Theory of Machine: Unleashing AI on your Users without Confusion

Imagine this: You've just downloaded a new AI-powered MasterChef app on your phone, excited to discover mouthwatering recipes tailored to your taste buds. With anticipation, you grant the app access to your dietary preferences and eagerly input the items you found in your pantry: a can of chickpeas, a bag of quinoa, and a jar of sun-dried tomatoes.

But instead of presenting you with a delicious, customized recipe, the app starts generating bizarre, inedible concoctions. It suggests a dish that mixes chickpeas with chocolate syrup, or a quinoa salad topped with an absurd amount of hot sauce.

You're left scratching your head, wondering what went wrong, and how to fix it.

Whether AI fails completely or slightly misses the mark, most users, feel the same: confused. They are guests in the world of AI. For them, the inner workings of AI are enigmatic and complex.

This isn’t a failure of AI. It’s a failure of design. AI is imperfect. We have to teach our users that AI isn’t perfect. This is where we meet our hero: the “theory of machine”.

A “theory of machine” is a mental instruction manual imprinted in the user’s brain that tells them how the AI works. This manual determines how we try to use AI and what we think is possible.

If how the user thinks it works ≠ how it actually works, we run into issues.

When the user’s instruction manual doesn’t reflect how the tool works, when the expectations do not meet up with reality, users are left baffled, annoyed, and confused.

Let’s go on a quick tangent and quickly dive into psychology research. We will bring it right back to AI in a heartbeat.

🧠 A Tangent into the Theory of Mind: Understanding Our Fellow Humans

Picture this: You're hosting a lively dinner party with close friends, and everyone's bringing their favorite homemade dish.

Your good friend, Peter, arrives with his legendary lasagna, a dish he's immensely proud of. The table is filled with various dishes, from your aunt's savory chicken casserole to your neighbor's tangy mango salad. But as the evening progresses, you can't help but notice that Peter's lasagna remains largely untouched.

This is where the theory of mind comes in. We try to understand, what other people expect, what other people feel, and what other people want.

You start to imagine how Peter might be feeling: crestfallen, embarrassed, or even questioning his cooking skills. As a thoughtful host, you decide to take the lead.

As you put yourself in Peter's shoes, you tap into his thoughts and feelings, allowing you to make a well-informed decision.

By understanding Peter's wants and needs, you can effectively address the situation and ensure a positive outcome for everyone involved.

How can we use the theory of machine to design AI features and tools

During childhood, we gradually learn to understand the thoughts and feelings of others, like assembling a complex puzzle piece by piece. Extended to the theory of machine, this means that the early interactions with AI features and tools have to be designed with intention.

As a designer, developer, or product manager, it's your job to make sure users grasp how the AI works and how to interact with it. Guide users towards an accurate theory of machine, mirroring AI's inner workings without delving into technical complexities.

Most of the time, how the user thinks it works ~ how it actually works is the sweet spot.

Each interaction adds a piece to the puzzle, helping users understand AI better.

How can you give the user an instruction manual without giving them a tutorial of AI?

• Metaphors. Describe your AI in the terms of a metaphor.A common example is to explain AI as a master chef. It takes in raw ingredients (data), follows a recipe (algorithm), and then presents you with a delicious meal (output).This makes it abundantly clear that the quality of the output is determined by the data. Not even the greatest chef can create a succulent Rattatouli with spoiled peppers.

• Prototypes. A prototype is a representative example that showcases the typical behavior of an AI model, highlighting the typical relationship between input and output pairs.Our masterchef AI can whip out fantastic recipes for whatever ingredients you have lying around. We could tell our user a story that shows a prototypical example:You come home after a long day, and you're craving a tasty homemade meal. You take a peek inside your fridge, and you find a bunch of cherry tomatoes, some fresh basil, a piece of mozzarella, and half a box of spaghetti. With a sense of adventure, you feed these ingredients into the AI, and it gets to work, using its culinary prowess to craft the perfect recipe. Almost instantly, it comes up with a classic Italian dish: Spaghetti Caprese.

• Counterfactual. You're using an AI to predict how tasty a meal will be based on its ingredients. Illustrate the AI's behavior for the user by tweaking the salt levels in a dish, like adjusting the volume on a radio, to help them understand the input-output relationship.By playing around with these scenarios, the user gets a feel for how the AI behaves and learns how the data influences the outcome, all without diving into the complex mechanics behind it.A counterfactual is a hypothetical scenario (”what if”) in which certain conditions or elements are altered.

To-Do: Get started on formulating your user’s theory of machine

Understand how users think it works, what they expect from it, and how they want to interact with it:

1. Understand the status quo: What is the current instruction manual your users have of your AI? How do they think it works? How are they interacting with it?

2. Understand the AI: How does the AI actually work? How would you interact with it with all your knowledge? Is there a discrepancy?

3. Bridge the gap: Can you think of a metaphor, a counterfactual, or a prototype that shows the user how to interact with your AI?Guide the user with given examples or counterfactuals. Allow the user to experiment. Explain a guiding metaphor for how to think about the tool.

Sources

1. J. Logg - The Psychology of Big Data- Developing a 'Theory of Machine' to Examine Perceptions of Algorithms.

2. J. Logg - Algorithm Appreciation- People Prefer Algorithmic To Human Judgment.

3. N. Rabinowitz - Machine Theory of Mind.

4. G. Stephens - Speaker–listener neural coupling underlies successful communication

5. P. Turner - A Psychology of User Experience

6. F. Bosco - The complexity of theory of mind