closetly

Our app is designed to help you pick out outfits based on the clothes you already
have, and it works for everyone, regardless of gender. It's super cool because it
doubles as a clothing store, showing you items from different brands that match your
personal style. You can even filter outfits for different occasions, like casual
hangouts, going out with friends, or special events with dresses, bags, and shoes
included. The whole point is to save time, especially for busy students who often
spend way too long trying to figure out what to wear every morning. Instead of
standing in front of your closet for ages, you just input your clothes, shoes, and bags
into the app, and it generates a bunch of outfit ideas for you. This way, you can save
time in the mornings and even wake up a little later because all you have to do is
press a button for outfit suggestions.

But, like any project, there are some challenges we need to think about. One big
issue is getting users to accurately label their clothes. If they don’t describe their
items right, the outfit suggestions might not be very helpful. For example, if someone
calls a blue shirt “light blue” or doesn’t specify what type of pants they have, it could
mess things up. Plus, if someone has a small wardrobe, they might end up getting
the same outfit ideas over and over, which can get really boring really fast.

Another challenge is that everyone has different styles, so making an outfit generator
that works for everyone is pretty tricky. Not everyone wants to dress the same way,
and figuring out how to make personalised suggestions is definitely a task. On top of
that, building an algorithm that knows how to match colours and stay on top of the
latest fashion trends sounds super complicated. We want the app to be stylish and
relevant, so we’d have to keep it updated regularly.

Partnering with clothing brands could also be a hassle, especially when it comes to
managing what’s in stock and making sure users get accurate recommendations. It
can be tough to keep up with all that. And we need to think about keeping users
engaged over time, fashion changes quickly, and we want the app to feel fresh and
exciting. Some people might also have concerns about sharing their wardrobe info
because privacy is a big deal nowadays.

With so many fashion apps out there, we really need to find a way to stand out from
the crowd. We want our app to be unique and offer something different than what
others do. Lastly, it’s super important that the app is easy to use. If it’s too confusing
or complicated, people might just skip it altogether. We want it to be user-friendly so
everyone can enjoy it. And of course, let’s not forget that building and maintaining
the app can get pretty expensive, so we’ll have to think about how to manage those
costs.
Overall, our app has a lot of potential, but it’s important to keep these challenges in
mind as we develop it. I really think it could make a big difference for people trying to
figure out what to wear each day.
Exercise 2: Create Your Own AI Artwork
25 marks

Objective: Use two different AI art generation tools to create an image,

experimenting with different prompts and reflecting on the process. You will create
three versions of your artwork by revising the text prompts and observing how each
version changes based on the prompt modifications.

Steps:
1. Plan Your Artwork:
· Brainstorm an idea: Think of a creative theme, scene, or concept you
want to visualize. It can be anything—fantasy, nature, a futuristic city,
etc.
· Write down the description you will use as the base for generating
the artwork. Keep it clear but creative.
2. Use Two AI Tools:
· Choose two different AI tools for your artwork creation from the list
below:
1. DALL-E (now known as Craiyon):
https://www.dall-efree.com/user/image-generator
2. Midjourney: https://www.imagine.art
3. Artbreeder: https://www.artbreeder.com/create
4. Stable Diffusion: https://openart.ai/

3. Experiment with Prompts:

· Enter the initial prompt into both tools and generate your first image.
· Modify and revise the prompt at least two more times to create
different versions of the image. You can:
o Change wording or add details.
o Adjust styles or elements in the description (e.g., colors, lighting,
objects).
o See how the prompt changes affect the final outcome.
· Create 3 iterations per tool by experimenting with different prompts for
each image.
First- An ice-cream truck in front of a school for children to get ice cream
Artbreeder- OpenArt-

Second- An ice-cream truck labelled (Icy) in front of a secondary school for children
to get ice cream in the afternoon time

