25%

Let me ask you a question.

Manufacturing as a percentage of GDP – what is the highest that this number has been in India’s case? And bonus question: in what year?

Maybe you are a Modi supporter, and your guess is that we have achieved the highest number sometime in the last ten years. Or maybe you are a fan of Manmohan Singh, and believe that we achieved the highest number while MMS was the Prime Minister. Well, take a look for yourself (hover the cursor on the line for the years):

Play the “My Gormint Best and Your Gormint Worst” game to your heart’s content on your social media platform of choice, but the fact remains that all governments in India’s recorded history have failed in this specific regard. Give this dataset to any undergrad econ student and ask them to add a trend line to this series.

I went ahead and asked ChatGPT to add two, in fact. One for the entire dataset, and one for post 1991:

And please tell me – pretty please, do tell – how in the wide world have we managed to reduce the share of manufacturing in GDP at a faster rate post 1991? Now, that question is partly rhetorical, because part of the answer is the fact that services took off post-1991.

But even so. This is a deeply depressing, deeply enraging chart.

Why is it depressing and enraging?

Because the empirical data tells us that if we want India’s per capita GDP to go up over time, we should be increasing the share of manufacturing in GDP.

Either that, or we have to chart a path that no other country has before us. What kind of path? A path that shows the world how to grow GDP from relatively low levels while reducing the share of manufacturing in GDP

Note that the numbers (manufacturing as a share of GDP) are different because of different computational methodologies being used in both charts.

But regardless of how one computes the data, the fact of the matter is that we need to figure out a way to ramp up the share of manufacturing in GDP.

This pattern isn’t just empirical observation—it’s backed by decades of economic theory. From Arthur Lewis’s dual-sector model in the 1950s to more recent work by economists like Dani Rodrik and César Hidalgo, the consensus is clear: manufacturing has played a crucial role in economic development. It absorbs excess agricultural labor, drives productivity improvements, and creates the kind of stable, middle-class jobs that form the backbone of a prosperous society.

Opinions differ about whether manufacturing can (and will) continue to play the starring role it has in the past. But it remains a variable of crucial importance, and India is not well served by having manufacturing remain at around 10%-15% of GDP.

From a macroeconomic perspective, this manufacturing push is about more than just GDP growth. It’s about increasing labor’s share of national income and, crucially, developing a meaningful middle class in India. We can talk about increasing GDP all we like, but the real point is to pull people out of poverty, not just to make a big number bigger. A strong manufacturing sector, with its stable jobs and productivity growth, is our best (only?) shot at achieving this goal.

So what holds India back? Why have we remained incapable of pushing manufacturing up? Opinions differ wildly about the “correct” answer to this question, as you might imagine. But regardless of political affiliation and position on the economic spectrum, one can count on a few inevitable factors: rigid labor laws, skill gaps, land acquisition difficulties, and the rapid pace of automation in global manufacturing. Of these, the last is obviously out of our direct control. As regards the first three, we know these are problems, but we cannot get ourselves to Do Something About It.

But the more we delay, the stronger the very real (and already present) threat posed by automation in global manufacturing. One cannot be serious about meeting this threat without reforming our labor laws, meaningfully improving upon higher education and simplifying land acquisition. Perhaps most difficult of all, the mindset of the government and the bureaucracy has to change from that of an administrator to that of an enabler. Worse still, this has to happen at local levels of government. But cultural shifts are hard enough as it is – engineering a cultural shift across all levels government bureaucracy is like getting an oil tanker to do a wheelie.

So what is to be done?

Here is my proposal: set a madly ambitious target. 25% (say) of India’s GDP should come from manufacturing over the next 20 years. Specifically, one-fourth of our total cumulative output over the next two decades should come from manufacturing.

I am under no illusions – I think this target is not achievable. I also think that this target, for a variety of reasons, is best thought of in conjunction with the subsidiarity principle (about which there will be much more in a follow-up post). But assume with me, for the moment, that this is the target, no matter how implausible it sounds. Now ask yourself the question – in order to achieve this, what, exactly, needs to change in our country, and by when?

Forget 2029, and definitely forget 2047. Focus on one number, and one number alone: 25% by 2044. Is what we are doing helping us achieve this goal? If yes, go ahead and do it. If not, ask yourself why we are doing it. The correct time to ask and start answering this question was yesterday, but no matter. Now is an entirely acceptable substitute.

