A Response To Marc Andreessen’s ‘Why AI Will Save The World’

— By someone who is both a fintech startup founder and an artist

The world’s a stage and we are all artists – William Shakespeare; And the supreme gift of an artist is the knowledge of when to stop – Sherlock Holmes/Arthur Conan Doyle.

First, a quick little story about me that is relevant to my response. And also because you don’t know me like you know Marc 🙂

I am the founder of a profitable, bootstrapped Indian payment startup SabPaisa. I have also written a couple of paperback romantic thrillers one of which became a bestseller back in 2011. I am also a marketer, a science lover, and an aspiring comedian.

But the thing that is most relevant to this response — and where the recent seismic arrival of LLMs may affect me negatively — is that I am a songwriter too. A late bloomer who has created many songs but hasn’t properly put a single one in the public domain.

My Story Of An Unusual Songwriter

Ten years ago — when I was around 33 and part of another edutech company I had founded — I started crooning/recording songs into my mobile while on an evening walk one Friday. I did this for 9 years: crooning hundreds of Hindi, Hinglish, Punjabi, Urdu, even English songs across multiple genres into my mobile at all times of the day or night or midnight, whenever a song popped up in my head. Some songs were fully written, most only partially, but all were simultaneously written/composed in my head alone.

For all those years I didn’t know how to bring those songs to life so others could hear them too.

I knew even then I could write pretty good lyrics; others around me had told me the same. But I didn’t have a great voice. And neither did I have any training in music or songwriting or knew how to play any instrument (though I did try to learn to play Sitar for 6 months when I was around 13 in a music class).

So I didn’t have an idea how good were my crooned compositions. I was also running a startup which is a pretty hectic thing to do. Ergo…

Almost a year ago, I finally decided to do something about this talent god seemed to have given me. I got in touch with a young composer/arranger over Fiverr and worked with him to bring my songs to life. I learnt a lot from him, including the fact that I had already composed the songs and you didn’t necessarily need to know how to play a musical instrument to compose a song. And that you could do all that in your head simultaneously: create the lyrics and compose.

The outcomes of our collaboration surprised me, made me almost jubilant. When the first few songs came out, I discovered I had not known for decades of my life that I had raw, innate talent for creating beautiful or entertaining songs. For example, a song I had written/composed in my head in less than 30 minutes when a mild evening storm had hit Delhi — a dance/party number called ‘Dilli-Mumbai Toofan Song’ — was predicted to be a sure shot hit by whoever heard it among my family and friends. To tell the truth, occasionally I felt like a genius*.

With hundreds of songs already written, many it seemed with the potential of going viral and becoming hits, I was about to sail out among the stars in a musical ship, way beyond the moon.

And then the dream may have ended too soon: GPT 3.5 arrived in November. Less than 5 months after I had finally started doing at the young age of 43 what I thought I had been truly sent into the world for.

How do I feel? We will come to this question later.

The Three Camps Of The AI Battle

The recent arrival of powerful large language models like GPT 3.5, GPT 4, Claude, LLAMA, Bard etc has divided the world — among those who are either directly affected or know enough to care — into largely three camps.

Camp 1:  AI is going to be great for the world, so keep your hands off my AI.

Camp 2: AI is stupid and/or going to take away my/our jobs and hurt the poor and the marginalized, so regulate it strictly and better reject it or even outright ban it as I/we say.

Camp 3: Yes, AI may create a heaven on earth someday, but there is a pretty good chance it may kill us all instead, so regulate it strictly as we say.

Pretty much everyone I have listened to or read on this topic falls in one or the other or at most two of the camps. So currently it’s an intellectual and political slugfest between people in Camp 1 vs people in Camp 2 vs People in Camp 3.

But I am yet to read stuff written by someone — though I am sure they are out there — who feel they simultaneously belong in all three camps and are utterly conflicted in their feelings towards AI.

I am that someone, so I have decided to write this piece. And argue simultaneously from the viewpoints of all the camps.

What makes me a resident of the first camp? — I have been a tech enthusiast since I was a teen in high school and college. I have always felt and argued – starting with the first debate I took part in my 2nd year at IIT Chennai in the year 2000 – that technology, by enabling a relatively abundant world with lesser struggle for existence, has probably been the greatest among the forces of good for humanity. I had arrived at this conclusion by myself by reading the stories and novels of great Indian storytellers like Sharatchandra Chatterjee and Munshi Premchand and comparing their times and society and cultural beliefs to ours.

For example, technology has had a big hand in making liberal constitutional morality and democracy, mature and endure as the dominant moral and political system for our modern times. 

So I truly look forward to a future where AI, along with energy abundance achieved through fusion, leads us into a post scarcity society. This in a nutshell is the reason why I belong in the first camp despite being an artist – a fiction writer and a songwriter albeit a recent one – who may get hurt by the arrival of LLMs.

However, like millions of other artists, I also fear that AI will dilute or make redundant our talent for creating art (like storytelling or songwriting). It will do so by helping people — who may not have our innate or hard-earned-through-the-sweat-of-our-brows talents — create high-quality art with just a text or a voice prompt based upon a vague idea (and even those ideas can be generated by AI) — something that has already happened with image generation tools like Midjourney.

Here are three images I have generated with Midjourney with a simple one line text prompt: Einstein cutting his own hair, Cinderella wearing a gown made of fractal patterns and Socrates working while wearing casuals in a co-working setup. And I am pretty sure with a little bit of training on these tools, everyone today can become a defacto art creator.

But can we call that art? Can I call myself an artist in the traditional sense? Can I call myself a painter because despite being pretty bad with a pencil or a paintbrush, with a single line prompt to Midjourney I made Elizabeth Bennet and Mr. Darcy dance a tango in the middle of the road in the style of Georges Seurat or made Mirabai drink poison in the style of Raja Ravi Varma or expressed the emotion of anger on canvas in the style of Jackson Pollock. I even did a pretty good job of recreating Burj Khalifa in Monet’s style and invented a hybrid style called Quantum Cubism by trying to blend cubism with the concept of quantum superposition to depict musical instruments. 

It was a lot of fun, but I definitely don’t feel like an artist. More like an impostor, albeit one who’s having a ball playing with these tools.

As far as the third ‘AI is gonna kill us all’ camp is concerned, frankly I am a novice in the field and don’t know enough to take sides. But I do have a couple of theses that makes me an AI optimist at times, and at other times an AI doomer. You can say my legs are inside the third camp, but my body is outside of it.

Shit Is Complex: My Worldview & Philosophy In Brief

Before going into responding to Marc, let me quickly outline my worldview and philosophy that informs and colors everything I write.

If I have to describe it in three words, it would be: Shit is complex. And dynamic. And evolving. And often goes to shittier shit if people with differing ideologies and worldviews keep fighting each other instead of collaborating to tackle the shit. Ok, that’s twenty-six words.

Therefore in my opinion each of the three camps is simultaneously right and wrong. The camps are right when they are trying to put forward their own views; and they are wrong when they are trying to negate the other side’s view as unworthy of consideration. Like the blind men and the elephant, they are each touching one part of the elephant in the room and thinking they know the entire elephant. Or at least pretending they know the entire elephant.

So bringing Walt Whitman here and paraphrasing him: AI is also large, AI also contain multitudes, AI also contradicts itself.

Do I have a answer to all the contradictions? No. I haven’t gone to Hogwarts (yet), and I don’t know magic (yet). Like I said, shit/problems in our human world are always complex, dynamic and evolving. So will be the solutions to the shit.

It’s just that because of my somewhat unique situation in life, I happen to be straddling the shit identified by all the three camps: AI is very likely going to screw me as an artist; AI is very likely to help me as a tech founder and a citizen of modern times in a liberal democratic country; AI if not regulated intelligently may kill me (or my children or my grandchildren, if and when I have them).

So despite also being blind to most aspects of AI, I perhaps see a bit more of the AI elephant than others. And my job in this piece is to try to point out to each camp, especially Marc and others like him with the ability and power to influence shit, how the other camp may be correct too from its own vantage point. And then hope all of us can come together to find solutions to the multitudes of problems the current and future iterations of AI present to us.

All The Abundance In The World (Is Not Enough)

A few days ago, former Netscape founder and famous venture capitalist Marc Andreessen published an essay called why AI will save the world.

He touched upon multiple topics and ideas in his essay on which I would love to share my thoughts some day. But today is not that day. Today, to keep this response short, I will only grapple with the most important issue I see with his assumptions and assertions: his dismissal of the possibly bleak future of human-executed jobs in the age of AI.

Before I disagree with Marc, let me tell you where I agree with him: I too share his optimistic view of an AI enabled future, especially a future in which AI has helped us solved the energy scarcity by making fusion possible. That possibility alone convinces me to be bullish towards an AI enabled future.

One of the biggest problems the world has always faced is absolute material scarcity. It makes people physically suffer, sometimes in the extreme, and fight with each other usually making the situation worse. That’s why throughout human history the world has seen plenty of violence and fighting on and off (which likely destroyed many populations and civilizations that had advanced technologies and systems in many areas of life, and set back humanity time and again).

What has mitigated this sorry situation in modern times, at least to a certain extent? The relative abundance brought about by technology, along with the moral and political stability and balance brought about by liberal democratic norms which again can survive and flourish only in a state of abundance.

