top of page

How Not to Invest in Artificial Intelligence

Last month we had the chance to participate in the Creative Destruction Lab (CDL) event that took place in Toronto. This is an interesting program that combines three partners: investors that seek new investment opportunities, startups that seek funding, and the University of Toronto, which enhances its MBA by having its students take part of this very dynamic and exciting part of the economy.

It was fun, it was exciting, and we met a lot of very smart people. We also got to experience the odd process of how venture capital seeks new investments at the earliest stage of a company. They call this ‘pre-seed’ which is funny, because the term ‘seed’ was conceived for investments it meant to be ‘the earliest possible, when you have nothing but an idea’. Now we have pre-seed.

CDL aims to find good ideas lacking the business piece to create a company. It is no surprise then that previous business experience is not very valuable. On the other hand, a PhD on something related to AI is very valuable, because it indicates that the idea is probably on sound technical ground.

That was surprising to me; as an entrepreneur and investor, I would predict bad odds for startups with only an academic achievement from the founder. But the weirdest part is that this approach of valuing ideas over experience is widely used right now. There is a lot of funding going to AI and it does follow that pattern: investing in a PhD or a Professor of AI. Or even better if there are a few founders, all of them PhDs.

I’m guessing this is done with hopes that having such a team will present an attractive acquisition to a tech giant. Do some research, make sure it is noticed, and hope Google will call!

I can’t say there is something wrong with that, but I personally would never start a company hoping it will sell fast. A company is essentially an entity that creates a product that is sold to customers. If the product makes a difference, customers will pay for it. The transaction of selling the product creates a win-win situation: the customer is happier with the product than the money, while the company uses that money to improve that product and expand, and finishes with growing profits.

In my simple definition of a company, the product is really the foundation. No product, no company. Poor product, bad company. AI is basically a software product. An AI company must then have a good software product specialized in artificial intelligence that eventually makes a difference on something in which customers find value. Research alone won’t do this. Universities are good places for research, but are companies?

Software comes in all sort of types and shapes. I imagine that software can be categorized as ‘complex’ if many of the following apply:

  • The code base will be large (lots of lines of code)

  • The requirements are partially unknown and changing (modularize and encapsulate business logic)

  • Performance is a hard requirement (execute fast)

  • There is a batch component and a real-time component (backend / frontend)

  • Use of many tools, frameworks and libraries is a must (nobody can start from scratch)

  • More than one programming language is needed

  • There is a database (too much data to keep in memory all the time)

Okay, so AI is software, the complex type, and this will be the product of a company that wants long-term success. What kind of skills do we need to make this kind of product? Surely PhDs can’t hurt. Here is my take:

  • 90% is architecture and design

  • 9% is domain knowledge

  • 1% base science.

If I imagine this as food, I’d say carbs is the design, proteins is the domain knowledge and the last 1% is vitamins and minerals. This is not to say one should underestimate the vitamins! We can do with fewer carbs and proteins than ideal, but sustained lack of vitamins and minerals will kill us. Don’t forget to eat your salad.

In my last company, which made complex billing software, the 1% base science was accounting. That was the foundation of everything that was built after. But that did not mean we needed to hire an accountant. Once a few key concepts were understood, the bulk of the work was to know and understand how companies use billing (domain knowledge), and lots of architecture, design, and coding to get the product to do something useful.

If we had the accounting wrong, that would be disastrous. Yet, let’s not mistake quality with quantity.

If I needed a bridge built, I wouldn’t hire an army of physicists. Gravity is a key factor, sure, but to get the job done, I would hire architects and construction workers. The people who design and build the bridge use the work of physicists to do their jobs correctly, but they don’t need a physicist on staff. We trust these experts to do their job using the science developed by others, and they’d better have the physics right!

So, starting an AI company with founder who has a PhD but no business experience will make it tough to succeed. It is even worse if the idea is based on some yet-to-be-done research. The 1% science is very key indeed for the success of a product. If the 1% science that a product needs is not even well studied yet, then we are in trouble. Companies like Tesla and SpaceX are good counter examples, but we’d all agree that these are exceptions that make Elon Musk the awesome entrepreneur that he is.

I find it funny that when I started jBilling, nobody told me: “this idea won’t work, because the accounting research needed is already done and known.” But when I started Decisive AI, many told me: “you know, your idea doesn’t make sense, computers playing games has been extensively researched already.” Yes it has! And that is good, because when I start a product, I want the science to be solid. It is super challenging already to implement existing theoretical research into something practical.

A lot of capital is going to AI startups, simply because there are a couple of PhDs or a Professorship among the founders. They then go on and hire more PhDs and maybe an extra Professor. Imagine the meetings! I don’t know if we would ever want to hire a full time PhD for Decisive AI. There is no immediate need at least, as there is a lot of literature already out there (some decades old) on how to make computers intelligent so they can play games well.

These science heavy startups miss the fundamentals of product building. I’ve seen a few that don’t even have a specific idea of a product, yet have received millions of dollars in funding. But we have to keep in mind that product alone wouldn’t do either. A company is a healthy mix of many skills, like sales, marketing, administration, and most important: people management.

Create a great company culture with great people that create, sell and support a great product. That works!

bottom of page