Revolutionizing Space:

The Potential of Artificial Intelligence

AHMAD AL-DAHOUD1, MOHAMED FEZARI2, ALI-AL-DAHOUD3, DARAH AQEL3,

HANI MIMI3, MOHAMMAD SH. DAOUD4

1Faculty Architecture and Design,

Al-Zaytoonah University of Jordan,

JORDAN

2Dept. Electronics,

University Badji Mokhtar Annaba,

ALGERIA

3Faculty of Science and IT, Al-Zaytoonah University of Jordan,

JORDAN

4College of Engineering,

Al Ain University,

UNITED ARAB EMIRATES

Abstract: - Generative AI is a new branch of artificial intelligence, which creates fresh content using neural

networks and machine learning methods. Systems of generative AI can generate music, images, text, speech,

and other types of content by finding new styles in huge databases. The automation of tedious tasks through the

creation of personalized content, and the improvement of accuracy in difficult tasks makes generative AI

technology to transform a variety of industries, including gaming, advertising, and healthcare. There are many

types of generative AI models. Each has pros and cons of its own. Despite being a relatively young technology,

generative AI has many potential applications that make it a fascinating field to research. More research,

growth, and advancement in the future may be seen. Future potential uses for generative AI include improving

cybersecurity by identifying and preventing cyberattacks, creating human-interactive virtual assistants, and

creating intelligent robots that can do challenging tasks in various industries. As generative AI continues to be

developed, we should expect to see increasingly sophisticated applications in the years to come, which will

open up new opportunities for growth across numerous industries.

Key-Words: - Generative AI, Artificial Intelligence, Machine Learning, OpenAI models, ChatGPT, DALL-E,

and GPT-4.

Received: January 29, 2024. Revised: July 11, 2024. Accepted: August 9, 2024. Published: September 4, 2024.

1 Introduction

Artificial intelligence (AI) has achieved growing

momentum in its application in many fields with

limitless possibilities. It has changed every aspect of

businesses, industries, and lives by applying it for

different purposes including intelligent marketing,

fraud detection, and customer support. It was

applied to many important applications such as

natural language processing, [1], [2], [3], agriculture

[4] and stock marketing, [5]. Currently, it also

enables machines to utilize visual or textual data for

developing new content, and that is what is called

Generative AI, [6]. Therefore, Generative AI

represents a developing AI field that is concerned

with generating new content, such as images, music,

and text. In particular, it generates these new

contents based on the patterns recognized in the

training data that it has been applied to. Generative

AI can renovate many industries, including

healthcare, gaming, and advertising. Moreover, it

can transform industries by enhancing

personalization, creating recommendations,

automating repetitive duties, and generating unique

content that can be utilized to facilitate customer

engagement and satisfaction.

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

404

Volume 12, 2024

Fig. 1: Example of Generative AI, Transforming a Text into an Image

It also increases the accuracy in the health care

domain including drug discovery and medical

diagnosis as well as it helps in automating the time-

consuming tasks, where this can take less time and

decrease costs for industries. ChatGPT [7] and

DALL-E [8] are controlling the headlines as some

of the generative AI models.

Also, different types of models can be applied

for Generative AI, such as auto-encoders,

variational auto-encoders (VAEs) [9], generative

adversarial networks (GANs) [10], Boltzmann

Machines [11], and transformers [12]. Each type of

model has its benefits and restrictions. Selecting the

best Generative AI model is based on the task and

type of data being created. Generative AI is mainly

based on the use of advanced machine learning

algorithms and neural networks, including

transformers, autoencoders, VAEs, GANs, and

Boltzmann Machines, [13].

Whereas Generative AI is still a new

technology, its advantages and demands make it an

interesting field of development and research. In

addition, we are expecting to see more advanced

and innovative usages of Generative AI in the

coming years.

The OpenAI models such as ChatGPT, GPT-4

[14] and DALL-E are widely getting the attention of

people in the world of business, industries, and

content creation. In this research, we describe

Generative AI, the types of models that use

Generative AI, the test results on some Generative

AI models, and the dangers and limitations of using

Generative AI. The rest of the paper is organized as

follows: Section 2 defines Generative AI and

presents how it works. Section 3 describes the most

popular Generative AI examples. Section 4

illustrates the types of Generative AI models.

Section 5 demonstrates the main benefits of

Generative AI. Section 6 presents the most

important applications of Generative AI. Section 7

shows the dangers and limitations of Generative AI.

Finally, Section 8 concludes the paper and presents

the future work.

2 Defining Generative AI and How it

Works?

