In Episode 43 of Razorleaf’s Stay Sharp podcast, Under the Covers with AI, co-hosts Jen Ferello and Jonathan Scott take a deep dive into AI or artificial intelligence—or Jonathan’s preference, “augmented intelligence.” The duo focuses on how humans interact with it, what it can contribute to our jobs and our lives, and what you need to consider if you’re getting ready to use it.
Humans Are Still Critical
Jen and Jonathan start by agreeing that AI isn’t ever going work without the human element. Fundamentally, that’s because AI is mimicking what humans do—taking in data, processing it, and doing something with it—and AI will always rely on data being selected and input. The duo also argues that humans will also still need to make decisions, because ultimately humans are the ones responsible for outcomes. AI will help us in many ways, they believe, and more, it should push us to do as much continual learning as the machines are doing—“because it’s our intelligence we’re putting into AI, and that’s not artificial,” says Jen.
All About the Data
The most appropriate analogy Jen and Jonathan discuss for AI is that it’s like a child. From birth to about age four, like Jen’s grandchildren, it’s about data intake, and that’s the same for the first step in working with AI. Whether child or machine, you need quality data to generate a quality outcome—or as they say, “garbage in, garbage out.” Or as Jonathan discovered, bad words in, bad words out.
No matter what you’re asking AI to do, the quality of the data is paramount. The strength and quality of pattern recognition—which AI is significantly better at than humans—depends on the data you’re feeding it. Too much or too little is as big a problem as bad words, for example. “If you’ve limited the data set too much because you’re trying to keep it too focused, it may miss because you’ve limited its field of view. It may miss some things that are helpful to you in terms of its learning,” Jonathan says. “But if you include too much, it may be that it’s not too wide a field of view, but that you’ve included data that’s bad and going to mis-train it.”
Finding Patterns in Data
After training AI with data, the next step is turning AI loose to find those patterns. One of the surprising benefits of AI’s complex processing skills is emergent behavior, which means insights or properties that emerge from the interaction of different parts of a complex entity, but that wouldn’t be found in the constituent parts on their own. With AI, Jonathan explains, “we’ve set up a training model for it to learn in one way, but then that piece of code, that intelligence can expand upon that to learn and look for other patters in new ways that are not necessarily what was defined.” He adds that often AI will generate connections or conclusions that we don’t see, noting, “We thought it wasn’t logical, but actually it was very logical. We just didn’t see the connection.”
The Final, Critical Step
AI requires testing and validation. Like you have to keep testing and validating, or correcting, your kids as they grown and learn, you have to do the same with AI. “You have to bring it back to, are those relationships and patterns correct?” Jen says. “Is it accurate? Is it performing? Is it making decisions that are supportive of your business and what you need to do?”
You also need to test the “hallucinations” or inaccuracies that AI can sometimes present as truth. Jonathan has an interesting take on hallucinations, suggesting that they can be a good thing—citing that humans have always attempted to induce hallucinations for deeper insight or creativity. With AI hallucinations, however, especially in a technical context, you’ve got to be careful. “Hallucinations can actually get you to something that you could never have gotten to before, just as a possibility. And now that there’s that possibility, what’s real about that? What actually can be useful about that?” Jen adds, “But again, you have to have a human interaction…Is it helpful? Is it not helpful?”
Learn More About Making Sense of AI
The full podcast episode offers more details including using AI for advanced information gathering or internet searching, how AI can learn from data we don’t realize we’re giving it, the possible future of AI to help engineers do their jobs, and more.
Check out the full conversation in Stay Sharp Episode 43: Under the Covers with AI, and join us each week for a new podcast.
Follow the Razorleaf Podcast, Stay Sharp in Digital Engineering on:
