Interpreting science-technology pair as need-solution pair

[hyunjimoon]

1. Macro economic model of two stocks with the knowledge of science and technology are as below; causal model in the left and stock and flow diagram in the right (labor of scientist and technologist can be stock as well). I wonder whether the reference mode of science-technology would be closer to that of prey-predator or to that of actual-measured room temperature dynamics. Some analysis on feedback loops and delay:

• science-technology is 3R0B when all six policy parameters are positive. However, a model without any balancing loop is less realistic. It may be more practical to allow signs of $\phi, \psi$ to shift between negative (burden of knowldge, fishing out) and positive (recombination) to curb the exponential growth. Exponential growth is not supported empirical evidence (??). prey-predator is 2R3B (or 2B2R when birth and death rates are not influenced by the population is constant e.g. enough food and space in the environment. 3B2R means when there is more population they make more babies and die more e.g. enough food but limited space); actual-measured temperature is 0R2B. Among the eight system architecture, science-technology seems to resemble power dynamics / success to successful most.

• Material and mental delay have different structure in that input and output are usually modeled separately in the former. Tweaking the proportion of each six parameters would have equivalent effect of tweaking adjustment time and measurement delay for room temperature model and tweaking birth and death fraction in prey and predator model.

Process noise can be injected to represent unknown/unrepresented uncertainty. I still don't have clear understanding on how process noise in the generator (which we may be able find corresponding hierarchical generator e.g. hyper parameter sampled from half normal distribution) affects estimation of hierarchical model.

4. I wonder whether redrawing below (reference room temperature) on the axis of adjustment fraction (inverse of parameter) may make reference modes more cluster-able. It would be a hyperbolic space

[tomfid]

Regarding #2 (the plot from BATS), I've been wondering whether the modes here are really distinct. OSCCT and GR2DB probably differ in damping. But is SSHGR even more damped so it can't overshoot, or is it just that the time scale gets longer as the delay gets longer, so the peak in GR2DB ends too late in the model time horizon to be observed?

[hyunjimoon]

So it all depends on timestep and time horizon as they are measured by correlation -- let me link the code in a min!

[hyunjimoon]

https://github.com/hyunjimoon/DataInDM/blob/9adbd4eb93a51f7b951ae64a12d9009128562c46/Code_Commit/4_Auto/GeomPrior/aria/test_scripts/ists_test.py#L59

[hyunjimoon]

Chat with Yaman which may help us understanding:

Angie:

1. ists_instance.smooth(y) -> ists_instance.smooth(y, alpha): I am not sure which value I should give to alpha.
2. ists.covariance(y) -> ists.covariance(y, lag):adding lag for covariance is needed.

Yaman:
Thank you for the corrections.

• Alpha you mention is the alpha value used in standard expon smoothing. It is up to the user, typically between 0.01 and 0.20. Alpha is equivalent to 1/AT used in delay formulations (smoothing). (In time-continuous models, it is like (1/AT)*DT). Small alpha means strong smoothing of noisy data. Large alpha means
  very little smoothing (the output nearly mimicking the noisy input
  data).

You are right, autocovariance lag is a required user input (ists would compute the autocorrelations up to that lag).

[hyunjimoon]

vensim files for three models:
twostock_uc.zip

[tomfid]

Regarding the tech model, I like it. We had some similar components in models for DOE Office of Science a few years ago. This is a nice compact representation.

I don't think the temp control model is a good analog though. They're topologically similar, but the gains and nonlinearities are different. The temp control loops are negative, while the tech model is dominated by positive feedback.

[hyunjimoon]

Oh I see, this model stems from a pset for a class I'm taking below. Scott seems to be very interested in the tech2sci feedback.
15_357_Pset1_QA.pdf

[hyunjimoon]

Including the following paper which estimated tech2sci by collecting the data of science organization that couldn't use physical implementation/ experience due to catastrophe (I think , didn't read too carefully)
Baruffaldi22_Returns2PhysicalCapital4KnowledgeProduction.pdf

[tomfid]

It's possible that the "unexploited principles" loop is already implicit in the equation for dA. You could rewrite it as:

dA/dt = A^a * B^b * (B/A)^c * L^k
phi = a-c
lambda = b+c

I think the main implication is that 'a' (phi in the tech2 equation on the diagram) doesn't have a clean interpretation, as it mixes several effects.

[hyunjimoon]

'a' (phi in the tech2 equation on the diagram) doesn't have a clean interpretation, as it mixes several effects.

Might "mixed" be similar to "residual" from Solow's residual?

[tomfid]

In the sense that Solow's residual means "stuff other than capital and labor that we attribute to tech" you might say the same about the roles of A&B here.

Incidentally people usually think of the Solow tech as computers, software, etc. but it's also soft tech, like better market instruments or contracts, as well as human capital.

[hyunjimoon]

One of the most insightful comment from today's talk by Adam Jaffe was his comment that data is public good and is by definition non-rivorous and excludable but through data-basing efforts like I³ Tracker for Essential Open Innovation Datasets + spreadsheet, Adam is trying to push data from excludable to non-excludable.

[hyunjimoon]

synthesizing three diagrams (angie-tom's diagram + the last two from this thread) my conclusion is we need measurement 🦁beast!

---

@tomfid's tom under theory implem.mdl.zip contains one important question that needs to be answered: how does the entrepreneur know which direction to change an attribute in order to improve "quality". tom gave the example of sharper knife better than duller one as simple quality measure but argued in most cases quality measures are multidimensional and ambiguous. Good example is vensim - i have no idea whether target customer would want cleaner graphical interface with a few functions, or a complicated one with lots of functions. But this is a function of target customer which itself is a function of vision. vision comes from quality of imaginative understanding.

Focusing on measurement, ecosystem innovation policy makers should provide a good measurement infrastructure/guideline (analogy is for each entrepreneur's compass to work well, underlying magnetic field should be increasingly measured hence partially controlled).

Tom also shared his confusion on form and function, like @jeanbaptiste. Thinking of the knife example, form is function, but I don't think that's what you mean. Does this relate to product vs. process innovation, or ... ?

---