Generative AI is a sub-domain of artificial

intelligence, in which computer algorithms are

applied to create outputs that are the same as

human-created content such as texts, images, and

music, graphics, and computer codes. It allows the

use of existing content like texts, images, or audio

files to generate new reasonable content. In

particular, it enables computer algorithms to extract

the most important patterns related to a given input

and then utilizes it to produce a similar content.

Algorithms in Generative AI are applied to learn

from labeled training data examples. By exploring

the patterns within the training data examples,

Generative AI models can generate new content that

has the same features as the original input data and

appears reliable and human-like. It is based on the

use of machine learning and deep learning

techniques such as neural networks. To train a

Generative AI model, large amounts of data must be

given to a machine learning or deep learning

algorithm to learn useful patterns from this data.

This data could be texts, codes, graphics, or any

other type of contents that are suitable to the given

task. So, after collecting the training data, the

Generative AI model analyzes the patterns within

the data to find the main rules that control the

content. Then, the AI model adjusts its parameters

during the learning phase, to improve its capability

of simulating human-created content. If the AI

model can generate more content, then the created

output will be more sophisticated, accurate, and

believable.

3 The Most Popular Generative AI

Examples

We describe with examples the most popular GAI

developed by OpenAI like ChatGPT, GPT-4, and

DALL-E.

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

405

Volume 12, 2024

3.1 Chat-GPT of GPT-3.5

OpenAI introduced ChatGPT [7], an AI language

model built on the Generative Pre-trained

Transformer (GPT) architecture. It is designed to

respond to natural language inputs that resemble

those of a human, enabling conversational

interaction between humans and AI. ChatGPT can

recognize and respond to a range of themes, from

straightforward questions and answers to more

sophisticated conversations, thanks to the vast

database of textual material that it was trained on.

This model can learn from vast volumes of data and

produce high-quality responses because it employs a

deep learning method known as transformer

architecture. ChatGPT is capable of comprehending

context, identifying patterns, and generating

meaningful and logical responses to a range of

inputs. Its capabilities span abroad.

Example 1: use

Input: What is generative deep learning? Figure 1

shows the answer to that question from ChatGPT.

Fig. 1: “Answer to what is a generative deep

learning?”

Example 2: generating code.

Input: Generate code python bandpass active filter.

Figure 2 shows the answer to the previous question.

Fig. 2: ChatGPT response to « Generate code

python bandpass active filter”

Assuming an 8000 Hz sample frequency, the

preceding example involves creating a second-order

Butterworth bandpass filter with a 1000 Hz center

frequency and a 200 Hz breadth. To find the filter

coefficients B and A, use the signal.butter() method.

The bandpass filter is specified by setting the btype

argument to 'bandpass'. The frequency response of

the filter is then plotted using the signal.freqz()

function and displayed using matplotlib. You can

modify the filter specifications to design different

types of passband filters.

3.2 GPT-4

GPT-4’s, [14], is the same as its predecessor, GPT-

3.5, in which it generates its output in response to

natural language queries and other

requests. OpenAI says that GPT-4 can “follow

complex instructions in natural language and solve

difficult problems with accuracy”. In specific, GPT-

4 can resolve mathematical problems, write

programming codes, respond to questions, and tell

stories. Furthermore, GPT-4 can review and

summarize large amounts of content, where which

supports the business use cases and customers.

OpenAI examined the capability of GPT-4’s to

replicate information in a consistent order by using

many assessments, such as AP and Olympiad exams

and the Uniform Bar Examination. GPT-4 gained

different scores on AP examinations. The results of

running GPT-4 through standardized tests have

shown that the GPT-4 model can construct accurate

and current responses. GPT-4 forecasts which token

is probable to come next in a sequence of words. (A

token may be a word, numbers, letters, symbols, or

punctuations).

In addition, GPT-4 does not include information

more recent than September 2021 in its lexicon.

Google Bard represents one of GPT-4’s

competitors, which does have up-to-date data,

material, and information because it is trained on the

modern internet.

3.3 GPT-4 vs. ChatGPT

The OpenAI Company has mentioned that the GPT-

4 model can process more data and perform more

computations than the billions of parameters that

ChatGPT was trained on. It has also proven more

efficient in writing a huge diversity of materials,

such as stories and fiction. Additionally, GPT-4

achieves higher performance than ChatGPT on the

uniform tests stated above. Responses to prompts of

GPT-4 may be more accurate and simpler to parse.

Moreover, GPT-4 is preferred over GPT-3.5 in

decision-making, content summarization, and time

scheduling. OpenAI argued that GPT-4 is 82% less

likely to answer questions for banned content and

40% more likely to generate realistic answers.

