Many comparisons have been drawn between the movie industry and the pharmaceutical industry. Both of them are driven by hits that are difficult to predict beforehand and uncertainty and chance play a huge role in both industries.
Assessment of movies is always subjective. It is difficult to argue that Toy Story is more or less “important” than Titanic. This is not the case with the pharmaceutical industry. Most of humanity will agree that a cure for cancer is, in fact, more important than the cure for baldness or in-growing toe nails.
Bottom-up trial-and-error (or tinkering) will serendipitously solve arbitrary problems in the problem-space without regard to any prioritization that may be applied by us. We cannot ask the tinkering process to first give us a cure for cancer before giving us a cure for in-growing toe nails.
This means the strategies of massive diversification and tinkering that can be applied to other domains such as the movie industry and publishing need not be entirely applicable to the pharmaceutical industry. A priori knowledge of priorities should influence our strategy in some manner. I don’t quite know how, possibly through a mix of tinkering and top-down processes.
This is the point on which I disagree a little bit with Nassim Taleb on his FT piece: “Drug Research Needs Serendipity.” I do agree with all his points if the only goal of drug companies was to maximize return on investments (in which case a cure for baldness may be better than a cure for cancer), but I think drug companies and universities also have a higher goal in mind (curing important diseases) while trying to maximize ROI.
If top-down efforts have a non-zero chance of success, they need to be considered if the priorities for results are known beforehand.
To better understand this perspective, take the limit case where humanity is threatened with imminent extinction from a specific disease. Where would you allocate resources to find a cure against a deadline:
- Top-down process attacking the specific problem with a non-zero chance of success,
- Bottom-up tinkering in an infinitely large problem space, but also with non-zero chance of success of finding this specific solution,
- A mixture of both
Remember, you have finite resources. The answer is far from clear
They’re easy to learn and can easily have high impact and quickly shift the mood. Only a handful can go a long way.
I have an essay coming up that discusses intertemporal discounting. In it I show that the ideal future discount rate graph in highly uncertain “wild” environments will correspond to what I call “Arcsine Discounting.” Current studies show that we instinctively follow a different future discounting pattern (known as hyporbolic or quasi-hyperbolic discounting). Here is the essence of the essay: pay attention to the near-term and the far-term, but discount the mid-term pretty heavily. The last question you want to pay attention to is “Where do you see yourself 5 years from now?” Instead focus on the very near-term, day-to-day stuff, while keeping an eye much further out for those really high-impact events that have ridden the exploding conditional expected mean of fat-tailed processes.
The discount factor is the black curve below (so my chosen “arcsine” label is actually incorrect). The blue curve is more intuitive because it shows what we have to pay attention to in a world dominated by high peaks and fat tails.
More later.
There needs to be a word that describes people who are doing what may be called “university quality” work while not actually being in an university. “Independent scholar” is too pretentious, I would prefer something that is closer to describing what a PhD student does, but without connection to any university. Most of the candidate labels out there are trying too hard to be anti-establishment.
In spite of everything, I think labels are important to communicate with others and to load and compress entire concepts appropriately in ones brain. Labels are also important to develop and maintain a self-identity.
I think a particularly useful and high-impact heuristic to increase exposure to serendipity is to say “Yes” when you want to say “No.” (Caveats apply).
A great essay by Leonard Read: I, Pencil.
It walks us through the bottom-up process that goes into making a single lead pencil. I always marvel at how many things need to fall into place for even the most basic stuff to happen. A very entertaining read.
When I was at Microsoft it become clear that most projects these days are so complex that no one person can load all the relevant parameters into his or her head at any given time. It usually takes a team. Unfortunately, human communication is far from perfect, so a lot of the uncertainty gets shuffled into the communication channel. I see this as a potential source of kurtosis risk even for what may be a bottom-up process. I wonder if that may be because communication is often designed top-down, with information needing to be stuffed into pre-designed templates and brought up within company protocols. Does kurtosis risk gravitate to dimensions that are top-down?
One useful aspect of our brains being susceptible to hindsight bias and the narrative fallacy to smooth over the non-linearity of life is that it helps us get on with things. If we didn’t have the capability of smoothing out the big jumps, we’d be standing around awestruck at the technology and changes around us. Instead, we hop into enormous aircrafts without a passing thought to the number of things that had to fall into place before we could do that. And that’s probably a good thing.
They’re also great mechanisms to overcome negative Black Swans, of course. This has been discussed by Dan Gilbert and others, but I think nobody’s looked at how we’d react if we were constantly made aware of the positive Black Swans around us.
A stationary stochastic process is one that has a probability distribution that does not change with shifts in time (or space). Most of the math associated with stochastic processes requires the underlying process to be stationary for the math to work.
Unfortunately this is not necessarily true with socio-economic systems and there is not enough of a vocabulary to go with the notion of non-stationary processes. It is difficult to begin explaining the notion of stationarity every time I simply want to say that a recession in a modern environment need not resemble a recession in the last decade. The whole underlying process could have changed, along with the probability distribution. In an earlier post I talked about how systems built on top of one another will exacerbate the fat-tailed effects - that is nothing but a change in the distribution, a non-stationary process.
This applies to surprisingly homely notions which I will discuss later. A lot of problems can arise out of wrongly assuming a stationary process, even in down-to-earth topics.
An introduction to the Empirical Tripod and Epilogism since I’ll probably be returning to them in the future.
The Empirical Tripod is the methodology practiced by the empirical greeks that allows real-life decision making without relying on reasoning. The three legs are:
- Personal experience
- Observations recorded by others (minus the theory)
- Metabasis: cautiously going from the similar to the similar
Epilogism is the practice of going from the seen to the unseen. Remaining open to the idea that there may yet be a bigger blockbuster than the movie Titanic and a more deadly disease than the plague. One can learn from a data set by considering how limited it is.
Experience is a two-edged sword in the practice of epilogism. It does help seed the imagination, and an experienced operator can certainly imagine more scenarios than an inexperienced one, but the same experience can also make one overconfident and “overfit” the data, thus shutting out certain scenarios simply because they haven’t occurred before.
Epilogism is essential for true preparedness, but we are all limited by our imagination and experience. Any technique that helps move our minds into a state that can practice epilogism will be of tremendous help. I suspect thinking in terms of a “bizarro world” will help to a certain extent, but doing so with the right parameters is the challenge.
