## Decision making and explainability

My hypothesis is – the more you need to explain your decisions to people, the worse your decision-making gets.

Basically, instinct gets thrown out of the window.

Most of you who have worked in a company would have seen a few attempts at least of the company trying to be “more data driven”. Instead of making decisions on executives’ whims and will, they decide to set up a process with objective criteria. The decision is evaluated on each of these criteria and weights drawn up (if the weights are not known and you have a large number of known past decisions, this is just logistic regression). And then a sumproduct is computed, based on which the decision is made.

Now, I might be biased by the samples of this I’ve seen in real life (both in companies I’ve worked for and where I’ve been a consultant), but this kind of decision making usually results in the most atrocious decisions. And it is not even a problem with the criteria that are chosen or the weights each is assigned (so optimising this will get you nowhere). The problem is with the process.

As much as we would like to believe that the world is objective (and we are objective), we as humans are inherently instinctive and intuitive individuals (noticed that anupraas alankaar?). If we weren’t we wouldn’t have evolved as much as we have, since a very large part of the decisions we need to make need to be made quickly (running from a lion when you see one, for example, or braking when the car in front of you also brakes suddenly).

Quick decisions can never be made based on first principles – to be good at that, you need to have internalised the domain and the heuristics sufficiently, so that you know what to do.

I have this theory on why I didn’t do well in traditional strategy consulting (it was the first career I explored, and I left my job in three months) – it demanded way too much structure, and I had faked my way in. For all the interview cases, I would intuitively come up with a solution and then retrofit a “framework”. N-1 of the companies I applied to had possibly seen through this. One didn’t and took me in, and I left very soon.

What I’m trying to say is – when you try to explain your decisions, you are trying to be analytical about something you have instinctively come to the conclusion about, and with the analysis being “a way to convince the other person that I didn’t use my intuition”.

So when a bunch of people come up with their own retrofits on how they make the decision, the “process” that you come up with is basically a bunch of junk. And when you try to follow the process the next time, you end up with a random result.

The other issue with explaining decisions is that you try to come up with explanations that sound plausible and inoffensive. For example, you might interview someone (in person) and decide you don’t want to work with them because they have bad breath (perfectly valid, in my opinion, if you need to work closely with them – no pun intended). However, if you have to document your reason for rejection, this sounds too rude. So you say something rubbish like “he is overqualified for the role”.

At other times, you clearly don’t like the person you have spoken to but are unable to put your rejection reason in a polite manner, so you just reverse your decision and fail to reject the person. If everyone else also thinks the same as you (didn’t like but couldn’t find a polite enough reason to give, so failed to reject), through the “Monte Carlo process”, this person you clearly didn’t like ends up getting hired.

Yet another time, you might decide to write an algorithm for your decision (ok I promised to not talk about AI here, but anyways). You look at all the past decisions everyone has made in this context (and the reasons for those), and based on that, you build an algorithm. But then, if all these decisions have been made intuitively and the people’s documented decisions only retrofits, you are basing your algorithm on rubbish data. And you will end up with a rubbish algorithm (or a “data driven process”).

Actually – this even applies to artificial intelligence, but that is for another day.

## Darwin Nunez and missed chances

There is one “fact” I’m rather proud of – it is highly likely (there is absolutely no way to verify) that in CAT 2003-4 (scheduled for 2003; then paper got leaked and it was held in Feb 2004), among all those who actually joined IIMs that year, I had the most number of wrong answers.

By my calculations after the exam (yeah I remember these things) I had got 20 answers wrong (in a 150 question paper). Most of my friends had their wrong counts in the single digits. That I did rather well in the exam despite getting so many answers wrong was down to one thing – I got a very large number of answers right.

Most readers of this blog will know that I can be a bit narcissist. So when I see or read something, I immediately correlate it to my own life. Recently I was watching this video on striker Darwin Nunez, and his struggles to settle into the English Premier League.

“Nobody has missed more clear chances this season than Darwin Nunez”, begins JJ Bull in this otherwise nice analysis. Somewhere in the middle of this video, he slips in that Nunez has missed so many chances because he has created so many more of them in the first place – by being in the right place at the right time.