Yet, despite the numerous bounties of present day technology, the world still faces quite a bit of scarcity. Especially energy scarcity, which leads to other scarcities like food and water and creates numerous interdependent problems that spiral out of control in many places creating a gigantic amount of unnecessary suffering.

Therefore a world where energy is abundant thanks to fusion or other non-conventional energy resources, with a resulting abundance of material goods and services, will see a lot less absolute scarcity and a lot less fighting and suffering than what we see today.

So where is the problem then? Where do I see gaps in Marc’s essay?

The issue is that Marc has forgotten there’s no rose without a thorn. In his essay he only talks about the lovely fragrance of the rose he is smelling because he is going to get the rose of the AI enabled future without the thorns. The thorns in the materially abundant world created by AI will fall in other people’s kitty.

Thorns, what thorns? Marc may ask. After all in his rose-tinted view of the rosy AI enabled abundant future, there is a job here, there is a job there, there are jobs, jobs, everywhere. (Sorry Marc, not trying to be snarky, just couldn’t resist the opportunity to riff on the famous phrase).

You can’t actually fault Marc for thinking like this. History confirms he is mostly correct. Indeed, every single time in the last few hundred years, whenever technology and automation have changed the nature of production and transformed the type of jobs available, people have panicked. Turned out needlessly, at least in the long term.

Yes, many lives were damaged in the short term, as Henry Hazlitt’s himself mentions in his Economics In One Lesson (in the chapter called The Curse of Machinery) which Marc has cited in his essay. But in the long-term things did seem to change for the better, as the chapter illustrates using the example of the introduction of machines in the stocking industry.

If historical patterns repeat, the same may happen this time too. There may be a lot of chaos and pain in the near term for a lot of people displaced by the newly developed capabilities of the AI, people who would lose their jobs and livelihood. But in the long-term things would get back to being better than ever.

Yes, that may indeed happen. God willing it will be so. (Because I think the age of AI is inevitable; it’s always been impossible to stop the wheels of technological change barring wholesale destruction of civilization; you may slow things down a bit if you try hard, but you can never stop it, and the same is likely true for AI advances as well).

But I am not so sure history will repeat this time. I predict the AI enabled future, where we have solved energy scarcity, may be one of overflowing material abundance and prosperity, yet coupled with a big job scarcity unless we take specific steps to stop that from happening. Why?

To understand why — why Hazlitt or Bastiat (of The Candlemakers’ Petition fame) mayn’t be right anymore — I will need to suggest an entirely new framework (new as far as I know) with which to see things. I haven’t yet found a good, non-cringy name for it, so let’s go with ‘The Means & Laws of Transaction’ as a cringy, working title. (It’s a riff on the title of a chapter in George Eloit’s ‘Mill On The Floss’, one of my fav novels. Guess which chapter?)

The Oldest Job In The World

Let me now try to be a little edgy, controversial, rabble-rousing, provocateur. It’s the easiest way to compete in an attention economy 🙂

So, I will talk about the oldest job in the world. No, not that one, the other one.

The oldest job in the world, and also perhaps the most important one, is today under threat — I think for the first time in the history of the world — from an entity that is usually our ally: technology. I am sure even Marc is worried about this situation, like many others are including Elon Musk.

The job in question is not only the oldest and the most important, but also the most common one in the world: men and women coming together to have kids and raising them so human life can go on. A job important and yet difficult enough that human societies all over the world came up with the institution of marriage to streamline and optimize our performance on this job, and a few other related jobs.

That job is today under stress in many parts of the world of which US, Finland, and South Korea are big examples. And the arrival of AI is going to make the situation worse.

You may think I am talking nonsense. Having and raising kids, and getting into a marriage to do so, is not a job bro — you may be itching to tell me — it’s the foundation of society. And how does this relate to your disagreement with Marc anyway?

The framework I mentioned before — ‘The Means & Laws of Transaction’— will help me explain. And I have specifically chosen the the phenomenon of marriage decline and falling TFRs to illustrate my argument within this framework as they have elements that most succinctly show how the age of AI may result in a job apocalypse.

Illustrating The Means And Laws of Transaction

99% of jobs in the world depend for their existence upon three umbrella level actions: Production, Transaction, Consumption. (Excluding some jobs which people do for themselves where they both produce and consume stuff themselves without a transaction happening. Or a miniscule number of jobs that fall outside this broad framework).

For example, most jobs involve people wanting something, and someone producing it at a price people are willing to pay to make the transaction happen (as Paul Graham said “make something people want”). Some jobs involve people producing something and making others want it through sales or marketing at a feasible price and the transaction happens. Some other jobs, especially in banking and finance, exist to manage the risk of the entire process. Occasionally jobs don’t actually produce anything to be consumed, but only exist to make people money through arbitrage etc.

99% of traditional relationships and marriages that have happened till date on earth have been a job as described above — for both the partners, the wife as well as the husband.

For example if you are a single Indian man of marriageable age living alone, older relatives who are itching to arrange a marriage for you, will often ask you the dreaded question: why aren’t you marrying? Followed by another question: how do you manage to eat, who is cooking for you? Of course the implicit assumption is that you need a wife to cook for you.

From a traditional viewpoint, that was a sensible enough reason for a guy to marry. Cooking is a labor-intensive, difficult job*. And if you were getting a girl to do all that for you (among other things) by marrying her and promising her a few things in return, why not?

*(It is even more difficult in many places of the world especially in rural areas where poverty reigns and women have to cook without the help of modern technological implements along with doing associated tasks like procuring fuel or water. As a child myself, I have seen my grandmother and aunts cooking with wood and coal while in a village. And I shamelessly declare that if I had to choose between being forced to do that everyday four times a day vs going to a battlefield where I had a 99% percent chance of dying, I will not hesitate a minute before choosing the latter).

The act of cooking inside a traditional family thus falls neatly under the above framework. Why still one of the most important questions asked to the bride-to-be in an Indian arranged marriage setup is: beta, do you know how to cook?

In essence, the wife produces a meal, the husband or other family members consume it, and the transaction is governed by the institution of marriage where the wife has been promised a bundle of stuff/transactions in return e.g. the husband will provide the material with which the food will be cooked etc.

One important thing to point out here — and make you less angry with my cavalier, heartless, stone-cold treatment of the holy institution of matrimony — is that transactions inside a marriage are not always practical and mundane, like I have made them in the above example. I have done so only to I hammer in my point unambiguously (and to make you angry; I had said I will be edgy, didn’t I?; you were forewarned, so don’t complain).

Transactions inside a marriage — like in the outer world — fall along the Maslow’s Hierarchy as a crude approximation. Proposed by many others, including in this article in ‘The Atlantic’, it’s an important thesis that drives my argument — one among the 12 key theses discussed in this essay that together illustrate how the age of AI might lead to the job apocalypse scenario.

What are these transactions in a marriage and how do they fall along the Maslow’s hierarchy? (I have chosen Maslow’s Hierarchy to illustrate my crude approximation because Maslow’s Hierarchy is a pretty well-known framework that illustrates the point well).

I Say Maslow, You Say Maslow, Let’s Call The Whole Thing Maslow

First, let’s talk about the most obvious and perhaps the most important of the transactions, the raison d’être of most relationships and marriages, why straight people don’t want to marry someone of their own sex, and gay people don’t want to marry someone of the opposite: sex (and physical intimacy).

Sex (and physical intimacy) inside a relationship or a marriage, like many other means of transactions, maps on to multiple levels of Maslow’s Hierarchy. And like many other means of transactions, it can be both an object and a currency of transaction depending upon the context.

Sex is a primal human drive like other primal human/animal drives e.g. hunger and thirst, so it clearly goes on the first level as an object of transaction: people have sex with each other because they simply want sexual satisfaction.

It also maps to the second level as a currency of transaction. People use sex to obtain security in a relationship or a marriage illustrated by the fact that lack of physical intimacy is one of the top reasons for divorce.

It also obviously maps to the third level as a currency of transaction: physical intimacy is one of the key ways in which people give/get love and affection.

It also maps to the fourth level as a currency of transaction. Being sexually desired in an exclusive way, or not, by one’s partner, is often an important determinant of one’s self-esteem. This also reflects in people’s reaction towards infidelity: one of the biggest effects of infidelity is the devastation of the victim’s self-esteem.

Finally, sex, in certain spiritual disciplines in a limited context, is also a currency of transaction to achieve self-actualization. Now don’t debate this one with me; I am from the land of Kamasutra and Tantra.

That’s as far as sex is concerned. What about cooking since we have already talked about that skill a few paragraphs before? I will leave that exercise for you. You can begin from the top levels of Maslow’s hierarchy by remembering that working and cooking to put food on the table for their kids or partner is one of the easiest and most common ways for a parent to find meaning or purpose in their life; and is also often a key determinant of their self-esteem.

But what has the means/currency of transaction in a marriage mapping onto Maslow and his hierarchy got to do with the phenomenon of marriage decline and falling TFRs? And what has all of that got to do with AI causing a job apocalypse?

For that, we will need to look at 12 theses about the economic properties of a marriage, three of which we have just discussed in the preceding paragraphs, and then extrapolate them to discuss and prove our central thesis.

Thesis 1 (A general thesis applicable to marriage): Production and consumption happen when there is possibility of a transaction and a human job exists when all three happen with the involvement of a human on the production side. Of these three, for the continued existence of a human job or birth of a new one, the important conditions are therefore production with human involvement and transaction at a feasible price.

