Optimising financial processes

Posted on:

Which Flavour of AI Delivers More Value – Predictive or Generative?

It is always good to look at counter-cyclical arguments.

They help sharpen our critical thinking skills.

Many of you will know that I am fascinated by our cognitive biases and the corresponding characteristics of the “Hype Cycle” characterized by Gartner and the “Dunning Kruger Effect”.

Generative AI (GAI) has certainly captured the imagination, and investment, of businesses and speculators around the world.

To some extent, GAI has overshadowed the great progress and value delivered by more “classical” AI, and it’s focus on “data driven decision making”.

Eric Siegel described “classical AI” much better as “Predictive AI” in his recent article in in Forbes.

Whilst Generative AI attracts headlines, Eric argues that Predictive AI delivers greater value.

Which kind of AI should we focus on?

  • Generative AI, which produces writing, computer code, images, video and other content?
  • Or, Predictive AI, which focusses on data driven decision making in finance and accounting, procurement, marketing, fraud detection, risk management et al?

Eric suggests “rather than selecting an attractive technology and then searching for a problem, industry leaders advise to first identifying an important problem and then figure out how best to solve it.”

I couldnt agree more . . . 

University of Toronto management professor Mihnea Moldoveanu asserts “AI strategies fail because AI is a means, not an end.  

‘Do you have an AI strategy?’ makes as much sense as asking, ‘Do we have an Excel strategy?'” . . . .

The article argues that Predictive AI holds three main advantages over Generative AI.

  1. Predictive AI delivers higher returns than Generative AI by focusing on an enterprise’s largest-scale processes that consist of many individual decisions that are ripe for predictive optimization. UPS saves an estimated $35 million annually by optimizing package delivery planning with predictions of tomorrow’s deliveries.
  2. Predictive AI can operate autonomously, whereas Generative AI usually cannot. Whilst AI in general benefits from a “Human in the loop” (HITL), Predictive AI can be trained over time to drive decisions independently.
  3. Predictive AI is much cheaper and imposes a much smaller footprint than Generative AI. The machine learning models needed for Predictive AI are generally orders of magnitude lighter-weight than generative AI’s models. Large language models (LLMS), Generative AI models that produce text and code—normally consist of between 100 billion and 1 trillion parameters and are often trained over billions of pages, which consumes a great deal of energy. In contrast, predictive AI models often consist of only dozens to hundreds of parameters—rarely more than several thousand—and are commonly trained over only 100,000 or fewer learning cases. You can train it on your laptop.

Ideally, organizations should approach each operational problem with the most suitable technology to generate the greatest value.

This will probably result in us exploiting Predictive AI more, alongside Generative AI’s adoption.

Food for thought . . .

Eric Siegel’s Forbes article “3 Ways Predictive AI Delivers More Value Than Generative AI” is worth a 10 minute read, here . . .

I have talked previously about the “Dunning Kruger Effect” here . . .  and specifically related to Generative AI here . . . 

A brief perspective on “data driven decision making” can be read here . . .  

And, of course, Gartner’s famous “Hype Cycle” is illustrated below . . .

Thanks for reading . . . .