25%.

Let’s go.

Unitree Robotics: The G1

This is from a month ago, and it already feels… old?

China and Robots

In Which Garett Comes to Someshwarwadi (But Elinor Can’t)

I stay in Someshwarwadi. It’s a small little locality that lies between two other localities in Pune, Baner and Pashan. It is famous for a Someshwar temple that was built sometime in the 12th century (or thereabouts), and little else.

But one thing that it should be famous for is the fact that it will lie right next door to a marvel that is coming up in Pune – the Pune Metro. That alone is not enough to make it famous, of course, because one hopes that lots of areas in Pune will lie next to the Pune Metro.

What ought to make Someshwarwadi famous is the fact that commuters can soon get off a world class metro, and then not have footpaths to walk on. Yup, you read that right: all of Someshwarwadi has no footpaths.

Nothing gives economists greater pleasure than using jargon, so of course we have a name for this phenomenon. What phenomenon, you ask? That of building metros before attempting to build footpaths. Us economists, we call it isomorphic mimicry. I come from a city that can give this phenomenon much better names, especially in our own language, but we’ll leave that be for now.

But not having footpaths is problematic, of course. And not just because walking becomes an adventure sport with life and death consequences. It is also problematic because Garett has come a-visitin’.

Who is Garett? Why, Hardin, of course:

The tragedy of the commons is a metaphoric label for a concept that is widely discussed, and criticised, in economics, ecology and other sciences. According to the concept, should a number of people enjoy unfettered access to a finite, valuable resource such as a pasture, they will tend to over-use it, and may end up destroying its value altogether. Even if some users exercised voluntary restraint, the other users would merely supplant them, the predictable result being a tragedy for all. The metaphor is the title of a 1968 essay by ecologist Garrett Hardin. As another example, he cited a watercourse which all are free to pollute. The concept itself did not originate with Hardin, but rather extends back to classical antiquity, being discussed by Aristotle.

In Someshwarwadi, people do enjoy unfettered access to a finite, valuable resource: the side of the road. There are no footpaths, no “No Parking” signs, no nothing. So everything and everyone – cars, abandoned cars, people, garbage and stray dogs, among other things and beings – jostles for space.

And speaking of “other users who would supplant”, we have new members in this crowded club: school-buses.

These school-bus owners take advantage of the fact that there is an absence of no parking signs on this road, and solve for themselves the problem of where to park these large vehicles. Which is all well and good for them, of course, but is a less than desirable state of affairs for the rest of us in Someshwarwadi. Unfettered access for all results in a less than ideal outcome for all. Why, Garrett bhai was writing about Someshwarwadi while working in the United States in the 1960’s! Must have been a prescient sort of fellow.

Students of economics and ecology know what is coming up next. Here we go, they are thinking to themselves as they impatiently scroll through. Hurry up and get Elinor in here already!

It was long unanimously held among economists that natural resources that were collectively used by their users would be over-exploited and destroyed in the long-term. Elinor Ostrom disproved this idea by conducting field studies on how people in small, local communities manage shared natural resources, such as pastures, fishing waters and forests. She showed that when natural resources are jointly used by their users, in time, rules are established for how these are to be cared for and they become used in a way that is both economically and ecologically sustainable

But as you will see from the title of this post, it ain’t quite so simple. As I have mentioned in the title of this post already, Garett is very much in Someshwarwadi, but Elinor cannot make it here, alas.

And why is that? It is because the area in question also has a local, extremely well connected political presence that, er, “decides” everything. Enforcement of local customs in Someshwarwadi is a top-down thing, not a bottom up thing. So yes, rules are established, sure, but not by folks who are motivated to strive for solutions that “economically and ecologically sustainable”.

What we get are promises of footpaths and of no parking signs. Actual footpaths will be delivered as soon as we figure out how to solve the problem of isomorphic mimicry. Until then, parked buses and Mr. Hardin will be permanent fixtures of our little neighborhood.

Ah well. So it goes.

The Heresy of Recommending Rent Control

Ouch. It pains me to have to write that title, let alone continue to write a blog post about it. All of my training as an economist gone down the drain, for how can an economist possibly write a post about rent control without first loudly and definitively decrying it?