Thesis 2: Marriage is a bundle of transactions between partners and the most common and important job in the history of the world. As a corollary, traditional marriage especially — for most of history — has been mostly an economic arrangement between the husband and the wife.

Thesis 3: The means/currency of such transactions map crudely onto Maslow’s Hierarchy.

The Economic Properties Of A Marriage

Let me outline the rest of the theses about the properties of a marriage before going any further.

Thesis 4: Marriage has been one of the history’s biggest example of price control and artificial lowering of prices (set by the society) to enable transactions/weddings to happen. How?

As discussed before, marriage is a set of transactions spread across all levels of Maslow’s Hierarchy. Historically, because of chronic and widespread poverty and scarcity, the transactions that mapped to the last two levels – physiological and security levels – overwhelmingly dominated in the set; at most three level dominated supposing the marriage was a happy one. And of course, the poorer you were, the more the lower level transactions dominated.

The expectations of those entering the marriage market was also set accordingly. Not only because of the true dynamics of the situation, but also because the family and society worked to deliberately push the expectations of the market participants lower, especially for women, so that the transactions could take place no matter how bad the quality of products in the eyes of the actual consumer (as opposed to the buying decision-makers). In other words, to keep the weddings happening and the marriages going, families and societies continuously primed you to lower your expectations and accept lower prices for your work if you were the producer, and lower quality product if you were the consumer, even more than what was necessary. That’s how arranged marriages worked. And if you decided to rebel, they could excommunicate you or even feed you poison to keep other market participants in line.

Thesis 5: In the last two hundred to hundred to fifty to twenty years, the job of marriage has seen a significant price inflation as society slowly removed its price controls and people started marrying for love beginning with England and the US, then in the western hemisphere, and then to a lesser extent in the rest of the world. Technology, social media, and economic and societal freedom and equality (by disrupting the traditional gender roles) have further disrupted the transactions, especially in the last twenty years, and especially the ones happening at the lower levels of Maslow’s Hierarchy.

Going back to the example of cooking, single Indian men in urban areas don’t need to enter a traditional marriage these days just so they can eat well. As long as they have some money, they have a way out if older relatives hell-bent on getting them married ask them the question about who cooks for them if they don’t have a wife — they can say they have Swiggy or Zomato (Doordash equivalent for India). In short, technology in the form of on-demand delivery, has created for the single Indian male a quasi-traditional-wife substitute, at least as far as cooking is concerned. (Yes, you can let your imagination run wild regarding the possibilities here in the age of AI; Movies like ‘Her’ and ‘Ex Machina’ are a good place to start).

All of this has resulted in an escalating shift in the expectations of relationship and marriage seekers regarding the nature of transactions a good marriage should have. The price expectations have increased for all the transactions. And further, the expectations have shifted the transactions from lower to higher levels of Maslow’s Hierarchy to a significant extent. Previously men were happy if their wife was a good cook; now men need wives who can debate with them what Thoreau meant when he said most men live a life of quiet desperation. Previously women were content if their husband could put food on the table and not get drunk and treat her badly. Now women want men to do the bare minimum before they will tie the knot (and after).

Essentially, due to technology and changing societal mores and individual expectations, the institution of marriage is witnessing the phenomenon of crowding out in the lower levels of Maslow’s Hierarchy, especially the last two levels. People are being needed less and less to fulfill the transactions corresponding to lower levels; and when such transactions do happen, the average price at which they happen are gradually increasing. It’s not enough for a man to put food on the table to be marriage-worthy; they need to be fairly well-of. Women can’t just be a good cook; they need to be a nine or a ten.

Beyond that, for marriages to happen now, human beings need to be capable enough, intellectually and emotionally, to fulfill the transactions corresponding to the higher levels of Maslow’s Hierarchy. Because that’s where a lot of the transactions have shifted.

And human beings are finding it increasingly difficult to cope with this new and continuously changing dynamics of the marriage market and starting to refuse to apply for the job.

From Quiet Desperation To Screaming Desperation

In summary, there are four key reasons why people have gone from being quietly desperate to screamingly desperate seeing the change in job description of the marriage job.

Viewing through the production-transaction-consumption framework, one of those reasons belongs to the domain of production, one to transaction, and the other two belong to the consumption side.

What’s the first reason, which belongs to the domain of production? It’s that from an evolutionary standpoint, we are statistically more suited to both produce (and consume) stuff that correspond to the lower level of Maslow’s Hierarchy.

Pretty much everyone can potentially grow food and eat, dig for water and drink, sleep, fight and defend, have sex etc; but it’s more difficult for us humans to write a book about man’s search for meaning after surviving a holocaust in a concentration camp. So the more the transactions needed to do the marriage job well move up the Maslow’s ladder, the less people will feel able to cope with the demands of the job and will refuse to apply for it.

We have already discussed the second reason belonging to the domain of transaction: it’s the problem of inflated expectations because of the crowding out phenomenon. It has lead to the increase in the friction/price of transactions when they do happen, once again increasing the demands of the marital job and leading to people saying “no, thanks.”

The last two reasons why marriage rates are falling belong to the domain of consumption. And they are most salient to the central topic of this essay because they deal with the counter-intuitive phenomenon of the limits of consumption.

The Limits Of Consumption

Human history has mostly been a history of scarcity — often a history of crippling, devastating, gut-wrenching scarcity where the limiting factors were almost always on the side of production made much worse by the greed of people in power.

For this and other evolutionary reasons, human beings understand and can deal with scarcity intuitively. We easily understand the limits of production. But we (and our bodies) have difficulty understanding and dealing with abundance (the recent dramatic rise in obesity in certain parts of the world is a good example). 

Transaction and consumption being a limiting factor in an abundant world — instead of production — is therefore a counter-intuitive situation which the current minds, even the best of minds, are struggling to grasp. So we keep creating issues trying to deal with abundance the way we deal with scarcity.

Coming to the economics of relationships and marriage, the limits of consumption show up in the marriage market in two different ways with different properties and ramifications: limits pertaining to the lower levels of Maslow’s Hierarchy and limits pertaining to the upper levels.

The limits of consumption pertaining to the lower levels of Maslow’s Hierarchy are quite easy to imagine and grasp. For example, for most of human history, most of the human population didn’t have enough food and struggled to acquire and keep what it had. This problem still exists in many parts of the world though some of the wealthiest countries have been able to solve it to a large extent.

In the wealthier countries people, by and large, have enough food and are unlikely to want much more unless they are Kumbhakarna. Yes, they may still want to go to a Michelin three star fine dining restaurant or crave to eat Knipschildt Chocolatier’s Madeline. But the intensity of their wants is likely to be much lesser than someone who is starving and wants a little food to stop being so hungry. In other words, if you are starving, and someone offers to feed you regularly if you marry them, you are far more likely to say yes to them — no matter their other qualities — than if you are well-fed and someone offered to take you to a Michelin star restaurant regularly (unless the fine dining aficionado had other qualities to go along with being that rich).

To take another example, it’s great to go from zero to one TV, and from one TV to three or four large screen TVs in your house. But even the richest people in the world don’t have a hundred TVs in their mansions. And if you are a regular person, I think you are less likely than more likely to marry a person with hundred large TVs in their house thinking they might be crazy.

Consumption, at the lower levels of Maslow’s hierarchy, thus has its asymptotic limits. As a corollary, the more the material abundance in the world, the less is the perceived value of a transaction related to any specific material thing in a marriage situation. And the less the role of the lower-level, material needs in persuading someone to take up the marriage job.

But what about a private jet you may ask? One may not want to go from one TV to a hundred or a hundred thousand TVs as the world becomes more and more materially abundant. But there are other, much better ways to show you are a member of a hyper-abundant society and impress someone enough to make them say yes to marrying you. Like can’t we all, or at least a significant number of us, get private jets when the world becomes a cornucopia?

No, We Can’t All Get Private Jets (Or True Love For That Matter)

Leaving aside the problem of the perceived value of private jets in the marriage market when most people have a jet, we would have another much bigger problem if the world had billions of private jets. So… No, as long as all of us are on planet earth, we can’t all get private jets. Even if let’s say the price of a private jet became $1000 in today’s terms in a 100 years and we had limitless clean energy to fly them around endlessly.

What’s that problem?

With this question, we have finally reached the most important limitation affecting marriage and fertility rates and which may also cause the coming job apocalypse: it’s the hard limits on the means of human driven production and transaction, especially as we go higher up the Maslow’s Hierarchy, into and beyond Maslow’s third level.

For example, one of those hard limits (with respect to one of the means of production: space) is the hard limit of airspace. Because of which we can’t have billions of private jets flying around the world (hard limit of consumption) even if let’s say with the help of AI we could achieve an astonishing level of precision-flying 24×7 to avoid mid-air collisions. The problem of our beautiful, blue sky (and a starry starry one at nighttime) disappearing behind the asses of the multitude of private jets would still be there.

What other hard limits exist?

Let’s analyze by taking up once again our previous example of falling marriage and birth rates.

First of all, there are hard limits on the number of marriages possible (a hard limit on the means of production). At the maximum it can be the number of adult people in the world divided by two unless couples in a marriage are simultaneously practicing polygamy i.e. unless both spouses are in more than one marriage at the same time.

Then there are hard limits on the time we can give to our partners in a single day. A day on earth has twenty-four hours. And since human beings have to spend some of that time fulfilling basic animal drives like sleeping and shitting, there is a limit to the time we can give to our partners even if we are truly, madly, deeply in love with them.