Artbreeder- OpenArt-

4. Reflect and Compare:

· Choose one final image from your generated artworks that best
matches your idea.
· Compare the results from both AI tools:
o How did changing the prompt affect the image outcome?
The more details added in the prompt, the more accurate outcome
is received
o How did each AI tool respond differently to the same or similar
prompts?
OpenArt seemed to understand the prompt more and generated a
better image than Artbreeder even after adding more details
o What was surprising or interesting about the results?
That it generated an image on exactly how I imagined it to be when
writing the prompt
5. Discuss AI in Art:
· Reflect on your experience using AI to create art:
o What are the advantages and challenges of using AI in art
creation?
-Advantages of using AI in art, offering new styles and ideas, and
generating complex designs that might be difficult for humans to
create manually.
-Challenges of using AI in art include limited control over the final
result, the risk of the artwork not fully matching the artist's vision
 o How does AI’s interpretation of your prompt differ from what you
originally imagined?
It was surprisingly similar to what I imagined while writing the
prompt
o How might AI impact creativity in the future?

AI may improve creativity by providing new tools, but it could also

reduce the need for skills, making us question the role of human
creativity.
6. Deliverable:
· One image from each AI tool, showing the final result of your prompt
revisions.

OpenArt- Artbreeder-
· One-page report (around 500 words) discussing your creative idea,
the prompt modifications, the differences in the artwork produced, and
your reflections on the process.

Tip: Experiment with different prompt styles (e.g., descriptive vs. simple, abstract vs.
detailed) to see how the AI responds!

MAYMOUNA

Maymouna
Exercise2- reflection

Exploring AI in Art Creation

I imagined a simple yet vivid scene: an ice cream truck parked in front of a school,
capturing a happy moment as students gathered for a refreshing treat after a long day. I
used two AI tools, OpenArt and Artbreeder, to bring this concept to life, experimenting with
different prompts to see how each would interpret my vision. The results were both
enlightening and surprising, providing insights into the possibilities and limitations of AI in art.
My initial prompt was simple: “An ice-cream truck in front of a school for children to get ice
cream”. I used this in both tools to generate the first images. The results were decent, but
adding more details improved the outcomes. For example, revising the prompt to “An ice-
cream truck labelled ‘Icy’ in front of a secondary school for children to get ice cream in the
afternoon” generated an image in OpenArt much closer to what I had imagined, with the
label and afternoon lighting accurately depicted.

The differences between OpenArt and Artbreeder were significant. OpenArt responded well
to detailed prompts and produced images more aligned with my vision. For instance, the ice-
cream truck with the label “Icy” and the afternoon setting was spot on. Artbreeder, while
creating aesthetically pleasing images, struggled to incorporate the specifics, even when I
refined the prompt.
What surprised me most was how closely OpenArt matched my mental image. When writing
the prompt, I could visualize the truck, school, and children, and OpenArt’s output reflected
these elements well. Artbreeder, on the other hand, didn’t achieve the same level of
accuracy in interpreting my prompts.

This experience gave me a deeper understanding of the advantages and challenges of AI in

art creation. AI offers new possibilities for style, detail, and complexity, and can quickly
generate art that would take much longer to create manually. It also introduces new ideas or
perspectives, providing a creative boost.
However, AI has limitations. Users don’t have complete control over the final image, as the
AI’s interpretation can deviate from the original vision. Misinterpretation of prompts can also
occur, as I found with Artbreeder, which didn’t fully capture the scene even when I added
more detail.

AI will likely play a major role in creative industries, offering artists and designers new tools
to experiment with. While it can enhance creativity, it also raises questions about the role of
human creativity and traditional artistic skills. Despite this, I believe AI will always require
human input to guide and inspire its creations.

In conclusion, OpenArt produced results that were more aligned with my vision, while
Artbreeder offered an interesting but less precise alternative. This exercise helped me
explore the intersection of technology and creativity, showing how AI might shape the future
of art.

Overall Reflection Maymouna-

For this assignment, I focused on Exercise 2, where I explored AI-generated art using
OpenArt and Artbreeder. My goal was to visualize a scene with an ice cream truck parked in
front of a school, surrounded by happy students. This process highlighted both the strengths
and limitations of AI in art creation and emphasized the importance of my active involvement
in shaping the final results.