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

406

Volume 12, 2024

Figure 3 presents an illustration of GPT-2 as a

translator.

Fig. 3: An Illustration of GPT-2 as a Translator

3.4 Dalle-E

DALL-E (pronounced like "Dali") [8], is an AI

model developed by OpenAI, which stands for

"Denoising Autoencoder for Learned Language

Embeddings." It is a type of generative model that

can create novel images from textual descriptions.

DALL-E can recognize natural language

characterizations of images and create a new image

that matches the description. Mainly, it is a neural

network-based generative model for creating images

from textual descriptions, developed by OpenAI in

early 2021. The name "DALL-E" is said to be a

combination of Salvador Dali, the famous surrealist

painter, and the robot character EVE from the Pixar

film "Wall-E".

DALL-E is based on the GPT-3 architecture and

generates visuals from textual inputs by utilizing a

combination of transformers and convolutional

neural networks. The model can both generate new

pictures to fit a textual content description or alter

an already-present picture to fit a brand-new textual

content input. A considerable series of picture-

caption pairs was used to educate it. DALL-E is

terrific for its capability to generate remarkably

revolutionary and surreal pictures that defy the

expectancies of picture synthesis models. This is

because the model is not reduced by the laws of

physics, allowing it to produce images that are not

truly usable and consistent with the textual

description. For example, it can draw pictures and

drawings such as "snail made of harps" or "pizza

with a giraffe pattern". This makes DALL-E a

beneficial tool for both artistic content material

production, along with making visuals for novels,

movies, or advertisements, and scientific and

engineering programs, together with constructing

sensible representations of complicated machines or

organic systems.

Two deep gaining knowledge strategies that

DALL-E makes use of are a transformer-based

totally language version and a denoising

autoencoder. Using a huge dataset of picture-text

pairs, it changed into teaching to recognize the

relationships among textual descriptions and related

images. DALL-E is beneficial in some of the

approaches. For example, it can be applied to the

design, fashion, and entertainment industries to

produce creative designs or visual content for

motion pictures.

It can create images in different styles, from

photorealistic imagery3 to paintings and emoji. It is

capable also of processing, handling, and

rearranging objects in its images. Figure 4 and

Figure 5 present some photos generated by DALL-E

as outputs.

Example 1 of DALL-E 2

Input: cat with sunglasses on a boat in a sunny day

DALL-E response:

Fig. 4: Dalle-E response to “cat with sunglasses on a

boat on a sunny day”

Example 2:

Input: astronaut in virtual space

Fig. 5: Some photos generated by DALL-E as

Outputs

3.5 Other Examples of Generative AI

Models

Generative AI has achieved important

improvements in recent years, by using some tools

that attract public attention and create a sensation

between content developers and creators. Big

technology enterprises such as

Google, Microsoft, and Amazon, have released their

own Generative AI tools.

 Midjourney: released by San Francisco-based

research lab Midjourney Inc., Midjourney

understands and analyzes text prompts and

context to create visual content, like DALL-E 2.

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

407

Volume 12, 2024

 GitHub Copilot: is an AI programming tool

developed by GitHub and OpenAI. It helps

users to write codes faster and proposes code

completions for them in development

environments such as Visual Studio and

JetBrains.

 Cisco brings a Chat-GPT experience to

WebEx

Related to Gartner.com, articles, we believe that

Generative AI will increase and speed up designs in

many businesses as well as has the capability of

discovering new and effective designs. The output

of AI systems, that use Generative AI, may contain

high-value artifacts such as video, code, narrative,

and synthetic data. AI use cases for Generative AI

are increasing, particularly in five areas such as

parts design, drug design, chip design, material

science, and synthetic data.

4 Types of Generative AI Models

Generative AI models can be used to create new

content and samples that are the same as the original

data by sampling data from the complex

distribution. In summary, there are many types of

models utilized for Generative AI, each with its

benefits and weaknesses. Choosing the right model

depends on the type of the utilized data and the task

at hand. These models are all designed to address

certain problems and applications. The following

categories apply to these generative AI models.

4.1 Transformer-based Models

 Transformer-based models are a representation

of a deep learning framework that

revolutionized sequential data processing such

as natural language processing (NLP), [12],

[13], [15]. After being presented in the 2017

paper "Attention Is All You Need" by [12], the

transformer scheme has become an industry

standard for many NLP projects, [12].

 • The attention model provides the key to self-

attention, which is essential for dedicating time

for contemplation while the architecture of each

meaning is progressively being constructed.