But partners in a relationship or a marriage don’t just want time; they also want your attention. They want you to focus on them while you are spending time with them. Since human beings can’t focus all the time even if the person in question was Adonis or Helen of Troy or Rani Padmini herself — we are human after all — the amount of attention you can possibly give to your partner is even less, often much less than the time you can give them.

But partners don’t just want attention; they want a demonstration of love, affection or romance, through words, touch, cuddling or in myriad other ways (e.g. giving them chocolates or flowers). Or they want caregiving when they are sick.

But partners don’t just want a demonstration of love, affection or romance, or just caregiving; they want to feel desired by having sex. Or they want you to boost their self-esteem by giving them compliments.

And so on…

As we can see, the higher we go up the ladder of Maslow’s Hierarchy, the harder are the limits we encounter on the means of production and transaction like time, sex, intimacy, attention, affection, compliments etc. To give an example of the epitome of such a limit (for which the transaction maps to the highest level of Maslow’s Hierarchy), think people who won’t marry till they find their one and only true love, their soulmate. Or people who spend their lives singing soul-stirring poetry to romance their divine beloved.

We are now ready to talk about the central thesis of this essay: why it’s possible, even likely, that the age of AI might result in a job apocalypse and some ways we might prevent that from happening.

But before going there, let me quickly summarize and repeat all that I have discussed before using the example of marriage and its declining rates.

12 Economic Rules For Marriage (And Why Its Rates Are Declining)

Thesis 1 (A general thesis applicable to marriage): Production and consumption happen when there is possibility of a transaction and a human job exists when all three happen with the involvement of a human on the production side. Of these three, for the continued existence of a human job or birth of a new one, the important conditions are therefore production with human involvement and transaction at a feasible price

Thesis 2: Marriage is a bundle of transactions between partners and the most common and important job in the history of the world. As a corollary, traditional marriage especially — for most of history — has been mostly an economic arrangement between the husband and the wife

Thesis 3: The means/currency of such transactions map crudely onto Maslow’s Hierarchy. Also, because of chronic and widespread poverty and scarcity, the transactions that map to the last two levels of Maslow’s Hierarchy — physiological and security levels — have overwhelmingly dominated the set for almost the entire human history. And of course, the poorer you were, the more the lower level transactions dominated in the set.

Thesis 4: Marriage has been one of the history’s biggest example of price control and artificial lowering of prices (set by the society) to enable transactions/weddings to happen. That’s how arranged marriages worked. And if you decided to rebel, they could excommunicate you or even feed you poison to keep other market participants in line.

Thesis 5: In the last two hundred to hundred to fifty to twenty years, the job of marriage has seen a significant price inflation as society slowly removed its price controls and people started marrying for love beginning with England and the US, then in the western hemisphere, and then to a lesser extent in the rest of the world leading to disruption in the marriage market. Technology, social media, and economic and societal freedom and gender equality (by disrupting the traditional gender roles) have further disrupted the transactions, especially in the last twenty years, and especially the ones happening at the lower levels of Maslow’s Hierarchy.

Thesis 6: All of this has resulted in an increasingly escalating shift in the expectations of relationship/marriage seekers regarding the nature of transactions a good marriage should have: the price expectations have increased for all the transactions, and further, the expectations have shifted the transactions that are happening from lower to higher levels of Maslow’s Hierarchy to a significant extent. Essentially, due to technology and changing societal mores and individual expectations, the institution of marriage is witnessing the phenomenon of crowding out in the lower levels of Maslow’s Hierarchy, especially in the last two levels. People are being needed less and less to fulfill the transactions corresponding to lower levels; and when such transactions do happen, the average price at which they happen are gradually increasing. This has led to a drop in the marriage rates.

Thesis 7: Since from an evolutionary standpoint we are on an average more suited to both produce (and consume) stuff that correspond to the lower level of Maslow’s Hierarchy, the more the transactions needed to do the marriage job well move up the Maslow’s ladder, the less people feel able to cope with the demands of the job and refuse to apply for it. For marriages to happen now, human beings need to be capable enough, intellectually and emotionally, to fulfill the transactions corresponding to the higher levels of Maslow’s Hierarchy because that’s where a lot of the transactions have shifted. This is another reason that has lead to the drop in marriage rates

Thesis 8: Human history has mostly been a history of scarcity — often a history of crippling, devastating, gut-wrenching scarcity where the limiting factors were almost always on the side of production made much worse by the greed of people in power. For this and other evolutionary reasons, human beings understand and can deal with scarcity intuitively. We easily understand the limits of production. But we (and our bodies) have difficulty understanding and dealing with abundance (the recent dramatic rise in obesity in certain parts of the world is a good example). Transaction and consumption being a limiting factor in an abundant world — instead of production — is therefore a counter-intuitive situation which the current minds, even the best of minds, are struggling to grasp. So we keep creating issues trying to deal with abundance the way we deal with scarcity.

Thesis 9: The limits of consumption show up in the marriage market in two different ways with different properties and ramifications: limits pertaining to the lower levels of Maslow’s Hierarchy and limits pertaining to the upper levels

Thesis 10: Human production, transaction and consumption have both asymptotic and hard limits depending upon the context.

Thesis 11: At the lower levels of Maslow’s hierarchy, transaction and consumption have both hard and asymptotic limits.

Example of hard limit: Homo Sapiens can have sex or orgasm only so much, say unlike Panthera Tigris or Sus Domesticus. Or as explained above, we can’t all get private jets.

Example of asymptotic limit: It’s great to go from zero to one TV, and from one TV to three or four large screen TVs in your house. But even the richest people in the world don’t have a hundred TVs in their mansions.

As a corollary, the more the material abundance in the world, the less is the perceived value of a transaction related to any specific material thing in a marriage situation; and the less the role of the lower-level, material needs in persuading someone to take up the marriage job. In other words, if you are starving, and someone offers to feed you regularly if you marry them, you are far more likely to say yes to them — no matter their other qualities — than if you are well-fed and someone offered to take you to a Michelin star restaurant regularly (unless the fine dining aficionado had other qualities to go along with being that rich).

Thesis 12: This is probably the most important thesis of them all. At the upper levels of Maslow’s hierarchy, beyond level three, a lot of human production, transaction and consumption have pretty strong hard limits stemming out of the fact that human beings have only 24 hours in a day. Intimacy, attention, affection, compliments, care-receiving are some of the obvious examples if we talk about relationships and marriage.

Ok, I know I have been repeating this like a clockwork for the previous seven thousand words, but we are now finally, truly, undoubtedly ready to tackle the central thesis of this article: the coming of the age of AI may result in a job apocalypse if we don’t do something about it.

Lord Of The Clockwork Orang

Homo Sapien is a strange animal, a walking, talking, eating, drinking, complaining, judging, laughing, fearing, hoping, crying, loving, hating, singing, dancing, murdering, happening, fucking ape.

Like all other living things, Homo Sapiens desperately wants to survive, 99.9% of the time. But it is also the only animal that sometimes seems to mystifyingly defy the evolutionary drive, the only animal that at times deliberately doesn’t want to reproduce. And also perhaps the only animal that occasionally and randomly (not instinctively like some other animals) even refuses to survive and kills itself.

Why? Because this ape wants more than just survival or to spread its genes. It wants intimacy, attention, affection, commitment, compliments, fame, faith. It wants love. It wants to be happy. It wants to sing. It wants to dance. It wants not to suffer from loneliness or heartbreak, or die from it. It wants to be acknowledged. It wants to understand the universe. It wants to hate. It wants to dream. It wants meaning and purpose. It wants hope. It wants to envy. While wanting to dissolve itself into the infinity. It’s perhaps the only animal that can be perfectly described in two words: it’s complicated.

Homo Sapiens is also significantly better than other animals at problem solving to get what it wants from the environment it exists in. To solve its problems (and to fight over or negotiate or share with other members of its species the resources available), this ape has discovered or invented a multitude of problem-solving ways over its history of existence: frameworks, systems and institutions, tools and techniques, religions and ideologies. Fire is one such problem-solving way. Agriculture is another. Marriage is third. Hammurabi’s code is fourth. The Middle Path is fifth. Capitalism is sixth. Liberal Constitutional Democracy is seventh. A Limited Liability Company is eighth. Computers are ninth. Artificial Intelligence is tenth. And so on.

Sapiens seems to have done a pretty good job of problem-solving over last two hundred thousand years as a species. Evidenced by the fact that despite running into hiccups now and then, this ape has multiplied to over 8 billion members as of the year 2022 CE (another framework it has invented to measure the passing of time).

The problem-solving ways and systems Sapiens has invented to solve problems are across multiple domains and often exist and function simultaneously in an interconnected way. Some of them are of course mutually exclusive, like Liberal Constitutional Democracy vis-a-vis Communism, or all major organized religions vis-a-vis polyandry. But many are not, for example, marriage and capitalism and democracies and LLCs and computers can and do co-exist together: imagine a startup that sells a computer software which helps you plan your marriage and honeymoon in accordance with the demands of your legal and socials systems of the liberal constitutional democracy to which you belong.

Occasionally, Sapiens runs into a meta problem when the problem-solving way or system the ape has invented stops functioning as well as it had functioned before. Usually this happens because of the occasional arrival of disruptive new technologies also invented by the ape (technology is another name for the ape figuring out how to use nature’s laws to serve its own ends, sometimes to its own detriment).

