Introduction to Artifical Intelligence (AI)

Melissa van Schaik

m.e.vanschaik@saxion.nl

RESEARCH GROUP SUSTAINABLE & FUNCTIONAL TEXTILES

 Table of contents

01 What is AI? 04 Neural networks

Example ML AI in research
02 Simplified support vector machine for
classifying cats and dogs
05 Advantages and challenges

Example NLP
03 Understand how chatbots are designed 06 Q&A
 the ability of a computer program or a
machine to think, learn, make decisions,
recognize patterns and solve complex
problems similar to humans [1]
REAL WORLD EXAMPLES

Self driving cars

Digital advertisements in sports games [2]
@lilmiquela [3]

Chatbot

Medical diagnosis
License recognition
Can you envision any applications within
your area of expertise?
WHAT IS AI? [4]

ML
Artificial intelligence (AI) Subset of AI
Uses algorithms to detect patterns in
large data sets
Allows machines to learn and adapt

Machine learning (ML) DL

Subset of ML
Uses neural networks for in-depth data
processing and analytical tasks
Simulates how a human brain perceives
and understands the world
Deep learning (DL)
Generative AI
Subset of DL
Detect patterns and creates output
without explicit instruction
Generates text, images or code based on
provided input

Generative AI
EXAMPLE OF MACHINE LEARNING
SUPPORT VECTOR MACHINE (SVM) [5]

Recognizing cats vs dogs

 EAR LENGTH

WEIGHT

Each object is represented as a point in an n-

dimensional feature space, where n
corresponds to the number of features used
to describe the object.
The SVM algorithm then finds the hyperplane
that best separates the classes in this space.
It aims to categories all points of one
category (dogs) on one side of the hyperplane
and the other category on the other side of
the plane (cats)
The margin is the distance between the separating hyperplane and the
closest data points from each class, with the goal of maximizing this
margin to improve classification accuracy.

IN
G
R
A
M
TRAINING VS TEST DATA

POINT FEATURE FEATURE ... CATEGORY

1 0.8 0.7

2 0.5 0.3

3 ... ...

... ...

Supervised learning model: training data is used to train and optimize the model by
adjusting its parameters based on patterns and relationships in the data
Model will predict if next object is a cat or a dog
 SUPERVISED LEARNING

LABELED
TRAINING DATA

Cat Dog

Dog

Dog

Cat
 SUPERVISED LEARNING

LABELED
TRAINING DATA

Cat Dog
MODEL TRAINING

Dog

Dog

Cat
SUPERVISED LEARNING

LABELED
TRAINING DATA

Cat Dog
MODEL TRAINING PREDICTION

Dog

Dog

Cat

TEST DATA
SUPERVISED LEARNING

LABELED
TRAINING DATA

Dog Cat
Cat MODEL TRAINING PREDICTION

Dog Dog

Dog

Cat

TEST DATA
UNSUPERVISED LEARNING

UNLABELED
TRAINING DATA

INTERPRETATION PROCESSING

Makes clusters based on data

Try to recognize the underlying structure or patterns
Examples: Netflix, Spotify...
NATURAL LANGUAGE PROCESSING (NLP)[6]

a field of artificial intelligence that enables

computers to understand, interpret, and
respond to human language in a meaningful
way
Example chatbot
If the user says "What are your hours?", Identify specific words in the user’s input
the bot might respond with: Use regular expressions (predefined input)
to match user’s input
"We are open Chatbot uses templates to respond to the
from 9 AM to 5 PM."
user based on the identified keywords or
patterns
 INPUT USER

“I would like to order pizza.”

 INPUT USER PROCESS INPUT

“I would like to order pizza.” Searches for particular keywords or phrases.

“I would like to order pizza.”

 INPUT USER PROCESS INPUT SELECT RESPONSE

“I would like to order pizza.” “I would like to order pizza.” Based on the identified keywords or patterns, the
chatbot selects an appropriate response template.
The templates can include placeholders for dynamic
data, such as the user's name or specific information
(e.g., order numbers).
 "What size pizza would you like
to order?"

INPUT USER PROCESS INPUT SELECT RESPONSE SEND RESPONSE

“I would like to order pizza.” “I would like to order pizza.”

PROS & CONS

+ Simple - Lack of understanding

+ Consistent interactions because - Predictable: Responses can be repetitive, and
responses are predetermined the chatbot might fail with unrecognized inputs
+ Fast - Difficult to scale up: As the complexity of
conversations increases, the number of
templates can grow exponentially
NEURAL NETWORK[7]

a computational model inspired by the

human brain, consisting of interconnected
nodes that process information through
weighted connections and layers, enabling
tasks like pattern recognition and decision-
making
DEVELOPMENT OF NEURAL NETWORKS

INPUT

Data to learn from

DEVELOPMENT OF NEURAL NETWORKS

INPUT HIDDEN LAYER

The hidden layer has “neurons” that take the input data and
apply some math (like multiplying by weights) to find
patterns.
After processing the data, each neuron uses an activation
function to decide if the data should be passed on. It helps
the network learn complex patterns, like deciding what part
of the data is important.
DEVELOPMENT OF NEURAL NETWORKS

INPUT HIDDEN LAYER OUTPUT

Example: ChatGPT, Copilot

Uses multiple layers of
artificial neurons to process
and generate human-like text
based on patterns learned
from large amounts of data.
Helps understand context,
language structure, and
meaning, enabling it to
generate relevant responses.
Final result
BENEFITS OF AI

Efficiency Accuracy Pattern recognition

Can analyse large datasets and perform Reduces human errors Can identify patterns that might not be
complex calculations faster than humans immediately apparent to humans
e.g. “AI-enhanced image analysis significantly
reduces errors and accelerates diagnostic E.g. “AI's role in predictive analytics for early
processes, leading to quicker patient diagnosis disease detection and personalised medicine,
and reduced healthcare costs” [8] indicating a shift towards more tailored
healthcare approaches” [8]
REMEMBER THIS IMAGE?

