Week 2: The Machine Learning Spectrum and How it Affects Design

Understanding the different ways machines learn and why that matters for design

Feb 06, 2026

Introduction
The Three Machine Learning Styles
1. Supervised Learning
  1. Definition
  2. How it Works
  3. What it Means for Design
2. Unsupervised Learning
  1. Definition
  2. How it Works
  3. What it means for design
3. Semi-supervised learning
  1. Definition
4. Reinforcement Learning
  1. Definition
  2. How it Works
  3. What it Means for Design
Why this Matters to Your Work
Homework
Extra Resources
1. Additional reading
2. Interesting reads from the past week

When ChatGPT first launched I got into a few polarizing discussions around AI with my girlfriends who don’t work in tech. Naturally, a lot of them were skeptical about it—their younger siblings were using it to “cheat” in school and skeptics were warning people of doom and gloom. Most of them were rightfully worried that it would affect our brains and the overall job market. They vowed they would never use it.

One interesting observation I made in all of these conversations was that most of them didn’t realize they’ve already been using AI-powered products for quite some time. Spotify, Netflix, their banking apps, autocorrect on their phones… They’ve been “AI” users for years. When I’d share that with them, they’d be in shock.

AI has a different “ambassador” every decade, each basically being a subset of AI. Right now, the “ambassador” is GenAI, 10 years ago it was topics around machine learning, and in 10 years it will be a different subset. So it’s not at all surprising that it’s not clear to people what AI really is.

So what’s the difference? When we talk about AI today, what are we really talking about?

To simplify it, Artificial Intelligence (AI) is the broad concept of machines mimicking human intelligence to perform tasks. It is a field of computer science dedicated to creating systems capable of performing tasks that typically require human intelligence, such as reasoning, learning from past experiences, problem-solving, and understanding language.

Last week we learned about deterministic and probabilistic systems. We kept it simple, relating deterministic systems to the systems we have been most familiar with the past 20-30 years, like your desktop computer or a static website. We related AI to probabilistic systems, systems that can personalize experiences and provide contextual answers to questions, because the AI most of us use on a daily basis is inherently probabilistic.

Some AI, like expert systems in the medical field, rely on determinism more than others. However, probabilistic AI uses statistical models to analyze data, identify patterns, and make predictions, instead of relying on fixed rules. This type of AI is able to “learn” from data and improve its performance over time.

These types of changes in the backend are directly influencing our jobs as designers. Now more than ever we have to understand the relationship between frontend and backend in order to create effective experiences. So how do machines learn?

The Three Machine Learning Styles

1. Supervised Learning: Learning by Example

Definition

Supervised learning is a machine learning technique that uses labeled data sets to train AI models to identify the underlying patterns and relationships. The goal of the learning process is to create a model that can predict correct outputs on new real-world data.1

In order for the model to correctly execute a certain action, you first show it thousands of examples that it can learn from and then make “educated guesses” in the future.

How it Works

Let’s say you’re training your system to automatically filter email spam. You will give it tons of information on emails that have been marked “spam” or “not spam.” The machine will “study” these labeled examples and learn to make predictions on new, unlabeled data.

This is how most of them learn to make those educated guesses. They study labeled examples until they can predict labels for new data.

Remember the probabilistic examples from last week? Some of them are prime examples of systems using supervised learning:

Uber’s surge pricing is most likely trained on historical ride data, labeled with demand levels and corresponding optimal prices. It learns from past surge events such as concerts, conferences, weather patterns, etc. helping it determine future pricing in instances where demand for rides spikes.
LinkedIn connection suggestions gets trained on millions of examples where it knows “these people connected” vs “these people didn’t connect.
Autocorrect is trained on datasets where it knows what people meant to type (the label) vs. what they actually typed. Still weird people are meant to type duck…
Package delivery estimates are trained on historical delivery data labeled with actual delivery times. It learns patterns from weather, distance, time of year, route complexity, etc. to estimate delivery windows.

What it Means for Design

Knowing that your product and experience is powered by supervised learning helps you make design decisions that allow the system to support an experience with the things it learned.

You have two jobs as a designer:

1. Understand what experience the system’s knowledge is powering.

2. Understand what data the system is collecting for future learnings.

The design needs to support the experience, like in our email example where a spam folder needs to exist so the emails the system labeled as spam are still accessible to the user for viewing but organized in a separate folder. However, if your system is also gathering data for future supervised learning, it should also support that. So, giving users the ability to label their own spam through the UI allows the system to collect new data that’s labeled as “spam.”

2. Unsupervised Learning: The Pattern Hunter

Definition

Unsupervised learning uses machine learning algorithms to analyze and cluster unlabeled data sets. These algorithms discover hidden patterns or data groupings without the need for human intervention.2

You give the system unlabeled data and tell it to find patterns. It clusters similar things together, spots outliers, and discovers hidden structures and connections. Unlike supervised learning, there’s no “correct answer list” for it to learn from. The system just looks for structure in whatever data they have access to, making guesses about what patterns might be meaningful.