Because Sapiens is a social being, almost every method or system or institution it has invented, functions because of some underlying economics. Economics is essentially the system and study of ways in which the ape produces, transacts and consumes tangible or intangible stuff with other members of its species — stuff that maps to one of the levels of Maslow’s Hierarchy as discussed earlier. Economics is also how disruptive new technologies usually destabilize existing systems built by the ape and make them inefficient by their own prevailing standards (which could be — depending upon a specific ape’s point of view — a better standard or a worse one compared to the new standard, like the dynamics between tools of birth control and traditional marriage system). Technologies do that by changing the existing underlying economics of the systems.

When such a thing happens, members of the Sapiens species react with fear/anger, indifference or giddy acceptance based upon whether the new technology hurts them, does nothing for them, or helps them/makes them rich. Over a period of time the ape usually, irrespective of its initial reactions, adapts to the new technology and redesigns its institutions and systems to function as optimally as possible in tandem with the new technologies.

This cycle has happened time and again in the history of Sapiens: technology determines economics, and economics determines pretty much everything else that happens in the ape’s life because it shapes the frameworks, systems and institutions the ape needs to survive and occasionally flourish.

Technology in short, is and always has been, the lord of the clockwork orang, even though the orang may be its creator.

But, till now, despite displacing a certain number of orangs from jobs now and then — like when it made calligraphers, or coachmen, or weavers, or telephone operators, or photo studio workers obsolete — technology has never truly been the orang’s competitor. Because, till now, technology never talked. Not even like a stochastic parrot.

As a species, Homo Sapiens has possessed one fundamental distinction between itself and other animals or apes or any other entity on earth, why it has occasionally thought it has been made in the image of god. Because it alone had language. Because it alone could talk. Because it alone could write wacky poetry, deliver devastating puns, draw breathtaking beasts.

Well, not anymore. And that’s where the problem begins.

You might be getting bored, so let me tell you a joke.

Pythagoras, Descartes And Einstein Walk Into A Bar And Yell “E = m*(a^2 + b^2), Therefore We Exist”

No, not really. This is just a cool sub-title as I like cool sub-titles. But Descartes and Einstein indeed have something to do with the point I am about to make now.

In his legendary papers on ‘Special Relativity’ in 1905, Einstein first proposed what is now the most famous equation in the history of physics: E = mc^2. Basically Einstein showed everything in our universe, including all the atoms comprising all the matter in the universe, was equivalent to energy. Matter was essentially condensed energy. Since Sapiens is also a collection of atoms, as a corollary, it’s also condensed energy.

About two hundred fifty years before Einstein’s papers, Descartes, troubling himself over what really made Sapiens unique and made Descartes Descartes, declared: I think therefore I am. According to him, we human beings were not just atoms and energy; there was something different about us. Because we could think. Through words, through language, we could process data, shape information, invent ideas. We were special.

As it turns out, we may or mayn’t be special, but the information, language, ideas we produce in our brain aren’t. Our ability to think doesn’t make us special.

A chain of thought experiments, theoretical propositions and inferences and real life experiments, beginning with James Clerk Maxwell (Maxwell’s Demon), and followed by other brilliant scientists like Ludwig Boltzmann, Leo Szilard, Erwin Schrodinger (What is Life), Léon Brillouin, Ilya Prigogine, Rolf Landauer, Jacob Bekenstein, Stephen Hawking and many others, has advanced the hypothesis that information is nothing but negative of entropy (negentropy), and like matter, is equivalent to energy. In this view, information is essentially the fourth building block of the universe — along with matter, energy and spacetime (if we exclude dark matter and dark energy for now). We live in a universe which is a sum-total of atoms, energy and bits (information) and exists in a web of spacetime to fulfill the inexorable diktat of the second law of thermodynamics which is entropy maximization of a closed system. That’s all.

Life is a similar thing, and exists for the same purpose — a sum-total of atoms and bits running on energy existing in a web of spacetime to fulfill the demands of the second law of thermodynamics. And which is again not different from the form or purpose of non-life, except that life is more efficient than non-life at entropy maximization (exactly why it evolved out of non-life).

So Descartes should have said instead: “I am more efficient at entropy maximization than that chair — by ingesting information (negentropy) from my environment in the form of food, water, air, and expelling waste material dissipating energy in the process — therefore I exist. And incidentally, I can think too.”

Anyway, Sapiens has a non-life competitor now, an intelligence that is artificial. Which, like the rest of the universe, is a combination of atoms and bits and runs on energy while existing in spacetime, but worse: like Sapiens, it seems to talk and think like Descartes. And soon, if the current speed of advancement holds up, might think and talk better than Descartes.

Sapiens, the clockwork orang, may be screwed.

A Bumpy Ride To Utopian Joblessness

Now that we have established the possibility that AI may not be so different from us — the same old combination of atoms and bits fueled by energy — the possibility that it might take away our jobs from us begins to feel more real, isn’t it?

Now I am not saying AI will soon become human-like or surpass us. (And I am not saying it won’t either). I am far from an expert in this field, evolution has had a billion years to work on us, carbon and silicon are two different substrates etc etc. I am just saying there exists a fundamental similarity between human intelligence and artificial intelligence as the work of scientists from Maxwell to Hawking has told us. And we should be prepared for surprises either way, especially given the way things are going right now.

Let’s take a look at the key arguments of Marc and others, why they think catastrophic job losses won’t happen. As Marc says in his essay:

“But, using the principles I described above, think of what it would mean for literally all existing human labor to be replaced by machines.

It would mean a takeoff rate of economic productivity growth that would be absolutely stratospheric, far beyond any historical precedent. Prices of existing goods and services would drop across the board to virtually zero. Consumer welfare would skyrocket. Consumer spending power would skyrocket. New demand in the economy would explode. Entrepreneurs would create dizzying arrays of new industries, products, and services, and employ as many people and AI as they could as fast as possible to meet all the new demand.

Suppose AI once again replaces that labor? The cycle would repeat, driving consumer welfare, economic growth, and job and wage growth even higher. It would be a straight spiral up to a material utopia that neither Adam Smith or Karl Marx ever dared dream of. “

Marc explains how, till now, technology has replaced people from jobs only temporarily, not permanently, as new type of jobs, often better ones, arrived to take place of the old jobs. He refers to Hazlitt’s article to make his point who in turn gives us the powerful examples of stocking industry and cotton-spinning industry to make his.

Marc and Hazlitt are right, at least historically. In the past, when machines displaced people from jobs, they also often simultaneously brought down the cost of production by orders of magnitude. This happens primarily for three reasons. One, machines are often orders of magnitude faster and less error-prone than humans. Two, machines need to be paid in initial investment, energy, and maintenance while human beings need to be paid for their time which is usually costlier. Three, the friction of managing machines is much less, sometimes close to zero, compared to the friction of managing human beings.

With costs coming down drastically — e.g. the cost of producing a page of written material or a book after invention of the printing press — prices also drop drastically making demand for the product shoot up. A large number of new jobs, far more than the number that was lost, are created to fulfill the additional demand. Not only that, such technological efficiencies both free up and increase the capacity in the economy creating demand for more, often unforeseen products and services through second or higher order effects, e.g, the vast increase in the size of the publishing industry led to the creation of a new job called literary agent.

Then where am I disagreeing with Marc?

The first mistake Marc makes in his argument is extrapolating from history: he asserts his thesis was true in Hazlitt’s time, so it must be true now. But we have already seen how technology, economy and political changes have brought significant upheavals in our systems and institutions — using the example of decline of marriage rates — making our time quite different from Hazlitt’s. Theses that were true in Hazlitt’s time mayn’t be true now.

The second big mistake Marc makes is forgetting that his entire paragraph can remain unchanged except for a couple of small additions/deletions in the last line, and the situation would still lead to job apocalypse. His mistake originates from forgetting to differentiate between production/consumption and a job where a human is involved.

If I had to rewrite the paragraph for him and make a small edit in the last sentence: Someday in the future we are living in a cornucopia where rate of economic productivity growth is absolutely stratospheric, far beyond any historical precedent, and prices of existing goods and services have dropped across the board to virtually zero, such that consumer welfare has skyrocketed along with consumer spending power, and new demand in the economy has exploded so that a small number of entrepreneurs are creating dizzying arrays of new industries, products, and services, and employing as many AI and as few humans as they could as fast as possible to meet all the new demand. It’s been a straight spiral up to a material utopia that neither Adam Smith or Karl Marx ever dared dream of.

So, given powerful enough artificial intelligence systems, a material utopia can exist without much involvement of humans in producing that utopia in the form of jobs.

Marc can still say, what’s the problem as long as we do have a material utopia. Humans can after all exist by only consuming; we are animals aren’t we? And does an animal, like say a deer, always produce the food it consumes?

But then Marc would have to agree to UBI which he doesn’t seem to be in favor of now. Because humans would need a bit of money to pay for goods and services no matter how ridiculously low the prices in the coming material utopia. Plus I have a strong feeling UBI won’t solve a large number of problems that may arise in our material utopia.

Anyway, before we arrive in that not-so-certain material utopia, we have to traverse a long, bumpy road. And it’s important to take stock of things that might happen while we are getting there.

