New Tech produces economic and political disruption at scale. This was the effect of Naval fleets to project power, electricity, the steam engine, and rail.

AI can be the most disruptive tech of the 21st century because it is a general-purpose tech (not limited to say hauling loads). AI also has zero distribution costs (being digital). But the value AI brings to the world is distributed unevenly. Let's consider AI's geopolitics.

Geopolitics is the study of the effects of Earth's geography on politics and international relations. It provides context and improves decision making at the macro (and sometimes not so macro) level. Geopolitics focuses on states and countries, but for the geopolitics of AI, Companies are a better unit than countries. Why? Because in AI we care about talent and data. Both are mobile and malleable (when compared to mountains and rivers) but still have some geo-specificity (German data may not be very useful in North Korea).

According to Kai-Fu lee's 'AI superpowers' book, there are seven AI giants (the equivalent of nation powers): Google, Facebook, Amazon, Microsoft, Ali Baba, Tencent, and Baidu. Beyond those seven, there's a dramatic drop in companies' AI capabilities. All world companies outside the 'seven giants' put together have fewer AI capabilities than any single giant. Let's call them the AI-poor.

The gap between AI-rich and AI-poor is widening extremely fast, because AI is a winner-takes-all game: if one company cracks the autonomous vehicle problem, it wins the entire market. There’s no point for an AI-poor to ‘clone’ it, and have an ‘also ran’ technology that is second best: if ‘number one’ gets a successful trip 99,9% of the time, and ‘number two’ gets only 95%, that makes ‘number two’ not viable. A car company will buy ‘number one’s’ product because we are talking about human lives. They need to buy the technology from an AI-rich. We will explore the consequences of this technology dependency in this article.

What makes these AI-rich companies different? The seven AI giants have (1) Talent (2) Data and (3) infrastructure. Plus the seven AI giants are all platforms. The Platform business model is the most successful business model in the 21st century. A platform is a business model that creates value by facilitating exchanges between two or more interdependent groups, usually consumers and producers. In order to make these exchanges happen, platforms harness and create large, scalable networks of users and resources that can be accessed on demand. Platforms create communities and markets with network effects that allow users to interact and transact. Platform companies have far higher profits and growth than any other. For example, Google around 2016 (according to a VP of search, personal communication) had 21% yearly growth and  20% profit. These numbers were similar for Microsoft, and completely out of range for most non-platform companies, particularly enterprises. Platforms are also very difficult to displace, their network effects building an effective moat.  

AI, with its zero distribution cost (digital), plugs perfectly into platforms. The seven giants use their AI and data advantage to try and enter every industry: health, HR, finance, retail, banking. It’s important to understand that even before we consider AI, these seven giants are completely different animals compared to the incumbents that reign in every vertical. The incumbents are often ‘linear companies’, not platforms, and often they are not very far in the digitalization scale. Their business model didn’t change from what was successful in the 20th century. This makes it very hard for them to take advantage of AI.

But is AI really so valuable?

To answer this question Take DeepMind, a British AI company that was bought by google in 2014. DeepMind's algorithms saved enough of Google’s data center electricity costs to pay back the purchase price in the first year. Since then DeepMind has been in the news because they solved problems that most considered impossible, including beating the human champion at the game of Go.

DeepMind’s last breakthrough helps finding 3D structure of proteins. Scientists have identified more than 200m proteins but structures are known for only a fraction of them. Traditionally, the shapes are discovered through meticulous lab work that can take years. Alphafold, DeepMind's algorithm, managed to find structures and nearly two-thirds were comparable in quality to experimental structures. This was one of the grand challenges in biology. Alphafold matters because proteins define and power ALL life functions. It would vastly accelerate efforts to understand the building blocks of cells and enable quicker and more advanced drug discovery.

Andrei Lupas, the director of the Max Planck Institute for Developmental Biology in Tübingen, Germany, said he had already used the program to solve a protein structure that scientists had been stuck on for a decade.

'AI-poor' companies and countries have no choice but to buy AI from someone

Because the 'winner takes all' dynamics of AI, it's tough for an 'AI-poor' to create state-of-the-art AI in-house. They often don't have talent, data, or infrastructure. There’s no way around: they have to import AI.

The 'seven giants' strategy is to sell AI through automation. According to Kai-Fu Lee, if AI is the new electricity, they are the utility companies. They are installing 'the grid' to satisfy demand. Let's use Google as an example of how this strategy pans out.

Google has Tensor Processing units (TPUs), a technology that makes AI computation far cheaper than the previous generation (GPUs). It's in Google's interest that any 'AI-poor' consumes AI through Google cloud services. Because this is an eminently scalable business, google needs to make AI as easy to consume as possible. They are investing in how to simplify usage, expand use cases, educating businesses. The goal is that any AI-poor country or company can consume AI/cloud from them even with limited resources in talent or infrastructure.

Most of the AI-poor countries have political leaders who understand how risky this tech dependency is and try to minimize it at all costs. This is where Element.AI (And Canada!) could have been just the ticket!

How Canada, and Element AI, could have maintained an equilibrium

Canada has the most deep learning researchers per capita in the entire world, and some of the best labs.

The country saw their ridiculously abundant researcher pool as an opportunity. They funded companies and changed visa policies to make sure Canada would become an AI-rich country. One that would provide services in AI to the AI-poor countries. Stopping the brain drain to the US was an extra layer of goodness.

Canada has excellent relationships with the G7. All G7 members (excluding the US, which hosts 4 of the 7 giants) have a skill deficit in AI. They would have been happy to buy Canadian AI, as they cannot buy AI from the US nor China: it would be a geopolitical mistake to be more in their hands than they already are.  Canada could have become a competitor to the two AI superpowers (US and China) if it played its cards right. This is extraordinary: most other countries would kill to be in that position as their industries' revenues dwindle down.

