85% of any intellectual work is …

#lang/eng

Mechanical extension

My experience with modern and not so modern technology instilled in me a belief that Computer1, Internet, Google2 are complex systems3, that had been made by man for man, to help people outside the system. Observing my own behaviour in the above-mentioned systems, I noticed that it is of a mechanical nature, that gave way to the replacement of men by automation.

In pre-Google times, to find anyone’s service number I had to use the phone book, which before finding its way to a phone booth4 had to be compiled by somebody, edited, printed and delivered. In pre-Google times, to find an article that I could use in my publication I had to visit a library, ask a librarian for a certain publication and, if it wasn’t being used by somebody else at the time, read it and reference it later in my publication.

Nowadays there is no such nonsense: when I need to find the number of a restaurant, I ask Google5, which as an answer opens a map, provides the location of the place, phone number, website address, and plots the fastest way to get there. Working on publications for journals I don’t need libraries anymore6; Google Scholar knows almost everything about printed, pre-printed and published works7.

Computer, Internet, Google have become a sort of mechanical extension of mine, that obediently delivers the information needed for my work. The only thing that is required of me is to know the question, and the answer is usually several clicks away.

In the paradigm of mechanical extensions that used to exist, the complex systems created by men to help men outside the system were more of a help than something to be helped. With Google we were used to asking “What is the answer?” type of questions. But with AI the paradigm shifted fundamentally, giving ground to a new type of question: “What is the question?”

Cognitive extension

Since “the dawn of the machines8” the issue of extending the capabilities of an ordinary man has been raised multiple times. For one, Poincare was the first to speculate about the “What is the question” level of symbiosis. I follow suit and pose the very same question, but whereas Poincare only speculated about the future, I have experienced it firsthand, and can venture the opinion that “What is the question?” type of question moves symbiosis from the pure mechanical domain to the cognitive, truly cognitive.

In pre-AI times, to think about interacting with a computer as with a colleague that has competences similar to ours required more effort on the part of the man. With the rise of AI, the competences of the machine far surpass those of any colleague9. Hell, it surpasses the competences of any Nobel prize winner, living or dead. What used to be a mechanical extension10 evolved into a cognitive one11.

The graph that the reader enjoyed in the previous note12 is the result of me extending some cognitive capabilities to a machine. I asked the machine specific questions to help me clarify it, gave the machine three publications, asked it to read specific parts that contained the answer13 and to think how we can present the data in such a way that it will be perceived at a glance by the reader.

In just a couple of minutes it produced the required result14. This is the type of cognitive extension that researchers could only have dreamed about 5 years ago. I remember myself jealously looking at the Jarvis-Stark partnership and dreaming that some day …

Christmas has come early, so to speak, and in the light of the opened possibilities I wish to speculate on the average work of a researcher. How much time we need for different steps of research.

I am not the first one asking this question. In 1957 J.C.R. Licklider made a claim15 that there are not enough studies of the time-and-motion analysis of the mental work16 that the researcher participates in while doing technical and scientific work. Licklider decided to log his activities as a researcher for a whole spring and summer17 of the year 1957.

The analysis of scientific work helped Licklider to uncover the empirically proven conclusion that 85% of research work was spent not on thinking but on getting into position to think, which included the following:

  • Make a decision to learn specific information.
  • Looking for this information after the decision is made.
  • Reading and otherwise digesting material.
  • Hours on plotting graphs or instructing an assistant how to plot.

Licklider argues that after getting all the visuals, the answer is obvious, but to get there, to have something to think about, a scientist needs to spend hours and hours.

I do not have Licklider’s data to check his claim, but I have something much better: the chronometry of Lyubishev, which I can use to calculate the amount of pre-thinking work done by the Russian biologist18.

85% or not? That is the question

I cannot take Licklider’s words for granted; if anything, life taught me not to trust anybody, so I decided to scrutinize the figures that I have from Lyubishev. It turns out that there are four sources that are worth a second exploration19.

The diary from 1918-192220 — I don’t remember if it has any account of chronometry; my best recollection is that it is what it is – a diary, which it proved to be. Bergson, vitalism, Poincare, the non-euclidean geometry. No account of how much time he spent classifying bugs and how much on writing about them.

In the end I stopped at Lyubishev’s expanded version of the guide addressed to young researchers21, written in 1951. As an example for students to follow, he puts into this work the whole chronometry from one of his years, and this is gold for our analysis, that’s what we are after. The whole ledger book of time spent on that and that and that.

