Valohai ML platform logo

What if Edison Was a Data Scientist?

The Inventor

Thomas Alva Edison stormed out of a classroom at age seven - after being called "addled" - and never really returned. Homeschooled by his mother, what he lacked in formal education, he made up for in perseverance. All 1093 patents later - for such wonders as the microphone, telephone receiver, stock ticker, phonograph, movies, office copiers, and incandescent electric light - we can conclude that his mother did something right.

Young Thomas Edison

Thomas Edison at the age of 14

To invent, you need a good imagination and a pile of junk.

Thomas Edison

One of the first books she made him read was the School of Natural Philosophy, explaining how to perform chemistry experiments, which transformed young Edison into a mad scientist. He reproduced every single experiment in the book, turning their cellar into a chemistry lab. Amazingly, the house never burned down.

But it wasn't easy. Edison worked hard to make money by selling magazines on the train to finance bottles, chemicals, wires, and other required hardware. Even after securing a respectable monthly salary, skipping meals became commonplace to afford more equipment as his appetite for more elaborate experiments grew. Edison even hustled permission to use the train's baggage car to feed his obsession. Fast-moving trains and volatile chemicals do not mix well, and unlike his family home, the baggage car did not survive it.

The Invention Factory

Adult Thomas Edison

Inventing is never easy.

I have not failed.
I’ve just found 10,000 ways that won’t work.

Thomas Edison

His entire life, Edison was a pioneer, a first mover. Still, our romantic idea of a lone mastermind fools us. Edison didn't invent the light bulb - the very archetype of an invention - or most of the other 1092 patented items with his name. They were slowly developed by teams under his supervision, slightly improving well-known ideas that originated elsewhere.

Instead, Edison's greatest invention was the process of invention itself.

Early on, he realized how the sheer logistics of experimentation limited progress. Securing the funding, acquiring the materials, and doing all the manual work himself, capped the rate of experimentation. To break through this bottleneck, he would need to industrialize the process of invention - like his best buddy Henry Ford industrialized the process of car manufacturing - and build an invention factory.

The layout of the Edison’s invention factory complex c. 1914

The layout of the Edison’s invention factory complex c. 1914

Thomas Edison and senior staff at Menlo Park invention factory

Thomas Edison and senior staff at Menlo Park invention factory

In one of his letters to investors, he vowed to create at least one small invention every week and a large, society-changing one every six months. This would be achieved by offering a team of master experimenters and craftsmen a shared facility with every tool, science book and material known to man on-premise with a clear process to follow.

For example, his invention factory conducted 10,296 experiments before finalizing the design for its nickel-iron storage battery, which charged twice as fast as any other battery on the market. The fact that we know the exact number of experiments speaks volumes.

Every team meticulously documented progress or even a whiff of an idea. The invention factory even hired special bookkeepers to gather and organize the trail crumbs of the mad scientists. Edison didn't want to waste any resources on redundant experiments or litigating ambiguously documented patents, both mistakes that had cost him dearly.

A page from Edison's notebook c. 1881

A page from Edison's notebook c. 1881

Innovation puts pressure on tooling and infrastructure, too. Edison’s father - a master builder - cast unique flooring to stabilize the sensitive instruments, and for the incandescent light, they designed new kinds of electric meters, light switches, and insulated wiring before any actual experiments could even take place. Whether you are a 19th-century innovator or a 21st-century ML pioneer, the flexibility to improvise is a must.

Now, let’s take a time machine to the year 2022!

1920

1930

1940

1950

1960

1970

1980

1990

2000

2010

2020

2022

Introducing
The ML Factory

Thomas Edison and senior staff at Menlo Park invention factory

Genius is 1% inspiration and 99% orchestration.

Thomas Edison reborn as a Data Scientist

Just like Edison would ask new questions in the context of the physical world, as pioneers of machine learning, we are asking the computer questions no one ever asked it before. Going a well-beaten path is easy, but being a first-mover implies a high initial failure rate.

The modern version of young Edison experimenting alone in the cellar is a data scientist with a laptop. A single laptop can do one experiment at a time and mostly sits idle, just waiting for someone to pour coffee on it. The modern version of the invention factory is a team of data scientists armed with a modern MLOps platform.

For an ML pioneer, the rate of experiments equals the rate of success.

Experimentation rate is optimized by experimenting faster and experimenting in parallel. Edison made the teams go faster by having everything they needed on-site and parallelized it via facilitating hundreds of experiments side-by-side, like a 19th-century grid search.

The modern invention factory makes experiments effortless by removing the friction of manual technical busywork and allows massive parallelization in the cloud by having thousands of compute instances as readily available as a single one. Improvising custom tooling is also easier for the 21st-century pioneer surrounded by modern APIs and the safety of a containerized environment.

There is no need to hire an army of bookkeepers to follow Edison’s every step in the modern ML factory, either. The knowledge base automatically builds up into the MLOps platform whenever he embarks on one of his manic invention streaks. Edison, the data scientist, is eventually more interested in proving that the model works in production than gearing up to fight the patent wars like his 19th-century counterpart.

The Layout of an ML Factory

Productivity Stimpaks

Smart Orchestration

Pioneers experiment at scale. Valohai enables data scientists to launch thousands of experiments on cloud or on-prem environments with zero DevOps support.

Productivity Stimpaks

Developer Core

Pioneers utilize the latest technologies. Valohai’s developer-first approach makes adopting new frameworks and integrating with any existing tools a breeze.

Productivity Stimpaks

Knowledge Repository

Pioneers build on knowledge on top knowledge. Valohai allows teams to improve continuously by automatically storing all experiments, metrics and models in a shared repository.

What It Looks Like in Practice

I’m in. I need to know more.

You’ll be in good company.

AiforiaBasemarkFloyGreenSteamIceyeJFrogKonuxLevityLifebitMaytronicsNeurolabsOCELLOMHUOnc.aiPath RoboticsPreligensReveal BiosciencesSAMASharperShapeSpendeskSyngentatwitterYousicianZesty