Long ago when I used to be a regular quizzer (nowadays I’m rather irregular), in finals I wouldn’t get stressed if our team missed a lot of questions (either with other teams answering before us, or getting something narrowly wrong). That we came so close to getting the points, I would reason, meant that we had our processes right in the first place, and sooner or later we would start getting those points.

In general I like Nunez. Maybe because he’s rather unpredictable (“Chaos” as JJ Bull calls him in the above video), I identify with him more than some of the more predictable characters in the team (it’s another matter that this whole season has been a disaster for Liverpool -I knew it on the opening day when Virgil Van Dijk gave away a clumsy penalty to Fulham). He is clumsy, misses seemingly easy chances, but creates some impossible stuff out of nothing (in that sense, he is very similar to Mo Salah, so I don’t know how they together work out as a portfolio for Liverpool. That said, I love watching them play together).

In the world of finance, losing money is seen as a positive bullet point. If you have lost more money, it is a bigger status symbol. In most cases, that you lost so much money means that your bank had trusted you with that much money in the first place, and so there must be something right about you.

You see this in the startup world. Someone’s startup folds. Some get acquihired. And then a few months later, you find that they are back in the market and investors are showering them with funds. One thing is that investors trust that other investors had trusted these founders with much more money in the past. The other, of course, is the hope that this time they would have learnt from the mistakes.

Fundamentally, though, the connecting thread running across all this is about how to evaluate risk, and luck. Conditional on your bank trusting you with a large trading account, one bad trading loss is more likely to be bad luck than your incompetence. And so other banks quickly hire you and trust you with their money.

That you have missed 15 big chances in half a season means that you have managed to create so many more chances (as part of a struggling team). And that actually makes you a good footballer (though vanilla pundits don’t see it that way).

So trust the process. And keep at it. As long as you are in the right place at the right time a lot of times, you will cash on average.

## Stable Diffusion and Chat GPT and Logistic Regression

For a long time I have had this shibboleth on whether someone is a “statistics person or a machine learning person”. It is based on what they call regressions where the dependent variable is binary. Statisticians simply call it “logit” (there is also a “probit“).

Now, in terms of implementation as well, there is one big difference between the way “logit” is modelled versus “logistic regression”. For a logit model (if you are using python, you need to use the “statsmodels” package for this, not scikit learn), the number of observations needs to far exceed the number of independent variables.

Else, a matrix that needs to be inverted as part of the solution will turn out to be singular, and there will be no solution. I guess I betrayed my greater background in statistics than in Machine Learning when, in 2018, I wrote this blogpost on machine learning being a “process to tie down coefficients in maths models“.

For “logistic regression” (as opposed to “logit”) puts no such constraint – on the regression matrix being invertible. Instead of actually inverting the matrix, machine learning approaches simply focus on learning the terms of the inverted matrix using gradient descent (basically the opposite of hill climbing), so mathematical inconveniences such as matrices that cannot be inverted are moot there.

And so you have logistic regression models with thousands of variables, often calibrated with a fewer number of data points. To be honest, I can’t understand this fully – without sufficient information (data points) to calibrate the coefficients, there will always be a sense of randomness in the output. The model has too many degrees of freedom, and so there is additional information the model is supplying (apart from what was supplied in the training data!).

Of late I have been playing a fair bit with generative AI (primarily ChatGPT and Stable Diffusion). The other day, my daughter and I were alone in my in-laws’ house, and I told her “look I’ve brought my personal laptop along, if you want we can play with it”. And she had demanded that she “play with stable diffusion”. This is the image she got for “tiger chasing deer”.

I have written earlier here about how the likes of ChatGPT and Stable Diffusion in a way redefine “information content“.

And if you think about it, almost by definition, “generative AI” creates information (and hallucinates, like in the above pic). Traditionally speaking, a “picture is worth a thousand words”, but if you can generate a picture with just a few words of prompt, the information content in it is far less than a thousand words.

In some sense, this reminds me of “logistic regression” once again. By definition (because it is generative), there is insufficient “tying down of coefficients”, because of which the AI inevitably ends up “adding value of its own”, which by definition is random.

So, you will end up getting arbitrary results. ChatGPT often gives you wrong answers to questions. Dall-E and Midjourney and Stable Diffusion will return nonsense images such as the above. Because a “generative AI” needs to create information, by definition, all the coefficients of the model cannot be well calibrated.