With the use of token-pair-level attention and

importance scoring, this would determine the

sequence of how the input tokens are shown.

Consecutively the model concludes stressing

over the most significant data when it is in the

prediction mode.

 Among all transformer-based neural network

models, one could argue that the BERT model

(Bidirectional Encoder Representations from

Transformers), which was introduced by Google

in 2018, is the most widely applied model up

until now. Large amounts of text are being

converted from raw text into a transformer

model, which is trained with a transformer-base

model and masked language modeling method.

BERT stands out as its performance has been

commendable, recording top scores and getting

the best answer orientation among other NLU

tasks such as question answering, and natural

language inference.

 XLNet (bootstrapping multilingual intermediate

modeling via meta-learning), T5 (text-to-text

transfer transformer), and GPT (generative pre-

trained transformer) are the other sets of well-

known transformer-based models.

 By enabling the models to recognize distant

syntactic relationships and comprehend word

context, transformer models have together

brought about a dramatic shift in natural

language processing (NLP) systems. In the field

of NLP, these models may be used for a wide

range of tasks, including machine translation,

text categorization, and language model

creation. Neural networks for natural language

processing produce the predictions for the

ChatGPT and GPT-3.5 OpenAI models. For

everyone who is trained in big data, the process

remains the same until these patterns are

identified. Additionally, they can quickly notice

the interactions with sequential data, which

makes them effective for jobs involving text

production.

4.2 Example: Vision transformers (VITs)

[12], [15], [16]

One kind of deep learning model that is utilized for

picture categorization tasks is called Vision

Transformer (VIT), [12], [15]. It applies

transformers over patches of the image, where

patches are supposed to be independent and

distributed. The VITs components are self-attention,

positional encoding, and multi-head self-attention.

The positional encoding component is based on

finding the position of an entity in a sequence of

input tokens, learning the relative distances between

patches, and recognizing the spatial structure of an

image. The self-attention method specifies the

significance of each patch and estimates the

correlation and context between the other patches.

Multi-head self-attention allocates several self-

attention blocks to one account for many types of

interactions between the patches and integrates them

into a single self-attention output. The main

architecture of VIT is illustrated in Figure 6. A WSI

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

408

Volume 12, 2024

is transformed into a series of patches, where each

patch is associated with its positional information.

Learnable encoders transform each patch and its

position into one embedding vector called a token.

An extra token is given for the classification

procedure. The transformer encoder takes the class

token along with the patch tokens as inputs to

estimate multi-head self-attention and then outputs

the class and the learnable embedding of patches.

The output class token is used as a slide-level

representation for the last classification of the

model. The transformer encoder contains many

stacked identical blocks. Each block includes multi-

head self-attention and MLP, coupled with a layer

normalization and residual connections. The

multiple self-attention heads and the positional

encoding are useful for integrating spatial

information and increasing the context and

effectiveness, [17] of the VIT technique over other

techniques. However, there is a limitation in VITs,

in which it is considered to be more data-hungry,

[15].

Poorly supervised methods offer many

advantages. Eliminating the manual annotations

decreases the data preprocessing cost and minimizes

the bias and interrater variability. Therefore, the

models can be simply applied to huge datasets, for

various tasks. As the models can learn from the

whole scan free, then they can classify the

predictive features even if the regions were assessed

by pathologists. These weakly supervised methods

achieve a good performance and propose that many

tasks can be resolved without expensive manual

annotations. Figure 6 presents the vision

transformers architecture application in X-ray image

analysis for COVID-19.

Fig. 6: Vision Transformers Architecture

Application in X-ray Image Analysis for COVID-19

Detection

4.3 Generative Adversarial Networks

‘GAN’

Generative Adversarial Networks (GANs) [10]

consist of two neural network models, a

discriminator, and a generator. The discriminator

estimates the authenticity and the quality of said

data, while the generator generates data. After a

while, both neural networks perform their roles

effectively, and more accurate outputs are generated

by them.

In the adversarial training process of the GANs,

the generator and discriminator are trained together

on a given data. In this process, the generator

attempts to generate data that fools the

discriminator, while the discriminator attempts to

recognize whether the given data is fake or real. The

output of the generator is then tuned depending on

the discriminator's feedback, and the process

remains until the generator can create data that is

similar to the real data. There are several

applications of GANs, such as data augmentation,

image synthesis, and anomaly detection. GANs have

also been applied to produce real images of

buildings, faces, and cities. They have proven their

worth in the healthcare domain, as they can be

utilized to create synthetic medical images and data,

which can then be used for training machine

learning methods for disease diagnosis and

treatment. Figure 7 displays the GAN Architecture.

