fjuerfilis

I don't have quite the same critical perspective as the blogger, but I think there's a certain misguided attitude underlying the phenomena observed by Shor.

Yesterday or the day before I was listening to the radio and someone with a physics background was talking about something (I think quantum entanglement) and started asserting that physics has basically figured out almost everything. This is probably a somewhat unfair paraphrase, but not too unfair.

What irritated me about it was the assumption that, if most of your predictions are correct, your model is almost entirely correct, and just needs to be tweaked a bit. This is certainly true some of the time, but sometimes those little empirical cracks are what brings down a major paradigm, and leads to another one, one that has the same predictions as in 99% of the cases, but in the other 1% has totally different predictions with very different implications.

This carries over to grant funding, etc. in that the prevailing community often assumes that what they're doing is fine, and all that's left are these little empirical tweaks. That's certainly helpful some of the time, but it seems to dominate too much. Academics needs to leave more room for people to fail at high rates with good ideas, to increase those small percent of times they succeed wildly.

For one, it seems to assume the algorithm involved is somehow inferentially perfect, which is curious. A lot of this literature is circular, in that it assumes credit from citation patterns, which are flawed, which then taints the whole thing. For another thing, it seems to assume researchers aren't manipulating the perceptual processes being referenced in the article, which is obviously isn't true (that is, that they're not engaging in dishonest hijacking of the credit system).

I know I'm in the minority, but I object to the idea of responsibility actually, and prefer to think of change in a kind of "neurobehavioral engineering" or transformative justice sense. That might seem pathological or even psychopathic (which is ironic because I'm anything but); I just mean that I think responsibility (like randomness maybe, at least with reference to free will) is sort of a red herring. It's one of the reasons I'm interested in free will issues, because I think the notion of responsibility (outside of a very strict causal sense, as in "this geological formation is responsible for this waterfall") is misguided, and also shifts focus away from change efforts, toward retribution. To me, punishing an individual for a crime out of retribution makes as much sense as punishing a car for breaking down; I'd prefer to see the individual "fixed" in the same way I'd prefer to see my car fixed. The primary obstacle to the former from my perspective is lack of knowledge, which will diminish rapidly with time whether we as a society want it or not. I think one of the biggest challenges we will face as a society in the next 200 years (assuming we don't disappear or devolve into a dark age) is how to integrate advances in neuroscience and psychology into our sense of justice and responsibility. E.g., if you could change a person completely for the better, is withholding that change unethical? What's the point of retributional justice then? Aren't you just shooting yourself in the foot, societally speaking?

This is all tangential to the paper, but I think issues of free will become critical when you are faced with the possibility of total change in an individual.

People with [psycho]somatic symptom disorders, with medically unexplained symptoms, are more likely to report negative side effects in response to placebo, for example.

When I've read about it, the research is often focused on perceived negative symptoms rather than lack of therapeutic effect, but that's just my impression and I suspect both happen.

In fact, one of the problems with the Myers-Briggs is that it is continuously brought up in these discussions, even when experts in the area of behavioral individual differences dismiss it because it is lacking in evidence (and it has never really been dominant, or at least not for decades). There's all sorts of model-testing that lends support to some models (e.g., the Big Five, which is mentioned in the essay) and not the M-B, in terms of its internal empirical characteristics and predictive properties. And they do involve falsifiable predictions of multiple sorts.

The problem is that people complain about nonsense such as the M-B being nothing more than a money-making consulting scam, but then don't take that assertion seriously, in the sense that they assume the consultants are scientists.

It's as if con artists were selling perpetual motion machines, and HR departments started buying them, and then we started complaining about physics being non falsifiable, rather than about HR departments and business administration not understanding physics. It's all strawman arguments.

As an exercise, for example, I recommend someone searching for modern basic research using Myers-Briggs uncritically in mainstream psychology journals. You probably won't find it except for as some kind of deceptive ruse in an experimental protocol.

As for measuring behavior, any measurement throws away information. That's the tension: weight, BMI, blood pressure, temperature (under what pressure?), etc. The problem isn't in the measurement, it's in how that measurement is used and interpreted, and how much information is thrown away.

