Unit 4 - Generative Artificial Intelligence

Timeline of Generative AI
Generative Al has been getting better and better over the years. Now, generative Al can do lots
of cool stuff like writing text, making pictures, and creating new things. It's been a long process
of learning and making things better, but now we can see all the amazing things it can do.

1956: John McCarthy coined the term Artificial Intelligence.

1964: First functioning generative AI chatbot ELIZA was created.

2011: IBM's Watson was introduced.

2013: Variational autoencoders (VAE) were created.

2014: Creation of Generative Adversarial Networks (GAN), which were a breakthrough in

Generative AI.

2015: Diffusion models were introduced, representing a novel approach to generative modelling,
example: Tensorflow

2016: Google DeepMind’s AlphaGo was introduced.

2017: Deep learning architecture-transformer was proposed.

2018: OpenAI introduced Generative Pre-trained Transformers (GPT)

2019: GPT2 was introduced.

2020: GPT3 was introduced.

2021: Open AI launched DALL-E, an AI platform designed to generate images from textual
descriptions.

2022: Two notable AI image-generating tools, the open-source Stable Diffusion and Midjourney
were introduced and ChatGPT was introduced.

2023: OpenAI released GPT-4, an advanced version of Generative Pre- trained Transformer
series. Also Microsoft Copilot (previously Bing Chat), Google Gemini (previously Google
Bard), Adobe Firefly, Meta Llama were introduced.

Page 1 of 5
Answer the following questions:

Q1. What do you understand about Generative Artificial Intelligence? Give any two
examples.
Ans: Generative artificial intelligence (AI) refers to the algorithms that generate new data that
resembles human-generated content, such as audio, code, images, text, simulations, and videos.
This technology is trained with existing data and content, creating the potential for applications
such as natural language processing, computer vision, and speech synthesis.

Examples of Generative AI:

Art: Generative AI is being used to create unique works of art.
Music: Generative AI is being used to create new music, either by composing original pieces or
by remixing existing ones.

Q2. Write the difference between Generative Al and Conventional Al.

Ans:
Generative Al Conventional Al

Goal Generative Al creates new content which Conventional Al analyses,

mimics the original content. This content processes, and classifies data. It
includes images, text, music, or other works to improve the accuracy,
forms of media. precision, recall, and speed within
the scope of the defined task.
Training Generative Al models often trained using Conventional Al models are
techniques such an generative adversarial typically trained using supervised,
networks (GANs), variational unsupervised or reinforcement
autoencoders (VAES) or autoregressive learning techniques.
models.
Dataset Generative Al models typically require Conventional Al models rely on
large amounts of diverse and smaller, more curated datasets that
representative data to learn effectively. are tailored to the task at hand.

Output Generative Al output is fresh, innovative, Conventional Al produces more

and often unexpected. predictable output based on existing
data.
Applications Generative Al used art, music, literature, Conventional Al is used in banking,
gaming and design. healthcare, image recognition, and
language processing.

Q3. Write any two AI tools each for the following-

a) Generative AI image generation tools
i) Midjourney
ii) Magic Studio
iii) Pebblely

b) Generative AI text generation tools

i) ChatGPT
ii) Notion AI
iii) Compose
Page 2 of 5
 c) Generative AI audio generation tools
i) FineShare
ii) Boomy AI
iii) Playlist AI

Q4. Explain the different types of Generative AI.

Ans: Generative Al comes is a variety of forms, each with unique advantages and uses. Some of
the most typical varieties are as follows:
1. Generative Adversarial Networks (GANs): GANs are neural networks that work to
produce fresh data. It is made up of two neural networks which work together to create
realistic synthetic data.
These two neural networks are:
a) Generator Network: It produces the data that is as close as possible to real data.

b) Discriminator Network: It analyses the data and provides feedback, i.e. it takes real data
and the data generated by the Generator as input and attempts to distinguish between the
two.
Some of the examples of GANs are as follows:
 It can create portraits of non-existing people.
 It can convert images from day to night.
 It can generate images based on textual description.
 It can generate realistic video.

2. Variational Autoencoders (VAEs): This is another class of generative models. To

produce fresh data, VAEs learn the distribution of the data and then sample from it.

Some of the examples of VAFs are as follows:

 It can generate new images like the given training set.
 VAE's can produce new text that follows the same style and structure as the training data.
 It can be used for composing new music pieces or creating sound effects, music
composition etc.

3. Recurrent Neural Networks (RNNs): RNNs are a special class of neural networks that
excel at handling sequential data, like music or tort. They excel at tasks where the order
of the data points is important, as they can remember previous inputs and use this
information to influence current outputs.

Some of the examples of INNs are as follows:

 It can generate novel in the style of a specific author of genre.
 It can predict the next character or word in a sequence.
 It can be used to predict future values in a time series.

