When you hear backpropagation, you probably think of machine learning, neural networks, and intimidating math. But even if the concept is new to you there’s no reason to worry. Because if we look closely, backpropagation isn’t just a computer science algorithm for machine learning.
No, backpropagation acts on the philosophy of learning through feedback, and thereby has a lot in common with design thinking.
In this article, I compare design thinking to machine learning to make complex concepts from computer science more graspable. I translate the logic of backprop (backpropagation) into design thinking language, and I illustrate how both follow the same idea: iterative improvement through feedback loops. In the latter half of the article I explain more machine learning concepts, the “bias”, “cost function”, what is “overfittig” and “underfitting”, as well as “activation functions”.
And what seems incredibly complicated or simply unknown to you now will be a little bit more clear and relatable by the end of this article.
Step 1: Define the Problem (Forward Pass)
In a neural network, information flows forward from input to output through various layers to produce a prediction.
In design thinking, there are phases instead of layers that a design problem flows forward through.
Both processes start by making a guess.
The neural network guesses the output; the designer defines who the user is to eventually guess what they might need.
This is the forward pass where action is taken based on what is known.
Step 2: Measure the Error (Testing & Feedback)
Once the network makes a prediction, it compares it to the real answer. The difference between the two is the error.
In design thinking, the error is identified during user testing and validation.
That feedback is crucial because it is how both systems learn what didn’t work.
- In ML: The network sees how far off its prediction was.
- In design: You see how well your prototype fits users’ needs (and where it fails).
Interestingly enough, because of their iterative nature both see failure not as the final judgment, but simply as useful information.
Step 3: Trace the Error Back (Reflection)
This is where Backpropagation comes into play. The algorithm sends the error backward through the layers of the network to see which parts contributed to the mistake. Concretely this refers to the nodes within each layer and the strength with which they influence the outcome.
Design thinkers do the same thing when they reflect:
“Where did our assumptions lead us astray?”
It could be that the results from a quantitative consumer study that was carried out in another geographical market (e.g. US) were overestimated and do not to apply to (let’s say) european consumers. That’s when it is learned that this specific artifcat should determine the other artifacts in subsequent design phases/layers less strongly.
Therefore, this is the learning loop. You trace the problem back to its source to identify where to adjust the design.
Step 4: Adjust the System (Iterate and Refine)
Once the neural network knows which “weights” caused the error, it updates them — slightly.
This is the iteration phase in design thinking: you tweak the prototype, refine the interface, or change the messaging.
Importantly, the network doesn’t throw everything out.
It makes small, calculated updates — one gradient step at a time.
That’s the secret to both AI and design: small improvements, continuously applied, compound into intelligence and innovation.
Step 5: Repeat the Loop (Continuous Learning)
Backpropagation isn’t a one-time correction. It’s an endless cycle:
predict → measure → reflect → update → repeat.
The same holds true for design thinking. You empathize, define, ideate, prototype, and test — again and again — until the solution aligns naturally with real-world needs.
The point isn’t perfection. The point is convergence through iteration.
let’s take this a step further, more technical..
The Loss Function
The loss function measures the error between the model’s output and the perfect solution during ML-training.
source: https://www.ml-science.com/loss-cost-function
Two high-error scenarios
Underfitting
If a machine learning model underfits, it is too simple to capture the data’s underlying patterns, resulting in poor performance on both training and new data. To fix this, you can increase the model’s complexity, add more relevant features or train the model for longer. (source)
a model that can not tell the difference between objects with wheels is too simple and underfits
In Design Thinking, underfitting would be a solution that is too general and does not solve a specific problem. Think of a park bench that look like a bench, feels like a bench but is too high to sit on or that soaks up water so it’s always wet.
Overfitting
If a machine learning model overfits, it performs exceptionally well on its training data but fails to generalize and make accurate predictions on new, unseen data. This is because it has learned the training data too precisely, including random noise and specific details, instead of the underlying general patterns. An overfitted model is unreliable for real-world applications and is characterized by low bias and high variance. (Source)
A machine learning model that knows the training data too well can not generalize it’s judgement to new data very well. It overfits
In Design Thinking, overfitting might refer to asolution that is too scientific and only works in the lab. For example: The usability tests of the new ERP software yield excellent results in controlled environments however, as soon as someone uses the software at a real desk it turns out that successful task completion is low due to the distracting office environment, or because at the start of the workflow the user does not know how to export data from an existing CDP into the new ERP.
Designers are bias
In AI, the bias is a constant. In design, it’s a consciousness.
We’ve heard about AI’s bias. A core limitation that comes all sorts of ethical and qualitative implications. In Design Thinking the bias is held by the designers. They add their assumptions and worldview into artifacts and MVP’s withot knowing it. Ideally, just like in science the bias is something we are aware about and that is reflected on so it can positively influence the learning experimentation. Eventually, bias is also gut feeling (like deciding when to close the double diamond) and one of the core non-tangibles factors when it comes to innovation and visionary ideas.
Bias is not bad… it simply “is”. And that’s crucial to take into account.
Takeaway
Backpropagation and Machine Learning may sound technical, but it’s really not that far off from design thinking and what designers already do:
- Prioritize Testing Ideas.
- Listen deeply to feedback.
- Learn what’s not working.
- Adjust intentionally.
- Keep looping and learning.
Or, as a neural network might put it:
“Minimize your loss function — one iteration at a time.”
How this is done in machine learning and the explanation of gradient descent and other algorithms that put backpropagation into action are topics for next time…
gradient descent to reduce the error. source
Comments ()