The pitch of the whole lecture: code is how we make a machine execute it; design is how we make a human feel it. HCI is not decoration bolted on at the end — it is the part of software engineering that decides whether the thing gets used at all.
01Why HCI matters
The business case
- 63% of large software projects go over cost. Managers give four user-related reasons: users requested changes; the team overlooked tasks; users did not understand their own requirements; user–developer communication and understanding was insufficient.
- Coders jump too quickly into detailed design. The result is founded on incorrect requirements, has an inappropriate dialog flow, is not easily used, and is never tested until it is too late.
- The slogan: pay a little now, or pay a lot later. A redesign after implementation is expensive; a paper prototype thrown away is cheap. This is exactly the traditional waterfall vs agile / rapid prototyping contrast — high financial risk and time cost once, versus low risk spread over many cheap iterations.
- Usability sells. Product reviews emphasise usability (Consumer Reports, Telles 1990: “the interface has become an important element in garnering good reviews”). Customers have used related products and can download trial versions including competitors'. Features matter less: today's users are impatient and intolerant of bad design.
What bad design actually costs
- Money. Costly errors in serious systems — financial institutes, incorrect billing, systemic failures.
- Lives. Medical, air traffic control, aviation, nuclear.
- Safety and security. In-car navigation, face-ID hacks, RFID skimming.
- Trust. Confidence in an institution (a stock exchange, an airline) is itself a usability outcome.
HCI is not just interface design. In the KLM/Expedia case, a traveller tried to upgrade with air miles, the website offered no way to pay with miles, he abandoned the session, and his whole reservation was silently cancelled. Four separate systems each behaved “correctly”. The failure lived in the customer journey, the organisation of work, task delegation and business processes — not in a button. “Computing for the real world” means dealing with conditional truth, semantics and pragmatics: research precedes computing.
The lecture also makes a philosophical point with modus ponens: if vitamin B is good for hair, and this shampoo has vitamin B, then it is good for my hair — empirically false, because the truth condition is “iff vitamin B is metabolised from food”. Shampoo is not metabolised. Code written on folk beliefs behaves the same way. The parallel to code:
// folk belief if (hasVitaminB) { goodForHair(); } else { noEffect(); } // informed belief: only metabolised sources count if (fromFood && (metabolised || absorbed)) { benefit(); } else { noEffect(); }
02Three disasters to quote
Examiners love these because they compress the whole course into one sentence: the official cause was human error; the actual cause was poor HCI.
Mizuho / J-Com, Tokyo Stock Exchange (2005) — fat-finger syndrome
- A broker meant to sell 1 share at ¥610,000 and instead sold 610,000 shares at ¥1.
- Mizuho tried to cancel three times; the exchange does not cancel executed erroneous orders.
- Losses around US$225–331 million, confidence in the exchange damaged, a rival software maker's shares up 11%.
- Fat-finger syndrome = the tendency of stressed traders in fast-moving electronic markets to press the wrong button. Small targets + time pressure + stress + no confirmation step.
American Airlines 965, Cali (1995) — no feedback
- The crew typed ROMEO when they meant ROZO; the flight management system accepted it and displayed no confirmation of the chosen beacon.
- The aircraft turned toward the wrong beacon and flew into a mountain. 159 on board, 4 survived.
- Official cause: “pilot error”. Actual cause: poorly designed HCI — the system gave no informative feedback about the selected target, and the interaction was not designed to catch a plausible slip.
Three Mile Island (1979) — false feedback and alarm overload
- A lamp indicated that a valve was closed when in fact it was open — the indicator was wired to the control signal, not to the valve position.
- About 1,500 alarm indicators; a silencer that was either fully on or fully off; a caution tag physically obscured the relevant light.
- Design remedies named in the lecture: offer informative feedback, reduce short-term memory load, offer error prevention, support internal locus of control. Better control panels would have prevented it.
03What HCI is
The lecture gives four definitions; the first is the one to reproduce.