Fig. 7: The GAN Architecture

4.4 Autoencoders

One kind of neural network design used in

unsupervised learning is the autoencoder. By first

mapping the input data to a lower-dimensional

representation and then mapping the lower-

dimensional representation back to the original

input, known as the decoding, they are intended to

learn a compressed representation of a given input

data, known as the encoding or latent space.

Numerous data kinds, including text, music, and

images, can be used to train autoencoders.

The following are the main uses for autoencoders:

 Image Compression and Reconstruction: By

using autoencoders for image compression and

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

409

Volume 12, 2024

reconstruction, massive image datasets can be

stored and transmitted more effectively.

 Anomaly Detection: By highlighting data points

that the model is unable to correctly reconstruct,

autoencoders can be used to find abnormalities

in datasets.

 Feature extraction: Auto-encoders are capable

of efficiently identifying data and extracting

features, which can be utilized as input for

machine learning models.

 Dimensionality reduction: dimensionality of

high dimensional datasets can be reduced by

using auto-encoders, which facilitates the

visualization and study of complicated data.

 Data Generation: New data samples can be

produced by auto-encoders. To do this, samples

are selected at random from the sample space

and then decoded.

In brief, autoencoders are widely used in many

fields, such as computer vision, signal processing,

and natural language processing. They are a helpful

instrument for unsupervised learning and may help

uncover patterns and insights in massive

and complex datasets. Figure 8 displays a denoising

model utilizing an autoencoder.

Fig. 8: Illustration of autoencoder as denoising

model

4.5 Variational Autoencoders

An encoder and a decoder are used by variational

autoencoders (VAEs) [9] to produce material. After

receiving input data—such as text or images—the

encoder compresses the data before sending it out

again. The decoder then uses this encoded data to

reassemble it into new data that bears a resemblance

to the original input data.

One class of generative models known as VAEs

is used to learn a condensed representation of input

data. With the addition of a probabilistic model that

allows them to execute data sampling from the

compact representation, they are regarded as a kind

of autoencoder.

VAEs are trained using unsupervised learning

and can be applied to many data types, such as

images, text, and sound. Figure 9 presents an

illustration of VAE with the latent space defined by

mean and variance.

The primary applications of VAEs include:

 Data Compression: VAEs can be used to encode

data into a lower-dimensional representation

compress it and then decode this data back into

its original structure.

 Anomaly Detection: VAEs can be used to detect

anomalies in datasets by comparing the

reconstruction error of a given data sample with

the reconstruction error of the training data.

 Semi-Supervised Learning: VAEs can be used

for semi-supervised learning tasks by

incorporating label information into the model

training process.

 Data Augmentation: VAEs can be utilized to

create new data samples that can be applied to

increase the training data for other machine

learning models.

 Image and Video Generation: VAEs can be

trained to generate new images and videos by

sampling from the compressed representation of

the input data.

Fig. 9: VAE with the Latent Space Defined by Mean and Variance

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

410

Volume 12, 2024

4.6 Multimodal Models

Multimodal AI, [17], [18], [19], [20], [21] is a new

AI pattern, in which different data types (numerical

data, image, text, and speech) are integrated with

multiple intelligence processing algorithms to attain

superior performance results. Multimodal AI often

achieves better performance results than single-

modal AI in several real-world applications.

Multimodal deep learning is a new area, in which

algorithms learn from data of multiple modalities.

For instance, a human can use both hearing and

sight senses to recognize an object or a person.

Similarly, multimodal deep learning is based on

developing similar capabilities for computers.

Multimodal models can analyze and process many

types of input data, including audio, text, and

images. They join different modalities to generate

more developed outputs. Some examples of these

models are such as DALL-E 2 and OpenAI’s GPT-

4, which also accepts image and textual data inputs.

Multimodal deep learning has many important uses,

such as:

 Automatically creating illustrations of images,

such as captioning for blind people.

 Looking for images that suit text queries (e.g.

“find for me an image of a yellow cat”).

 Generative art system that creates images from

text illustration (e.g. “create a picture of a

yellow cat”).

They can be used to create new data samples

that look like the original data by performing data

sampling from the complex distribution. In

summary, many types of models are applied for

Generative AI, each with its strengths and

weaknesses. Choosing the right model depends on

the data type and the task at hand.

5 Benefits of Generative AI

The main benefit that Generative AI offers is

efficiency, whereas with this benefit, Generative AI

allows businesses to systematize specific tasks and

direct their effort, time, and resources on more

significant strategic objectives. This reduces the

labor costs and maximizes the operational

efficiency.

