When designing for AI, there are three essential parts of the interaction we need to consider: the input, the output, and the reward.
The input is the information a user sends to the machine.
The output is the machine’s response back to the user.
The reward is what the user does in response to the output—essentially, a second input that either reinforces or challenges the system.
These interactions can be overt or hidden. Overt inputs and rewards are intentional: typing into a chat box, clicking “thumbs up,” or asking for a regenerated response. Hidden ones are unintentional: behavioral data like how long a user lingers, where they drop off, or what they ignore. Both feed the system and shape future outputs.
Why the Use Case Matters
The use case ultimately determines how we design the inputs, outputs, and rewards. For example, take ChatGPT’s chat. The real purpose of the product isn’t just to answer questions—it’s to learn how people interact with conversational AI and train the model.
- Input: a user types into an open text box with little guidance.
- Output: the AI responds with text.
- Reward: the user can copy the output, rate it up or down, ask for a regeneration, reply with a new prompt, or simply leave.
The design is beautifully simple, but perfectly suited for the use case: learning from interaction at scale.
Exploring Just the Input
Let’s imagine a hypothetical case: “find the best restaurant near me.” There are endless ways we could design the input alone:
- A plain text box where the user types: “find the best restaurant near zip code 10001.”
- A text box with built-in suggestions to guide phrasing, paired with a hidden input that auto-detects location.
- A palette builder where users select favorite cuisines, plus a hidden location input that’s bundled into the prompt.
- Even an upload flow where users share photos of past meals they loved, combined with location detection.
Each of these designs changes the interaction—and changes what the AI learns in return.
The Bigger Opportunity
None of this is new in principle, but what’s new is the power of today’s language models. The possibilities for interaction design are richer than ever, and our responsibility as designers is to create experiences that are transparent, valuable, and joyful.
So whether you’re designing inputs, shaping outputs, or defining rewards, remember: you’re not just shaping an interface—you’re shaping how people and machines learn from one another.
Have fun designing. 😃