AI Glossary / MLOps and deployment / Beginner

Human-computer interaction

Definition

Human-computer interaction, or HCI, is the study and design of how people interact with computing systems. It examines people's goals, abilities, activities and environments alongside the interfaces and system behavior that support them. Its aim includes making technology useful, understandable, accessible and appropriate for the people affected by it.

What it means

HCI is broader than arranging buttons or choosing attractive colors. It asks whether a system helps people accomplish meaningful tasks, whether they understand its feedback, and what happens when something goes wrong. The field draws on computing, design, psychology and social research. For AI products, HCI includes how people interpret suggestions, correct errors, maintain control and understand the limits of a system whose outputs may vary or be uncertain.

How it works

A typical HCI process begins by understanding users and their context through observation, interviews or other research. Designers create possible interactions, often as prototypes, and evaluate them with representative tasks and participants. Findings guide revisions. Measures can include successful completion, errors, time, accessibility barriers, comprehension and subjective experience. No single measure captures the whole interaction. Evaluation should reflect realistic circumstances, including interruption, limited attention, unfamiliarity and the use of assistive technology where relevant.

Three practical examples

• A scheduling tool is redesigned after users repeatedly mistake a tentative suggestion for a confirmed booking.
• An AI writing assistant makes proposed changes visible and gives the writer a clear way to accept, edit or reject them.
• A self-service form is tested with keyboard navigation and screen-reader users so that essential actions do not depend only on a mouse or visual layout.

Think of it this way

Think of designing a doorway by considering the people who must pass through it, what they are carrying and whether they can tell which way it opens. A technically strong door can still be difficult to use. The analogy is limited because software interactions evolve over time and can involve information, uncertainty and collaboration rather than a single physical action.

Worked example

Suppose ten participants each try to change a delivery address in a prototype. Six finish correctly without help, two finish only after assistance and two fail. The unassisted success rate is 6/10 = 60%, while assisted completion is 8/10 = 80%. Reporting only 80% would hide the support required. If a redesign lets nine of ten new, comparable participants finish unassisted, that is encouraging evidence, but the small samples and differences between participants limit the conclusion. Review observed errors and task conditions alongside the percentages before claiming a general improvement.

Illustrative case study

Fictional case: Maple Assist builds an AI tool that drafts replies to customer enquiries. In testing, staff sometimes assume a fluent draft has already been checked against the customer's account. The team changes the interface to show what information the draft used and places review actions beside the text. It tests whether people notice missing information and correct the draft before sending. This fictional case illustrates interaction design and evaluation, not a claim that a label or interface change guarantees accurate decisions.

Common misconceptions and mistakes

Do not equate usability with visual polish or assume that designers can reliably predict every user's behavior without evidence. More engagement is not always better if people are spending extra time recovering from confusion. User satisfaction can coexist with unnoticed errors, especially when AI output sounds convincing. Accessibility also cannot be reduced to one font-size setting. Finally, testing only experienced colleagues can miss problems faced by new users, people with disabilities or people working under different conditions.

Business use

Businesses use HCI to identify friction, reduce avoidable support needs and create products that fit real workflows. Begin with a task outcome and a representative user group, then test whether the product supports that outcome. For AI, measure both useful assistance and inappropriate reliance. Plan for correction, cancellation and recovery, and make system state understandable. Keep research findings tied to design decisions so later teams understand why an interaction works as it does.

People also ask

Is HCI the same as user-interface design?

Interface design is part of HCI, but HCI also studies behavior, context, evaluation and broader effects.

Does HCI apply only to screens?

No. It includes voice, physical devices, gestures, collaborative systems and other forms of interaction.

Can a highly accurate AI system still have poor HCI?

Yes. People may misunderstand its purpose, miss its errors or be unable to correct its actions.

Check your understanding

A new interface reduces task time but doubles the number of incorrect submissions. Has it clearly improved usability?

Answer: No. Speed is only one measure. The team must consider correct completion, error consequences and recovery effort before judging the overall interaction.

Related terms

• Usability
• user experience
• interaction design
• accessibility
• human-centred design
• mental model
• human-AI interaction

Learn next

Choose one realistic task in an existing product and observe several representative people attempting it without coaching. Record what they expected, where the interface differed from those expectations and how they recovered. Turn the observations into specific design changes, then evaluate the revised interaction with fresh tasks or participants where possible.

More in MLOps and deployment

• Golomb coding - Golomb coding is a family of lossless variable-length codes for nonnegative integers.
• Gradient Checkpointing - Gradient checkpointing, also commonly called activation checkpointing, is a training technique that reduces stored intermediate activations by recomputing...
• Grammar-based code - A grammar-based code represents data using a formal grammar that reconstructs the original sequence through reusable rules.
• Incremental encoding - In string compression, incremental encoding represents a string partly through shared content with a preceding string, commonly storing their common-prefix...
• Interlaced video - Interlaced video represents a picture through fields containing alternating sets of image lines, commonly grouped as two fields per frame.
• Key Value (KV) store - A key-value store is a data-storage system that associates a key with a value and supports retrieving that value using its key.

Sources

Primary institutional references: ACM SIGCHI, https://sigchi.org/about/ ; ACM/IEEE Computer Science Curricula 2013, HCI knowledge area, https://www.acm.org/binaries/content/assets/education/cs2013_web_final.pdf . SIGCHI page opened and curriculum search excerpts consulted; full context-specific and editorial review remain pending. Numerical and organisational examples are illustrative.

Previous: Human-cognitive bias Next: Human evaluation