Stylized non-PR facts about generative AI:
- genAI lowers dramatically the cost of knowledge work outputs.
- genAI knowledge work output is undistinguishable in quality from domain expert output by low-information consumers.
- genAI knowledge work output is of worse quality than domain expert output.
(These facts are either already accepted by the reader or impossible to argue within the scope of these quick notes.)
Therefore, inside organizations:
- It becomes competitively necessary for any domain expert producing knowledge work for a non-expert consumer to use genAI.
- The quality of internal knowledge work inside organizations becomes lower.
- The marginal impact of expertise inside organizations becomes lower.
- It becomes rational for organizations to invest less in expertise.
There are similar dynamics in the external market:
- Products with a larger component of genAI knowledge work input are perceived by low-information consumers as having equivalent quality.
- It becomes competitively necessary for companies selling to low-information consumers to maximize the use of genAI in their products.
- Low-information consumers are shifted to lower-quality equilibria; cost dynamics depend on market competitiveness.
- However, high-information consumers perceive products using genAI as having worse quality. This drives companies serving them to not use genAI, making them less competitive with genAI companies.
At the simplifying limit the market coalesces into extremes:
| High quality, high price, no genAI | |
| Low quality, low price, full genAI |
These are made stable by the fact that genAI is brittle: it can seem high quality for a given sample, but fail catastrophically at any arbitrary point in the future. So
- You can't cut costs for a given level of quality by adding genAI for something sold to high-information consumers.
- Not using any new available genAI tool for something sold to low-information consumers is not competitive against somebody that does.
Every model is wrong, etc. Yet rather than talking about AI exposure by sector and activity it might be more relevant to talk about intrinsic divergence dynamics inside them. The cost/quality tradeoff is always present, but by lowering costs so much and exacerbating the difference between high-information and low-information consumers genAI pushes market differentiation into almost discrete poles of high-information consumers served by high-expertise, no-genAI companies, and low-information consumers served by low-expertise, genAI-intensive companies.
It's arguable that deciding which one to try to be is the most important deliberate decision an organization can make right now. (Similar dynamics and choices apply, mutatis mutandis, to individuals.)