INPUT HIDDEN LAYER OUTPUT

Prompt: “Can you create a diagram of a neural network Prompt: “Can you make it more like a diagram?”
with an hidden layer, easy explained, and with a few Adapted from [9].
simple steps?” Adapted from [9].
MICROSOFT’S CHATBOT TAY

Designed to engage people in

dialogue through tweets or direct
messages, while emulating the
style and slang of a teenage girl

Source: Adapted from [10].

Is someone using an AI tool?

Prompt: “Can you provide a picture of a researcher?”

Prompt: “Can you provide a picture of a lawyer?”
Prompt: “Can you provide a picture of a CEO?”
 MY RESULTS

Prompt: “Can you provide a picture of a Prompt: “Can you provide a picture of a Prompt: “Can you provide a picture of a CEO?”
researcher?” Adapted from [9]. lawyer?” Adapted from [9]. Adapted from [9].
CHALLENGES

Bias Transparency Data privacy Source

& accountability & security ambiguity
AI tools can give incorrect, Be transparent how AI was Limit access to sensitive data, AI tools scans the internet for
incomplete or biased used throughout your research and anonymize data since it is information, but the accuracy
information [12] Academic journals about AI [11]: not clear what happens to the of sources cannot be
Own knowledge is needed to To improve readability prompts and data input [12] guaranteed [12]
identify incorrect information Cannot replace authoring Always verify content by
tasks such as drawing reading academic papers and
scientific conclusions, cross-checking references to
research design ensure reliability
Authors are responsible for
original work
Not allowed for creating
images, only when
generative AI is part of the
research design e.g.
biomedical imaging
CONCLUSION

Be transparent, responsible, thoughtful

and critical in using AI tools
Be aware of ethical concerns
Expertise and knowledge is necessary to
detect incorrect, biased or incomplete
information
Assess the generated texts against other
sources

Adapted from [13].

 THANK YOU!
YOUR QUESTIONS AND INPUT ARE WELCOME

Melissa van Schaik / m.e.vanschaik@saxion.nl

REFERENCES

[1] R. O. Mason, “Ethical issues in artificial intelligence,” Encyclopedia of Information Systems, pp. 239–258, 2003. doi:10.1016/b0-12-227240-4/00064-2
[2] P. Wiesmayer, “Targeted perimeter advertising for live TV broadcasts thanks to AI,” IO, Mar. 07, 2021. https://innovationorigins.com/en/targeted-perimeter-
advertising-for-live-tv-broadcasts-thanks-to-ai/
[3]“Instagram.” https://www.instagram.com/lilmiquela/?hl=en
[4] L. P. Zhuhadar and M. D. Lytras, “The Application of AutoML Techniques in Diabetes diagnosis: Current approaches, performance, and future directions,”
Sustainability, vol. 15, no. 18, p. 13484, Sep. 2023, doi: 10.3390/su151813484.
[5] C. M. Bishop, Pattern Recognition and Machine Learning (Information Science and Statistics). 2006. [Online]. Available: https://dl.acm.org/citation.cfm?id=1162264
[6] F. Scarcello, “Artificial intelligence,” in Elsevier eBooks, 2019, pp. 287–293. doi: 10.1016/b978-0-12-809633-8.20326-9.
[7] J. West and M. Bhattacharya, “Intelligent financial fraud detection: A comprehensive review,” Computers & Security, vol. 57, pp. 47–66, Mar. 2016, doi:
10.1016/j.cose.2015.09.005.
[8] L. P. Zhuhadar and M. D. Lytras, “The Application of AutoML Techniques in Diabetes diagnosis: Current approaches, performance, and future directions,”
Sustainability, vol. 15, no. 18, p. 13484, Sep. 2023, doi: 10.3390/su151813484.
[9] ChatGPT. (GPT-4). OpenAI. Accessed: Sep. 26, 2023. [Online]. Available: https://chat.openai.com/chat
[10] O. Schwartz, “In 2016, Microsoft’s racist chatbot revealed the dangers of online conversation,” IEEE Spectrum, Jan. 05, 2024. [Online]. Available:
https://spectrum.ieee.org/in-2016-microsofts-racist-chatbot-revealed-the-dangers-of-online-conversation
[11] “The use of AI and AI-assisted technologies in writing for Elsevier,” www.elsevier.com. https://www.elsevier.com/about/policies-and-standards/the-use-of-
generative-ai-and-ai-assisted-technologies-in-writing-for-elsevier
[12] A. Bosshardt, “Handreiking ChatGPT bij Saxion,” Saxion University of Applied Sciences, Enschede, February 2023.
[13] A. Tiulkanov, “Is it safe to use ChatGPT for your task?”, University of Oslo, Oslo, January 2023. [Online]. Available: https://www.vtpi.org/avip.pdf

Unless otherwise stated, all images are licensed under Canva.