bits.

FREE HISTORICAL FIGURES OF COMPUTING PROFILE

J. C. R. Licklider

1915–1990 · Psychologist and director of ARPA’s Information Processing Techniques Office

J. C. R. Licklider

Psychologist and director of ARPA's Information Processing Techniques Office · 1915–1990 · United States

J. C. R. Licklider argued that computers could participate in a continuing process of inquiry. A user should be able to inspect a result, change a hypothesis, and ask another question without repeatedly arranging a separate batch job. As a research administrator, he helped support the people and systems needed for that style of computing. He directed IPTO, an office within ARPA, not ARPA as a whole.

A research proposal, not a prediction checklist

Licklider trained in psychology and worked on problems of perception and human interaction with technical systems. His 1960 essay Man-Computer Symbiosis set out a programme for close interaction, including time-sharing, memory organization, languages, and input/output. Interactive systems already existed; his argument concerned a broader, more useful partnership rather than the invention of every form of human-computer interaction. IEEE biography, 1960 paper

The paper reports an informal study of his own work. He estimated that roughly 85% of his thinking time went into preparation such as obtaining data, plotting, or converting representations. He explicitly questioned whether his experience was typical. It is evidence for the problem he wanted to investigate, not a measured universal law about scientists or a proof that computers eliminate 85% of intellectual labour. 1960 paper, section 3

An interactive tool must preserve the connection between a question and its answer. Faster arithmetic alone is insufficient if input preparation remains slow, output is unreadable, or the interface hides the assumptions of the calculation. Conversely, a convenient interface can make a wrong model easier to use. Interaction changes opportunities for investigation; it does not establish the truth of the investigation's conclusions.

Worked mechanism: feedback latency and shared processors

Consider an original teaching scenario. A scientist needs eight sequential iterations, each depending on the preceding result. Every iteration requires two minutes of human inspection, five seconds of computation, and either a 30-minute batch wait or one second of interactive overhead.

The batch workflow takes 8 × (1800 + 120 + 5) = 15400 seconds, about 4 hours 17 minutes. The interactive workflow takes 8 × (1 + 120 + 5) = 1008 seconds, or 16 minutes 48 seconds. The processor performs only 40 seconds of computation in either case. Removing delays between dependent steps can matter far more than accelerating those 40 seconds. This is a deliberately chosen model, not a measured claim about an historical installation.

Time-sharing can make such interaction economical by letting multiple users share a processor. In a simple round-robin model, five continuously runnable tasks each receive 20 milliseconds of computation followed by one millisecond of switching overhead. A complete cycle takes 5 × 21 = 105 milliseconds. Useful processor time is 100/105, about 95.2%.

The fifth task begins at 84 milliseconds if the first begins at time zero. That delay is not zero, and more runnable tasks increase it. Terminal users who are thinking or waiting for input are not necessarily runnable tasks; using the number of logged-in users as the queue length can be badly misleading. Real operating systems add priorities, I/O waits, interrupts, and memory constraints. Time-sharing does not guarantee that every user feels alone on the machine.

From an office to a research community

IPTO began in 1962. Licklider's initial leadership supported interactive-computing research; Ivan Sutherland followed him, and Robert Taylor became director in 1966. Taylor's and Lawrence Roberts's networking work should not be collapsed into Licklider's own directorship. DARPA office history, Kleinrock's participant history

Licklider and Taylor's 1968 paper, The Computer as a Communication Device, discusses computers as a medium for collaborative work and communities. Read it as a proposal with social and technical assumptions. Describing a desirable network is different from specifying packet formats, routing, reliable transport, or operational governance. ARPANET and the later Internet are related developments, not interchangeable names for an already complete design. 1968 paper reprinted in the DEC memorial volume

The organizational lesson is that research infrastructure includes people, shared tools, and durable support as well as processors. This is an interpretation of the programme, not evidence that funding alone produced its results or that one person authored all later achievements. Military sponsorship, university research, and eventual civilian uses can coexist in the same institutional history.

How to evaluate an interactive system

Suppose an interface reduces response latency from two seconds to 200 milliseconds. A useful evaluation measures task completion, error rates, recovery from misunderstandings, and whether users notice uncertainty. Measuring only clicks or processor utilization could reward a system that produces more activity but poorer conclusions.

The right comparison depends on the workload. Independent overnight jobs may benefit little from an interactive display. A sequence of dependent design decisions may benefit greatly. The useful distinction is the structure of the task and its feedback requirements, not an absolute ranking of batch and interactive computing.

Exercises

  1. Scheduling trace. Eight always-runnable tasks get ten milliseconds each, followed by one millisecond of overhead. Task 1 starts at time zero. When does task 8 first start, how long is a full round, and what fraction of processor time is useful? Explain what changes if only two tasks are runnable.
  2. Speedup ceiling. In a workflow, preparation occupies 85% of elapsed time and all other work occupies 15%. Suppose a tool makes preparation ten times faster while leaving the rest unchanged. Calculate total speedup and its limit if preparation became instantaneous. Explain why Licklider's own estimate does not establish these gains for every user.
  3. Design an evidence test. A product team claims its interactive dashboard improves scientific reasoning because users finish plots faster. Propose a comparison that could distinguish reduced clerical effort from improved inference. Include one confound and one observation that would weaken the claim.
Compare your reasoning with the worked solutions
  1. Each slot lasts 11 milliseconds. Task 8 starts after seven slots, at 77 milliseconds. Eight slots take 88 milliseconds, with 80 milliseconds of useful work, so utilization is 80/88 = 10/11, about 90.9%. With two runnable tasks the cycle is 22 milliseconds and the second starts at 11 milliseconds. The fraction stays the same in this idealized saturated model, but waiting decreases. Blocked tasks should not consume empty ten-millisecond quanta in this model.
  2. Normalize the original duration to 1. The new duration is 0.85/10 + 0.15 = 0.235, giving speedup 1/0.235 ≈ 4.255. Even instantaneous preparation leaves duration 0.15, so the ceiling is about 6.667 times. These numbers assume unchanged task quality and no new work or overhead. Licklider's self-observation does not establish the fractions in another population, and faster preparation may change which problems people attempt.
  3. Randomly assign comparable tasks and counterbalance interface order to reduce learning effects. Measure plotting time separately from accuracy of inferred relationships, detection of misleading scales, and justification of conclusions. Prior statistical experience is a confound if experienced users disproportionately receive the new interface. Faster plots with more unjustified causal claims would weaken the claim about reasoning even while supporting a claim about clerical speed. An informative study reports both outcomes rather than treating speed as a proxy for understanding.

Sources and further study