So, nothing is wrong with getting a measure of emotional-behavioral state. The problem is overgeneralizing from that, across time or situations, overestimating its predictive information, failing to consider uncertainties or biases of measurement, and so forth. BMI is an imperfect summary of someone's physical health, but it does have utility. The danger isn't in BMI per se, it's in assuming it won't change, assuming things about the reasons for a given BMI, ignoring how any given BMI was calculated, and so forth.

I think at some point notions of intervention or change become critical. So, for example, I think implicit in the idea of free will is the idea that an individual can freely choose to change their decision in some repeatability sense, and this change cannot necessarily happen in any other way, such as from outside influence.

Just to make it clear, let's say there's two choices or choice types, A and B. A is "bad" and B is "good," in whatever sense (utilitarian, moral, whatever).

Key to the notion of free will it seems is that one could change, of their own volition, their choice, from A->B or B->A. In practice we can't rewind time or transport to possible worlds in a counterfactual sense, but we do talk about decisions as interchangeable and repeatable. E.g., someone can choose whether or not to take a drug at one point, and then, at a later point, choose again whether or not to take the drug. In one metaphysical sense, those decisions are not the same because circumstances have changed, but we treat them as the same.

I think the idea of free will is that someone could change their decision, to an extent that another person could not. That is, I cannot make you change your choice, only you can.

This is fine at some level, but what if someone wants to change their mind but cannot? E.g., a drug addict or someone trying to lose weight? What about something simpler, like decision under a lapse in attention? How do we define volition or lack thereof?

Or, maybe more importantly, let's say the external individual in question is omniscient. If an omniscient external individual cannot alter your behavior, why would you hold that individual responsible in a free will sense? That is, let's say someone truly wants to change, but an omniscient being could not even help and change them. Why should the non-omniscient individual seeking change be held responsible?

This is all stream-of-consciousness, but I think at some point the notion of free will starts to lead to counterintuitive problems and/or becomes very poorly definable. At some level I suspect it implies not only autonomy but complete agency, which is suspect.

The prior state idea seems more accurate, but even that (at least in its extreme state) doesn't seem to me to encapsulate the idea of free will because it disallows the idea of a decision being undetermined and free to decide.

I personally think the notion of free will is fallacious because it's poorly defined, or even undefinable. I admit I could be totally wrong about that, but I think lack of predictability leads to an illusion of free will. It's the same illusion as the god illusion, injecting agency as an explanatory mechanism when there is none, either through chaotic processes, true randomness, or epistemological weaknesses.

People seem to get fixed on ranked choice voting versus plurality voting (the current system), and what we really need is a movement to change the voting method to something more accurate.

But it's puzzling to me why the discussion seems to highlight ranking rather than approval voting. The only argument I've really heard against it is that people would get confused and not realize they can vote for more than one person. But it seems to me in that case you've just re-implemented plurality voting, and people would figure it out well enough. Just put out a big public ad campaign. People would learn.

1. My experience with the frontiers of open access science in my field have been concerning, in that it seems to feed even more off of the social media hype-TED-talk branch of academics that I see as just as much a problem. That is, with traditional closed publishing, you have lack of access to the public and lack of transparency which seems unacceptable, but with very open access platform things can turn into a petty social media popularity contest. Yes, you get your stuff out, but the dynamics can be very different and feel more like a glorified flamewar ala reddit or old-style USENET groups.

Related to this, I think the way peer review plays out in different fields is very different. So, arXiv stuff might seem normal in comp sci, math, stats, but if you ever want to see uptake of that in an applied field, it had better have a rubber stamp of peer review or no one will touch it.

2. Unfortunately, at many universities, at least in the US and the UK, the pressure and nonsense has moved (increased?) from publishing from prestigious journals into grants. That is, the power structures underlying this all are indirect costs (in the US) subsidizing universities. The problems with this are problems in the way grant monies are allocated and awarded, the kind of "slush fund" quality of the indirect costs that afford universities profits, and the vast inequities across disciplines and topic areas in terms of what requires funding and what is popular at any given time (some research, for example, just requires less money to conduct; this should be a good thing for research funding, but is a bad thing from the university perspective because there's less indirect money to siphon off of).