And the consequence of this is that however good these AIs get, however much data is used to train them, there will always be an element of randomness to them. There will always be test cases where they give funny results.

No, AGI is not here yet.

## Computer science and psychology

This morning, when I got back from the gym, my wife and daughter were playing 20 questions, with my wife having just taught my daughter the game.

Given that this was the first time they were playing, they started with guessing “2 digit numbers”. And when I came in, they were asking questions such as “is this number divisible by 6” etc.

To me this was obviously inefficient. “Binary search is $O(log n)$“, I realised in my head, and decided this is a good time to teach my daughter binary search.

So for the next game, I volunteered to guess, and started with “is the number $\ge 55$“? And went on to “is the number $\ge 77$“, and got to the number in my wife’s mind (74) in exactly  7 guesses (and you might guess that $\lceil log_2 90 \rceil$ (90 is the number of 2 digit numbers) is 7).

And so we moved on. Next, I “kept” 41, and my wife went through a rather random series of guesses (including “is it divisible by 4” fairly early on) to get in 8 tries. By this time I had been feeling massively proud, of putting to good use my computer science knowledge in real life.

“See, you keep saying that I’m not a good engineer. See how I’m using skills that I learnt in my engineering to do well in this game”, I exclaimed. My wife didn’t react.

It was finally my daughter’s turn to keep a number in mind, and my turn to guess.

“Is the number $\ge 55$?”
“Yes”

“Is the number $\ge 77$?”
“Yes”

“Is the number $\ge 88$?”
“Yes”

My wife started grinning. I ignored it and continued with my “process”, and I got to the right answer (99) in 6 tries. “You are stupid and know nothing”, said my wife. “As soon as she said it’s greater than 88, I knew it is 99. You might be good at computer science but I’m good at psychology”.

She had a point. And then I started thinking – basically the binary search method works under the assumption that the numbers are all uniformly distributed. Clearly, my wife had some superior information to me, which made 99 far more probable than any number between 89 and 98. And s0 when the answer to “Is the number $\ge 88$?”turned out to by “yes”, she made an educated guess that it’s 99.

And since I’m used to writing algorithms, and  teaching dumb computers to solve problems, I used a process that didn’t make use of any educated guesses! And thus took far many more steps to get to the answer.

When the numbers don’t follow a uniform distribution, binary search works differently. You don’t start with the middle number – instead, you start with the weighted median of all the numbers! And then go on to the weighted median of whichever half you end up in. And so on and so forth until you find the number in the counterparty’s mind. That is the most optimal algo.

Then again, how do you figure out what the prior distribution of numbers is? For that, I guess knowing some psychology helps.

## The Twelfth Camel

In a way, this post should write itself. For those of you with context, the title should be self explanatory. And you need not read further.

For the rest I’ll write a rather small essay.

The story is of the old Arab who died leaving his eldest son half his wealth, the second a fourth of his wealth and the youngest one sixth. The wealth in question turned out to be 11 camels.

With 11 being a prime number, how could this will be executed without any of the camels being executed? An ingenious neighbour came in and lent his camel. Now there were twelve. The three sons respectively received 6 ($\frac{12}{2}$), , 3 ($\frac{12}{4}$) and 2 ($\frac{12}{6}$) camels respectively. One camel was left over – the neighbour’s, who took it back.

This is mathematically inaccurate, since the sons received fractions of their father’s wealth slightly different from what he had intended. However, in general in life, this parable of the twelfth camel offers a useful metaphor.

In engineering, this is rather common – you have systems such as a choke, for example, to enable systems to get started from a “cold start process”. The choke comes in only at the time of startup – once the thing has started, it plays no role.

However, it has its role in normal life and business as well. For example, after a bad breakup, you might rebound to a “stop gap partner”. You know that this is not going to be a long term relationship, but this partner helps you tide over the shock of the bad breakup, and by the time this relationship (inevitably) breaks up, it has achieved its purpose of getting you back on track. And you get on with life, finding more long term partners.

Then, when the company is in deep trouble, you have specialists who come in to take over with the explicit goal of cleaning things up and getting the company ready for new ownership. For exanple, John Ray III has recently taken over as CEO of FTX. His previous notable appointment was as CEO of Enron, soon after that scandal had broken. He will not stay for a long term – he will just clean things up and move on.

