Optimization and Generative AI

Since #ChatGPT was released, the world got crazy with #GenAI and #AI went from taboo to buzz.

Gartner for IT reminded us in 2024 that AI is wider than GenAI.

All those AI techniques are very useful alone but we can also hybrid them. In this article we deal with mixing optimization and #genAI

In 2022 my first experiment with OpenAI was to solve my famous zoo and buses example with #ChatGPT and #CPLEX : I used a LLM to generate an #OPL #CPLEX model.

Within #watsonx #AI IBM you can do all this since #CPLEX is part of #watsonx

And then within watsonx you can solve this. (OPL and python)

This is very useful but with watsonx code assistant we can do even better. Within #VSCode 微软 we can write: "300 kids need to travel to the London zoo.The school may rent 40 seats and 30 seats buses for $500 and $400 . How many buses of each size do you use to minimize cost? Write a docplex cplex mip python model for this"

The watsonx code assistant generated model is

and the result is

CPLEX and optimization deal with scarcity. Optimization experts are scarce resources and with generative AI building model will be easier. Of course, this is naive and the need for human experts won't vanish but I guess human experts will focus more on hard issues which really require deep expertise. Any thought about that ?

NB:

Generating optimization model is the main use case when mixing optimization and generative AI but let's not forget other successes:

• Generate data with synthetic data
• Document optimization model
• Translate model (From OPL to python for instance)
• Foundation models time series
• Ease extract structured data from IT systems
• What if analysis through natural languages queries

INFORMS ROADEF #ORMS #CPLEX DecisionBrain Datatilt Cubewise Monaco Digital