I began with a simple prompt: “an ice-cream truck in front of a school for children to get ice
cream.” While the initial outputs were interesting, they didn’t fully match my vision. I realized
that I needed to refine my prompts to achieve better results. Adding details like “An ice-
cream truck labelled ‘Icy’ in front of a secondary school in the afternoon” improved the
outcomes, particularly in OpenArt, which captured the scene more accurately with the label
and lighting matching my expectations. This made me appreciate how much control I had
over the process through careful adjustments.

The most rewarding part of the exercise was the active experimentation. I wasn’t just
inputting random ideas but constantly refining my prompts and learning from each iteration.
The quality of the images depended on how much effort I put into guiding the AI. While
OpenArt responded well to my changes, Artbreeder struggled to capture certain details,
which was frustrating but part of the learning process. This pushed me to think more
creatively and adapt my approach.

In conclusion, this exercise taught me how to engage with AI tools in a more interactive way,
balancing the technology’s limitations with my artistic vision. It was both a technical and
creative challenge, showing me that while AI can be a powerful tool, human input is essential
to shape the final product.

Exercise 3:Text summarization

https://www.thenationalnews.com/health/2024/06/26/who-about-a-third-
of-worlds-population-at-risk-of-disease-due-to-lack-of-exercise/

AI accuracy comprehensivenes clarity consisten Context

s cy understandin
g

quillbot Quill bot Quillbot was really The The The

captured the comprehensive it summary summary summary did
main ideas didnt leave out any was really was keep the
correctly just key information clear and structure same
like the and it focused on used d really meaning and
article the main topic simple well in shared main
terms like one idea with the
the article paragrap article it felt
h like the
summary
was just a
quicker
version
 hypotenuse Hypotenuse Hypotenuse did The It felt like The
did capture leave a few details summary it wasnt summary
the main out and focused was clear connectin missed the
idea too. on the little details but the g the main point of
more than the terms sentence the article
main topic used were and it but it did
more was keep the
complicate confusing same idea
d than the
article
itself

Reflection on Using QuillBot and Hypotenuse

I tried out two AI tools for summarizing articles: QuillBot and Hypotenuse. My goal was to
see how well each tool could pick out the main ideas of a text and keep things clear. This
experience helped me learn about the strengths and weaknesses of these AI summarizers.

Experience with the Tools

Using both QuillBot and Hypotenuse was interesting. QuillBot was easy to use. I could
quickly input the article, and it generated a summary in just a few seconds. Hypotenuse was
a bit trickier to navigate and took more time to figure out. Even so, both tools helped me
understand the content better and showed me how AI can make summarizing easier.

How Well Each Tool Summarised

When I looked at the summaries from each tool, I noticed some big differences. QuillBot did
a great job of capturing the main ideas of the article. It provided a complete summary that
included all the important points and stayed focused on the main topic. The summary was
clear and used simple words, making it easy to understand. Plus, it was well-structured, so it
felt like a shorter version of the article.

Hypotenuse also captured the main idea, but it had some issues. While the summary was
clear, it often focused too much on small details instead of the big picture. As a result, some
key information was missing. The language in Hypotenuse's summary was more
complicated, which could confuse readers. Sometimes, the sentences didn’t connect well,
making it hard to follow what it was trying to say.

Which Tool Was Better?

Overall, QuillBot did a better job of capturing the main ideas. It was accurate, clear, and
covered everything well. The way QuillBot connected the sentences helped me understand
the content better. In contrast, Hypotenuse's summary, although it kept the same ideas, did
not tie them together as effectively and felt more confusing.

Surprises and Challenges

One surprising thing was how different the two tools were in terms of clarity. I thought both
would provide high-quality summaries, but QuillBot's summary was much easier to read. A
challenge I faced was figuring out how to use Hypotenuse properly because its interface was
more complicated. This made it harder to get a good summary right away. The complicated
language in Hypotenuse’s summary also took more time to understand, which made the
experience less enjoyable.