And sometimes the role of the twelfth camel is rather more specific. Apart from “generic cleaning”, the temporary presence of the twelfth camel can be used to get rid of people who had earlier been hard to get rid of.

In sum, the key thing about the twelfth camel theory is that the neighbour knew all along that he was going to get back his camel. In other words, it is a deliberate temporary measure intended to achieve a certain set of specific outcomes. And the camel itself may not know that it is being “lent”!

## Size, diversity and social capital

Starting off with a “global” statement, life is full of tradeoffs.

When we listen to stories as kids, we think of “battles between good and evil”. When we watch sport, there is “our side and their side”. From stories, we are usually conditioned to “battles” where one side is superior to the other, and there is a clear “favourite” to root for.

Real life is not so simple. A lot of times, you have battles between two sides that are both bad (a lot of elections, for example). And frequently you come across situations where there are two “good” things of which you can only have one (and so the tradeoffs). One of the more famous of these is the “impossible trinity” of international economics.

In a completely different context (which I’m writing about today), two desirable things that are a tradeoff are diversity and social capital. The general theory is that the more diverse a society gets, the lower is the social capital (google is impossible in providing good links on this, so maybe ask ChatGPT).

The theory here goes that you fundamentally trust people like you and mistrust people who are not like you. The more homogeneous a society is, the more there are “people like you”, and so the more you trust. And if everyone trusts one another much more, there is more social capital.

A recent conversation and observations makes me wonder if social capital is also related to size. More specifically I’m thinking of size as in number of people in an institution.

One observation we had when we went for our reunion last week is that the campus is a lot quieter than in our time, and that a lot more rules are followed in letter rather than just in spirit (no, not that kind, but I’m talking about rules about that also).

For example, the basketball hoop in L^2 has been removed. While we pitched up a net and played tsepak in BEFG Square (where we always played it), current students informed us that they can’t play there because playing there is now banned. Students mostly hung out in their wings rather than in the common areas. After our first evening there we assumed all the students were out on holiday, while it turned out only the first years had a term break then.

I still remember my first night in IIMB (in 2004). I stayed in G Block, on one side of the aforementioned BEFG square. A bunch of people were playing Tsepak until late in the night, which meant I didn’t get good sleep (and for the rest of the week we had our “orientation” which meant I couldn’t sleep well then either). A few days later I just joined these people in playing Tsepak and making noise. “If you can’t stop them, join them”, was a perfect way to go about things back in the day.

What I remember is that, with a batch size of ~200, our social capital was pretty good. While there were some students who occasionally displayed elevated levels of conscience, we largely stuck by one another and tolerated one another (and, of course, massively trolled one another). Disagreements and fights always happened, but were largely resolved among us by dialogue, rather than inviting external parties.

With a much bigger batch size now, though, from what we were told, it appears it is not so simple to resolve things using dialogue and mediation. And people frequently take to inviting external parties. And the expected result (crane-mongoose effect) happens.

That some people want to study when others want to play now means that the former complain against the latter, with playing within the hostel being banned. Some people want to enforce the rules on spirits, and they bring in external parties, and the law is invoked in letter rather than spirit.

When social capital dwindles, in some ways, the minority rule comes into play – when there is a small but vocal minority that wants things a certain way, that becomes the way for everyone else as well. (With high social capital, the majority might be able to convince the minority that they need to be more tolerant)

Yet again, this is not a one way street. You can also argue that when social capital is too high, the minorities can tend to get oppressed (since their views don’t count any more), and so a high social capital society cannot be inclusive.

And so yeah, the Baazigar principle is there everywhere. To get something, you need to give up something. To get a more inclusive class, you need to be less majoritarian, and that means less fun on the average. When you have lots of intolerant minorities (a consequence of diversity), those “intolerant rules” get applied on everyone, and the overall payoff reduces.

A few random thoughts to end:

1. It’s not just the class size, it’s also the fees. We paid ~ ?300,000 over 2 years as tuition fees. Many of us (who sat for campus placements) made almost twice that (post taxes) in our first year of graduation.

Students nowadays pay ~ ?2,500,000 , so they are a lot more conscious about getting their money’s worth. And being able to study.