But Gulzar Natarajan says that rent control may well be needed in markets, and trust me on this, GN knows a thing or twenty about the practice of economics. So let’s find out why he says what he does!

  1. Unhindered, unfettered markets will “play out” the best outcome over time, but the key words in there are “over time”, not “best outcome”. Here’s a fun thought experiment for you to think about: if you had to choose between letting markets play out the best outcome for vaccine procurement during the second wave, or having the government step in and buy up vaccines left, right and center, what would you choose?
    Now, you might say that vaccines in the second wave was a special case. Healthcare markets in India, you might say, should have no intervention, and “the energy” of the free market should be allowed to guide outcomes. The poor in our country in the future will thank you for your brave call, but what about the poor in the country today?
    Note that everybody in the country was poor when it came to access to vaccines during the second wave.
  2. What do you want from a market? Maximal coverage at minimal prices such that the market remains a viable, self-sustaining phenomenon, or maximal profits? Well, depends on who is answering, and for which market. And if you honestly think that both answers are necessarily the same (even if in the long run), come with me to Agra. I own this awesome building, and I think you might be interested in buying it.
  3. “Everyone knows what to do, but the problem is to win elections after doing it”. Study political economy, not just economics.
  4. “Theory should not become the enemy of the practical. This might be a case of saving capitalism from the ideologues of free-market capitalism”
    Strong words, perhaps, but I found myself nodding in agreement.
  5. “Experts and commentators should wake up to the reality of a world where market interventions are inevitable. Public debates should be anchored around those second-best interventions that generate the least distortions and inefficiencies. Instead of ignoring realities, public policy discussions should be conducted in this constrained space.”
    Fit your theory to the world you live in, in other words. Fitting the world you live in to the theory you like is rather problematic in practice.
  6. Studying economics without working in public policy is a little like watching YouTube videos about cooking without ever stepping in the kitchen. That’s where the heat is, and you’ll have to face up to that heat sooner or later. That’s just how it is.

What Do You Mean by “General” Intelligence?

AGI is either already here, or just around the corner, or ain’t never gonna happen. Regardless of which camp you fall in, you will have heard these three assertions time and time again over the last two years.

But what is AGI? Here’s Wikipedia:

Artificial general intelligence (AGI) is a type of artificial intelligence (AI) that matches or surpasses human capabilities across a wide range of cognitive tasks. This is in contrast to narrow AI, which is designed for specific tasks. AGI is considered one of various definitions of strong AI.

So we define artificial intelligence not in terms of what it is, but in terms of what it exceeds. Note, also, that Wikipedia has done a neat little sleight of hand here – a type of AI that matches or surpasses human capabilities across a wide range of cognitive tasks. This avoids us having to get into a discussion of what human intelligence is!

But let us hold our blog post to a higher standard, and try and define AGI in terms that stay constant throughout. And so rather than try and define AGI in comparative terms, how about we try and define GI (A or otherwise)?

What is, in other words, General Intelligence?

A generally intelligent entity is one that achieves a special synthesis of three things:

A way of interacting with and observing a complex environment. Typically this means embodiment: the ability to perceive and interact with the natural world.

A robust world model covering the environment. This is the mechanism which allows an entity to perform quick inference with a reasonable accuracy. World models in humans are generally referred to as “intuition”, “fast thinking” or “system 1 thinking”.

A mechanism for performing deep introspection on arbitrary topics. This is thought of in many different ways – it is “reasoning”, “slow thinking” or “system 2 thinking”.
If you have these three things, you can build a generally intelligent agent.

So embodiment, plus System 1 and System 2 thinking. That is the necessary, but not sufficient condition for GI. Without these three, you simply cannot have GI.

But even with these three being present, you will not have GI… unless you are able to “coherently execute the above cycle repeatedly over long periods of time, thereby being able to attempt to optimize any objective.”

The first of these is robotics, a field that I am increasingly more curious about (as you may have been able to see, given the number of videos I have been sharing this year about advancements in robotics). The second of these – what is referred to as “intuition”, “fast thinking” or “system 1 thinking” here – is simply what we think of as chatbots today. That is kind of true, kind of “meh, not really true”, but it is good enough for our purposes in this blog post.