This allows us to check Licklider’s claim: how much time is spent on getting into position to think. Here is the data, organized, sorted and commented:

INFOGRAPHIC · A TIME-MOTION LEDGER HOW A RESEARCH YEAR IS ACTUALLY SPENT A. A. Lyubishev’s own logbook · 1,564 h of «category-1» work · teaching & admin set aside 01 — OF THE RESEARCH WORK ITSELF · ~1,250 H LICKLIDER 1957 · 85% PREP · 60% WRITING · 37% 3% POSITION TO THINK disputed — position, or thinking? THINK 60% if writing-up counts as thinking · 97% if it counts as output — Licklider’s 85% sits inside the band. 02 — WHERE THE HOURS WENT · ACTUAL HOURS Identifying species 207 Cutworm report (writing) 198 Reading the literature 174 Flea-beetle paper (writing) 173 Methodology write-up (writing) 108 Fieldwork: collect + sort 88 Other prep (biblio, stats…) 84 Off-plan systematics 81 Processing measurements 54 Observations & records 51 Original thinking + notes 36 The biggest single task — identifying species (207 h) — outweighs writing the very paper it feeds (173 h). PREPARATION DWARFS THE THINKING — BY A LOT. Pure ideation is a sliver: about 3% of the research year. The other 97% is getting into position to think — reading, collecting, identifying, plotting, and writing it all down. 60% prep · writing = thinking 97% prep · writing = output Licklider 85% · inside the band AI now collapses that 97%. What it can’t do is own the questions — that stays human. getting into position (prep) writing-up (contested) thinking (ideation) Источник: А. А. Любищев, «Руководство для начинающих научных работников» (Ульяновск, 1951) — годовой бюджет времени за 1948 г. Метрика — часы по категории-1, без преподавания и переписки.

Where Licklider, as far as I can judge, cut his work into pure clerical and cognitive, Lyubishev took a far more granular track of logging. He split his life into two categories: lectures, reviews, the secondary reading and scientific work. Both categories have the reading, the writing and the thinking intertwined creatively, and this is the hardest part. To compare Lyubishev’s numbers with Licklider’s I need to merge both parts of Lyubishev’s life in such a fashion that the data produces a trustworthy result. So I need to be honest with the reader; if anything, it is my interpretation of Lyubishev’s routine.

Reading, bibliography work, collecting in the field, mounting the specimens, working on measurement — this is all work that he had done during the year, but the largest single item of the year – determining the species, which is 207 hours, more than writing the paper the bugs were collected for. Every step he did before writing is preparation, getting into position to think, and the rough estimation: collected/identified/determined 207 hours of general field work and wrote the paper in 198 hours makes 60% of the research year.

Before we assume that Licklider was wrong, let’s look at the numbers from a slightly different angle. The paper, the cutworm report, took nearly 480 hours together (with every other activity). And here is the turn that I wish to emphasize: if we consider writing as getting into position to think22, then Lyubishev was getting into position to think 97% of his time, which puts Licklider’s 85% squarely into the band23.

This actually, tangibly proves that preparation dwarfs the synthesis, the pure ideation, meaning thinking as thinking in the sense that most of us perceive the process is a sliver of some three +- percent.

AI supported research and cognitive extension

Back to the point. To process 4 sources, even if I had read them previously24, I would need, I don’t know, maybe a day. Reading the numbers of the actual report another day or three, organizing the data and plotting charts, drawing infographics another couple of days. All in all it would demand a week of my time and a great chunk of cognitive resources.

Having Claude in the sidebar of my Obsidian lifted this burden. I sat myself down to write this paper at 4:46 AM; now it is 7:20 AM and I have almost finished. The paper is 2500+ words long. Instead of spending weeks analyzing the data and plotting charts, I asked Claude to read 4 specific papers and return to me with his findings25; while Claude was doing it, I continued writing/thinking. In 10 minutes it returned with the data I needed, and only one mistake that I didn’t fix and left for the reader to find intentionally26 27. I told Claude that the data was exactly what I was looking for and that now I wished to organize the findings in such a fashion that later I could present the reader with clean and comprehensible visuals as well as a short explanation; the machine once more set to work.

In another 15 minutes it returned with categorization and visuals, which completely satisfied me.

I steered AI in the best possible kind of way28. It wasn’t a “here are sources, fetch, attaboy” way of interaction. Every decision has been mine: “read the materials”, “present data in that and that way”. “Let’s create such and such visualization.” Claude never decided what mattered. And together we managed in 25 minutes instead of a week, not bad, aye?