How it Works

Let’s say you have a restaurant app and want to create a curated list of restaurants with niche names like “bougie girls brunch” or “NFL brewskis for the boys.” You would have your model learn through unsupervised learning. The system would look at user patterns of restaurant selections, times and days of booking, types of bookings, etc. and cluster selections together. Later, when another user selects a certain restaurant, they will get suggestions of related restaurants from that cluster, even if the restaurants might not serve the same cuisine as your original selection.

Examples from last week that use unsupervised learning:

Spotify Discover Weekly uses collaborative filtering and clustering, by finding patterns and clusters in listening behavior without being told “these songs go together.” It discovers taste neighborhoods from unlabeled data.
YouTube doesn’t manually categorize every video into topics like “Lo-fi Hip Hop” or “Mechanical Keyboard ASMR.” It clusters videos based on who watches them together, creating hyper-specific categories that emerge from viewing patterns.
Your bank groups transactions into categories like “Dining,” “Entertainment,” “Travel” by finding patterns in merchant names and transaction types, not because someone labeled every Starbucks purchase as “dining.”
Amazon’s “Frequently Bought Together” discovers purchasing clusters by analyzing millions of shopping carts and finding items that appear together more often than random chance would predict.

What it Means for Design

Unsupervised learning surfaces insights humans might miss, but those insights aren’t always meaningful. The system will find patterns, but it is your job to translate algorithmic patterns into human-relevant categories and make sure the platform UI can handle changes without the UX changing.

This is where a good understanding of effective information architecture comes in. You need to test whether discovered patterns actually help users accomplish their goal and then design interfaces that let users easily access and trust these groupings.

3. Semi-Supervised Learning: The Hybrid Approach

The name says it all but here’s the definition

Semi-supervised learning is a branch of machine learning that combines supervised and unsupervised learning by using both labeled and unlabeled data to train AI models for classification and regression tasks.3

Many platforms and systems combine the two and update as they grow. In design terms, understanding where your product lies and what the product needs vs. what the user needs is essential.

4. Reinforcement Learning: The Trial-and-Error Student

Definition

Reinforcement learning is a type of machine learning process in which autonomous agents learn to make decisions by interacting with their environment.4

The system tries different strategies, sees what happens, and gradually shifts its behavior toward whatever gets rewarded.

How it Works

Netflix is a great example of a system that utilizes reinforcement learning. Let’s say you’re on your Netflix homepage looking for what to watch. The system recommends titles based on what you’ve watched and rated in the past (supervised learning), but then it tests which images and arrangements get you to click certain titles, learning over time what works and what doesn’t. That’s reinforcement learning.

Another example of reinforcement learning is dynamic ad bidding. Advertising platforms learn how much to bid on ad placements by trying different bid amounts and getting rewarded when ads lead to conversions. The system explores what happens if it bids higher and analyzes results. It optimizes spending over time through trial and error.

What it Means for Design:

Reinforcement learning means the system is goal-oriented and it optimizes for whatever reward signal you define. This can be a powerful experience enhancer but can also very easily make things go to shit.

If you reward a customer service chatbot for closing conversations quickly, it’ll learn to give short, unhelpful answers. If you reward it for customer satisfaction scores, it might learn to be so agreeable it never enforces policies. The reward function IS the product behavior.

Your job is to define success metrics that actually align with user needs. You need to design feedback mechanisms that give the system good reward signals and also build safeguards so the system can’t optimize itself into problematic behavior. This is where error prevention is crucial. Being very clear on what the product is optimizing and how that could backfire will help build and design a successful reinforcement learning product.

Why This Matters for Your Work

More and more experiences are about personalization, which increases the need to create dynamic experiences, instead of static. When we design AI experiences we have to become very familiar with the relationship between frontend and backend and be clear on what interactions the system requires.

Design discovery questions should include:

Which ML approach are we using?
What training data exists or needs to be created?
What is the long term goal for the system?
How will the system improve over time?
What user feedback is needed to do that?
What could go wrong if the system optimizes incorrectly?

Understanding what you’re working with helps you set requirements for the experience you’re designing. If you know the product needs a dashboard that will have different personalized categories of recommendations, what type of design will handle that effectively? What character limits will you set for the category names? How will the categories be represented visually? UX and UI requirements are shifting from linear journeys towards accommodating dynamic, contextual journeys.

UX Homework: Identifying machine learning types

The exercise this week is something you can do on your commute to work or during your breaks.

Open your phone and take a look at the apps you use daily. For each one that feels “intelligent” try to determine if it utilizes:

Supervised learning - making predictions from labeled examples
Unsupervised learning - finding patterns in your data
Reinforcement learning - optimizing results or the whole system through trial and error

Once you identify the ML type, ask: How does this influence the user experience? Where are the feedback loops? What’s being optimized? How could it fail?

The goal is not to become an ML engineer, but to develop pattern recognition for how systems learn so you can design better interactions around their learning processes and the experiences they can power.

Have a friend you think would love this? Send it over so they can join in on the exercise.