The spearhead of Canada’s strategy was Element AI. A company with about 500 employees world-class-level at deep learning, rivaling the concentration of talent in DeepMind. A company with Joshua Bengio, one of the Godfathers of the field, as a cofounder. A company with >300 million USD in funding over 4 years. Element AI became the self-appointed representative of Canada’s AI sector. A company that couldn’t fail. Or could it?

Element AI, designed to avoid Canadian talent leaving for the US, was just bought by ServiceNow, a Californian company. Element AI was ServiceNow’s fourth AI acquisition in 2020, following Loom Systems, Passage AI, and Sweagle.

Having the full support of the government didn’t help finding a business model that worked. And they withered in the vine. This is a historic moment. The implosion of element AI marks the end of an era. One where Canada could raise to match the AI sophistication of the US and China, and serve the AI-poor countries.

Effect for the AI-poor: the weak, now weaker

There were two opportunities for an 'AI-poor’' country to catch up: DeepMind and elementAI. Deepmind went to google, enlarging an already tremendous advantage in AI and data, and the UK must have lamented their decision to let that happen ever since.

ElementAI went to a US company before they could produce anything of significant value, but they were still a gigantic whale that could have ‘fed’ an AI-poor country for a decade.

The chances of another AI company of that caliber forming anywhere outside the US and China are virtually zero. The ‘AI-poor’ world has missed the last opportunity to create a stronghold. The G7 (minus the US) will have to attach themselves to one of the two AI superpowers, in a deal that would get progressively worse as the value AI provides grows compared to traditional industries such as manufacturing.

I can see two possible scenarios.

Pessimistic scenario

  • Even with lots of open source libraries and models, the 'AI-poors' manage to not get value out of AI on their own. Their dependency on the big 7 keeps going up.
  • Every AI-poor country is a tech colony of the US or China. And at this point there's no way out (after DeepMind and ElementAI are off the market, and local talent is not enough to defend any possible local AI or data advantage).
  • The market value of '20th century economy' (manufacturing) keeps going down, profits are slim, and the purchase power of industrial-era countries dwindles versus that of AI rich countries.'
  • AI nationalism and cyber neo-colonialism explains many geopolitical transactions in the 21st century. Political realignment depending on who provides your AI.
  • Talent and data centralizes on a few companies in the US and China.
  • AI-rich countries will keep invading vertical after vertical (example: apple health) and the incumbents will be powerless (Example: Alphafold from DeepMind and similar discoveries displace pharma and biotech companies. From this point on, all major discoveries happen in one of the big 7 AI giants).

Optimistic scenario

  • The seven AI giants produce open source libraries and write papers. Data stays in the AI-poor country of origin. While there are no researchers, there are enough tinkerers in AI-poor countries to benefit from advances and don't need to buy AI from the giants. Still big dependency on libraries: if the AI-rich companies stopped sharing open source code, papers, and pretrained models, AI-poor countries would go back to a precarious situation.

  • AI value moves from invention to implementation (by an army of tinkerers that are not world-class but ‘good enough’ once the AI-rich solve the big, scary problems). Mass electrification exemplified this process. Following Thomas Edison’s harnessing of electricity, the field rapidly shifted from invention to implementation. Thousands of engineers everywhere began tinkering with electricity, using it to power new devices and reorganize industrial processes.

  • This army of tinkerers fits more the 'incremental innovation' approach of Europe. For example, Germans are masters at iterating on an idea (someone else invented and published) till perfection. Germany didn't invent the car, but perfected production to a level that allowed Germany to gain the most economic value from the invention. If Germany would use the same approach to AI, this would be a reason to have hope. For an example take the German company Deepl. Deepl does machine translation better than google translate (probably on a shoestring budget), which is an amazing achievement: How many companies can beat Google at their own game? And in AI no less? Sadly companies like Deepl are the exception. Germany's economy is still very much focused on manufacturing, with no signs of refocusing so that companies like Deepl would be more common and competitive.

  • Germany benefits the most from perhaps the most coveted of all AI-powered products: fully autonomous vehicles. And surprisingly, companies like Uber (which IPO’d on their supposed benefit from autonomous cars) actually don't benefit that much from the technology! When/If this devilishly complex problem (driving) is solved, it completely changes Uber's business model (not owning cars) into a different one (owning robot cars) that’s much less profitable and where they have no expertise in. German manufacturers though can own a float of cars they rent instead of selling. And as such, they are going to benefit the most when/if self driving software becomes viable, even if it’s not invented in Germany. This would imply an ‘AI-poor’ company (a german car manufacturer) transitioning into a platform business model, which would just in itself be a gigantic improvement.

  • The giants are happy to make AI easy to use so that more companies can buy cloud services, which is how they make their money. The direct uses of AI on their consumer products (gmail autocompletion, google photo search) is not making money. But it helps selling cloud infrastructure.

  • Germany still requires well-trained AI engineers, the tinkerers of this age, to apply and tweak deep learning to each vertical. These engineers are not easy to attract when the US and China have the culture to pay a big multiple for their services. What if they matched the salaries to attract the talent from AI-rich countries? Then the local workers in pretty much every other profession would feel humiliated and there would be social unrest. So growing their home talent is the best option: create engineers that accept the local salaries but are still good enough to make progress. There are no signs that Germany is investing in education (anything digitalization is very much 2nd class even after years of investing; let alone AI). But investing in AI education appears to be an extraordinarily lucid strategic decision that must be about to happen.