From Open Atoms To Closed Bits

There are a few important lessons we have learnt from the 12 economic theses for marriage that also apply while we are on the road to wherever the age of AI is gonna take us, whether we are on the road to utopia or we are on the road to perdition.

The first lesson is that technology till date has often pushed people out from jobs, that from a transaction/consumption perspective mapped to the lower levels of Maslow’s Hierarchy (e.g. tilling fields or fetching water or weaving baskets or making arrows), to jobs that mapped to the higher levels of Maslow’s Hierarchy (e.g. teaching, design, sales, marketing, HR, customer support, management, entertainment etc). Also, technology has also mostly pushed people out from jobs where from a production viewpoint they primarily manipulated atoms (physical objects and stuff), into jobs where they primarily manipulated bits (data, information, stories).

While the two ways technology has pushed people out from traditional jobs are not the same, there is a high overlap between the two ways: jobs that from a transaction/consumption perspective map to the lower levels of Maslow’s Hierarchy often primarily involve manipulating atoms; while jobs that map to the higher levels of Maslow’s Hierarchy often primarily involve manipulating bits. Manipulating bits in this context doesn’t only mean jobs related to computers and information technology though that came much later in human history; it means jobs related to any work that deals with manipulating information, language, and ideas: data, words, emotions, lines on a paper etc. For example, Buddha or Socrates or Michelangelo or Gandhi or Einstein or Charlie Chaplin had jobs where they primarily manipulated bits.

A lot of jobs, like caregiving or nursing or painting or songwriting or movie-making, are a significant mix of the two: you need to manipulate both atoms and bits simultaneously to do a good job. Also, a huge number of bit-manipulation jobs are in industries whose end-to-end goal is atom manipulation, e.g. making/selling Coca Cola or F-16s.

There are a couple more nuances here. Even when we talk about manipulating atoms and bits, there are jobs where atoms and bits can be manipulated relatively passively as per narrowly defined, clear, step-by-step instructions. For example, if we talk about jobs where we manipulate atoms step-by-step using narrowly defined, clear instructions, they would be making a pencil or solving a Rubik Cube. And if we talk about jobs in which we manipulate bits step-by-step in a similar fashion, they would be the ones in which we create, read, manipulate, update or delete data using step-by-step logical rules e.g. a cashier who counts notes and gives the required sum to the customer, or an entry level accountant who makes simple accounting calculations on her notebook and passes the data to her superior, or the data entry operator who enters data into the computer as per clear instructions etc.

Then there are jobs which are also accomplished by manipulating atoms and bits, but in ambiguous situations where you actively need to make complex decisions to achieve an optimal balance of multiple objectives that clash with each other. Michelangelo sculpting Pieta is an example, involving manipulating atoms in a creative, ambiguous situation. So is the design of CCTV tower by Rem Koolhas. So is figuring out plumbing leaks within your walls. And what is an example of manipulating bits in ambiguous situations to achieve multiple objectives? Jobs where we need to analyze, debate, brainstorm, evaluate, decide, create, synthesize, teach, persuade or entertain — like writing Kamasutra, delivering Gettysburg address, or creating Schindler’s List. Or starting a company named OpenAI only to make it closed-source.

That’s from the production side. How is it from the transaction/consumption side?

Michelangelo sculpting Pieta is an example of atom-manipulation (from production side) that from a transaction perspective maps to multiple levels of Maslow’s Hierarchy, while mapping to the highest two levels from a consumption perspective. (Michelangelo in return for making Pieta got money as well as self-esteem, fame and spiritual self-actualization, and maybe even a ticket to heaven; while the guy who commissioned Pieta, the consumer Cardinal Jean de Bilhères, probably got a significant boost in self-esteem as well as admiration from others and perhaps spiritual satisfaction even). CCTV tower design by Koolhas is also an example of atom-manipulation that maps to multiple levels for both transaction and consumption. Figuring out plumbing leak maps to multiple levels from a transaction perspective while mapping to the lowest level from a consumption perspective.

On the bit-manipulation side, writing Kamasutra maps to multiple levels from both transaction and consumption perspective; while making Schindler’s list maps to multiple levels for transaction and the highest two levels for consumption. And so on.

But how this is related to the topic of job losses?

Let’s go back to the example of cooking that we discussed earlier when talking about failing marriage rates.

Anyone Can’t Cook Ratatouille, Yet

Cooking — let’s say making a French dish like Ratatouille — is primarily atom manipulation on the production-side which maps to multiple levels of Maslow’s Hierarchy on the transaction/consumption side. But when compared to stocking manufacture — though they are both primarily atom-manipulation — cooking is different in multiple significant ways — the reason why we haven’t been able to automate Ratatouille manufacture the way we automated stocking manufacture more than a hundred years ago and maybe why there are no inspiring animation movies about cute members of other species making beautiful embroidered stockings that remind you of your mom’s embroidery.

One, the mechanical process of cooking is far more ambiguous and complex than making a stocking, because there exist thousands of dishes and tens of thousands of ambiguous ways to make them while there are only few ways to make stockings. Two, wearing stockings is not emotionally significant or an important familial ritual the way cooking is. Three, you would have no problems buying a three-year-old packaged stocking while you are very unlikely to buy three-year-old packaged food. Four, machines can make stockings at a speed which far outpaces humans but machines are unlikely to make tasty Ratatouille much faster than humans. Five, if you were obsessed with stockings and if you had money, you may decide to buy 1500 pairs of stockings and use and throw them at the rate of four a day for the next 365 days just for the fun of it (don’t try this by the way; we are yet to reach the AI cornucopia and even if you had the money for it, our mother earth can’t handle that much waste). But even if you were the biggest fan of Ratatouille in the world and missed your mom’s Ratatouille like crazy, you are unlikely to eat Ratatouille 4 times a day for breakfast, lunch, evening tea and dinner for the next 365 days.

And so on.

All these ways in which cooking is different from stocking manufacture, and more, has prevented machines from taking over human jobs till now. But with the rise of LLMs and more advanced AI, I think things are about to change.

The Day Androids Will Dream Of Electric Ratatouille

This essay is in danger of becoming a short book. So I will have to make things boring by compressing a large amount of information in the coming paragraphs as I move at a faster pace. Please forgive me.

Machines haven’t taken away jobs from human permanently in past times because two things have happened when machines have taken over a certain set of jobs, e.g. stocking manufacture: one, humans started consuming more of the thing so new different kind of jobs arrived; and two, the economy expanded in other unforeseen ways so even more new different kinds of jobs arrived. So things after a brief bout of pain returned to being better than normal.

But what if all the new jobs were also taken over by machines? And human production (occasionally) and transaction/consumption (more often) finally began hitting its limits? (At least until we enter a hybrid society and the day arrives when androids finally start dreaming of electric Ratatouille and start turning into consumers too).

Marc will say that won’t happen. And I sincerely wish that’s true. But as we already know simply wishin’ and hopin’ doesn’t always led to things (or weddings) happenin’.

I will now outline 12 different ways machines have impacted jobs, and taken over jobs. Plus why machines haven’t taken over some jobs till now but may take them over in the future rendering much of humanity jobless. Which may not be so bad if it’s the kind of joblessness that allows us oodles of time to try and sabotage our best friend’s wedding the Julia Roberts way, unless we have simultaneously reached a state where no one is marrying anyone anymore.

Rage Against The 12 Ways Machines Can Take Your Job

Like we saw with relationships/marriage, machines displace people from jobs in two ways. One, when they displace people from production and crowd out people from jobs mapping to lower levels of Maslow’s Hierarchy while pushing them into jobs up the Maslow’s Hierarchy from a production as well as transaction/consumption perspective. Two, when this leads to a situation where eventually jobs in the upper levels begin to hit the limits of production (occasionally) and transaction/consumption (more often) limiting the number of jobs available.

With the recent arrivals of LLMs, a third situation has begin to develop where machines have started crowding out people on the production side even from jobs that map to higher levels of Maslow’s Hierarchy.

Let’s first talk about the ways in which machines take away jobs by displacing Sapiens from production, whether from jobs that map to lower levels of Maslow’s Hierarchy or jobs that map to higher:

1) Day Zero of machine taking over a human job was perhaps the day the wheel was invented. That’s the first way in which machines take over jobs done by Sapiens — by being much more efficient as a means of production. Cars are another example. Guns are a third. The industrial revolution has a plethora of such examples.

2) Stocking manufacture or cloth spinning automation or printing press or railways is the second way machines take over human job, by not only replacing humans, but replacing them en masse. However, the limitation is that the product in question must be one for which mass production is acceptable. Most things these days fall in the acceptable category — our homes are full of mass produced stuff — except for a few things, e.g. certain kinds of food.

Till the arrival of computers, these two were the only ways in which machines took over human jobs — which were jobs limited to atom-manipulation. And as explained earlier, the machines posed only a temporary threat to overall number of jobs after which things returned to being better than normal in the long run.

Then came the computers and the era of machines doing bit-manipulation arrived.

3) This famous pic of Margaret Hamilton standing next to the software that put human beings on the moon is a good example of how computer took over many human jobs that involved bit-manipulation or some combination of atom and bit-manipulation. It made typewriters and typists obsolete; it made traditional cameras and photo studios obsolete; it made traditional telephones and telephone operators obsolete. But beyond that, it made things possible that perhaps may not have been possible otherwise. Like word-processing. Like social media. Like an astonishingly clear image of the universe. Like creating music without musical instruments, 3D movies, or development of vaccines in less than a year.

