The Problem of the Notion of a [Diversity] Science

Replace the “diversity” tag above, with your favorite tag, be it Black, Green, Gay, Feminist or whatever. For much of science the problems are clear. What is “gay” number theory, or “feminist” crystallography besides nonsense.

A friend pointed me at this essay on First Things entitled, The Myth of Scientific Objectivity. There is much to unpack there, but I’d thought I’d offer a few thoughts. My notions of philosophy of science and how science works to put my cards on the table are much influenced by Michael Polanyi especially this book, Personal Knowledge. I think some of the insights from that book would do to criticize and quiet the problems that arise such as Mr Wilson’s example, a ‘feminist’ sociologist examining the “good” features of divorce, which requires ignoring much of the obvious. Mr Polanyi points out that much of science is, contrary to popular notions, a process which we can’t explain but have to learn for ourselves. One of the features of this explanation of how science works it that there is an essential step which Mr Wilson doesn’t mention.

Mr Wilson points out that the scientific process is not the abstract inductive or deductive process, but one of a collection of personal insights for which the advocate of that insight then gathers data to support and convinces other that he/she is correct. I think the part missing here is that the person who has this insight has become, through years of work, skilled at the ways of thinking and methods in solving problems in the particular field of research that their insight is not uninformed but instead based on a collection of personal history and knowledge in that same field. The aesthetics of what comprises good science in any particular field is taught and learned and makes an essential feature to the progress of science.

Diversity in and of itself has impact on fields of science, as you would expect, only as much as the social aspects of human life are the within the scope of inquiry in that branch of science. If you are studying how flagella propel microorganisms in fluids, then your notions of gender and race exactly irrelevant. But within sociology, psychology, and such arguably have contributions that might be possible from other social points of view. But those insights gleaned from those fields likely are as impermanent as the social conditions in which they are implanted. One the other hand, inquiry into the nature of elliptic curves over the rational numbers … not so much. The insights gleaned will not fade as social conditions change nor will the truths discovered be dependent on any features and facets of  human society.

I might note, that there is a good counter argument to Mr Polanyi’s ineffable nature of scientific knowledge, in that computer science and programming may be an answer to what is and what is not ineffable. See for example, this text. If you can teach a computer to do the thing you are trying to explain how to do, then you understand it at a level which is no longer ineffable. Your program is the explanation.

Leave a Reply

Your email address will not be published. Required fields are marked *

3 comments

  1. Boonton@gmail.com says:

    If you are studying how flagella propel microorganisms in fluids, then your notions of gender and race exactly irrelevant.

    I don’t think anyone actually would disagree with this statement, but is this statement important or is it a red herring?

    Let’s say that studying flagella is a very important thing. Say we have established an Institute of Flagella Study, IFS, which has 10,000 people working at it trying to unlock the flagella’s secrets.

    Would diversity in this operation matter? I suspect it does. Given thousands of universes, I think you would find the universes where most of the 10K scientists were hired based on connections to Jerred Kushner’s friends and family, their output would be less impressive than universes that followed some type of Google like process of trying to find the most flagella minded of scientists.

    Might it be possible an IFS staffed by unqualified cronies might get more lucky than dedicated scientists? Sure, but if we’re going to be with odds we should reject that long shot and go with the better odds.

    Now that we excluded that option, what about all the options where IFS is staffed by scientists but what type? Should they all just be the ‘best of the best’? Would it matter if they ended up 99% male or white or whatever?

    Well interesting thing I heard about search and learning algorithms. They get better if you introduce noise. Here’s a thought experiment from Nicholas Christakis. You have an environment of hills, valleys and one tall mountain. You randomly place 4 people handcuffed back to back and blindfolded in it and tell them they must find the highest point. One simple method they could use is take turns taking one step forward in each direction. If that step increases their elevation, take another one, if it doesn’t try a different direction. The only problem with this method is that they are likely to end up on top of one of the hills and not the mountain…unable to move since any step will lower their elevation. Introduce some ‘noise’ such as being open to taking ten steps downwards and you’ll greatly increase their ability to map the entire terrain and find the actual highest point.

    Our hypothetical IFS is presumably trying to find the largest secrets of the flagellum. Unfortunately we do not already know the flagellum’s secrets so when they tell us they have found it, we cannot know if they really did or if they just got themselves on top of a minor hill.

    In this sense then diversity is a management tool added to ensure the team is ‘stirred up’ and has people one it whose views will clash and generate ‘noise’ in their discovery process.

    We can note the US Constitution does this too, except it uses geographic diversity. Back in the day it might have been the case all the smartest people were clustered near Philadelphia, NY and Boston. Nonetheless the Constitution didn’t allow all of them to just be in the gov’t, the House and Senate forced geographic diversity…if South Carolina had dolts and PA had 10 geniuses, that didn’t matter PA only got 2 Senators which meant the Senate couldn’t just be ‘the best of the best’. Back then ‘identity politics’ probably was deeply tied to geography so what we really had and still have was diversity.

    I think where people falter here in this discussion is resorting to scientism in an area where science has actually spoken very little. The top 10,000 thinkers on the flaggellum may not be the optimal team to study it. When people complain about diversity measures at places like Google, they often make an assumption that ‘science’ has proven ‘the best engineer’ is the one who should get the job….but if you’re hiring hundreds of engineers you aren’t really focused on any one, instead you’re trying to find the optimal team and that’s not necessary as simple as selecting the top test score or grades or whatever easy metric someone might pull out.

