A Briefing for Boards on (Generative) AI

In this writing, I am espousing that we are at the dawn of a new supercycle in technology. A supercycle is a period of sustained economic growth that starts with a strong demand for something once nuanced that eventually becomes a commodity. Initially, demand for the resource of the supercycle can exceed what producers can supply, which can cause prices to rise before they eventually fall exponentially. Supercycles can last years or even decades. The last two supercycles in technology were the Internet and the microprocessor. Each supercycle has a defining moment when a flywheel product is introduced that triggers an arms race for something that was once nascent. With the Internet, the flywheel product was the web browser and with the microprocessor, it was the personal computer, both starting two arms races that resulted into two multi-decade supercycles.

Much like the Internet was nascent before the web browser and the microprocessor was nascent before the personal computer, artificial intelligence has been nascent until LLMs.

Flywheel products that have started arms races before ChatGPT tend to have three characteristics that trigger multi-decade supercycles. The characteristics of flywheel products are (1) they prove that significant risks can be removed from the investment category, (2) they offer a benefit for every business, and (3) they offer an improvement for every consumer. Assuming that the benefits of artificial intelligence to every business and every consumer are forgone conclusions, lets look at why capital will increasingly flow into artificial intelligence as an investment category.

Removing risks from investing in artificial intelligence

Until ChatGPT, most investments in artificial intelligence were knowingly risky. It required too much capital to generate a product that could service a large enough market to warrant putting the capital at risk. The risks involved in investing in artificial intelligences before ChatGPT were (1) too expensive to build, (2) too narrow in their resulting breath of intelligence, and (3) too imperfect to productize. ChatGPT proved that you can invest in artificial intelligence without these three risks. It proved that you can capitalize building really large models, you can expect to have a resulting wide breath of intelligence, and you can package and productize imperfection. The lessons from the kickoff of the artificial intelligence supercycle lie in examining how and why each of these three risks are being removed, and leveraging the learnings to do good while doing well.

Too expensive to build. The combined cost of cloud computing and powerful chips has fallen below a critical threshold. Thanks to significant advancements in Hypervisor technology, the cost of clouding is free-falling. ChatGPT was trained and delivered by investing under one billion dollars while IBM Watson consumed more than ten billion dollars over its lifetime. The cost of “large” models (language or not) will fall below one hundred million dollars by the end of 2023, and below ten million dollars by the end of 2025.

The “too expensive to build” risk is disappearing.

Too narrow breath of intelligence. The sophistication of the application of mathematics to build artificial intelligence models continues to increase exponentially.  Starting with the Hinton paper on Multilayer Perceptrons in 1985, to the Vaswani paper on Transformers in 2017, the artificial intelligence models have been moving exponentially from being good at one thing, to being good at many things. The breath of the intelligence that can now be built has progressed past a critical threshold. Foundational models can now be used to deliver a wide range of intelligences. By the end of 2025 the marketplace will be saturated with commoditized foundational models of text, images, sound, video, code, and documents with breaths of intelligences ten times wider than the models we use today.

The “too narrow breath of intelligence” risk is disappearing.

Too imperfect to be productized. The genius in OpenAI and ChatGPT is not only in the economic efficiency of training and the sophistication of the model architecture, it also lies in the productization, packaging, and distribution of the artificial intelligence. Before ChatGPT most bits of artificial intelligence seemed like sophisticated science experiments because of how they were productized, packaged, and distributed. Artificial intelligence pre-ChatGPT was productized to be perfect at something specific, packaged as software to be embedded into other software only by skilled engineers, and accessed through applications with complex user interfaces that need to be downloaded and installed on powerful machines. ChatGPT was productized to be imperfect by design, packaged as a chat application where the chat is a simple prompt provided by any consumer, and distributed through the Internet only requiring an internet connection and a device with a web browser.

The “too imperfect to productized” risk is disappearing.

Boards must be briefed on the changing nuances of the three risks discussed above to appreciate the impact and opportunity of the artificial intelligence supercycle to our businesses and markets. Investments will continue to flow into artificial intelligence as an investment category because it is getting less expensive to train increasingly larger models, the application of mathematics to stitch multiple model architecture together is improving exponentially, and the know-how to productized, package, and distribute artificial intelligence is improving and being democratized to product managers globally at breakneck speed.

