Innovation Roundup (( by Hamid Ghanbari, MD ))
A newsletter about innovation, technology and empathy in medicine (8/19/19)
Hi,
I wanted to thank all of you who have been reading my labor of love that is this newsletter. The response has been tremendous and I hope to continue publishing it. If you enjoy reading this newsletter, please share it with someone else who might enjoy reading it.
I am also going to add a section on what I have been learning. This section is mostly notes from books and articles that pertain to statistical thinking and machine learning. I hope that you would find this helpful as I document my learning journey.
Best,
Hamid
What I learned this week
In Ts’ui Pen’s work, all the possible solutions occur, each one being the point of departure for other bifurcations. Sometimes the pathways of of the labyrinth converge. For example, you come to this house; but in other possible pasts you are my enemy; in others my friend.
The Garden of forking path is a picture incomplete, but yet not false, of the universe such as Ts’ui Pen conceived it to be. He believed in infinite series of times, in a drizzly growing, ever spreading network of diverging, converging and parallel times. This web if time- the strand of which approach one another, bifurcate, intersect or ignore each other through the centuries- embraces every possibility. We do not exist in most of them. In some you exist and not I, while in others I do not. and you do not, and in yet others both of us exist.
Garden of Forking Paths- Louis Borges
The first time I read this short story I did not fully grasp its basic meaning. For those of you who read Borges, this is not an unusual experience. There is a certain ambiguity in his stories that allow for multiple interpretations. I think that is part of the magic of Borges and why I keep going back to his stories.
I recently came across this story again when it was mentioned in a paper by Gelman and in McElreath’s excellent book. In this story the main character finds a book in which all events can happen at the same time. It is a strange and fantastic short story. It also seems to be the perfect metaphor for understanding Bayesian inference.
Bayesian inference and Garden of Forking Paths:
To make good inference about what actually happened, it helps to consider everything that could have happened. In that sense, Bayesian analysis is a garden of forking data in which alternative sequence of events are cultivated. As we learn about what did happen, some of these alternatives are ruled out. What remains is only what is logically consistent with our knowledge.
We cannot travel down every path in the garden of forking data. If we could, we would know what information to ignore and which to pay attention to. This would make Bayesian analysis unnecessary.
Probability theory is essentially counting the way things can happen. We count the ways through the garden of forking data that could result in the observed outcome in our small world of observed data.
Our goal is to figure out what is most plausible given some evidence about the content of the data.
We may have prior information about the relative plausibility of each path because we may have knowledge of how the the garden was constructed.
Plausibility is also probability - they are non negative (zero or positive) real numbers that sum to one.
Take the following logical argument:
If A is true, then B becomes more plausible
B is true therefore,
A becomes more plausible.
In our reasoning we depend on prior information to help us evaluate the plausibility of a new problem. In our daily lives, this reasoning process goes on unconsciously, almost instantaneously. We call it "common sense ".
As the new becomes available (assuming that the new data is independent from previous data), the number of logically possible ways for a conjecture to produce all the data up to that point can be computed. We just multiply the prior count by the new count.
Let me know your thoughts.
HG
References:
Probability, Random Variables and Stochastic Processes
Probability Theory: The Logic of Science
Department of Empathy
How Life Sciences Actually Work- This is a very interesting piece by the excellent Alexey Guzey. I like the structure of the essay which follows an observation that leads to a naive conclusion. Alexey does a nice of job of mapping reality the best he can using interviews from the people in the front line of science. Here is an excerpt:
“NIH’s biases make it very hard to fund methodological research”
Observation: NIH doesn’t like to fund purely methodological studies (e.g. development of better software)
Naive conclusion: it’s impossible to get funded for methods development by NIH
Reality: you can dress up methods development for NIH, e.g. by providing a concrete biological goal for which insufficient methods are the bottleneck and show NIH that you will be able to achieve concrete progress on something that matters to them using your better methods
Again, yes, doing this introduces obvious frictions and inefficiencies and skews what scientists work on towards things that are easy to dress up for NIH, rather than things that they believe are most important. But the amount of frictions and inefficiencies is way less than we would naively expect.
I Found the Outer Limits of My Pro-choice Beliefs- An American OB-GYN writes about her experience with Israel’s abortion law
Usually the discussion of abortion is longer and more wandering. At first, the patient may feel unsure of where she stands. As we talk, she may return to the subject and ask more questions. Conducting this conversation requires as much surgical skill as operating on a pregnant uterus. There is no right answer, only one that is less wrong for each patient. This is an almost impossible conversation—and one that doctors like me must have every day.