  2. Mark says:

    Boonton,

    I mentioned flagella recalling a talk I’d heard a physics colloquium decades ago, the researcher noted that given the Reynolds number (fluid viscosity and length scales) flagella twirling are about the same as a boat with a propeller floating in peanut butter in which the propeller was rotating once an hour. It isn’t inherently obvious which direction the boat will go.

    Well interesting thing I heard about search and learning algorithms. They get better if you introduce noise. Here’s a thought experiment from Nicholas Christakis. You have an environment of hills, valleys and one tall mountain. You randomly place 4 people handcuffed back to back and blindfolded in it and tell them they must find the highest point. One simple method they could use is take turns taking one step forward in each direction. If that step increases their elevation, take another one, if it doesn’t try a different direction. The only problem with this method is that they are likely to end up on top of one of the hills and not the mountain…unable to move since any step will lower their elevation. Introduce some ‘noise’ such as being open to taking ten steps downwards and you’ll greatly increase their ability to map the entire terrain and find the actual highest point.

    Alas the diversity you want is inter not intra cranial, i.e., “diversity qua noise” inside one brain not between many. What diversity gives you in groups at first order is loss of trust.

  3. Boonton@gmail.com says:

    “What diversity gives you in groups at first order is loss of trust.”

    So what? I work for a company that has maybe 100,000 employees worldwide. Logically that means at some point it must have had 75,000 and it was decided to add 25,000 more.

    The simple diminishing returns theory would say the first 75,000 hired would be the top 75,000 in the world, and the next would be ranked 75,001 to 100,000. If we are hiring, say, chess players on Yahoo we could simply go down their roster. But is that the dynamics at play in selecting 25,000 more people?

    First of all, there is clearly no such roster. Optimizing something like a firm’s profit would entail combining lots of variables so even figuring out the mix of types of jobs is a huge challenge (i.e. how many sales people versus scientists). Even if you lock in the mix there is no obvious way to get the optimal ‘right people’.

    Second there are multiple biases to contend with. Top management says hire 25K more people, but they cannot actually do it. The people who have to decide which people to hire are spread around the world at multiple levels of the organization. What are their incentives? Well when I look around one big incentive people have is to get their friends and family jobs. I’m not talking corruption here, just inclination. People are even told to use networking like college and LinkedIn to find jobs. More than a few times I’ve seen one person leave to take a job at company B and a year later there’s three or four more people who go from company A to B. Ask yourself in a company how many people went to college, HS, together? How many worked at companies together previously? How many had friendship or family relationships before starting at the company? If the staff of the company was selected by a pure, perfect AI to be just the absolute best combination of people in the entire world I would say odds are the combination would have fewer rather than more such connections.

    This isn’t new and companies do put policies in place to counter such biases. HR depts. require that jobs be posted publicly. Outside recruiters are paid to bring candidates in. Larger companies will do job fairs and such to ensure mixing things up.

    Here’s my challenge to you, identify a single organization that suffers from ‘over diversity’ inflicted by policy? Google, for example, is something like 70% male and 30% female. Do we have any reason to believe, despite the claim of the engineer who wrote the memo, that an ‘optimized’ workforce that disregarded diversity would end up, what, 80% male instead? A modest bias leverages itself over an organization so any particular manager might have a small bias (say inclined to ‘overvalue’ people that look like him by 5%…doesn’t impact when there’s a clear difference in quality between two candidates but it does come into play when the difference is small but still there) but that will scale up to a larger than 5% deviation from the supposedly ‘ideal’ workforce.

    In sum I would say:
    1. Like the people trying to find the highest point blindfolded and handcuffed to each other, organizations have a difficult time finding the optimal set of employees to hire. They are very, very, likely to end up not at the peak but on a small hill where it seems like they found the highest point because any one step results in losing ground.

    2. Left to their on inclinations, there’s other motivations and biases that come into play. At least in the hypothetical the team of people want to find the highest point, but in an organization people have other motives like “who do I like working with more” or “what type of team makes me look like a better manager”.

    3. Not always but often these biases will break along racial/gender lines. Why? Because these things are social constructs and by definition social constructs divide people into groups that are given different experiences and views.

    4. Your trust argument might be valid but can you demonstrate it is an issue with actual objective measurements? is Google less ‘trusting’ because its diversity operations resulted in 70% male rather than…what?…75% male? I suspect if this is a real factor, it is only going to be measurable if a diversity operation is very dramatic. If Google is ‘naturally’ 75% male and a strict diversity operation forces it to 50% male then maybe you will start to see serious trust problems. But I doubt you can find any real life examples of such ‘extreme diversity’ that are non-trivial.

    4.1 I do think theres a ‘alt-right white snowflake issue’ at play. A lot of the hysteria is over almost invisible ‘slights’. If Google ‘should’ be 75% male and its diversity programs result in 70% male, if you are a male of high quality it is almost impossible to have suffered any gender based discrimination from google. At best, you would suffer possibly if you were an average male who happened to land in competition with a female and the difference in quality between you two was almost imperceptible. In other words, this is less about real discrimination and more about perceived entitlement issues by Trump types.