With this backdrop, what actions should boards take?

Navigate the ship through the storm. No doubt the next 12-18 months will include peak hype and the trough of dissolution. Competitors will rush ahead, uncalculated acquisitions will be done, regulations will be integrated, and unknown unknowns will emerge unexpectedly. Weathering the storm requires immediate action to engage partners. Partners in the areas of (1) strategy, (2) ecosystem, and (3) execution. At the dawn of a supercycle, strategy becomes the most critical virtue of a corporation. Engage strategy partners, and invite them to be close advisors to the board and the executive team during the storm. The waters will be rough before we all get to smooth sailing on the other side of the storm.

Get the right people on the bus. The winners of the supercycles win on talent before they win their markets. Over the next three to five years the composition of talents needed to compete will change dramatically. Getting the right people on the bus will require that boards and executive teams tap into networks. Networks that open access to (1) talents, (2) partnerships, and (3) acquisition and divesture opportunities. As the supercycle takes off, getting deeper involved in talent networks will become paramount. Talent networks can usually be found around conference/event communities, geographic communities, and academic communities.

Starting with the future in mind. It is almost impossible to predict who the winners will be this early in a supercycle. Depending on when and how we look at things, at one point IBM would have won personal computing in the microprocessor supercycle, and at another point, AOL would have won web browsing in the Internet supercycle. Now we know that Microsoft won personal computing and Google won web browsing. Boards should tap into futurists. Futurists can be found in (1) thought leadership communities, (2) analyst ecosystems, and (3) technological historian niches. When a supercycle starts futurists can help us see the markets multi-dimensionally and offer us a high degree of optionality on the future instead of forcing us to bet on who and how it will be won prematurely.

Roadmap the business model evolution. The companies that win in supercycles tend to win because they took the time to reflect on business models, and lean into the thesis that there are business model evolution opportunities that will emerge in the midst of a supercycle. Boards should take the time to engage with world-class digital transformers to assist in road-mapping a business model evolution. Transformers can come in three types. Those that understand (1) the evolution of the technology supply chains, (2) how to take advantage of blue-sky economic opportunities, and (3) how technology marketplaces, ecosystems, and platforms evolve in supercycles. Digital transformers that have worked in multiple industries and geographies tend to be more valuable to boards and executive teams those that have not.

Prepare to compete with new currencies. Every supercycle changes the currency of competition. Artificial intelligence will increase the importance of trust as a currency in commerce. Boards must prepare to compete on trust, in addition to today’s table stakes such as experience and engagement; trust will take center stage. At the helm of corporations, boards will need the advice of ethicists to compete on trust. Corporate virtues such as (1) risk/reward calculations, (2) transparency, and (3) regulatory compliance, are some of the core tenets of trust. Ethicists usually do not assist in making ethical choices, but instead contribute to increasing the ethics index of every employee and hence the entire organization.

Summary

The answer to the question of whether or not we are at the dawn of a supercycle is not as important as we think. The price of being wrong on a supercycle is too high. Take advantage of the moment and we can win almost anything, miss the opportunity and we could lose almost everything. History is littered with very smart and capable executive teams that missed the timing on supercycles. In artificial intelligence, the price is as high as it was with the Internet and the microprocessor.

We must study and understand the nature and changing nuances of the artificial intelligence supercycle as it dawns. Investments will continue to flow exponentially into artificial intelligence because several critical thresholds have been met which are significantly reducing the risk of investing in the category. We must pay keen attention to the characteristics of the flywheel products, and learn to leverage the new best practices emerging in the disciplines to productize, package and distribute artificial intelligence responsibly to consumers globally.

I offer the five actions above for boards and executive teams (1) navigate the ship through the storm by leveraging partners, (2) get the right people on the bus by leveraging networks, (3) start with the future in mind by leveraging futurists, (4) roadmap the business model evolution by leveraging digital transformers, and (5) prepare to compete with new currencies such as trust by leveraging ethicists.

Richie Etwaru