Open-sourced blueprints for civilization
Using wikis and digital fabrication tools, TED Fellow Marcin Jakubowski is open-sourcing the blueprints for 50 farm machines, allowing anyone to build their own tractor or harvester from scratch. And that's only the first step in a project to write an instruction set for an entire self-sustaining village (starting cost: $10,000).
The whole point I want to make is that ambiguity is the essence of meaning. Here I mean “meaning” in the sense of the profound, the artistically deep, the spiritually relevant – not in the crude sense of “pointing” that I previously examined. The story of the Garden of Eden, The Shining, your Hamlet and Bend Sinister, the evidence of their profundity (that is to say, ambiguity) lies in the multitude of interpretations they support. To me the empty tomb at the end of the Gospel of Mark is more ambiguous, hence more artistically and spiritually interesting, than the fan theories worked out in the other gospels, although these are also beautiful and ambiguous in their own ways.
Secular Cycles- This is why I read everything Scott Alexander writes. I am always amazed about his critical appraisal of topics that are seemingly outside of his area of expertise (psychiatry). Don’t be fooled by data and fancy graphs!
Seeing patterns in random noise is one of the basic human failure modes. Secular Cycles is so prima facie crackpottish that it should require mountains of data before we even start wondering if it might be true… the chapter above on the Roman Principate included 25 named figures and graphs, plus countless more informal presentations of data series, from “real wages of agricultural laborers in Roman Egypt during the second century” to “mean annual real land rents for wheat fields in artabas per aroura, 27 BC to 268 CE” to “imperial handouts per reign-year” to “importation of African red slip ware into the Albegna Valley of Etruria, 100 – 600”. This book understands the burden of proof it is under, and does everything it can to meet it. Still, we should be skeptical. How many degrees of freedom do T&N have, and is it enough to undermine their case?
Department of Innovation
Innovation skill set comes through learning—first understanding a given skill, then practicing it, experimenting, and ultimately gaining confidence in one’s capacity to create. Innovative entrepreneurs in our study acquired and honed their innovation skills
Associating
Questioning
Observing
Experimenting
Networking
AI Algorithms Need FDA-Style Drug Trials
With each click, the algorithms learn to personalize the feed to their users’ tastes, thereby reaping profits for their owners. But the designers made a simple mistake: They assumed that human tastes are fixed. In reality, algorithms applied to malleable humans can have drastically different and pernicious side effects on a global scale. They modify our tastes to make us ever more predictable, edge us toward extremes, and ultimately erode civility and trust in society. It is time we stop blithely allowing this and create the digital equivalent of drug trials.
Prescription label for machine learning models
Art of the week
Adolph Von Menzel (1815–1905): In a Railway Carriage (After a Night's Journey), Art Institute of Chicago
Department of Artificial Intelligence
The “inconvenient truth” about AI in healthcare
However, “the inconvenient truth” is that at present the algorithms that feature prominently in research literature are in fact not, for the most part, executable at the frontlines of clinical practice. This is for two reasons: first, these AI innovations by themselves do not re-engineer the incentives that support existing ways of working. Second, most healthcare organizations lack the data infrastructure required to collect the data needed to optimally train algorithms to (a) “fit” the local population and/or the local practice patterns, a requirement prior to deployment that is rarely highlighted by current AI publications, and (b) interrogate them for bias to guarantee that the algorithms perform consistently across patient cohorts, especially those who may not have been adequately represented in the training cohort.
The test of an intelligent machine is whether it's intelligent enough not to reveal its intelligence.
In this multi-institutional diagnostic study of 3214 patients, a machine learning model was designed to achieve an accurate patient-specific risk score for pulmonary embolism diagnosis. The model was successfully evaluated in both multi-institutional inpatient and outpatient settings. The best-performing model achieved an AUROC performance of predicting a positive PE study of 0.90 (95% CI, 0.87-0.91) on intrainstitutional holdout data with an AUROC of 0.71 (95% CI, 0.69-0.72) on an external data set from Duke University Medical Center; superior AUROC performance and cross-institutional generalization of the model of 0.81 (95% CI, 0.77-0.87) and 0.81 (95% CI, 0.73-0.82), respectively, were noted on holdout outpatient populations from both intra-institutional and extra-institutional data.
Thank you again for reading my newsletter.
If you enjoy reading this newsletter, please share it with someone else who might enjoy reading it.
Talk Soon,
Hamid