You Might Also Like: A Better Way to Money

**Janelle Shane** (0:00)
Those articles started coming out, where they said, oh my god, it said it was self-aware. What do we do? I'm like, I can get it to say that it's a squirrel. That doesn't mean that it's a squirrel.

**Jennifer Borget** (0:14)
Welcome back to A Better Way to Money. I'm Jennifer Borget. A recent survey by Experian found that nearly 62% of millennials have already turned to an AI chat bot for financial advice. And honestly, it kind of makes sense. These tools are free. They're available at 2 a.m. when you're spiraling about whether you're saving enough. And they answer in complete sentences without making you feel judged. But here's the thing. AI is a pattern matching tool trained on the Internet. Not everything you read on the Web is true. And when it comes to your personal financial life, your income, your debt, your goals, your family, how well does the Internet really know you? Today, I'm joined by Janelle Shane, research scientist, author of You Look Like a Thing and I Love You, and the brain behind AI Weirdness, the blog that has been documenting AI's strangest, most confidently wrong moments for years. She's one of the clearest thinkers on what these systems are actually doing, where they genuinely help, and where trusting them too much can lead you down the wrong path. If you've ever gotten a clean and confident AI answer to money questions and thought, okay, I think I'm good, this conversation's for you. Before we get into it, if you're navigating a big financial moment right now, a new job, a new relationship, or a new baby, we have a free Family Finances Workbook waiting for you at northwesternmutual.com/podcast.
All right? Let's dig in.
So Janelle, you started out studying engineering, and you ended up becoming one of the most widely read voices on what AI can and can't do. Can you take us through how that happened?

**Janelle Shane** (1:48)
I actually started out with AI, believe it or not.
Went to a talk as a high school student trying to choose which school to go to. And there was a professor, Eric Goodman, at Michigan State University, Go State, where he gave this talk about all the machine learning algorithms they were using in his lab. And what really struck me was that they would set it to solve these different kinds of problems, like come up with a shape for this flywheel part for this mechanical device. And it would come up with something that solved the problem, but it would be very weird and like nothing any human would ever have designed. And sometimes it would come up with something that would technically solve the problem, but not actually be a valid solution because of some technicality. Why did it do that? Oh, I don't know.
That really struck me. And so I joined that lab as an undergrad doing research. That is kind of where I started originally, was in machine learning, evolutionary algorithm. There's a bunch of names for different aspects of this thing that we're calling AI right now.

**Jennifer Borget** (3:00)
And how many years ago was that?

**Janelle Shane** (3:01)
I was in 2002, but fast forward a few years to, I think 2015 or so.
When I was in graduate school, I'd started a blog with just some pictures from the research lab like, hey, this particular experiment we did came out awful, but it looked kind of cool. So let me post a picture of that online so that somebody can get some use out of what all our taxpayer dollars paid for. And so I already had this spot ready to go when I came across the first neural net generated text I had ever seen. And this was a guy named Tim Brew who had generated cookbook recipes. And it was a very tiny neural net. So the recipes were mostly incoherent but recognizable, but it would ask for stuff like shredded bourbon or water that had been chopped and then rolled into cubes. And I was left so hard, I don't think I could even see for a while. Like it was just tears streaming down my face. And then once I'd read them all, there weren't any more.
So then I had to, okay, what did he use to make those? Can I download this? What other kinds of data could I feed in there? And so pretty soon I was generating weird names for guinea pigs, like Fuzzable and Pop Chop or weird paint colors, like Turdly and Stanky Bean, and really unappealing paint colors. Because it was just, well, these letters seem to go together. Probabilistically, let's try this. So it turns out I wasn't the only one who thought that these were kind of funny. So to my surprise, people actually started reading and sharing this blog.