Generative AI suggests several benefits that can

help in transforming various industries and

improving efficiency and personalization. Some of

the main benefits of Generative AI are:

 Creative content creation: Generative AI can

help to create new and unique content, including

images, music, and text, that can be used in

various industries, such as advertising, gaming,

and art.

 Time and money savings: Companies may save

time and money by using generative artificial

intelligence for automated repetitive and time-

consuming processes such as data analysis and

production of content.

 Money and Time: Businesses may use

generative artificial intelligence to save money

and time by automating repetitive and time-

consuming tasks like data processing and

generation.

 Increased precision: artificial intelligence

examines huge amounts of data and finds

relevant patterns that people normally miss.

This can help in increasing accuracy in a variety

of activities that include drug discovery and

medical diagnostics.

 Novel uses: Generative AI creates new

opportunities and uses in several sectors,

including gaming and healthcare, that were

either impractical or too difficult to implement

in the past.

In summary, Generative AI offers several

benefits that can help in transforming industries and

improving efficiency, personalization, accuracy, and

creativity.

Generative AI proposes additional benefits to

businesses and entrepreneurs, such as:

 Creating new ideas, content, or designs.

 Typing, developing, testing, and optimizing

computer code.

 Formulating templates for articles or essays.

 Improving customer support with virtual

assistants, dialog systems, and chatbots.

 Streamlining the procedure for gathering and

expanding data so that machine-learning

algorithms may use it.

 Enhancing decision-making through data

processing and analysis.

6 Additional Domains in Which

Generative AI is Employed

Artificial intelligence that creates new content

through algorithms, such as writing, photos, movies,

and music, is known as generative AI. Its uses are

numerous and spread across several industries, such

as:

 Healthcare: The application of generative AI in

health expedites the search for new treatments

and reduces the time and expense of research.

Medical imaging may benefit from the usage of

generative AI in healthcare. Researchers can

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

411

Volume 12, 2024

design novel medicines with the use of AI

models, which can produce new medicinal

molecules and forecast their efficacy.

Generative artificial intelligence is used to

create 3D medical images for treatment

planning and improving diagnosis.

 Marketing: Advertisers can organize

personalized campaigns and modify material

according to the preferences of their clients

using Generative Artificial Intelligence. Product

recommendations and advertisements are

examples of the previous point. Artificial

intelligence algorithms may produce marketing

messages that are both audience-targeted and

appealing to the target audience because they

are trained on consumer data.

 Education: Many teachers use generative

Artificial Intelligence models to assess and learn

materials that are tailored to the unique learning

preferences of each student.

 Finance: financial analysts that are based on

generative artificial intelligence are used to

examine market trends and predict the direction

of the stock market.

 Virtual colleagues: Generative fake insights

may be utilized to create human-like virtual

colleagues that can comprehend and answer

normal dialect requests. Virtual colleagues who

take after individuals can upgrade client

fulfillment. In addition, it can raise the esteem

of virtual associates for an assortment of errands

counting planning arrangements, and reacting to

requests.

 Climate scientists use generative AI models to

study the environment to forecast weather

patterns and comprehend climate change.

 Generative AI may be used in creative sectors

such as music, design, and painting. Utilizing

inspiration from previously composed music, a

computer program may create new music.

 Video games: Characters, environments, and

levels may all be created with generative AI.

Game developers may provide more original

and captivating game material in addition to

saving time and money,

 In conclusion, generative artificial intelligence

contains a wide range of applications and the

capacity to change many businesses by creating

modern substance and improving proficiency

and personalization.

7 Dangers and Limitations of

Generative AI

Many difficulties raised by generative AI require

consideration. One of the problems is that it might

spread false information or sensitive or dangerous

content, which may hurt individuals or companies

and endanger national security.

Policymakers have given careful thought to

these linkages. For instance, in April 2023, the

European Union imposed new copyright regulations

on enterprises that develop generative AI tools.

These regulations require the disclosure of any

copyrighted content utilized in the development of

these tools. These regulations would reduce

intellectual property exploitation while also

promoting ethics and openness in the field of

artificial intelligence development. Moreover, these

rules protect the content creators from mimicking or

plagiarizing their work by Generative AI tools.

Automating the Generative AI tasks will have

effects on the workforce and employees, in which

those affected employees, are required to be

reskilled or upskilled. Furthermore, Generative AI

tools can magnify biases existing in the training

data, where this causes problematic results that

spread stereotypes and harmful beliefs. ChatGPT,

Bing AI, and Google Bard have all stirred up

controversy by creating damaging outputs since they

have been released. All these issues should be

resolved over time as Generative AI develops.

New Roman as here.

8 Conclusion

In summary, generative AI is a powerful artificial

intelligence tool that has the potential to

revolutionize a wide range of industries through the

production of fresh, original content, enhanced

personalization, cost and timesavings, improved

accuracy, increased efficiency, and the creation of

new opportunities. Generative AI can create text,

images, music, and other types of content that are

useful in industries like marketing, gaming,

healthcare, and the arts by utilizing machine

learning and deep learning algorithms. Even though

generative fake insights (AI) is still moderately

unused and has a few disadvantages, just like the

requirement for enormous volumes of preparing

information, its potential benefits and employments

make it a beneficial field for advance consider and

headway. Overall, the prospects for generative AI

show up shrinking since the innovation is still

creating, and we expect to see more cutting-edge

and inventive applications of generative AI in the

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

412

Volume 12, 2024

future. However, there are other deterrents to

overcome, just as the gigantic sums of preparing

information required, the plausibility of

predisposition and moral issues, and the necessity

for the created fabric to be straightforward and

comprehensible.

Hence, future investigation on the conceivable

risks of generative AI is fundamental. As generative

AI innovation is created, it is basic to form beyond

any doubt that AI is utilized morally by lessening

predispositions, boosting responsibility, expanding

straightforwardness, and helping with information

administration. An adjustment between human

interaction and assignment computerization for

generative AI is essential to maximize its benefits.

On the other hand, any unfavorable effects on the

workforce have to be decreased or evacuated.

Furthermore, we hypothesize that combining

several generative models—like VAEs and GANs—

produced reliable multimodal outcomes. Finally, we

performed a thorough analysis to determine how

missing data and poor supervision affected

multimodal learning. In terms of large amounts of

missing data, we investigated the possibility that the

suggested VAEVAE and VAEGAN models perform

better than the other generative AI models.

Subsequent research in multimodal data may

examine the outcomes of utilizing similar concepts

in the design of videos, where each frame comprises

text, audio, and visual elements. Further study and

development in this area will certainly provide new

chances and solutions for people, companies, and

society as a whole.

References:

[1] Kanan, T., Mughaid, A., Al-Shalabi, R. Al-

Ayyoub, M. Business intelligence using deep

learning techniques for social media contents.

Cluster Comput 26, 2023, 1285–1296.

[2] Aqel, D. and Hawashin, B., Arabic relative

clauses parsing based on inductive logic

programming. Recent Patents on Computer

Science, 11(2), 2018, pp.121-133.

[3] Radford, A., Narasimhan, K., Salimans, T.

and Sutskever, I., Improving language

understanding by generative pre-training.

2018, [Online].

https://openai.com/research/language-

unsupervised (Accessed Date: April 28,

2024).

[4] Igried, B., AlZu’bi, S., Aqel, D., Mughaid, A.,

Ghaith, I. and Abualigah, L., An Intelligent

and Precise Agriculture Model in Sustainable

Cities Based on Visualized Symptoms.

Agriculture, 13(4), 2023, p.889.

[5] Mukherjee, S., Sadhukhan, B., Sarkar, N.,

Roy, D. and De, S., Stock market prediction

using deep learning algorithms. CAAI

Transactions on Intelligence Technology,

8(1), 2023, pp.82-94.

[6] Dwivedi, Y.K., Kshetri, N., Hughes, L.,

Slade, E.L., Jeyaraj, A., Kar, A.K.,

Baabdullah, A.M., Koohang, A., Raghavan,

V., Ahuja, M. and Albanna, H., “So what if

ChatGPT wrote it?” Multidisciplinary

perspectives on opportunities, challenges and

implications of generative conversational AI

for research, practice and policy. International

Journal of Information Management, 71,

2023, p.102642.

https://doi.org/10.1016/j.ijinfomgt.2023.1026

42.

[7] Brown, T., Mann, B., Ryder, N., Subbiah, M.,

Kaplan, J.D., Dhariwal, P., Neelakantan, A.,

Shyam, P., Sastry, G., Askell, A. and

Agarwal, S., Language models are few-shot

learners. Advances in neural information

processing systems, 33, 2020, pp.1877-1901.

[8] Daras, G. and Dimakis, A.G., Discovering the

hidden vocabulary of dalle-2. 2022 arXiv

preprint arXiv: 2206.00169,

DOI:10.48550/arXiv.2206.00169.