An afterword

I would like to conclude this note with a phrase from Agent Smith: we are “only human” after all. A noisy, narrow-band thinking machine, with multiple deficiencies of the neural system29. The thinking machine that is usually trying to run multiple processes in a distraction-saturated environment. Whereas AI’s thinking is fast, ours is slow. It is impossible for a biological brain to adjust to the speed of a synthetic one. That is where the figure that 85% of research work30 is preparatory work comes into play.

Yes, we can’t adjust to the speed of the synthetic brain, but modern researchers can leverage what it has to offer to get into position to think faster.

BIO

🧠 theBrain mapping

ID: 202606120446 Source:: Friend:: Child:: Next:: The Great Enabler

Footnotes

  1. Capitalization here is intentional, to highlight the register ↩

  2. Not only as search engine but as an infrastructure of services ↩

  3. Complexity it this context, means all of the mentioned system above has more than two elements and each element influence in one way or another the system and themselves. ↩

  4. in Russia we had “Yellow pages” reference phone book. Thick as hell with mostly outdated phone numbers and addresses ↩

  5. In Russia it is mostly Yandex, because it understands local business much better. ↩

  6. Only when I need to cite specific PhD thesis, which are usually not in the open. To be honest in the past 10 years I needed not a single time to cite any PhD thesis in my work ↩

  7. Books, articles, pre-prints, etc. ↩

  8. I think the creation of Enigma by Alan Turing might be a good baseline ↩

  9. I love a mental experiment of Dario Amodei: Imagine, say, 50 million people, all of whom are much more capable than any Nobel Prize winner, statesman, or technologist. The analogy is not perfect, because these geniuses could have an extremely wide range of motivations and behavior, from completely pliant and obedient, to strange and alien in their motivations. But sticking with the analogy for now, suppose you were the national security advisor of a major state, responsible for assessing and responding to the situation. Imagine, further, that because AI systems can operate hundreds of times faster than humans, this “country” is operating with a time advantage relative to all other countries: for every cognitive action we can take, this country can take ten. What should you be worried about? ↩

  10. Go, fetch, return, find ↩

  11. Lets think together. Help me understand the concept. Draft the plan of the essay. ↩

  12. And the one he will enjoy in this note ↩

  13. It doesn’t mean that I cannot give it publication without looking into them. The machine will manage similarly outstanding ↩

  14. As I told a friend of mine “It one-shotted the task” ↩

  15. Licklider J. C. R. Man-Computer Symbiosis // IRE Transactions on Human Factors in Electronics. 1960. № 1 (HFE-1). C. 4–11. ↩

  16. ”Poor guy”. So little he knew about Lyubishev who since 1916 had been doing exactly what Licklider tried only in 1957 ↩

  17. As Lyubishev, btw, but the latter did this in 1910 first time. ↩

  18. Obviously, as I have my own logs, I can use them, but prefer to use a well documented account of a “made man”. ↩

  19. Because I forgot almost everything I’ve read. Or not everything, but most of it anyway. The first time my goal was the system, now I need the exact numbers that this system produced. Different goals – different reading. ↩

  20. Любищев А. Дневник А. А. Любищева за 1918–1922 гг. / А. Любищев, 2002. ↩

  21. Любищев А. А. Материалы в помощь начинающим научным работникам: учебное пособие к спецкурсу / А. А. Любищев, под ред. Л. Л. Каталимов, Ульяновск: УГПИ им. И. Н. Ульянова, 1991. ↩

  22. Because I for one can’t think in my head, I need to write. Notes, books, articles. This note – is thinking, or better, this note is a thinking process out of which only ONE idea is born. And I can’t do it solely in the head. Writing this note I was putting myself into position to think. The note in my case serves as a visual, that after completion provides actual material that I can think about. ↩

  23. The figures, for the curious: of roughly 1,250 hours of actual research in 1949 (teaching, administration and correspondence set aside), undisputed preparation comes to about 60%, the contested writing-up about 37%, and pure ideation about 3%. The “conservative” reading counts the writing as thinking and gives 60% preparation; the “maximalist” reading counts it as output and gives 97%. All of it taken from the 1949 budget printed in the 1951 «Guide». ↩

  24. This time the goal of reading was different ↩

  25. Mind this that I read everything myself, I know the contents of what Claude was reading, I didn’t outsource it, I delegated “lifting” part of the work. ↩

  26. It is minor and doesn’t poison the data ↩

  27. Can you find the mistake that I left? ↩

  28. Asking, sort of, “What is the question” type of questions ↩

  29. Compared to an AI ↩

  30. And judging by actual numbers 97% (or 60%, depends where you stand) ↩

Узел связи // Comm Station

TG: ОФЛАЙНBBS: ОФЛАЙН
Secure P2P Neural Link // Digital Garden v2.5