AI Redefined commits to better training models under new CEO

AI Redefined (AIR) is the developer behind an open source Cogment framework that makes it simpler for humans to train virtual agents created using AI models. Today the startup named G. Craig Vachon to be its CEO in place of Dorien Kieken, who will now assume the role of president.

Vachon is the founder of seed investment firm Chowdahead Growth Fund and has raised more than $1.6 billion in private equity, with investments in more than 30 companies across seven countries. He also serves on the Board of Directors for Yseop, provider of a natural language AI, and as a special advisor to the CEO of SupplyShift, a supply chain transparency platform used by Walmart and Amazon.

AI is a work in progress

As AI continues to evolve, it’s becoming clear humans will not be at the center of a lot of decision-making for much longer, Vachon told VentureBeat. The problem is that today’s AI models lack ethics, morality, and, for that matter, common sense. An AI model in a car may recognize a bouncing ball on a street, but — unlike a human — it will not reason that the presence of the ball means there is likely a child nearby unless it’s specifically trained to do so, Vachon noted.

Similarly, an AI model won’t recognize that a pizza order made to a 911 number might be a plea for help from someone being held hostage, he added.

Cogment provides a platform for humans to train AI models in a way that goes beyond simply recognizing different types of cat images, he said. For example, the framework provides the orchestration capabilities needed to create simulations involving multiple actors and enable humans to train virtual agents. “We want to keep the human in the loop,” Vachon said.

This approach to AI training will result in models that better augment the capabilities of humans rather than simply replacing them in the name of economic efficiency, Vachon added. There will eventually come a day when every individual has an AI model that has been trained to make optimal decisions and recommendations based on their personal preferences, he said. The challenge will be training those AI models to make decisions and recommendations that go beyond an analytics calculus.

More empathetic AI

In many ways, the ability to infuse AI models with some sense of empathy will prove crucial as AI models are embedded deeply within digital business transformation processes. For every user who engages with a chatbot online to resolve an issue, there is another who prefers to engage with a human. Reducing the cost of customer service will require speech-enabled virtual agents that can not only empathize with customers but also understand the lifetime value of a satisfied customer. That virtual agent also needs to be sensitive to the fact that continually upselling individuals with a limited income will create a political backlash once enough customers complain to their local government representative.

It may be a while before AI is pervasively applied, but machine learning algorithms are already being employed in ways that are raising concerns from the local statehouse to capitals around the world. Organizations employing AI models will ultimately need to provide more transparency into how those AI models are trained, especially as more people understand that AI is at its core an instance of complex mathematics that can be gamed to favor one outcome over another.