Summing up, this is how machines have replaced humans in various jobs till 2022, often for the better.

This was the case till stranger machines arrived in our world that could talk and paint and sing, even if just like a stochastic parrot.

With machines starting to talk, paint, sing etc., we have entered a new era in the history of humanity which may lead to a significant change in the nature of the relationship between technology and Sapiens. An era in which there are many more ways in which machines will start replacing Sapiens in various jobs.

4) One way that has already started happening — displacing people from jobs, ranging from ideation to content creation to graphic design to marketing and software development (and which might replace me as a songwriter too) — is that machines are now doing the same old bit-manipulation but of a much more complex kind. This new way of bit-manipulation involves not only the execution of step-by-step logical commands like earlier computers, but beyond that creation of — for all practical purposes provided you know what you are doing — original content, even creative stuff that was previously solely the domain of humans. Even if the work the machine does is nothing more than pattern matching to generate coherent and contextually appropriate content (or ideas or plans), businesses and consumers aren’t likely to care as long as the job is getting done.

This is what I meant earlier when I said that with the recent arrivals of LLMs, machines have started crowding out people even from jobs that map to higher levels of Maslow’s Hierarchy.

One huge problem that I foresee here is with respect to entry-level jobs where people spend on an average 3-7 years of their time. This is the time a lot of freshers get trained and pick up the skills — work-related skills as well as behavioral skills — that are necessary to perform well in the world of work. With LLMs having the potential to gobble up a huge percentage of entry-level work in bit-manipulation jobs, how are the freshers going to get trained in real-life scenarios?

The situation becomes worse in jobs that involve atom-manipulation. Even if there is a lesser chance of AI taking over these jobs in the short-term (the long-term is still an open question), there are already many domains where robots have made their entry. Now almost all of atom-manipulation jobs require you to pick up specialized skills in your early years while working with your hands, so getting hands-on proper training at the entry level is supremely important and can be the difference between life and death.

But how are freshers going to get the necessary training in the early years if machines crowd them out, and if they don’t, how will those jobs function over long-term unless robots begin to do everything?

One adjacent bigger problem is this fits neatly into the negative fly-wheel framework. The more the freshers lack training and can’t do the job well, the more organizations will be tempted to find ways to automate the work so the need for freshers is eliminated. Over a period of time, the same situation will arise at the junior managerial levels with same issues and same outcomes. And so on and so forth till the organizations hit a limit and the world finds itself in a soup before some kind of course-correction happens.

5) LLMs can not only spit out original content (like expressing an emotion in the style of a famous artist) through pattern matching, they have also gained the ability of what can only be described as thinking. For example, based upon my Hindi lyrics, chatGPT here is telling me scene by scene how to shoot the music video for my song ‘De De Aam De’ (Gimme Gimme Gimme Mangoes) by matching the letter as well as the spirit of the lyrics of the song with the type of shots I suggested it and the village setting in which the video is being shot (I shot the video in my ancestral village in Bihar, India).

chatGPT has thus reduced the need for a storyboard/screenplay writer for short videos. Hook it up to a regular camera setup to relay step-by-step instructions to the camera; and you may not also need a cinematographer.

Contrast this thing with the machine that automated stocking manufacturing (or MS Excel). The stocking maker contraption could never truly compete with a person. Because though it could act by itself in a very limited, narrow and well-defined way once it was switched on and fed energy, it could never think by itself. It did exactly as it was told in a clear, step-by-step fashion; but it could never have told you what to do next in a complex, ambiguous situation by being able to think even better than you. By combining thinking and acting, even if they may work like stochastic parrots, these new type of machines seem to behave like real people and have become the first entity in this world that can truly compete with Sapiens.

6) A lot of people I have come across online, especially artists or content writers, are scared that the LLMs will take their jobs because LLMs don’t need to be paid. True, that’s one of the key reasons job losses will happen. But there is another big reason why LLMs will cause significant job losses while drastically shrinking the size of the firms by eliminating the need of people at junior as well as managerial levels, especially middle-managerial levels: through huge reduction in managerial time, stress and overall friction (search, negotiation, coordination etc).

For example, recently I was able to create a detailed report on the various risks faced by startups across multiple stages (how I started) by myself over a couple of hours, something that would have earlier taken me days to get it done with help of my team including a significant amount of time that would have gone in delegation, management and back and forth. I did this while simultaneously watching Netflix.

LLMs, especially advanced agents that may come soon, may create a situation wherein for plenty of tasks at entry, associate, junior and middle-management levels, it becomes more time-consuming and stressful to delegate tasks to others and train/manage/coordinate/QC to get it done (even leaving aside cost advantage) than do it yourself with help of AI. So that’s the path many competent people at higher levels in organizations will take lessening need for people at the lower levels, more and more, unless those people are super-competent.

7) Let’s go back to our example of cooking vs stocking and the dish Ratatouille and bring in a dystopian touch. In the eponymous movie, the fearsome critic Anton Ego is finally moved to change his opinion on whether ‘anyone can cook’ because the Ratatouille he is served by our movie’s protagonists reminds him of his mom’s cooking when he was a child. Cooking and eating, as the movie shows, can be an intensely emotional activity, akin to relationships.

Beyond cooking and eating, two of the central ways in which human being express and transact emotions to fulfill their emotional needs is through language and touch (which are also often tightly enmeshed with the act of remembering). And till we had a world in which only people could talk, we also had a world where for most people, only other people and stuff created by other people (if we leave pets and appreciation of nature aside) could fulfill your emotional needs.

But that’s not true anymore.

Before machines could talk to us in a seeming natural way, we were unlikely to accept them in roles and jobs that had a high degree of emotional component, because expressing emotions through language is a key part of being human (the movie Ex-machina brings out this difference in a stark fashion). We were unlikely to let robots enter our homes to do things like cook for us or care for our sick partner.

But what if along the lines of the movie ‘Her’, someone made a robot that talked to you like your mom while cooking, cleaning or taking care of you in a momly way? Would Anton Ego accept ‘her’ among the ‘anyones’ that could cook and let her enter his home to cook for him Ratatouille like his mom? Perhaps he would. (I had started writing a novel about ten years back in 2013 called ‘Eternal Starshine Of The Spotless Mind’ based on a similar concept. I think it’s time to finish it.)

Anyway, this is another significant way in which machines with language capabilities would take jobs — mapping to higher levels of Maslow’s Hierarchy — away from humans.

8) The eighth way machines would be allowed to take more and more of human jobs is tangentially related to the point made above. And it’s not about some enhanced capability of machines, but changing behavioral and cultural mores. As machines begin to enter and become a part of higher number of jobs becoming even more ubiquitous in our lives than they currently are, the psychological and cultural barriers to their acceptance will come down further at increasing speeds. Today many people may not accept a psychotherapist bot, but a generation later they likely will. And many introverts already do, even today, preferring bots to humans.

9) Machines with language capabilities will enter jobs, where they previously couldn’t, in another significant way. Before the arrival of LLMs, it wasn’t possible to create machines — at least low-cost ones that could be given to millions or billions of people — that would work not for the entire world, but just for you. Anything that worked for you, e.g. MS Word or Excel or your laptop or your phone, worked for the rest of the world in the same way.

To put it in business jargon, it wasn’t possible to create low-cost bespoke machines. For example, you couldn’t have a psychotherapist bot that was made just for you, tailored to your specific life history, your specific behavioral quirks, your specific hopes and dreams etc.

Now that will become possible.

A regular human psychotherapist has between 15-40 clients and sees upto 6-9 of them every day. That person needs to know and remember life histories and details of those many clients, and switch between those many clients while helping them. That person is also not available to you unless at a high cost for a very limited time with prior appointment.

Compare that to a bot made just for you, tailored to your specific life history, your specific behavioral quirks, your specific hopes and dreams, remembers everything about you, has infinite patience, is trained on pretty much every single psychotherapeutic intervention known to humankind, and is available to you 24×7 at almost zero cost. Who you might choose, at least on a regular basis?

In a large number of areas — like psychotherapy, teaching, movie-making, songwriting etc — the inevitability of creation of bespoke, thinking, talking machines, is one of the biggest threats to people’s jobs.

Also, this is another example of machines crowding out people from jobs that map to higher levels of Maslow’s Hierarchy because jobs where bespokeness is needed are usually jobs that map to the higher levels of Maslow’s Hierarchy, both from a production as well as transaction/consumption perspective.

10) Another limitation that currently prevents machine from taking over people’s jobs, as mentioned above, is their relative high cost. This high cost is bearable if we are mass-producing things that will also be consumed at scale e.g. a stocking manufacturing machine producing stockings. Or machines that are part of mass-producing TV or washing-machines, if we have to take the example of machines involved in mass-producing other machines.

But with cost of making machines going down, as energy becomes more and more abundant and cheap, it will now be possible to create low-cost bespoke machines tailored to you and your specific needs e.g. the psychotherapy bot.

That sums up number of ways machine will replace humans from jobs by taking their place in production across all levels of Maslow’s Hierarchy.

And we arrive at the final two ways in which machines will displace people from jobs. They will do that by creating a situation, similar to marriages, where we begin to hit the limits on stuff that map to higher levels of Maslow’s Hierarchy from the production/transaction/consumption side.