“HCI is concerned with the design, evaluation and implementation of interactive computing systems for human use and with the study of major phenomena surrounding them.”
- CHI 1985: “an input language for the user, an output language for the machine, and a protocol for interaction.” — useful when the question is about dialog.
- IBM Design Concepts: “the parts of the computer that you see, hear, touch or talk to; the set of all things that allow you and your computer to communicate.”
- One-liner: a discipline concerned with the design, implementation and evaluation of interactive computing systems for human use.
Note what the definition makes examinable: design, evaluation and implementation are three separate activities, and “for human use” is the constraint that makes it HCI rather than software engineering.
04A short history — and what it explains
The history is not asked as trivia. It is asked as a shift: from many users servicing one machine, to one user with many machines, to machines tending to the user's needs.
| Era | Machines / ideas | Role of the user |
|---|---|---|
| Batch processing | IBM 285 tabulator (1936), COLOSSUS (1943), Mark I (1944), ENIAC (1946, 19,000 tubes), IBM SSEC (1948) | No interaction; requests processed group-wise; the human is an appendix to the machine |
| Time-sharing | CTSS (MIT, mid-60s), Stretch (IBM 7030), TSS360, BASIC, JOSS | Gave the illusion of having one's own computer; dramatically increased accessibility and created the immediate need for HCI |
| Ergonomics (1956–63) | Console ergonomics, job control languages | Operator as a physiological being with limits |
| Interactive experience (1964–71) | Interactive terminals, command shells, speech synthesis, Programma 101 (Olivetti, 1965 — “first PC”) | Interaction starts to happen; commands replace job submission |
| Design rules (late 60s–70s) | Personal computing, dialogue rules, Videotex / Minitel | Rules for conversation with a system, not just for running it |
| AFIPS demo (1968) | Engelbart: mouse, one-handed chorded keyboard, windows, hypermedia, shared files, video conferencing | The workstation as a communication medium |
| The user becomes important (1980–87) | Xerox Star 8010 (1981), Apple Lisa (1983), Macintosh (1984), ACM SIGCHI (1982) | Users are studied, tested, and given guidelines |
| User takes centre stage (1988–) | Ubiquitous computing, embedded software, IoT, social robots, companions and coaches | The machine adapts to the human |
The intellectual landmarks you should be able to attribute
- Vannevar Bush, “As we may think” (1945) — the Memex. Store all personal books, records and communications; retrieve by indexing, keywords and cross-references; annotate with margin notes; save a trail (a chain of links) through the material. Intended as external memory and the forerunner of hypertext and the web — designed on microfilm, never built.
- J. C. R. Licklider, “Man–Computer Symbiosis” (1960). Human brains and machines coupled tightly; men set the goals, formulate hypotheses, determine criteria and evaluate; machines do the routinizable work. Its stated short-term goals read like a product roadmap: time-sharing, electronic I/O of symbolic and pictorial information, real-time interactive programming, large-scale information storage and retrieval, speech recognition, natural language understanding, heuristic programming.
- Douglas Engelbart (1962, 1964, 1968). “Augmenting human intellect”: increase a person's capability to approach a complex problem, gain comprehension and derive solutions. Produced the first mouse and the 1968 workstation demo.
- Ivan Sutherland, Sketchpad (1963). Hierarchical pictures and sub-pictures, master pictures with instances (object orientation), constraints maintained through changes, icons, copying, light-pen input. Also 3D graphics and head-mounted displays.
- Alan Kay (1969) — Dynabook. A self-contained knowledge manipulator the size of a notebook; the cardboard prototype of the laptop. Ted Nelson (1974) — Computer Lib/Dream Machines: computers for people, deep computer literacy, against centralisation.
- Xerox PARC (1973) — Alto. Personal workstation, bit-mapped display, mouse, windows, menus, scrollbars, Ethernet. Xerox Star 8010 (1981) commercialised it.
The Xerox Star, as the textbook case of GUI design
- First commercial personal computer for “business professionals”; first comprehensive GUI.
- Familiar conceptual model (simulated desktop); recognising/pointing rather than remembering/typing; property sheets; WYSIWYG; a small set of generic commands; high consistency and simplicity; modeless interaction; limited customisation.
- First system based on usability engineering: inspired design, extensive paper prototyping and usage analysis, usability testing with potential users, iterative refinement.
- Commercial failure: $15,000 (IBM had just announced a cheaper machine), limited functionality (no spreadsheet), closed architecture (no third-party applications), perceived as slow (it was not), and slavish adherence to direct manipulation.
Apple Lisa (1983) copied the Star, was somewhat cheaper ($10,000), and also failed commercially. Macintosh (1984) succeeded with “old ideas, well done”: sharp pricing ($2,500), no trail-blazing, corrected the Lisa's mistakes, a market that was now ready, a developer's toolkit encouraging third-party software, and interface guidelines enforcing consistency between applications.
05Pioneers, settlers, and the economics of usability
- Before, the computer cost more than the people. Today, people cost far more than computers, so applications can afford to trade processing power for human usability.
- New ideas often do not make it into products straight away.
- Pioneer systems developed innovative designs but were often commercially unviable; settler systems incorporated the well-researched HCI designs many years later — and made the money.
This is also why “usability engineering is software engineering”: the cost of a change grows with every stage it survives, so the cheap place to be wrong is a sketch, not a shipped system.
06HCI is multidisciplinary
Each discipline contributes a lens, and the exam likes matching a scenario to a lens.
| Discipline | Contribution | Example from the lecture |
|---|---|---|
| Computer science | Systems, graphics, interaction techniques | Engelbart's 1968 workstation; RFID Guardian firewall against RFID skimming |
| Artificial intelligence | Inference, perception, autonomy | A self-driving car that flags a pedestrian as a “false positive” |
| Engineering | Devices, materials, ergonomics of hardware | Mouse design; input hardware |
| Cognitive psychology | Memory, attention, perception, error | Memory overload causes password insecurity: too simple, self-related, reused, forgotten |
| Ergonomics / human factors | Anatomy, anthropometry, physiology, workload | Single-handed Maltron keyboard; VR gait-recognition suit |
| Sociology | Groups, networks, organisations | How people establish social networks; social-media analytics for strategy |
| Communication | Media, cues, presence | More cues to face-to-face communication supposedly yield better relationships |
| Anthropology | Culture, ritual, taboo | Calling-line identification may be disliked in non-Western cultures; the Bluetooth “Charming Burka” |
| Linguistics | Language structure and use | Speech recognition, speaker identification, large language models |
| Industrial design | Form, affordance, aesthetics | Washable keyboards; physical controls |
07Who does what: human vs machine
| Let humans focus on | Let computers handle |
|---|---|
| Sensing low-level stimuli · pattern recognition · inductive reasoning · multiple strategies · adapting · creating | Counting and measuring · accurate storage and recall · rapid and consistent responses · data processing · calculation · repetitive action |
The trend line: computers are moving up into pattern recognition and artificial creativity, which is precisely where human claims to uniqueness used to sit (chapter 2 returns to this with deep learning's “mistakes”).
The lecture's closing summary, in one sentence per bullet:
- HCI leads to cost reduction (less redesign), it is what sells, and it avoids disasters — financial, aerospace, nuclear, automotive.
- Historically the focus shifted from humans servicing systems to systems servicing humans.
- Visionaries come up with the next-generation interaction modes; cherish your geeks.
08Concept check
One best answer each. Check, read the explanation, move on — the feedback is the revision.
Question 1A large software project goes over cost. Which set of reasons is the one the lecture attributes to managers?
Question 2Which statement is the ACM SIGCHI definition of HCI?
Question 3American Airlines 965 crashed after the crew selected the wrong beacon. What is the HCI reading of the failure?
Question 4At Three Mile Island, which mechanism best explains why operators misread the plant state?
Question 5Time-sharing is historically important because…
Question 6Why did the Xerox Star 8010 fail commercially despite being the first comprehensive GUI?
Question 7How is Fitts' law related to the fat-finger syndrome at the Tokyo Stock Exchange?
Question 8“Users in a non-Western culture may dislike calling-line identification.” Which discipline does this observation belong to?
09Exam-style questions
Written in the shape of the past-paper short questions. Answer in the answer book first, then open the model answer.
“Given the KLM/Expedia example, is HCI just about interface design?” Answer the question, then justify your position by tracing the failure through at least three layers of the system.
Show model answer
No. HCI is concerned with the customer journey, the organisation of work, task delegation and business processes — anything that shapes the user experience. Interface design is one layer of it.
- Interface / screen. The Flying Blue system offered no option to pay for the upgrade with miles, and did not state that upgrading with miles is impossible. The user formed a goal the interface could not satisfy.
- Inter-organisational workflow. The ticket was sold by Expedia; KLM cannot service tickets sold by an agent. That is a business rule, but the user experienced it as a system failure. The KLM site should have checked with the agency and warned about impossible actions.
- State visibility across systems. The Expedia reservation was cancelled, but no channel told the user. He reasonably assumed the untouched reservation still stood and discovered the cancellation 30 hours before boarding.
- Recovery. Fixing it required a phone call and a seat that happened to be free; the cost was time and money. A confirmation/notification step would have made the state change visible and reversible.
Marking note. Full marks for “no” + at least three distinct layers + the point that a business rule which the customer must not have to know is still an HCI problem. Mentioning that usability is a property of the whole socio-technical system, not of the pixels, earns the top band.
Two accidents are officially attributed to human error: AA965 (wrong beacon) and Three Mile Island (misread valve). For each, name the design principle that was violated, the mechanism by which the operator was misled, and one concrete redesign.
Show model answer
| AA965 | Three Mile Island | |
|---|---|---|
| Principle violated | Offer informative feedback; error prevention | Offer informative feedback; reduce short-term memory load; error prevention |
| Mechanism | ROMEO/ROZO is a slip. The FMS accepted the entry and showed nothing about which beacon was selected, so the wrong intention was executed with no chance of recovery. | The lamp was driven by the control signal, not the valve position, so a false “closed” was displayed. ~1,500 alarms and an on/off silencer made the true signal indistinguishable, and a tag obscured the indicator. |
| Redesign | Display the selected beacon by name and show the resulting track on the navigation display; require confirmation when the entered identifier is close to another valid one. | Sense and display the actual valve position; prioritise and group alarms; allow graded silencing; keep indicators physically unobstructed. |
Marking note. The mechanism must mention what the human believed and why. Naming “human error” without the design mechanism scores zero on the HCI part — that is the whole point of the case.
Compare the Xerox Star (1981) and the Apple Macintosh (1984) as pioneer and settler. What are the three lessons a startup should take from the pair?
Show model answer
Pioneer. The Star introduced the desktop metaphor, WYSIWYG, property sheets, modeless interaction and — critically — the first usability-engineering process. It failed at $15,000 with no spreadsheet, a closed architecture and a reputation for slowness.
Settler. The Macintosh reused the ideas at $2,500, did not trail-blaze, fixed the Lisa's mistakes, arrived when the market was ready, and multiplied its value with a developer toolkit and interface guidelines that made third-party applications consistent.
- Being right early is not the same as being viable. Price, functionality and an ecosystem decide adoption, not the elegance of the interaction model.
- Process is the durable asset. The Star's usability engineering — paper prototypes, usage analysis, testing with real users, iteration — is what later products inherited.
- Consistency scales through guidelines and toolkits. A platform's usability is the consistency of applications you did not write.
Marking note. Any three defensible lessons with evidence. A pure chronology (“Star 1981, Mac 1984”) earns at most one mark.