2. In general, cultures change over time, so coming back after 15 years and complaining that “things aren’t the way they used to be” isn’t very nice. So yeah, this blogpost can get classified in the “not so nice” category I guess (not like I’ve ever been known for niceness)
3. I wonder how much changed during the pandemic, when students were off campus for nearly a year, and had severely curtailed interactions even once they were back. With a 2 year course, it only takes 1 batch to “break culture”, making the culture far more malleable. So again I’m wrong to complain.
4. All that said, it’s my duty to pontificate and so I’ll continue to write like this

## Girard and reunions

Thanks to my subscription to Jim O’Shaughnessy’s Infinite Loops podcast, I have been exposed to some of the philosophy of Rene Girard. A few times, he has got philosopher Johnathan Bi on the show, to talk about Girard’s philosophy.

Bi has also done a series of YouTube lectures on Girard’s philosophy, though I haven’t watched any of them.

In any case, Girard’s basic thesis (based on my basic understanding so far) is that we are all driven by “mimetic desire”, or a desire to mime. This means we want to do things that others want to do.

So you see an instagram post by a friend who has gone to Sri Lanka, and you want to go to Sri Lanka as well. Your cousin has invested in Crypto, so you want to invest in crypto as well. Everyone in your class wants to do investment banking, and so you want to do that as well.

(actually now that I think of it, I was first exposed to mimetic desire by a podcast episode sent by my school friend Hareesh. In a way, Bi’s appearance on Infinite Loops only enhanced my liking for this philosophy).

This is yet another of those theories that “once you see you cannot unsee”. You see mimetic desire everywhere. Sometimes you copy the actions of people who you want to impress (well, that’s how I discovered Heavy Metal, and that has now turned out to be my most-listened-to genre of music, because it turned out I like it so much).

The theory of the “mirror neuron” is unclear (at least I’m yet to be convinced by it), but either by gene or by meme, we are conditioned to mime. We mime people’s actions. We mime their desires. We do things because others do them.

As the more perceptive of you might know from my previous post, we had our 16th year IIMB reunion this year. Not many turned up – about 30 from my class (2006) and 45 from the class of 2005 (thanks to covid both our 15th year reunions had been postponed, so we ended up having our 16th and 17th year reunions respectively).

It was an amazing experience. I don’t know what it was, but I liked it far more than the 10th year reunion.  One major thing was the schedule – the 10th year reunion lacked a focal point on the main day (I’ve written about it) because of which we were rather scattered around campus. The 10th year reunion also had a much more formal structure, with “sessions” which meant we had less time to chat.

This time round, the Saturday schedule was very good – an interaction with the current director RTK from 10 to 11, and then NOTHING. That interaction was enough of a focal point to get us in one place, so everyone was accessible.

Then, fewer people having turned up meant we ended up having deeper conversations. We spoke about life, philosophies, kids, spouses, divorces, other people’s divorces, random gossip and all such. Absolutely no small talk, and infinitesimal work talk, and that made it more satisfying.

This morning, Bi tweeted again about Girard and mimetic desire.

One of the corollaries of Girard’s theory is that people get into conflict not when they are different but when they are similar. And mimetic desire means that people will try to become more similar to each other, and that increases conflict.

If not anywhere else, that is true in a business school, especially one where the class is rather homogeneous. Mimetic desire means everyone wants the same jobs, the same grades. And so they compete. And get into conflict.

16 years post graduation, we have drifted apart, and not in a bad way. Over this period, a lot of us have figured out what we really want to do and what we really want, and understood that what we want is very different from what others around us want. Not really being around our former peers, we have no desire to mime them any more, and that has freed us up to do what we really want to do, rather than just signalling.

And so, when we meet at a reunion like this, we are all so independent that we just never talk about work. There is no sense of competition, and we just focus on having fun with people we went to school with. The time apart has helped us get out of our desires to mime, and so when we get together, we compete less.

Maybe I should read / understand more philosophy. Or is this desire just mimetic?

## Dhoni and Japan

Back in MS Dhoni’s heyday, CSK fans would rave about his strategy that they called as “taking it deep”. The idea was that while chasing  a target, Dhoni would initially bat steadily, getting sort of close but increasing the required run rate. And then when it seemed to be getting out of hand, he would start belting, taking the bowlers by surprise and his team to victory.

This happened many times to be recognised by fans as a consistent strategy. Initially it didn’t make sense to me – why was it that he would purposely decrease the average chances of his team’s victory so that he could take them to a heroic chase?