4. Autoencoders (AEs): These are Neural networks that have been trained to learn a
compressed representation of data. They work by compressing the data into a lower-
dimensional form (encoding) and then decompressing it back to its original form
(decoding). This process helps the network learn the most important features of the data.

Page 3 of 5
 Some of the examples of AEs are as follows:
 It can help in cleaning up noisy images to produce clear and highly realistic samples.
 It can help in compressing high resolution images for efficient storage and transmission.
 It can create artistic images based on learned features from famous paintings.
 It can help in drug discovery by learning and generating molecular structures that have
desirable properties.

Q5. How Autoencoders and Variable Autoencoders are similar to each other?
Ans: The similarities between Autoencoders and Variable Autoencoders are as follows
i) Both AE and VAE are neural network architectures that are used for unsupervised
learning.
ii) Both AE and VAE consist of an encoder and a decoder network. The encoder maps the
input data to a latent representation, and decoder maps the latent representation back to
the original data.
iii) Both AE and VAE can be used for tasks such as dimensionality reduction, data
generation, and anomaly detection.

Q6. Mention the difference between Autoencoders and Variable Autoencoders.

Ans:
AE VAE

Basic Function Neural network model that learns to Similar to AE but incorporates
encode input data into a compressed probabilistic elements to learn a
representation and then decode it back latent space representation of
to the original data. input data.
Reconstruction Loss Minimizes the difference between the Same as AE but also includes a
input data and its reconstructed reugularizer.
output.

Handling Overfitting Can suffer from overfitting due to the Less prone to overfitting due to
fixed encoding structure. the probabilistic nature.

Applications Image compression, denoising, feature Data generation, unsupervised

extraction. learning, anomaly detection.
Training Complexity Relatively simpler training process. More complex training process
due to the inclusion of
regularization terms.

Q7. What are the advantages of using Generative AI?

Ans: Some of the benefits of Using Generative AI are:
i) Creativity: Generates novel and imaginative outputs that can inspire new ideas and
solutions.
ii) Automation: Streamlines repetitive tasks such as content creation, data generation, and
design iterations.
iii) Personalization: Generative AI can be used to create personalized content for individual
users based on their preferences and behaviors, such as customized product
recommendations or personalized news articles.
Page 4 of 5
 iv) Efficiency: Generative AI can automate content creation processes, which can save time
and reduce costs compared to traditional manual processes.
v) Innovation: Facilitates breakthroughs in fields requiring complex simulations, such as
physics and engineering.
vi) Scalability: Generative AI can be used to generate large volumes of content quickly and
efficiently, making it a scalable solution for business.

Q8. Mention the Limitations of using Generative AI?

Ans: Limitations of using Generative AI are:
i) Data Dependency: Generative AI models require large amounts of high-quality data for
training. Poor quality or biased data can lead to inaccurate or biased outputs.
ii) Computational Resources: Training generative models, especially complex ones like
GANs, requires significant computational power and time, making them resource-
intensive.
iii) Ethical Concerns: Generative AI can create realistic fake content, such as deep fakes,
which can be misused for misinformation, fraud, and privacy violations.
iv) High Cost: Developing and maintaining generative AI models can be expensive due to
the need for specialized hardware and expertise.

Q9. Discuss some famous Generative AI tools.

Ans: Some famous Generative AI tools are:
1. ChatGPT : It is an AI language model designed to understand and generate human-like
text based on the input it receives.
2. Gemini: The Gemini is Generative AI project by OpenAI, aiming to explore large-scale
AI models that generate high-quality and diverse text.
3. Artbreeder: Artbreeder is a web-based tool that enables users to generate new images by
combining different GAN models. Users can select and combine different GAN models
to create new and unique images.

Q10. Discuss the ethical concerns associated with Generative AI.

Ans. Ethical considerations for Generative AI include:
1. Ownership and Attribution: There are questions about who owns the content generated
by generative AI. This is particularly relevant in creative fields such as music, literature,
or art, where generative AI can create original works that blur the lines between human
and machine authorship.
2. Job Displacement: Automation of creative tasks by AI could potentially displace human
jobs in creative industries, raising concerns about economic impacts and job loss.
3. Data Bias: Generative AI can replicate and amplify existing biases present in the data
used to train the model. This can lead to harmful or discriminatory outcomes, especially
if the generated content is used in high-stakes applications such as hiring, loan approvals,
or criminal justice.
4. Privacy: Generative AI can potentially be used to generate sensitive personal
information, such as credit card numbers, social security numbers, or medical records.
This could be used for malicious purposes.
5. Misinformation: Generative AI can be used to create fake news or deep fakes, which can
be used to spread misinformation and manipulate public opinion. This can have serious
consequences for democracy and trust in institutions.

Page 5 of 5