The difference between generalist and specialist is yet another distinction that's not what it seems to be.

 

— 〈  1  〉—

``When I feel pessimistic, I see the divergence between research lines and the impressive number of problems that are still open. When I feel optimistic, I see their remarkable coherence; and I dream we might be near great unifying solution to problems, like in the past: with all the machinery ready, trying a number of similar equations.'' — ROV04

A joke is that the generalist knows less and less about more and more. Ends up knowing almost nothing about almost everything.

The joke continues: the specialist supposedly knows more and more about less and less. Ends up knowing almost everything about almost nothing.

And this joke seems to make sense. Even to some generalists and specialists.

The perspective here is that of a budget. Viewpoint is worth 80 IQ points (KAY93). Which also means: Viewpoint is worth plus — or minus — 80 IQ points. Seems to me to be more often minus and not plus.

Scarcity thinking is sometimes appropriate: here it is not.

The real distinction is speed.

 

— 〈  2  〉—

Why is that?

Memory is active and reconstructive.

Comprehension, understanding, is coherence. So: compression (CHA03, DEU11, HAR54, HAR65).

That is how a small program can describe and generate enormous knowledge (WOL02, CHA03). Just has to be the correct program.

Our memory stores not so much stores bits of information than realizes programs which taking inputs from the environment and generate appropriate actions giving pragmatic or embodied knowledge (PRI71); and the question about how many bits the brain stores as memory as knowledge is misleading.

The more a brain understands, comprehends, the more coherent its knowledge. Each part related to and allowing inference to other parts. The more it can generate, the less it will store (FOE65, PRI66).

Our behavior has much tacit knowledge (JAM07).

Memory is mostly reconstruction — very little storage (BAR32). It is not despite the fact that memory is reconstructive that we can later remember some of what we had apparently forgotten and could not earlier remember (BAL13), but because of it. Different environment, different input, different output. We are mobile and active.

There is sometimes overwriting of the gluing function ... the program that reconstructs and classifies ... by experience that contradicts our world model ... by later experience. Therefore it's not really time since remembering or learning but later activity that makes us forget if we do forget.

For example: sleeping halts forgetting, but not because of anything special about sleeping: merely inactivity (KOE40). Being inactive but not sleeping leaves intact the same performance in remembering. Subsequent activity lowers recall performance (BRO71).

And that allows a content hashing relations gradient and therefore surprisingly fast mental search and recall based on content. Rather than where the content is stored. We can often rapidly say, despite the slow chemical operation of much of the brain, if we know this or that or know it not (KAN88).

That is also why the individual who controls what work they do each day and how much work they do and exactly where they do it will often have much, much better memory performance than somebody who cannot self select like that. Somebody working all day in place where they are requires to work rather than working at their leisure, will have much worse memory, in practice, than somebody else working in comfort. One reason why somebody who must work for a living is at a severe disadvantage in science. It's not primarily because they are tired by their work; significant irrelevant experience and lack of control over environmental input is a major source of decline in memory. Without sufficient memory they cannot absorb as much information as quickly.

If you know much about only one material, you will yet not have the general toolbox nor breadth of knowledge to perform much inference, prediction, compression, because there simply isn't enough coherence. Random details cannot be compressed. They must be stored. You quickly run out of memory. And that's your limit.

Rather if you get sparse but coherent information is compresses faster than you take in further information and additional information collapses into true predictive models of behavior faster and faster; there is no limit. Abundance. However compression requires processing to decompress. You must temporarily generate a record to use in any special case. And this delay occurs in almost every case. They're all special.

 

— 〈  3  〉—

The generalist can know more about more — but slowly. Each specialist knows less about less — but quickly.

More a system knows the less it needs to store as a record (CHA03, DEU11). But reconstruction requires processing. That requires time and that is the real trade off.

A specialist has limited but deep knowledge about something and this is not sufficient to be compressed or comprehended. It must be stored. Well: it can be rapidly recalled.

The generalist may know several general laws and can get staggeringly precise predictions. But must do a lot of work making the special case prediction each and every time; the probability that the specialist has stored an equally precise solution to an arbitrary problem is much, much less.

However the specialist, if he knows his answer, and while it is most probably of lower precision, will present it much more quickly.

A mathematician will generate all formulae for a subject on the spot in a minute; he presumably knows several general procedures and different subject matters to produce any formulae of arbitrary length and accuracy; the specialist, if he knows the answer, and he probably does, so long as the question is in his field, will tell you the answer in five seconds. Just most cases he does not know the answer. And cannot anymore learn the answer without enormous further effort beyond his current means. Moreover the precision with which he remembers solutions to problems is limited: he must store most answers. But for the same reason he will give them quickly.

The generalist approach will be to make a predictive model. Afterward test the model with gathered information. Then gather new information: truth dips into facts and goes into theory and then dips into facts and meanwhile the facts are interpreted in light of truth (JAM07, HAR54, HAR65, FOD68).

Making such models requires far more up front cost; and this sunk costs generates abundance forever after.

Without sufficient cases there is no coherence along which true inferences can pass

Suppose there's a competition.

Given an extra half an hour, the real generalist will always win. There's no competition. If there is not the extra half hour and leisure, the generalist will almost always lose. There's no competition.

Also: the generalist exists only above a sufficient threshold.

Below that threshold you have neither a generalist nor a specialist. No: there you have a student.

The student knows less and less about more and more — until he knows enough — and then suddenly he knows more and more — about more and more.

Somewhere there are rocks. Elsewhere there are thresholds.

The question is scale. It's not precise learning budget that is significant, but thresholds in your learning budget. You want to be above thresholds; and if not that, then a specialist.

Engineers work far more quickly than scientists, but get stuck sooner, or spend more time on hard problems. And less time on easier problems.

Specialists in the sciences work far more quickly than the generalists, but get stuck sooner, or spend more time on hard problems. And less time on easier problems.

Best of all if they work together as single units. (Then work actually gets done.)

 

— 〈  4  〉—

Let's talk about scale.

Why not?

Let's consider some dilemmas. And consider how we can consider them. Because viewpoint is worth 80 IQ points (KAY93).

Imagine two games where, when two individual meet, they do something. So in the first game they give each other cake. In the second game they punch each other. Else they're all randomly walking.

As you get more participating actors the difference in outcome in the two games becomes greater. One has more persons eating more cake; more happiness. The other has more persons getting hurt; more unhappiness. But keep adding persons playing each game and they become more and more the same. Despite the major difference in game dynamics.

Everybody is piling up and probability of having to feed or punch somebody who has to feed or punch somebody and get fed or punched first goes up for an increasing number of players faster than number of players grows. Which paradoxically results in very large games almost nobody being fed and almost nobody being punched; just millions standing around waiting. The same.

The change in outcome contributed by change of scale, or buffering of outcomes, or ..., may be greater than change in outcome contributed by change of all other dynamical variables or even transformation and change of the system dynamics.

 

— 〈  5  〉—

This is the general rule of thresholds. Considering the mind and behavior of a single individual we get the rules of the mind. The difference between generalist and specialist and student.

Considering the minds and behavior of many individuals we get the rules of society.

Most questions of budget versus abundance and strategy are fundamentally questions of scale.

©2018 tibra.
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License  . . .   . . .   . . .    Text and images: @tibra. @communicate on minds.com