a) Requires very little training data
(b) Always provides factual and highquality output
, (c) The output can be nonsensical or biased (d) Eliminates human creativify !L,t- \.
22. \A/hat is an ethical concern with generative AI ?
(c) Faster computing speeds (b) Protecting individual privacy
dr
(c) Lower hardware costs (d) Reduced energy u$age
G
23. Which type of AI just reacts to current stimuli ?
P
(a) Narrow AI (b) Reactive machines A
A
(c) Limited memory AI (d) Generative AI
R
24. What type of generative AI is used for photorealistic image generation ?
T
(a) vAEs (&) GANs (c) Transformers (d) Diffusion models

25. Generative AI can be applied to which of these fields?

(a) Drug discover (b) Computer-aided design
(c) Creating synthetic training data (d) All of these
26. Generative AI essentially leams the behind the training data.
a
(a) Rules & Patterns -- (b) KeYwords
E
(c) Programming logic (d) Mathematical equations
27. \ /hich is NOT a potential ethical issue with generative AI?
(a) Automated malicious code generation (b) Deepfakes and mishJormation
5L
(c) Increased energy efficienry (d) Privacy violations
53-
28. Limited memory AI is commonly used in:
(a) Game playrng (b) self-driving cars (c) Computer vision (d) Both (a) and (b)
29. What type of generative AI can text?
Senerate new
o

(a) vAEs (b) Transformers c) GANs (d) Diffusion models

1-
30. An AI tool that learns from existing compositions to create new music is:

(a)AIVA(b)DeepA:t(c)GANPaint(d)AlDungeon
31.. Whose principles guide ethical AI development ?

(a) OECD
united Nations (b) (c) world Bank (d) European union
L
32. Generative AI requires oversight.
(a) No (b) Limited (c) Extensive (d) Human
 Session 2: INTRODUCTION -O 3a'._:-- 261
,11. Generative AI learns from a -.,r{c 54. A Generative AI limitation is biased or

l.i.
arfiount of existing data.
CANs consist of _
low-quality ourput. (T/F) rl
(a number) cLrn-.i.rr:ii..i 55. Narrow AI is very general intelligenee.
(l/F)
neural nctworks.
56. Reactive machines have memory. (TIF)
3i. --- are great at generating tert.
57. Limited memory AI makes predictions. (T/F)
3{r. modeis learn data distribnh,,ri-L> -r
58. Generative AI learns rules of the data. (T/F)
--
generate images.
59. Chatbots use limited memory AI. (T/F)
.',,'. is a tool that generates images .rlit.L igi.l
-*-
prompts. 60. VAEs learn surface patterns. (T/F)
1", 61. Diffusion models generate from data
AI Fslforr.lr speclfic, narroir :a.ks.
distributions. (T/F)
3! -.--- machines onlv react to cnrrent stimnli.
62. Transformers can write human-like text. (T/F)
"llr" AI can learn and make predictior-rs.
6?. Generative AI raises no ethical issues. (T/F)
ii --_ The provides principles for ethical AI 64. VAEs leam the underlying
pattems in data. fl/F)
development.
65. Transformers like GpT-3 can generate
i: Cencratir c .AI requires _-- supervisiorr. human-like text. (T/F)
+r. _- AI is a type of AI that can create new 66. What is generative AI ?
data like irnages, tex! and audio.
67. tNhat are the tluee main types of Artificial
+-i. Generative AI learns from data ancl identifies intelligence? Which category does Gerrerative
AI belong to ?
45. A Generative Ai tool that helps filrn_rnakers 68. Name three types of generative AI models.
create originai scores is _ . 69- What is a GAN ?
tt$. One ethical concern with Generative AI is the 70- How
do diffusion models work ?
creation of _ news.
71- What does a tool like DALL-E do ?
47. Syrthesia is useful for creating training
72- List a benefit and a limitation of generative AI.
with virtual presenters.
73" IA/hat type of AI do self-driving cars use ?
48. A key limitation is that Al-generated output
may not be fuIly 74. Name some (z) image creation generative Af
or factual.
tools, (ll) text creation generative AI tools,
49" Generative AI classifies existing data. (T/F)
(lii) music creation generative AI tools, (lo)
50. Generative AI leams from a small dataset. (TIF) video creation generative AI tools
51- Generative AI essentially leams the rules of the 75. How is
generative AI different from conven_
data. (T/F) tional, regular AI ?
52. GANs have two neural networks. (T/F) 76- l//hat are some ethical concerns with generative
53- Transformers generate images. (T/F) AI?

Pn"qcrIcAl AsSIfiNM ENT

1" Image Generation. Tools llke DALL-E, Stable Use one of these tools to create text
Dffision, and Midjourtiey :use text prompts to descriptions for various plays which are
generate photorealistic images and artworks. participating in the Arrrual School Theatre
Use one of these tools to create poster for your Festival.
school's annual day. .) Video Synthesis. Using generative AI tools
3, Text Generation. Large language models iike that can generate entirely new video clips from
CPT-3 can write human-like text on almost any texf ereate a short video of a ',dog riding a
topic. sktteboard".
I