Why AI is hiring problem, not just a technology one

This article summarizes an episode of Alex Kantrowitz’s video series featuring Joelle Pineau, chief AI officer at Cohere.

Joelle Pineau, chief AI officer at Cohere / Photo credit: Alex Kantrowitz Podcast

Joelle Pineau, chief AI officer at Cohere, believes using AI is a hiring and team structure problem, not just an automation task. She thinks the technology will help new employees and put mid-level workers’ jobs at risk.

Current models struggle with memory and reasoning

Pineau sees gaps in what AI can do today. These gaps stop models from reliably doing multi-step tasks.

• Memory: The main challenge is not storing information, but finding the right piece of information at the right time from large datasets.
• Reasoning: Today’s methods are inefficient. Models can’t plan at different levels. They can’t switch between big goals and small steps.

Current AI struggles to connect a high-level plan with the small steps to get it done. This inability to switch between big ideas and small details stops models from breaking down big goals into smaller tasks.

Pineau notes, “The part that the reasoning models don’t do is to plan at different levels of temporal granularity… In technical terms, we call it hierarchical planning. That’s really hard to do, that decomposition and keeping the information relevant as you go back and forth.”

World models are needed to build reliable agents

These problems with reasoning limit what AI agents can do. To build useful agents, models must first understand and predict what will happen when they act.

• Purpose: A world model helps an AI predict how its actions will change its surroundings. Pineau says this is necessary for building agents that act on their own.
• Types: She points out two kinds of models, which are physical world models for robots and digital world models for web agents that do things like online banking.

Before an AI can be trusted to act for us, it needs to understand cause and effect. This internal ‘world model’ is the difference between a useful tool and an unpredictable one.

“World models are absolutely essential when you want to build agents,” Pineau argues, “because these agents are going to take actions which are going to change the world. You want to be able to predict these effects… whether you’re building robots… or agents getting deployed on the web.”

Companies are not using AI’s full potential

While researchers are making breakthroughs, companies are not fully using existing AI technology. Pineau sees a gap between AI’s capabilities and how businesses apply it.

There is a difference between AI’s potential and its practical use. Many paying customers choose “good enough” models that balance performance and cost, leaving better AI unused.

Getting AI to work inside a company is another hurdle. Company rules and disorganized data systems block agents from getting the information they need.

AI creates a new power balance at work

Control over AI is a strategy for company safety