We Don’t Have All The Time In The World

11) Imelda Marcos, the former first lady of Philippines, was discovered to possess 3000 pairs of shoes after she and her husband stole billions of dollars from the Filipino people. Maybe, someday in the future, when we are living in the AI created cornucopia, a descendant of Imelda would be able to simultaneously possess 30000 pairs of shoes without stealing anything from anyone.

Which is a slightly excessive but handleable situation. 30000 shoes can be stuffed in a closet, if you have a big enough closet.

But not time; that you can’t. You can’t keep 30000 hours in a closet the same way you can keep 30000 shoes, and take out one or more hours for use later. Even if AI helps us extend our lifespans, and we all live till a ripe old age of 30000 years, we will still have just a moment at any given moment. Just an hour at any given hour. Just a day on any given day.

That’s a fundamental limitation of our universe which can’t be breached. This is something which even Adam Sandler can’t change, at least in real life.

So how many proper full-length movies you can watch in a given day? 30000? Nopes. 3000? Again nopes. 300? Naah. 30? Umm… no, unless they are short movies. 3? That’s possible, if like me you are in your high school and skipped school to happily go downtown to watch Independence Day, Nutty Professor and Eraser back-to-back in a single day.

But very few people in the world are jobless high-schoolers with movie-mania on a Ferris Bueller’s movie day out spree. And even if let’s say you were a movie addict, as long as you have a family, one movie a day is mostly hitting the limits of movie-watching unless you want your partner to start yelling at you.

But what about bespoke content you may ask? Like with psychotherapy bot, AI can give us the ability to create movies that are made just for us, catering to our specific taste, our specific emotional triggers and sense of humor, our specific hopes and dreams. That way even if 5% of current world population watches a new bespoke movie a day, that’s 400 million movies a day that need to be made to satisfy the movie appetite of the world. Which is like a bespoke movie heaven totally possible with AI! (If our population stays constant which is unlikely’; it’s likely to start decreasing by 2100).

Two problems with that thesis.

One, movies are often social experiences. Like watching sports, like sex, they are an experience best enjoyed with other people. Like you can masturbate as much as you want, but sex is best enjoyed (by most people) when two people are doing it together. Unless of course you are Caligula. Which is why movie theatres earn often more money from selling popcorn, hotdogs and soda than they make from selling tickets.

Two, even if let’s say people add social to movies by treating it like playing a video-game and co-creating it with friends, it will still be like doing a barbecue with friends. Not a business which gives people jobs. Yes, a lot of cooking and talking and laughing and having fun goes on during a barbeque, but all that is for self-consumption for you and your friends and doesn’t give anyone else a job except for the small number of people providing the ingredients or the utensils/grill. An equivalent entity, where movie-making is concerned, will be the small number of large technology platforms that will provide the movie co-creation platform/service, like Netflix provides streaming platform/service today. But how many number of people will that employ compared to today’s movie business, especially given all the other ways AI is going to take away jobs?

So what happens when AI gives us the capability of making 400 million movies every day for releases across theatres or on streaming? Will we suddenly develop the god-like ability to slow time down or even stop it in its tracks so we can watch all those movies?

No, we won’t. The same way even though the number of published books in the world increased from a little more than a million in 2000 to close to 5 million in 2020 thanks to technology making publishing easier (even before the rise of LLMs), the per capita time spent reading daily has slightly decreased from 2014 to 2017, And I am sure it has now decreased even further with the rise of short video apps like Tiktok and Insta Reels.

Which is pretty much to be expected in an attention economy if you think about it for more than a minute. With multiple kind of content jostling for our attention that has a hard limit imposed from outside by the unchangeable laws of the universe, and more stuff being innovated to compete for the limited attention pie, the attention that can be given to any particular format will obviously go down. And with the rise of LLMs, and the expected exponential increase in the volume of content across some formats for now, and all formats in the near future, the situation is only going to get worse.

This is a similar situation to where the rise of technology and changing social mores is making people self-sufficient enough to see marriage just as an option, not a necessity, while other relationships are starting to compete with marriage which faces similar hard limitations fundamentally arising from the finiteness of time and human attention.

12) We have now reached the last way how AI is gonna affect the availability of jobs for the worse.

There are two types of jobs in the world: jobs in which the distribution of income follows the power law; and jobs in which it doesn’t (at least till now).

An example of a job in which income distribution follows the power law is that of an actor. Or a startup founder. Or an artist. Or a novelist: 1% of novelists make 99% of the money; 99% of novelists make 1%.

Contrast that with the job of a plumber. Or a nurse. Or a bureaucrat. Or a doctor. Or even a chef (who doesn’t go on TV or write a bestselling book because those are power law jobs). While these jobs also have some degree of uneven distribution with regards to how much money people in these jobs make, the Gini coefficient is far less for these jobs as compared to jobs that follow the power law.

The non power-law jobs, from a security perspective, are much safer than power law jobs because there are far less situations where a plumber becomes a multi-millionaire while his friend is failing to make the rent. Not so with actors. Or startup founders.

You may have noticed a clear pattern in power law vis-a-vis non-power law jobs. Non-power law jobs mostly satisfy four criteria: they usually fulfill our needs corresponding to the lower-levels of Maslow’s Hierarchy; they are often service jobs that can’t be automated away usually because of the need for human interaction and touch (at least till now); they also can’t be scaled up using automation, i.e. machines can’t do it 20 times faster than a human even if machines took over the job; it’s difficult to create a personal brand in these jobs, and even if you do so, that doesn’t make much of a difference to your pay.

On the other hand, power law jobs mostly map to top levels of Maslow’s Hierarchy. And in power law jobs, your income will show a stark difference based on your ability to develop a personal brand that can take a significant share of the people’s finite time, attention and admiration, or not. If you can, you will become a millionaire or more; if you can’t, you will fail to pay rent and will have to move on to do something else.

What is AI gonna do? It’s gonna push pretty much all the jobs — whether based on time/attention/admiration or not — into the power law category. That’s because in our AI cornucopia, machines will be ubiquitous in all the jobs doing 99.99% of the work. And the small number of people who will make 99.9% of the money will be the 0.1% of people that still have jobs building and managing AI (at least till the arrival of AGI) and the top shareholders of companies that provide the machines doing all the jobs.

The Man, The Machine, The Meaninglessness

We have already discussed earlier how Sapiens is a strange ape that wants more than just survival or to spread its genes. Because this ape wants more than just survival or to spread its genes. It wants intimacy, attention, affection, commitment, compliments, fame, faith. It wants love. It wants to be happy. It wants to sing. It wants to dance. It wants not to suffer from loneliness or heartbreak, or die from it. It wants to be acknowledged. It wants to understand the universe. It wants to hate. It wants to dream. It wants meaning and purpose. It wants hope. It wants to envy. While wanting to dissolve itself into the infinity. It’s perhaps the only animal that can be perfectly described in two words: it’s complicated.

And this very ‘it’s complicated’ status humans have compared to other living beings may cause complications in the AI enabled cornucopia where humans have plenty to consume — at least at the lower levels of Maslow’s Hierarchy — but little to produce because most of the jobs have been taken away by AI and automation.

The character of Agent Smith expresses this rather darkly in the movie ‘The Matrix’ while ignoring all the breathtaking beauty, the awesome songs, the four-hankie tear-jerkers this complicated ape has also created when motivated in the right fashion.

So what now? What should we do to make sure AI doesn’t take all our jobs, to prevent meaninglessness in the life of an animal that even kills itself in order to feel that its life has some meaning?

Should we ban AI? Should we regulate it heavily?

Should we just let things happen as it has never really been easy or even possible to stop the flow of technological change, especially that which directly increases overall productivity and the possibility of making more money even if you think the change is bad? (The why is obvious if you understand the concept of Prisoner’s Dilemma. And no, nuclear technology is a bad counterexample vis-a-vis low or moderately powerful AI applications that will mainly be responsible for all the job losses; conceptually AI is quite different).

Especially one that makes it possible to usher in a golden age of material abundance while practically removing hunger, homelessness, disease and survival driven fights and wars from the world.

One may be tempted to ask why should people that can’t accept and live with the fact that the universe might really be a cold, meaningless, uncaring place — with nothing and no one being special or in any way meaningful — be considered more important than the people who are dying of hunger, cold, famine, diseases and wars today? And even if we are in the absolutely worst-case scenario where humanity slowly dies off as a species devastated by its meaninglessness, thanks to AI there may still be perhaps a much greener world where other living beings continue to survive, reproduce and play till something drastic happens or till earth gets swallowed by the sun.

I don’t know the answer to that. I am a walking, talking, thinking ape, and walking, talking, thinking apes — courtesy the ‘blind men and the elephant’ problem — have proven to be notoriously bad at trying to control the future using power; and have often done far more ill than good sometimes leading to unbelievable damages when prescribing inflexible rules and solutions in the name of god or righteousness or efficiency, also due to this pesky thing called ‘unintended consequences’.

To put it in short, diagnosis is easy, prescription a million times trickier.

So I don’t know what to tell you. Like Arjuna in the battlefield of Mahabharata, I am paralyzed.

A Happy Sisyphus In The Matrix

I do have a few thoughts nonetheless, paralyzed though I may be. And then again there is the question of X-risk or not-kill-every-one-ism as it is being referred to in layman-ish terms.

But this essay has become too long. So I will share my thoughts on both the above questions later in another, much shorter essay called ‘A Happy Sisyphus In The Matrix’.

Till such time, adios.