AUSTIN — If we want humans to trust artificial intelligence (AI), then we need to teach the machines empathy, according to John Roese, CTO and president of products and operations at Dell Technologies.

Roese joined two other Dell Technologies’ companies CTOs on a panel at last week’s Dell Technologies Summit: Dell Boomi’s Michael Morton and RSA’s Zulfikar Ramzan. Boomi is a data management company that lets businesses integrate and transfer data between cloud and on-premises applications. RSA is a security company whose founders pioneered public-key cryptography.

The three CTOs discussed three big problems in the data era: what does infrastructure look like in an artificial intelligence (AI) driven, real-time data environment? What roles will humans play in an AI-driven, real-time data pipeline? And what will the security risk model look like in this brave new AI world?

These are three really, really smart guys (not so much comedians as they were quick to admit). And, as such, they all had some really interesting ideas about AI. But one, in particular, that caught my attention was empathetic AI. It’s a topic that Roese said he’s “been pretty vocal about in the AI world.” In fact, he’s participating in a panel related to this topic at SWSX in March 2020.

Advances in AI won’t happen unless humans can trust machine learning, Roese said. And right now, there’s a huge trust gap.

“AI is a black box to most human beings, and we don’t trust it because we don’t understand what is going on inside of it,” Roese said, adding that “your suspicion is well rounded. The AI is working based on the data it’s given. It doesn’t contemplate your emotions, your happiness. It’s basically taking a dataset and coming up with the best possible answer without any context of if it.”

Humans, on the other hand, usually have this context. And as such, our decisions may not always reflect the fastest, most linear approach to solving a problem. But they do – or at least, they should — take into account how a given decision will affect other people. In other words: humans have empathy. This means we need to integrate ethical considerations into the machine learning data set to ensure that the AI ultimately makes a decision with learned empathy that pleases humans.

“AI systems work based on extra stimulus,” Roese said. “You feed that data streams, and those data streams fundamentally influence the model.” But this also means people can feel AI systems bad data — like biases.

Roese suggests incorporating sensors and biofeedback into the machine learning model so the AI can learn the emotional state of the humans it’s interacting with. “What if I'm able to measure your biofeedback? I know your stress level, your heart respiration rate, or perspiration rate? And if that’s a factor in how the model works? So it learns that if I choose option A, even though it gives me a good outcome, it candidly pisses off the human.”

Obviously businesses that are trying to engender trust with their customers, partners, and employees that interact with these AI systems. These companies don’t want to make the humans angry, and this is why empathetic AI is so important.

So we teach the machine to choose option B instead. “That might be slightly more convoluted, but it will have a preferred outcome,” Roese said. “The human being will have less stress and it will be happier. And once you understand that the AI is thinking on your behalf, it’s considering your environment, you start to trust it more.”