What we are missing is the third of these – a mechanism for performing deep introspection on arbitrary topics. The author of the blog post we are talking about here feels that the ability to do System 2 thinking by machines is about two to three years away:

There is not a well known way to achieve system 2 thinking, but I am quite confident that it is possible within the transformer paradigm with the technology and compute we have available to us right now. I estimate that we are 2-3 years away from building a mechanism for system 2 thinking which is sufficiently good for the cycle I described above.

Actually, he says, it is not the ability to “do” system 2 thinking in order to achieve any given objective. It is the ability to be able to do System 2 thinking… and also to have the ability to figure out when to use it, and when to stop using it. Can the agent synthesize a plan, and work at it, or choose (because it realizes the objective is too hard) not to work at it?

… and all this leads the author (J Betker, of OpenAI) to predict that we are about 3 years away (best case scenario) from AGI. Five years at the most, but no more, he says.

I have no clue if the timeline makes sense or not. But reading this essay helped me become a little bit clearer about how to think about the concept of general intelligence.

WATTBA: Gene That Spreads Cancer Through the Body Identified

Mostoslavsky and colleagues first compared gene expression patterns in primary versus metastatic tumors in mice with pancreatic cancer or breast cancer. After identifying various genes whose expression increased in metastatic tumor cells, the researchers silenced each gene individually.

In these experiments, silencing the Gstt1 gene had no effect on primary tumor cells from mice, but it stripped metastatic cancer cells of their ability to grow and spread. It also blocked cell growth in two metastatic-derived human pancreatic cancer cell lines.

That is from Mass General, and I do not think I will read a more cheerful article today. I am very far away from being an expert in this field (or any field, for that matter, for I strive to be truthful), but it does seem as if we are making progress in terms of trying to figure out how to stop the spread of cancer.

I got this from The Not Boring Newsletter, and if you haven’t already, you should consider subscribing. As Packy so memorably put it (and you’ll have to forgive me the profanity, but I happen to share Packy’s views on the matter):

The next step would be developing a therapeutic that can safely and effectively silence the Gstt1 gene.

Fuck cancer. Fuck metastasis. Go science.

Speaking of subscribing to his newsletter, there are two other medical discoveries that Packy speaks about, the first of which will help us cure hereditary deafness in young children (seriously, WATTBA!), and the second will “modestly slow cognitive declines in patients with early stage Alzheimer’s”.

Now, that’s the good news: modestly slow cognitive declines. And you might think that it isn’t much of good news, because there is only a modest amount of slowing down. But it gets worse, because this was the good news. The bad news is worse: the drug comes with significant risks, including bleeding and swelling in the brain.

If you take the benefits and the costs into account, this seems like… not good news at all, right?

Well, that depends, because remember, one of the magic questions you should always be asking is “Relative to What?”.

And for a patient with a diagnosis of Alzheimer’s – a horrible disease with no known cure currently – yes, even this development is good news, and potentially worth the risk.

The question was whether that 4½ to 7½ slow-down was worth the relatively substantial risks of the drug. Alzheimer’s is a nasty, sad disease that currently has no cure. The committee approved the drug because, well, it ultimately decided that in the context of Alzheimer’s even a slight slow down of decline outweighed potential risks.

Donanemab still requires final FDA approval, but when and if it that occurs, the drug should become available to early stage Alzheimer’s patients. Of the current 6 million Americans currently suffering from the disease, approximately 50% are in early enough stages that donanemab could have a positive impact.

We love to see a common sense risk-benefit analysis from the FDA that takes both the risks and the benefits into account. Let’s get this drug live!

Opportunity costs, as any student of economics will tell you, are everywhere. Is it worth risking brain swelling and bleeding in order to stop a common cold? Definitely not. But is it worth it to stop (even if by a modest amount) Alzheimer’s? Trickier question, and we may forever be unsure of the correct answer – but surely we can all agree that it isn’t “Definitely not”.

WATTBA for the win!

Apologies for the long radio slience – I was away in Manipal in the month of May, and then got sucked into a whirlwind project that turned out to be a whole lot of fun.

But I’m back in the saddle now, and hope to stay there for the foreseeable future.

