Of course, both scientific approaches of history and myth are the work of fabulists. eg Jung, Campbell. The point is to examine the myth and then history as the source of causal illusions.
“The myth is the prototypal, fundamental, integrative mind tool … to integrate a variety of events in a temporal and causal framework.” Merlin Donald
That's folk science, what Donald is describing (he admit this in Origins of the Modern Mind).
Remember that the causal framework must be evaded to reach scientific correlations, where multiple contradictions can lead to knowledge. Myth and history were addictive hiccups that trapped humans in way simplistic explanations.
We evade this "plain English" silliness, like economics, or go bust.
Note that in orthodox microeconomic theory, price is equal to the marginal value of the last exchanged unit. To use the above example of food:
> What's the value of food? If you have none you die, so the value is quit of high, but the price is much lower than that because there are many competing suppliers.
The first calories of the day, the ones that prevent you from dying, have a very high subjective value - but you pay them at the value of the 3000th calorie of the day, the extra drop of ketchup on your fries, which has a very little value.
And thus of course average value x volume is very different from (marginal value of last unit) x volume.
If they had absolute perfect performance at zero cost, you would not need a radiologist.
The current "workflow" is primary care physician (or specialist) -> radiology tech that actually does the measurement thing -> radiologist for interpretation/diagnosis -> primary care physician (or specialist) for treatment.
If you have perfect diagnosis, it could be primary care physician (or specialist) -> radiology tech -> ML model for interpretation -> primary care physician (or specialist.
If we're talking utopian visions, we can do better than dreaming of transforming unstructured data into actionable business insights. Let's talk about what is meaningfully possible: Who assumes legal liability? The ML vendor?
PCPs don't have the training and aren't paid enough for that exposure.
To understand why, you would really need to take a good read of the average PCP's malpractice policy.
The policy for a specialist would be even more strict.
You would need to change insurance policies before your workflow was even possible from a liability perspective.
Basically, the insurer wants, "a throat to choke", so to speak. Handing up a model to them isn't going to cut it anymore than handing up Hitachi's awesome new whiz-bang proton therapy machine would. They want their pound of flesh.
Let’s suppose I go to the doctor and get tested for HIV. There isn’t a specialist staring at my blood through a microscope looking for HIV viruses, they put my blood in a machine and the machine tells them, positive or negative. There is a false positive rate and a false negative rate for the test. There’s no fundamental reason you couldn’t put a CT scan into a machine the same way.
Pretty much everything has false positives and false negatives. Everything can be reduced to this.
Human radiologists have them. They can miss things: false negative. They can misdiagnose things: false positive.
Interviews have them. A person can do well, be hired and turn out to be bad employee: false positive. A person who would have been a good employee can do badly due to situational factors and not get hired: false negative.
The justice system has them. An innocent person can be judged guilty: false positive. A guilty person can be judged innocent: false negative.
All policy decisions are about balancing out the false negatives against the false positives.
Medical practice is generally obsessed with stamping out false negatives: sucks to be you if you're the doctor who straight up missed something. False positives are avoided as much as possible by defensive wording that avoids outright affirming things. You never say the patient has the disease, you merely suggest that this finding could mean that the patient has the disease.
Hiring is expensive and firing even more so depending on jurisdiction, so corporations want to minimize false positives as much as humanly possible. If they ever hire anyone, they want to be sure it's absolutely the right person for them. They don't really care that they might miss out on good people.
There are all sorts of political groups trying to tip the balance of justice in favor of false negatives or false positivies. Some would rather see guilty go free than watch a single innocent be punished by mistake. Others don't care about innocents at all. I could cite some but it'd no doubt lead to controversy.
In that scenario, the "throat to choke" would be the primary care physician. We won't think of it as an "ML radiologist", just as getting some kind of physical test done and bringing it to the doctor for interpretation.
If you're getting a blood test, the pipeline might be primary care physician -> lab with a nurse to draw blood and machines to measure blood stuff -> primary care physician to interpret the test results. There is no blood-test-ologist (hematologist?) step, unlike radiology.
Anyway, "there's going to be radiologists around for insurance reasons only but they don't bring anything else to patient care" is a very different proposition from "there's going to be radiologists around for insurance reasons _and_ because the job is mostly talking to patients and fellow clinicians".
They didn’t say there wouldn’t need to be change related to insurance. They obviously mean that, change included, a perfect model would move to their described workflow (or something similar).
HackerNews is often too quick to reply with a “well actually” that they miss the overall point.
This paper proposes that idiosyncratic firm-level shocks can explain an important
part of aggregate movements and provide a microfoundation for aggregate shocks. Ex-
isting research has focused on using aggregate shocks to explain business cycles, argu-
ing that individual firm shocks average out in the aggregate. I show that this argument
breaks down if the distribution of firm sizes is fat-tailed, as documented empirically.
The idiosyncratic movements of the largest 100 firms in the United States appear to
explain about one-third of variations in output growth. This “granular” hypothesis sug-
gests new directions for macroeconomic research, in particular that macroeconomic
questions can be clarified by looking at the behavior of large firms. This paper’s ideas
and analytical results may also be useful for thinking about the fluctuations of other
economic aggregates, such as exports or the trade balance.
Isn’t “making acceptable tradeoffs” (optimization) while “meeting requirements & constraints” (under constraints) simply optimization under constraints? I fail to see how this is about sufficient _but not optimal_ solutions.
The dictionary defintion of optimal is just 'best'; the connotation that 'optimal' means not perfect, but satisfying multiple constraint, is engineering jargon. In fact, with 'optimal', people will often mean 'Pareto optimal' (a design where there is no change that would improve the solution on all dimensions/constraints).
Pareto optimality also implies that multiple designs may exist in a 'draw', where design 1 is better on dimension A, but worse on dimension B, and design 2 flips those. These designs are on a pareto front. Making the trade-offs explicit helps readers of the design document who come along later to choose a different trade-off when it later turns out that the relative importance of the dimensions involved changed.
I generally read "optimal" to mean "optimal according to some objective function". In the case of a technical design, the objective is not something you can readily formalize, because there are so many considerations.
For example, suppose you are designing a network service to meet some use case. Yes, it needs to meet the use case, but there are other things you care about like simplicity of the code, latency, cost to operate the service, and others.
For any design you propose, there might exist an alternate design that is strictly better by some criteria-maybe it works exactly the same but has much better latency characteristics. If an oracle gave you this design, you might agree it was more optimal and choose to implement it, but I don't think it means you failed if you come up with a (sub-optimal) design that solves the problem within the constraint.
