The return of the apprentice

The death of the privileged Stanford graduating class.

Photo by Andrea Piacquadio

Introduction

In this essay, I will talk about a return to the apprenticeship system of the Middle Ages.  

In this new-old workplace, masters will work with apprentices and AI tools.  

The combination of masters, apprentices and AI tools will have a profound impact on society.

High-paying tech jobs will turn into normal-paying jobs with more opportunities for more people, and a better fit to our day-to-day needs.

The Arrogance of power

"Power tends to confuse itself with virtue and a great nation is particularly susceptible to the idea that its power is a sign of God's favor, conferring upon it a special responsibility for other nations - to make them richer and happier and wiser, to make them, in other words, more like itself. If God has a chosen people, whose theology is it that insists they must prevail over all others? This is the arrogance of power."

J. William Fulbright:"The Arrogance of Power" 

The book is a collection of essays that Fulbright wrote during the Vietnam war.

We are on day 54 of the war with Hamas. 175 hostages remain in Gaza in the hands of Hamas.

This war marks the start of a chapter that is likely to affect millions of lives around the world, and not just in Israel.  It marks the arrogance of the professional Israeli military establishment. It marks coincidentally, the arrogance of big American tech companies.

But my thoughts today are not about Vietnam or Gaza or on the future of Israel.

My thoughts today are on how AI will affect the future of work for the entire world.

Since the Industrial revolution, new technology has always resulted in job creation. 

When the steam engine was invented, new jobs were created for steam engine mechanics. 

When the first computer mainframes were released in the 1950s, new jobs were created for programmers, analysts, operators and maintenance technicians.

In the past 20 years, society has seen creation of an elite class of highly-paid and highly skilled tech workers feeding a $16 Trillion industry of mobile, cloud computing and online commerce.

In this essay, we’ll show how AI changes this equation forever.

1. New job creation from new tech is dead

2. The hard shift from head count to hardware

3. The shift to a master-apprentice-AI tooling model of work

The future of work: The return of the apprenticeship system

New job creation from new tech is dead.

We are seeing rapid acceleration of changes to the job market  in the past 3 years since OpenAI released  GPT 3 in June 2020. 

More than 224,000 workers in the US-based tech companies have been laid off in mass job cuts so far in 2023 up by 240% from 2022 (93,000 US tech employees were fired in 2022).

There is a seductive train of thought that goes something like this:

Sundar Pichai  announces that Google just fired 12,000 people in order to focus on AI (and compete with Microsoft-owned OpenAI and Amazon-owned Anthropic) .

Maybe Google hires 12,000 other people, or they will go work for the competition?

It turns out that you don’t need that many people to develop a GPT (Generative Pre-trained Transformer).  Anthropic currently employs only 160 people.  

Google announces that they would invest up to $4BN in their competitor Anthropic.   Google has a Google AI group with large investments in hardware and software, but not that many people.

Google is hedging their bets it seems.

What happens to tech head-count?

OpenAI (the original creators of ChatGPT) has 770 people.

If we extrapolate (a dangerous exercise at any time) from Anthropic - we see a 2 orders of magnitude reduction in the workforce needed to develop and operate AI systems

The world requirements for AI are satisfied by less than 1500 people.

The hard shift from head count to hardware

OpenAI spends approximately $700,000 daily to operate ChatGPT. This is money spent on cloud computing; mostly on special-purpose AI hardware made by Nvidia.

In Aug 2023, Nvidia said heavy demand from cloud computing services and other customers for chips to power A.I. systems caused revenue for its second quarter, which ended in July, to jump 101 percent from a year earlier, to $13.5 billion, while profit surged more than ninefold to nearly $6.2 billion.

Nvidia is like the guys selling shovels to the miners during the California gold rush.

They don’t care if the miners find gold.  

As a matter of fact - better the miners should not find gold. As long as companies like OpenAI and Anthropic are training GPT models, they’re selling shovels. A lot of them.

The shift to a master-apprentice-AI tooling model of work

The apprenticeship system was most prominent during the Middle Ages, particularly in Europe, from around the 5th to the 15th centuries. This model was a cornerstone in trades such as blacksmithing, carpentry, masonry, and various other crafts.

In this system, a young person (apprentice) would learn a trade by working under the guidance of a skilled employer (master) for a period, often several years. The apprenticeship was not only about skill acquisition; it also included aspects of moral and social education. The model ensured that skills, techniques, and knowledge were passed down through generations.

In the late medieval period and into the Renaissance, this system was formalized through guilds, which were associations of artisans or merchants who oversaw the practice and quality of a trade in a particular area. The master-apprentice model began to decline with the advent of the Industrial Revolution in the late 18th and early 19th centuries, as mass production and industrial manufacturing processes changed the nature of work and skill acquisition. 

As new job creation dies, the apprenticeship system will be reborn, aided by AI tools and the small number of master engineers that will survive by virtue of their design and human communications skills.

AI tools based on GPT models create a huge jump in productivity

Hundreds of AI tools (now called GPTs) boost productivity in almost anything.  Writing articles, code development, analyzing data, transcribing videos, producing voice-overs.  I subscribe to a daily newsletter that surveys AI tools based on GPT.  

I lost track of the creative ways people build applications using the OpenAI API.

One of these tools is Copilot.

Software development has changed forever with Copilot

GitHub Copilot is an AI-powered code completion tool that assists developers. It’s like having an expert programmer, with a perfect memory setting next to you at your desk.

The 1 thing that AI cannot do, is understand the needs of human beings.  

Not understanding the needs of humans means that it is not capable of translating needs into a product that helps people.

Not understanding human needs, means that AI is worthless at effective communication.

Master engineers excel where AI is worthless.  

Master engineers are good designers and communicators, capable of talking to human beings,  understanding their needs and leading teams to develop solutions for people.

In the summer of 2023, I worked with 2 apprentices and ChatGPT to develop a system for measuring performance of oncology clinical trials using baseball metrics.

The 2 apprentices; Daniel Lis and Hannah Berman were engineering college students going into their junior year in Boston and spending the summer in Israel in the Masa (Onward) program.

I developed the requirements with Dr. Denise Lepley who is a PhD in molecular biology and director of clinical data science at a West Coast biotech.  I consulted with Dr. Moni Shahar who is PhD in theoretical computer science on the algorithms.

The first 2 weeks I had calls with Denise in the evenings. In the mornings, I worked with Hannah and Daniel to develop a working system that processed over 600,000 lab records for over 3,000 patients in 4 colorectal clinical trials.   Assisted by Copilot, I developed the data pipeline that took from the raw data into a Snowflake cloud database in less than a week. In the last week of the project, Hannah, assisted by Copilot developed an app that visualizes the trajectory of a patient in a clinical trial; how well the drug works, how safe it is, and how much liver toxicity the patient had.

We developed a complete system in 6 weeks, for $2,000. 3 years ago it would have taken 3 months and cost $120,000.

Conclusion

We’ve seen in this essay that new job creation dies in the age of AI.  

We’ve seen how hardware replaces headcount in the age of AI.

As new job creation dies, the apprenticeship system will be reborn, aided by AI tools and a small number of master engineers that will survive by virtue of their design and human communications skills.

This will have a profound impact on society and education.

The privileged, high-paid tech class will be erased.  

High-tech will become low tech.

There will be more options in the workplace for more people at any age.

Our children and grand-children will have better, and healthier options, working as apprentices and contributing to society.

Instead of working full-time at being popular.