The impossible dream

There are two impossibly high standards kicking around the core of the sciences that really need to be changed: clinical trials and computational models (Turing machines).

George Bernard Shaw, "No question is so difficult to answer as
that to which the answer is obvious."

In the same way that a company or country shouldn't have rules or laws on the books that they don't plan to enforce, my pragmatic side says that science shouldn't have standards that cannot possibly be met. The illusion of a standard damages practice. "The perfect is enemy of the good."

On clinical trials, "everyone knows" that the "gold standard" is a double-blinded, placebo-controlled trial in humans of representative ages and sexes. Aside from all the other problems with these, one show stopper is simply that this "standard" fails to scale up appropriately. I don't know the exact number, but probably the US could afford to do something under 10,000 of these, maybe under 1,000 of these at any one time - at probably over a million dollars apiece. And there aren't enough subjects. And we really don't want to test on pregnant women. On the other hand, and again I don't have the numbers, there are probably easily 1000 times that many new foods, products, cosmetics that hit the market in a given year. This gap will only widen. And, there are very few old people who only take one drug. In fact, by picking such persons, the sample has already become unrepresentative.
And even if it didn't widen, there is little in the way of post-approval testing of the drugs in vivo, in social settings as actually used, so there are all these changed-context issues.

In computation and communication, the core model is the Turing Machine - which models all computation on the basis of a theoretical paper tape that is infinitely long, never decays, doesn't have noise, and we have infinite time to do the calculation. In all realistic situations however, we have finite and even short time, we have finite space, there is substantial noise, and the data is flawed. The right model for computation and communication using serial symbol strings should make those limits explicit. Then again we have the problem that the meaning of data depends on context, which was left out by Turing. Probably the minimal correct model would at least be the level of image-processing primitives, where 2-D images have redundant parallel information and much higher bandwidth (One picture is worth 1000 words.). The proofs of these being equivalent all assume infinite time and space and energy, which don't hold in the world I live in. Given finite and small time, space, and energy we have to ramp up.

By ramping up, I mean two things. First, communcation has to be massively parallel to even get near the bandwidth necessary to communicate a message before the message has become irrelevant, along with context. Second, almost everything has to be either in shorthand or totally unspoken (implicit) or, again , the mssage won't fit. The problems of people having differing implicit worlds is well known.

This is not an occasional problem - this is the shape of the keyhole through which our entire understanding of the universe must be pushed. And like the impact that shoving waves or light through a slot has on the resulting output pattern, this is NOT a neutal impact. In mathematical terms, we're convolving the signal with the fourier transform of the keyhole, or some such thing.

In practical terms, no one can write a 135 page paper any more. No one will read it.
The window is getting even smaller, not larger, as time goes on. Probably the largest paper
that scientists will read is down to 12 pages, and anything much longer than 6 is at risk of being ignored.

Not just scientists are hitting this distortion-creating limit. TV messages have to fit in 8 minutes or less. Student attention span is well under 55 minutes. Blogs (unlike this one) should really have thoughts that fit on less than one screenful, which can be read in 4 minutes or less.

Someone needs to compute the system transfer function of this keyhole effect, and hypothesize what thoughts simply cannot be discussed any more because of it. Anyone notice a loss of ability to discuss long-range planning or complex subjects with policy makers?

Everyone sees all these problems but mistakes them for local issues, local symptoms, and doesn't see the fact that they are so deeply embedded in our system of "science" and social discourse that the output of that system has become throttled, distorted, and missing entire octaves of harmonics.

I thank John Gall, author of Systemantics - How things Fail, for the insights into finding system problems in what look like collections of misattributed local problems.