The reason why I bring this up here is because I think the publishing problem and grant problem are kind of linked, but in a way that makes the open access issue kind of missing the target in terms of where the big problems with academics currently are. I worry, for example, that if you move to a model where everyone just publishes whatever anywhere, it turns academics into a giant flamewar, decreases the value of good research per se even more relative to grant receipts (because it's all fuzzy anyway and everyone can post anything to arXiv), etc.

I definitely think academic publishing needs to become totally open. But my current sense is that there isn't a good infrastructure to replace it in many fields, especially the biomedical fields, and pathologies in those areas will just sort of rip what's left of its integrity apart.

To answer your question: in some cases, transferring probabilistic reasoning about discrete states to continuous states, or vice versa, is sort of trivial. But sometimes it becomes somewhat controversial, and often this is because it's unclear how the scale of information maps onto the scale of data, to put it murkily. When it's discrete, you know your observations have specific possible values, which in itself is a chunk of information. But when it's continuous, you have an infinite range of values, and what constitutes "indifference" can be unclear.

Think of a scale (like a kitchen or bathroom scale), for example. The scale will have a certain accuracy within a certain range, and as you go further outside that range, the accuracy will decrease further and further. So, to make an analogy with the water-and-wine paradox, let's say I tell you "here's this container of walnuts, that could be anywhere from 0 to 100kg. What's the probability its mass as measured on this scale is in a certain range?" What would be adhering to the principle of indifference? You could argue it would be a uniform distribution over the numbers from 0 to 100, but you could also say it's somehow uniformly distributed over the distinguishable units of the scale, which will not be uniform from 0 to 100 because the meaningfulness of the scale's numbers will be compressed in its most accurate range, and stretched in its less accurate range, and there will also be issues with machine precision, etc. If you were counting walnuts though, you've kind of implicitly fixed the scale to the nonnegative integers, and assumed your ability to count doesn't get fuzzy at some point. So you assume uniformity on one scale is uniformity on another.

E.g., if you think of Jeffreys prior as being one scaling of indifference, even with a Bernoulli variable, indifference isn't necessarily scaled as a uniform: https://en.wikipedia.org/wiki/Jeffreys_prior.

I think this paradox involves (along with maybe being underspecified or poorly posed) somewhat related information-scale mapping issues, about how you define uniformity and what your "possibility space is" to be sort of Jaynesian about it.

Even if they bring it up to 85% accuracy with 50% base rate, by the time you are dealing with base rates that are more realistic, you're going to run into way more problems than just 1 out of every 7 people.

I don't expect either R or Python to go away either time soon, nor would I want them to, but I would like to see people moving to things like Julia and Nim, which have the same level of expressivity, but are much more performant. I have difficulty imagining many people saying "I love programming in R and Python, but don't like Julia or Nim."

I like Python but at least with stats/numerics there isn't a big reason to move away from R except for specific libraries (especially DL stuff) or front-end integration with web-land (and even then things like Jupyter mitigate against that).

I think what you've described is really the US healthcare system as a whole.

I really don't understand why this isn't more of a focus of discussion. The US healthcare system is so monopolized, and so lacking in competition and transparency, it's obvious why healthcare costs are skyrocketing.

I understand the outrage over this story, but I don't understand why it might be surprising to some. So much of the US healthcare system is driven by similar incentives, especially at private, for-profit hospitals (and even at nonprofit public hospitals).

Think about it this way. Take some fundamental human need. Now legislate and structure society so only a relatively tiny number of people with a specific certificate, based on program completion, not skills demonstration per se, can supply that need. Go even further and structure the regulation of that certification by people who already have the certificate, under the argument that they are the only ones who are in a position to judge it. Now reinforce the whole thing by FUD arguments that any other approach will lead to catastrophe, illness, and death.

What do you think is going to happen?

Almost everything about the US healthcare system is characterized by rent-seeking, monopolies, or power structures and lack of transparency. It's begging for corruption. If you wanted to design an economic sector so as to drive up costs, you'd end up with something that looks basically like US healthcare.