Sludges and Meta

What are sludges? They are the opposite of nudges.

Now, I know, I know. That begs the inevitable question, so here’s ChatGPT:

Ever felt like the universe was subtly guiding you towards better decisions? That’s a nudge! Think of it as a friendly elbow that steers you towards healthier or smarter choices without restricting your freedom. For example, putting fruits at eye level in a cafeteria encourages better eating habits.

Now, imagine wading through a swamp of paperwork and tedious processes just to get something simple done. That’s sludge! It’s the gunk that slows you down and makes life unnecessarily difficult. Think endless forms, confusing websites, and bureaucratic hoops.

In short, nudges help you breeze through life, while sludge makes you trudge. One’s your invisible buddy, the other’s the proverbial thorn in your side.

My favorite example, and one that I often use in class, is to say that The New York Times makes it easy to sign up to their newspaper. Visit the website, enter your credit card details, and you’re good to go. Canceling, on the other hand, is jumping through a lot of hoops. It is better and easier than it used to be – but back in the day, you had to call them (at US times!) on an international number to cancel. That’s pretty much a spot on example of “gunk that slows you down and makes life unnecessarily difficult”. And so that’s what sludge is.

And here’s another great example of what a sludge is:

The clue is in the first tweet itself – when the author says that “it’s intentionally designed to be highly awkward in order to minimse the number of users who will object to it”, that’s a pretty important clue.

And the rest of the Twitter thread explains how (and how difficult) it is to object to your data being used to train Meta’s AI models. Do go through it, regardless of whether you use Meta or not.

Here’s another great example from later on in the thread:

There are two “Calls to Action” in this picture. The first is to simply close this notification. Can you spot where the “Close” button is?

The second “Call to Action” is the Right to Object. Can you spot where the “Right to Object” button is?

Would you call it a button?

Is it easy to find?

Is it a coincidence that one call to action is prominent and impossible to miss, while the other is in normal sized font, hidden in the middle of a lengthy paragraph, and looks like other hyperlinks that are not calls to action?

Sure, maybe it is just a coincidence. But I don’t feel like buying the Taj Mahal today, thanks.

TMKK: the world will always be looking to trick you into making decisions you really didn’t want to make. That’s just life. And there will be times when you will be fooled. That too, is just life. But minimising the chances of being fooled is very much up to you – and studying behavioral economics helps you minimize those chances of you being fooled.

Mobiles and Cashew Nuts

Whenever I give a talk on behavioral economics (and sometimes I feel that is all I do!), I always tell the story of Richard Thaler and the big bowl of cashew nuts. Tim Harford interviewed Richard Thaler for the excellent Lunch with FT series, and while the restaurant they were at didn’t serve a bowl of cashew nuts, alas, they did serve almonds.



And so an excellent story was (re)born:

He served the notorious bowl to some guests while dinner was roasting in the oven, then watched everyone compulsively munch on the nuts and gradually spoil their appetites. So Thaler decided to remove the temptation by hiding the cashews in the kitchen. His guests thanked him.

It would be an unremarkable tale, except that such behaviour simply does not fit the rational economic model of human behaviour. Either eat the cashews or don’t eat the cashews, says classical economics, but don’t thank the person who moves them out of easy reach.

Reflecting on such stories helped Thaler create “behavioural economics” — a branch of the discipline that aims at psychological realism. Doing so also helped him with the equally difficult task of persuading other economists to take the behavioural view seriously.

Either eat the cashews or don’t eat them. And certainly don’t thank the person who moved them out of easy reach.

But that’s cashews. What about mobile phones?

Here’s how it works: Once guests arrive at a show and present their tickets, Yondr workers or venue personnel ask them to silence their phones and place them in Yondr pouches. The staffers snap the bags shut, and guests take them inside.

Devices can still receive calls and messages, but to check them, guests have to step outside the venue or go to a cordoned-off phone-use area to have the bag unlocked. After the event, staffers quickly unlock and collect bags as guests exit.” 

“Two schools in Torrington, Conn., started using Yondr bags a couple of years ago, paying $60,000 to use the pouches for roughly 2,000 students.”

Of course I visited the Yondr website. Who wouldn’t? Whereupon I discovered that there is now (of course there is) a Yondr tray for home.