[9] Kingma, D.P. and Welling, M., Auto-

encoding variational bayes. 2013, arXiv

preprint arXiv:1312.6114,

https://doi.org/10.48550/arXiv.1312.6114.

[10] I. Goodfellow, J. Pouget-Abadie, M. Mirza,

B. Xu, D. Warde-Farley, S. Ozair, A.

Courville, and Y. Bengio, “Generative

adversarial nets,” in Advances in Neural

Information Processing Systems 27, Montreal,

Quebec, Canada, 2014, pp. 2672-2680.

[11] Ackley, D.H., Hinton, G.E. and Sejnowski,

T.J., A learning algorithm for Boltzmann

machines. Cognitive science, 9(1), 1985,

pp.147-169.

[12] Vaswani, A., Shazeer, N., Parmar, N.,

Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser,

Ł. and Polosukhin, I., Attention is all you

need. 31st Conference on Neural Information

Processing Systems, 2017. arXiv preprint

arXiv.v,

https://doi.org/10.48550/arXiv.1706.03762.

[13] Basem S. Abunasser, Salwani Mohd Daud

and Samy S. Abu-Naser, Predicting Stock

Prices using Artificial Intelligence: A

Comparative Study of Machine Learning

Algorithms, International Journal of

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

413

Volume 12, 2024

Advances in Soft Computing and its

Application, 15(3), 2023, 41-53. doi:

10.15849/IJASCA.231130.03.

[14] Liu, Y., Han, T., Ma, S., Zhang, J., Yang, Y.,

Tian, J., He, H., Li, A., He, M., Liu, Z. and

Wu, Z., Summary of chatgpt/gpt-4 research

and perspective towards the future of large

language models. Meta-Radiology, 1(2). 2023.

pp. 100017,

https://doi.org/10.1016/j.metrad.2023.100017

[15] Dosovitskiy, A., Beyer, L., Kolesnikov, A.,

Weissenborn, D., Zhai, X., Unterthiner, T.,

Dehghani, M., Minderer, M., Heigold, G.,

Gelly, S. and Uszkoreit, J., An image is worth

16x16 words: Transformers for image

recognition at scale. 2020, arXiv preprint

arXiv:2010.11929.

[16] Akb Akbari, H., Yuan, L., Qian, R., Chuang,

W.H., Chang, S.F., Cui, Y. and Gong, B.,

Vatt: Transformers for multimodal self-

supervised learning from raw video, audio and

text. Advances in Neural Information

Processing Systems, 34, 2021, pp.24206-

24221.

[17] Bao, H., Wang, W., Dong, L., Liu, Q.,

Mohammed, O.K., Aggarwal, K., Som, S.,

Piao, S. and Wei, F., Vlmo: Unified vision-

language pre-training with mixture-of-

modality-experts. Advances in Neural

Information Processing Systems, 35, 2022,

pp.32897-32912.

[18] Varghese, E.B. and Thampi, S.M., A

multimodal deep fusion graph framework to

detect social distancing violations and FCGs

in pandemic surveillance. Engineering

Applications of Artificial Intelligence, 103,

2021, p.104305.

[19] Baruch, E.B. and Keller, Y., Multimodal

matching using a hybrid convolutional neural

network. 2018, (Doctoral dissertation, Ben-

Gurion University of the Negev).

[20] Wang, C., Wang, X., Long, Z., Yuan, J., Qian,

Y. and Li, J., August. Multimodal gait

analysis based on wearable inertial and

microphone sensors. In 2017 IEEE

SmartWorld, Ubiquitous Intelligence &

Computing, Advanced & Trusted Computed,

Scalable Computing & Communications,

Cloud & Big Data Computing, Internet of

People and Smart City Innovation, (pp. 1-8).

2017, IEEE.

[21] Mohammad Husien Almajali, Laith Nasrawin,

Faisal Tayel Alqudah, Ahmad Abdallah

Althunibat and Nasir Albalawee, Technical

Service Error as a Pillar of Administrative

Responsibility for Artificial Intelligence (AI)

Operations, International Journal of Advances

in Soft Computing and its Application, 15(3),

2023, 274-287. doi:

10.15849/IJASCA.231130. 18

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

The authors equally contributed in the present

research, at all stages from the formulation of the

problem to the final findings and solution.

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

No funding was received for conducting this study.

Conflict of Interest

The authors have no conflicts of interest to declare.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2024.12.40

Ahmad Al-Dahoud, Mohamed Fezari,

Ali-Al-Dahoud, Darah Aqel, Hani Mimi,

Mohammad Sh. Daoud

E-ISSN: 2415-1521

414

Volume 12, 2024