But then, thinking about it, the strategy seems fair – he would never do this in a comfortable chase (where the chase was “in the money”). This would happen only in steep (out of the money) chases. And his idea of “taking it deep” was in terms of increasing the volatility.

Everyone knows that when your option is out of the money, volatility is good for you. Which means an increase in volatility will increase the value of the option.

And that is exactly what Dhoni would do. Keep wickets and let the required rate increase, which would basically increase volatility. And then rely on “mental strength” and “functioning under pressure” to win. It didn’t always succeed, of course (and that it didn’t always fail meant Dhoni wouldn’t come off badly when it failed). However, it was a very good gamble.

We see this kind of a gamble often in chess as well. When a player has a slightly inferior position, he/she decides to increase chances by “mixing it up a bit”. Usually that involves a piece or an exchange sacrifice, in the hope of complicating the position, or creating an imbalance. This, once again, increases volatility, which means increases the chances for the player with the slightly inferior position.

And in the ongoing World Cup, we have seen Japan follow this kind of strategy in football as well. It worked well in games against Germany and Spain, which were a priori better teams than Japan.

In both games, Japan started with a conservative lineup, hoping to keep it tight in the first half and go into half time either level or only one goal behind. And then at half time, they would bring on a couple of fast and tricky players – Ritsu Doan and Kaoru Mitoma. Basically increasing the volatility against an already tired opposition.

And then these high volatility players would do their bit, and as it happened in both games, Japan came back from 0-1 at half time to win 2-1. Basically, having “taken the game deep”, they would go helter skelter (I was conscious to not say “hara kiri” here, since it wasn’t really suicidal). And hit the opposition quickly, and on the break.

Surprisingly, they didn’t follow the same strategy against Croatia, in the pre-quarterfinal, where Doan started the game, and Mitoma came on only in the 64th minute. Maybe they reasoned that Croatia weren’t that much better than them, and so the option wasn’t out of the money enough to increase volatility through the game. As it happened, the game went to penalties (basically deeper than Japan’s usual strategy) where Croatia prevailed.

The difference between Dhoni and Japan is that in Japan’s case, the players who increase the volatility and those who then take advantage are different. In Dhoni’s case, he performs both functions – he first bats steadily to increase vol, and then goes bonkers himself!

This is (hopefully) a quick post I’m dashing off from the sidelines of the Basavanagudi aquatic Center where I’ve brought my daughter for her weekly swimming lessons

I write this as I’m reading Eric Hoel’s post on why he is leaving academia. Basically he talks about all the “extra curricular activities” that an academic nowadays needs to do. Reviewing journals, being on student bodies and the like.

The other reason he quotes is about how restrictive academia is – again because he is being evaluated on multiple dimensions, rather than simply on the quality of his research and teaching.

Given all of this, following four years as an assistant professor at tufts, he has chosen to quit academia full time and become a writer of newsletters. and he writes that “being an academic is not so easy any more”.

I was reminded of an old post I’d written about Indian and American universities. American universities admit students based on “a holistic set of factors”. So your test scores are important but so are your sports and charity work and 10 different kinds of extra curricular activities and all that.

Indian universities (at least the ones I went to) are far simpler – they admit solely on the basis of test scores.

After reading some articles on how the US admission process was producing highly homogeneous classrooms at universities, id written a few years back on how the Indian system rocked – because admissions were based on a single criterion, there was tremendous diversity jn the classrooms on all other criteria.

Now based on Hoel’s post I’m wondering if the same is true of teachers as well – the more the dimensions on which we evaluate professors for recruitment and tenureship, the more homogeneous the professorial class gets. Instead if we were to evaluate professors on narrowly defined conventional criteria (teaching and research) we’ll get a far more richer and diverse professors body.

This, however, is easier said than done. Quality of research is usually evaluated based on the quality and quantity of papers, and papers necessarily go through a peer review process.

And if your peers are all those who have succeeded in the “selected by holistic criteria” game, then you will have to conform to some of their biases to get good papers published.

All this said, I’m hopeful that in the next decade or so we will have a bunch of new and privately funded universities which Yale universities back tk what they used to be – centres of research and teaching , with professors selected on their credentials on these axes only, and a diverse body of students selected hopefully on a a small number of axes (such as test scores).