What is a Yondr tray for home? Well, it is a tray in which you can keep your phone, and then you can lock the tray. So a box, really, and not a tray. A box tray, note, with all these features:

https://www.overyondr.com/home-tray

And how much does this tray cost? $249. Or if you prefer Indian currency, about twenty thousand rupees, give or take.

Truly remarkable, the power of commitment devices and behavioral economics. Also truly remarkable? Smartphones, and the impact they have upon us.

Chatbots Are Going To Be Awesome For Teachers

Arnold Kling says that he has no sympathy for a professor who worries about a student using a chatbot to cheat during an examination:

I have no sympathy for a professor who worries that students could use a chatbot to cheat on an exam. Chatbots and their relatives can be helpful for student assessment if you want them to be.

Not only do I agree, but I positively cannot wait to try out chatbots for both teaching as well as for evaluation. Worry? I’m positively giddy with excitement!

The only full fledged course I teach these days is the Principles of Economics course at the Gokhale Institute. And I only have two problems in teaching this course.

The first is that I have to cross University Circle in order to teach it. If you are not from Pune and are reading this, you cannot know what I’m talking about. If you are from Pune and are reading this, you cannot not know what I’m talking about.

The second problem that I have with teaching this course is that there are a hundred and fifty students who take this course. Which means, of course, that teaching this course becomes a series of speeches, rather than an actual class. Because relevant personalisation, and the ability to handle students doubts and questions becomes all but impossible across a hundred and fifty students.

Now, AI can do nothing about the first problem, because Pune’s traffic is a problem worthy of ChatGPT-50, let alone ChatGPT-5. But the second problem? Ah, there many treats await us educators.

https://chatgpt.com/g/g-jFQpQYY9D-socratic-challenger

This is a custom GPT that I built, and is freely available on the ChatGPT website. All you need to do is (if you haven’t already, that is) create a login ID for the website, and you should be able to try the Socratic Challenger for yourself.

If you’re not sure of how to proceed, try clicking on one of the examples shown in the screenshot. Here’s what my screen looked like when I clicked on the ATP example:

What this custom GPT does is that it acts like a personalised tutor. It is programmed to begin with a groan-worthy dad joke (a skill at which none can best me, bar none). Once that is out of the way, it asks you what you wish to learn. Feel free to specify the topic, the subject which contains the topic and your level of expertise – or any or none of the above, for that matter.

This custom GPT will then proceed to have a conversation with you over three “rounds” of questions. That is, based on what you say, it will push back and play devil’s advocate in such a way that it tests your understanding of a subject, and your ability to talk meaningfully about it.

At the end of these three rounds, it will give you a “score”, based on how well you did. It is (and will probably forever be) a work in progress, so feedback and tips on how to make it better are always going to be very welcome.

Is it perfect? Hell no.

Would I do a better job than it, if I were to engage in such a Q&A myself with a student? Almost definitely yes.

Will it do a better job than me across 150 students? What do you think?

I’m a little impatient by now with folks who insist on pointing out that teachers will always do a better job than will chat bots. One, that is not necessarily true. Second, even if it were to be true, it certainly isn’t true at scale.

It does not matter how good, or dedicated, or focused I am at conducting vivas. I will never be able to maintain quality across 150 students. And I do not care who you are, I know you will not be able to maintain the same level of quality across 150 students. The last few vivas before your lunch break, before tea in the afternoon and just before you wrap up for the day will all be substandard vivas compared to, say, the first three of the morning.

So not this particular Socratic Challenger, perhaps. Nor the one that Arnold Kling came up with. But there will be, sooner or later, a most excellent chatbot that is customized just so, to meet each of your student’s different needs. And this chatbot will do a better job than you will at assessing if each student in your class has truly gotten what was being explained in class.

Again, if the benchmarking is to be done for one student, AI will probably not be as good as your above-average teacher is right now. But if colleges want to enrol 150 students in a single class and call it a quality education, you’d have to be nuts to not use AI extensively for pedagogical and evaluation purposes.

No, it can’t replace you just yet, don’t worry. But also no, you can’t teach well without it, and that was true as of yesterday.

I am willing to die on this hill: chatbots are going to be awesome for teachers.

And they probably are awesome for teachers already.

Let’s go!