Predictive Models in Advertising

From weather models to advertising

Predictive models rely on a set of initial conditions to set them in the right direction. We see this in various settings, including weather models, flight simulations, differential equations in physical systems, and of course advertising.

The process of getting a converging solution takes time. Digital campaigns typically need a couple days to begin performing, especially in complex systems, such as DSPs, that have many parameters. Sometimes, changes in the targeting or KPIs midway confuse the algorithms and delay or completely destroy the convergence.

Initial conditions in AI

In the case of AI, it’s initial set of conditions is essentially facts and human observations. Without those, any generative algorithm is bound to produce garbage. In addition, flaws in those initial conditions ensures messed up responses that need to fact-check in the same manner in which academic publications are examined. This fact of course ensures that AI-generated content is not difficult to discover even though it may read well. However, it also proves that AI is merely a tool like auto-correct that has to be checked and used with caution.

New revenues for knowledge platforms

The importance of the initial conditions should allow data and fact aggregators, such as Wikipedia, Stack Overflow, Quora, and other niche providers to start monetizing their content in a fair and unbiased way. Perhaps Wikipedia will figure out how to generate revenue that remains unbiased towards any advertiser. In addition, many niche data depositories, such as Arxiv, will become increasingly important sources of facts. Given their clumsy format and the niche nature of their readership, AI could positive impact academic publications and open them up to a more general audience.

Regardless of how corporate AI efforts solve the initial condition challenge, fact-checking and AI-detection tools need to develop in parallel to ensure we don’t repeat and proliferate false information.