You as a commodity and GenAI as your reliable partner to get there.

In recent time we hear a lot of buzz about how generative AI is a commodity like say electricity and how it will revolutionize the way we live. People claim its a new Industrial Revolution we are facing. This makes it feel exceptionally important and heavily influences our decisions about adoption of the technology. The characterization of generative AI (in fact GPT based large language models and diffusion models) as commodity by its vendors and backers implies that the intelligence as implemented by generative AI systems is a commodity. Due to the possibility of large compute capacity buildout, given sufficient capital allocation, the intelligence becomes abundant and therefore cheap and highly available inviting everyone to outsource their own intelligence to AI providers.

So the story sold to the public is simple. There is an unprecedented transformation going on which changes everything and therefore means you should absolutely care and the result of it is the state where intelligence becomes a commodity which you can pull on demand just like you pull the electricity from the grid. So you should embrace and adopt that stuff now, effective immediately.

Or shouldn't you? In fact the statement that it will revolutionize the industry sounds rather incredible given our knowledge of computer science and cognitive psychology. But also empirical observations so far do not seem to support this statement, especially if we consider the generative AI value chain holistically and take a closer look at costs, potential legal and regulatory risks, data sourcing issues and of course actual performance of the SOTA models. Note though: Statistical learning theory tells us, that given sufficient amount of compute (model capacity) and data, it is in principle possible to build a system which approximates the process which generated that data, so for example human thinking and problem solving, sufficiently well. The quality of the approximation and associated costs may or may not be sufficient for anticipated breakthroughs and benefits. This will become clear in the next years and we do not further go in deeper discussion about it in this note.

So, we just assume artificial intelligence is the commodity. We want to point your attention to the implication, that in this case, using generative AI as the major value added activity or a core element in your job or business makes you and your work a commodity! Bit obvious, but apparently not to everyone. Now let's see whether you should care about this one.

So why do we think this may be a problem for you and your business. Lets recap first what a commodity is and why you might care, that you and your work become one. A commodity is a good or resource, whose individual instances are considered interchangeable by the market. Furthermore, a commodity is widely available on the market. As example consider electricity. It does not matter who sells it to you or where it has been generated and therefore it is interchangeable. There is also sufficient number of producers. This has an implication, that in a commodity market the competition is dominated by price. This usually leads to low margins. So if your work is a commodity, then your profit will certainly perish. This is also possible even if today your business or job throw strong margins. The genAI may create new conditions, which will disrupt your basis of competition.

You may ask, why would my work considered a commodity if I use genAI to accomplish it? Well, the simple answer is you substitute your labor and services your business runs on by a commodity service. That substitutes your unique way of doing business by a standardized way via a commodity service. So you loose differentiation and note, this is not unique to AI services. Lets try to dig a bit deeper into statement that genAI is a commodity. The message the AI labs try to convey by this is the AI replaces human intelligence at fraction of cost. The emphasis is on cost reduction, suggesting the capability remains the same as human or even better. So if humans can do stuff such that it is differentiated and AI substitutes human intelligence then whats the matter? We think it omits an important point which gets clear if we consider the way the genAI works and being delivered to you.

GenAI is basically a sampler from a complex probability distribution over token sequences, which has been obtained in an elaborate way from tons of scraped data. Naturally, such a sampler will provide average answers which bear little potential for differentiation. Moreover, this model is shared by all users. With other words the stuff you get out of genAI is the stuff your neighbor will also potentially and easily get. There is no consistent way to differentiate on genAI content and your work, which relies on genAI, will be an average at best. Next issue is the access to genAI. It is accessible to everyone who are willing to pay for it and the prices are really low. There are even completely free offers. So entry barrier is practically non-existent and that means there will be a lot of people like you providing the same average work. This is the commodity situation we described earlier. The abundance of average work will destroy the margins of those who provide it. And don't be too optimistic about sinking costs of genAI, there are reasons to believe, that they shall actually increase. So while genAI is like electricity, substituting your essential activities, labor in particular, by genAI makes your business like electricity - a commodity with fierce price competition and pressure on margins.

So now, what to do? Should you stop using genAI? Well it is up to you, but here is the advise: Do not rely solely on genAI, you have to be wise in how you use it (I know, it is not quite clear now what it exactly means, considering all the challenges). Be strategic about what you do, understand your basis of competition, differentiate your work, try at all costs to avoid getting into commodity market with fierce competition. For now I think the best way to do it in the context of our discussion here, is not to make genAI the core part of your work.