Conclusion

In conclusion, using QuillBot and Hypotenuse showed me how different AI summarization

tools can be. In my opinion QuillBot came out on top because it was clearer and more
accurate. This experience not only helped me understand how these tools work but also
showed me the importance of picking the right one for a specific task. I’m excited to keep
exploring AI and how it can help with summarising and other tasks in the future.

Overall reflection Alanood:

In this assignment, I took on the role of the leader, ensuring that each team member
understood their responsibilities and deadlines. I contributed significantly to brainstorming
ideas, which helped generate creative ideas and solutions for our tasks. I particularly
enjoyed the activities we engaged in, like the "Create Your Own AI" project.in Assignment 1,
where I worked on text summarization and the bias activity. Collaborating with my team
improved my understanding of these concepts and provided valuable insights into how AI
can be applied in various contexts. This experience taught me the importance of teamwork
and communication, as we learned from each other and combined our strengths to complete
the assignment successfully
Objective: Understand and apply your knowledge of different types of biases found
in AI systems through real-world examples and reflection.
Part 1: Identify the Bias
For each scenario below, identify the type of bias that is present. Choose from the
following list of bias types. Write 1-2 sentences justifying your choice for each
scenario.

Types of Bias:
1. Selection Bias
2. Confirmation Bias
3. Measurement Bias
4. Stereotyping Bias
5. Out-group Homogeneity Bias

Scenarios:

a. A hiring algorithm consistently prefers male candidates over equally

qualified female candidates because the training data was predominantly male.
Bias Type: selection bias
Justification: This bias occurs because the training data is not representative of the
actual population, leading to the algorithm favouring male candidates.

b. An AI system trained on data from a single urban area struggles to predict

traffic patterns in rural areas accurately.
Bias Type: selection bias
Justification: The AI system's training data is limited to a specific context (urban
areas), which means it lacks the necessary variety to accurately predict traffic
patterns in rural settings.

c. A sentiment analysis tool labels negative reviews from minority groups as

more negative compared to similar reviews from majority groups.
Bias Type: stereotyping bias
Justification:

This bias arises from the tool applying generalizations based on the reviewer's
demographic, leading to an unfair and inaccurate interpretation of sentiment.

d. A predictive policing tool focuses more on areas with historically higher

crime rates, reinforcing the same areas as high-risk.
Bias Type:confirmation bias
Justification:

This bias occurs because the algorithm relies on historical data that reflects previous
policing patterns, leading it to confirm existing beliefs about crime hotspots.
e. An AI model for diagnosing skin cancer is less accurate on darker skin
tones because it was trained mostly on images of lighter skin tones.
Bias Type:measurement bias
Justification:

This bias arises from the model being trained on a non-representative sample of skin
tones, leading to inaccuracies in diagnosis for individuals with darker skin.

Part 2: Case Study Analysis

Read the case study below and identify the types of biases present in the AI
system. For each bias, explain in 1-2 sentences why it is problematic in this specific
context.

Case Study:
A financial institution uses an AI system to approve loan applications. The training
data predominantly consists of applications from urban areas. The AI system is
found to approve fewer applications from rural areas. Additionally, the system tends
to favor applicants with higher educational qualifications, even when income levels
and credit histories are comparable. The model also shows a trend of approving
more applications from certain ethnic groups while rejecting others with similar
financial profiles.

Biases Identified:

1. Bias 1:selection bias

Explanation: The AI system's training data primarily represents urban
applicants, which results in it not accurately reflecting the financial realities
of rural populations.

2. Bias 2:confirmation bias

Explanation: The system’s tendency to favor applicants with higher
educational qualifications may reinforce existing biases in the financial
system, where education is often correlated with socioeconomic status.

3. Bias 3:stereotyping bias

Explanation: The tendency of the AI model to approve more applications
from certain ethnic groups while rejecting others, despite similar financial
backgrounds, reflects a stereotyping bias.

Overall information: