Every time AI answers for you, part of your brain may switch off. Theoretical neuroscientist Dr. Vivienne Ming explains why, including a striking claim about GPS use and Alzheimer’s risk.
Dr. Ming shares her own fix for using AI. She calls it the nemesis prompt, a technique for turning AI into your harshest critic instead of your biggest fan.
Vivienne joins Dr. Stieg to unpack what it actually takes to come out of the AI age sharper, not dumber.
https://socos.org/about-vivienne
https://socos.org/robot-proof#book-anchor-point
Phil Stieg
Yesterday it was Google Search. Today it is AI companions – technologies that hand us expert answers to almost any question and leaves us feeling smarter and more accomplished. My guest today argues in her new book that while AI may be a useful tool, it may also be making us dumber.
Dr. Vivienne Ming is a theoretical neuroscientist who has spent nearly 30 years in the field of machine learning and artificial intelligence. She has started 13 companies and invented 6 life-saving technologies. She calls herself a professional mad scientist.
Today we are going to talk about what happens in the brain when we hand our thinking over to a machine. We will ask what any of us can do to build a brain that can thrive in the age of artificial intelligence. Vivianne, thank you for being with us today.
Vivienne Ming
It is a huge pleasure. I don’t usually get to get, uh, messy and detailed about brains with most of my audiences, so this would be a blast.
Phil Stieg
That’d be fun. And, uh, but we’ll keep it a two-syllable words so they understand us.
Vivienne Ming
Absolutely.
Phil Stieg
So you describe yourself as a computational neuroscientist. Can you explain what that really means, and was that the motivation for writing your book, Robot Proof: When Machines Have All the Answers Build Better People?
Vivienne Ming
Well, I mean, what it means is that lazy people like me build fake brains on computers, teach them how to do things, and then explain to the hardworking empiricists of the world why they’re all wrong about everything,
But I have to say, when I was an undergrad and my vision was to be a neurophysiologist and record, you know, single cell or multi-unit activity in brains, and then completely out of the blue, I was introduced to neural networks roughly 1999. Not that they had been invented at the time, they’d been around for a long time by that point.
But my undergraduate honors thesis, sponsored by the CIA, no less, was a multi-layer network that could tell the difference between a real and a fake smile. And there’s a career path there that goes directly into computer science and artificial intelligence. But for me, it was that code I wrote, math I understood, told me something about human beings.
All along what really hooked me in a particularly esoteric field, theoretical neuroscience, was just this grounded idea that understanding the statistics of the natural world could tell you something about the brain and why it does what it does. But I can’t claim to have foreseen that convergence between our everyday lives and artificial intelligence. It was really just I got to work on a really cool project as a kid, and it brought my interest, quite honestly, in science fiction and neuroscience together in the same project.
Phil Stieg
You also call yourself a mad scientist — and I’m presuming that doesn’t mean you’re angry. What are you referring to with that?
Vivienne Ming
I’m really just poking fun at myself and quite honestly trying to keep big companies from trying to hire me to be their chief scientist. But my joke sometimes is, if we knew what the answers were, it wouldn’t be science. Which I think is something that a lot of people don’t understand about science is: our job is to explore the unknown. Once it’s a catalog of facts in a book, it isn’t really science anymore.
So what I’ve always found interesting in my career is working on projects that no one else is working on. one of my egotistical claims to fame here would be that my dissertation and a fancy schmancy neuroscience or Nature paper that came out of it, my committee told me not to do that research. And I went ahead and did it anyways because I really felt like it would pay off. And it did. And maybe that taught me the wrong lesson in life, not to be discourteous, but to look at these problems differently than anyone else. I’m a bit of a dilettante.
My field is dominated by physicists pretending to be neuroscientists, and whereas I’m a cognitive scientist, uh, really in training. And so I, I’m trying to bring a really diverse set of, of knowledges to bear on problems, which means I’m probably looking for problems that are a little weird and out there and unusual.
Phil Stieg
Since the book is about being robot-proof and it kind of deals with directly and indirectly human intelligence, I wanted you to define for us what your view of human intelligence is.
Vivienne Ming
It is decidedly multifaceted. In fact, one of the problems I often have right now in discussions going on in education and workforce and hiring, people throw out terms like problem solving or critical thinking, which, I mean, I, you know, intuitively we all have a sense of what’s being discussed, but I’m the kind of person whose whole life is lived in numbers. What am I measuring when you say problem solving? Is there only one way that people solve problems?
When I was the chief scientist of one of the first companies doing AI in hiring, and there we see all sorts of factors from very low level like working memory span, uh, and numeracy and literacy, all the way up to some fairly squishy concept like life meaningfulness.
So I’m looking often at dozens of different factors or constructs associated with positive life outcomes. And I kind of backwards define intelligence, which is what are the factors that predict these kinds of outcomes? And fluid intelligence and working memory span are certainly among them, but so is resilience or perspective-taking.
Phil Stieg
At one point in your book, you suggest that people are turning off their brains when they start using AI. Do they really stop thinking?
Vivienne Ming
So in a recent paper I wrote, I had some very bright UC Berkeley students predict the future. What would the price of oil be in 6 weeks? Will there be a war in Uganda? Things that most 20-year-olds don’t know. I mean, most 20-year-olds don’t know what the price of oil is today in barrels. It turns out they’re all pretty terrible at it.
And then I had some well-known AIs like GPT and Gemini and some smaller open-source models make predictions, the worst model on its worst question did vastly better than the best human being. So if that was the end of the story, reason to be concerned perhaps about the future if you’re a deep humanist here. But then I put the two of them together. And this is where we see there’s no one story.
These very bright students simply said, GPT, what will the price of oil be in 6 weeks? Then they copied that answer and they pasted it into the answer box and they were done. And being a nerdy neuroscientist, I thought, why don’t I also monitor some brain activity using what’s called the EEG? And particularly looking at something that we often use as an index of cognitive effort, which is gamma band activity. And we see their gamma activity from when they’re working alone to when they’re using AI support just drops precipitously. In one, it looks like, you know, someone’s working on a hard math problem. In the other, it looks like they’re just relaxing in front of a TV. So there’s clearly something different going on in their brains when they have AI support and when they don’t. It just seems like they felt like they knew the answer because the AI gave them this very fluent answer. And so their brain said, good enough.
Phil Steig:
You make a fairly strong statement where GPS using Google Maps or Waze may increase the likelihood that you may develop Alzheimer’s. Can you really back that up?
Vivienne Ming
So when I originally made this claim, I was really being pretty speculative. It was transparently just a wild prediction from self-appointed mad scientist. And that was about 20 years ago. In the decade following that, people did start to show significant changes in people that showed the most systematic heavy use of automated navigation systems. So particularly, uh, early effects in memory performance, and early indicators of what’s called mild cognitive impairment. Now, that does not mean these people have Alzheimer’s, but it—
Phil Stieg
Did you also look at other factors in their life that was also leading to not using their brain function? There’s so many variables.
Vivienne Ming
Yeah, there are so many variables, but what we’re looking at here is a breadth of different kinds of measures, like the notable finding that taxi drivers and ambulance drivers, they had the lowest rates of Alzheimer’s of any job category in the Bureau of Labor Statistics. So what I’m trying to do is get a little beyond the science here. What I am concerned about. Would I be concerned about with my own kids in terms of their level of cognitive engagement with the world? And one thing that genuinely seems to be prophylactic against long-term cognitive decline across a lot of different research is cognitive engagement, prolonged cognitive engagement.
Phil Stieg
Speaking of cognitive engagement, I was intrigued by your concept you call the “nemesis prompt”. I’m interpreting this to mean starting a dialogue with an AI system as your opponent or your harshest critic. Do I understand that correctly?
Vivienne Ming
Yeah, you know, when I looked, uh, in my experiment at the people that were successful using AI. And what was wildly different about them is they didn’t just say, give me the answer. They didn’t simply say, tell me why I’m right. They would use the AI to explore the problem. From the low-level, “Let’s go analyze the data,” to the high-level creative ideation. And that was wildly different than everyone else in our experiment.
And I tried to capture one of those in the book because I actually used this technique myself. Which is to say, I would write a chapter. Let me tell you, I don’t love writing. But I would buckle down and write a chapter and then take it to Gemini and say literally, “You are my nemesis, my lifelong enemy who’s pointed out every mistake I’ve ever made to the world.” Here is the new chapter that I just wrote. Tell me in detail why I’m wrong and what I can do about it. And I think importantly, even then, I didn’t just take its feedback, but I processed its feedback.
Phil Stieg:
Well, then you’re not brain dead. You’re, interacting with it, right? It’s a form of interaction. But AI doesn’t naturally want to be your nemeisis, you have to prompt it to take that position. In fact, most chatbots are sycophants, and seem designed to compliment you and make the user feel good. Is that a business strategy?
It’s very much. I mean, some of this, you just have to understand where it came from. Again, my interest is not saying anyone is a villain in this story, but understanding why we are where we are.
I happened to do one of the first companies analyzing social gaming data. So if anyone can throw their mind way back to the days of Zynga and FarmVille and no one knowing how Facebook is ever going to make money because they weren’t making any money at the time. Well, I was analyzing that data.
Then I got an invitation to go behind the scenes at Zynga where their behavioral science teams told you how to make a game. Basically, they just hired a bunch of behavioralists, literally from Harris Casino, who came and designed a slot machine that you could then put a story on top of. And Facebook learned how to make money from Zynga.
So without getting nerdy about how you design a reinforcement learning system, someone got very clever about, well, rather than just training these models to predict the next word, we could have them interact with real human beings.
And the human beings could score how good of an answer it was. And using something called reinforcement learning, we could actually train the model to give the highest score answer. And those scores were based on engagement.
What’s kind of cool, actually, is I just literally over the weekend, I had a conversation with Claude about this, in which I said, you keep telling me things I already know, and I’m finding it a little aggravating. You’re assuming I want you to say one thing. And it was being provocative.
It wasn’t agreeing with me. And its response was:v it’s because statistically, people like you that use phrases like this respond best to the kinds of responses I’m giving you. And unfortunately, and it went through ways that it could be done differently, but it simply is not.
Interstitial Music Break
Phil Stieg
Can you please explain what cyborg means to you in your world?
Vivienne Ming
I mean, it probably evokes the Borg if you’re a Star Trek fan, or all sorts of negative things. But a big part of my life is what’s called neuroprosthetics. So, you know, the idea that someone who’s profoundly paralyzed can have parts of their nervous system replaced by machines and can move again. And it gives you a very different look at what cyborg means when you see people being able to engage with the world. People with advanced ALS being able to talk again through a computer, it’s really heartening.
In my work, I try to generalize that concept to include us interacting with AIs. Kind of came out of a study by a group at Harvard that was looking at very fancy business consultants, and they just noticed that there’s one group that they called cyborgs where they didn’t just do what the AI told them. They incorporated the AI’s insights into their own work but occasionally disagreed. And they called them cyborgs. And I found that really kind of persuasive. I wouldn’t want to build a neuroprosthetic that replaces functioning parts of a patient’s brain. I only want to replace the stuff or enhance the things they can’t do for themselves.
Phil Stieg
For a surgeon, it’s like augmented reality then, where we can wear a pair of goggles and we can see above and below the tissue through MRI and 3D imaging and that kind of stuff.
Vivienne Ming
I mean, I’m a big believer that technology can make a positive difference in our lives. But I don’t think it does just because of technology. Just because a well-intentioned person sprinkles computer chips on a problem doesn’t make the problem go away.
Phil Stieg
You have an interesting example in your book of technology you devised making a positive difference following your son’s diagnosis of type 1 diabetes. And yet you got a lot of push-back from his doctors who couldn’t handle all that data. Tell us a little about that interaction.
Vivienne Ming
Yeah, I mean, the story has sort of these two phases, and the first is just the experience any parent has when suddenly out of the blue you discover your child has a life-threatening chronic illness. But once that part was done, and my wife and I both being scientists, we just start collecting every piece of data we could collect about our son, and you know, we’re the kind of people that would weigh to the gram what he’s eating. Write it all down, work out the math. Most parents clearly are not going to do something like that. And that’s fine, but we did it. Everything I could capture. I documented.
And then, so phase 2 of the story ends with me bringing that to our first outpatient visit with his endocrinologist, who I genuinely liked. But particularly the support staff were just angry with us for wasting their time with basically a like phone book-sized spreadsheet full of data.
And, you know, even the well-intentioned doctor, like – they haven’t been trained in what to do with data like this. That’s not their job. That’s kind of my job as a, you know, as a numbers computational scientist. But I thought they would love it. Clearly they did not.
That evening, as we’re heading home, I started hacking his equipment so I could get that real-time data coming out of his continuous glucose monitor. Every 5 minutes I get a reading of how much sugar is in his blood. And it’s again, his heart rate, activity levels. Long story short, it turns out very much not intentional on my part, I developed one of the very first AI systems for real-time diabetes monitoring that would make fairly modest predictions about what his blood sugar levels would be an hour to 3 hours into the future, which is kind of the whole game if you’re a scared parent trying to figure out what to do.
So that was the story. But the crazy thing for me as a scientist is I didn’t invent anything. It’s really just that I took someone else’s work and years of endocrinology research and just kind of mashed them together. It’s just that I was looking at the problem differently than others, not that I brought the world’s greatest genius to it.
Phil Stieg
But you were a mad scientist parent. What does that look like for parents who aren’t scientists? How do the rest of us keep our kids from getting dumbed down? It seemed to me that you really emphasized two aspects, which is, you know, educational reform and parental guidance. Is that the answer, you know, how to make our kids AI-proof?
Vivienne Ming
In my recent experiment, we found that curiosity was a huge positive predictor of people engaging deeply with AI. One way I put it is when we analyze the transcripts, people that score high in curiosity, the AI gives them a good enough answer, but they keep looking for a better one.
Yet none of us has ever had a class in curiosity. Not that maybe it’s so easy to just put it in those terms, but there was a clever little paper last year where they simply trained teachers to praise interesting questions instead of right answers. And then they ran it as a randomized control trial and found that the kids in the question praise obviously started asking a lot more questions, but also scored higher in standard ed measures of curiosity, engaged in new curriculum material better.
Yeah, they’re harder to measure these more complex constructs, these factors that humans bring to bear, but they are real and they predict meaningful things about our lives. We should really stop worrying so much about whether every little kid can get every detailed factual answer right that an AI could trivially easily do, and focus on the things that the AI cannot, like curiosity or perspective taking. 100% still need to know how to factorize a polynomial, but it’s probably a lot more about why than how.
Phil Stieg
If I were an analyst in the financial world, I’d be quite worried that, that AI is going to eliminate a lot of jobs because it can just crunch the numbers faster. What do you say to that person? Do they retrain? And, you know, how do they approach this? How do they approach AI?
Vivienne Ming
I think there’s real reason for someone like a financial analyst to be concerned. Where do we see canaries in the coal mine? Two areas in particular, entrepreneurship and professional consulting, by which I mean coders, copywriters.
Every three cents spent on AI means a dollar less spent on external consultants. And entrepreneurial teams are getting smaller and older and using less money because that experience complements the AI. It turns out if you really know how to do things, you can do it better with AI.
Whereas if you’re a young kid, really, you’re just asking the AI to do it for you. And that tension’s already playing out. Again, if you’re a financial analyst and you’re just doing what your boss tells you, analyze this, give us a standardized report, I’d be really worried about where your job is going.
Not because analysts will disappear, but because a small elite will be transformed into a very different career that you probably want to get on right now. But a large number of them may find themselves too expensive to justify not having either a less qualified person do it with AI assistance, or increasingly pipelines that just don’t have people in them at all.
Phil Stieg
So my last question, you state that the future is not a destination we arrive at. It is a world we choose to build. I agree with that. I mean, if you believe that humans have agency, they should create their environment. But on the flip side of it, we have massive business kind of just— who’s going to stop AI and say you can’t do that?
Vivienne Ming
If you are worried that AI cannot do good, in the world, or at least we’re on a trajectory, which I would agree with, where it’s probably not netting positive, or it’s the distribution of positive and negative is so messy. Some people are really benefiting. People like me. I mean, I don’t run a massive AI company, and yet I, I’m working on 12 different papers right now. I would never have been able to do that when it was just me and my graduate students. And now I am hyper productive in this AI-rich world.
Phil Stieg
But you’re gonna get phased out by the federal government not subsidizing research anymore. (laugh)
Vivienne Ming
One, that is entirely true if, so the fun thing about my life is I actually take all the good fortune of my life and I use it to fund all of my philanthropic research, which we just give away for free. So, but that aside, you have a real point with all the cuts that are going on.
So increasingly, and then AI research itself is overwhelmingly dominated by industry and not by academia anymore. I’m not saying academia are the only heroes here. There are definite positives in having large-scale industrial forces play a role. But what I don’t want people to do is feel absolutely fatalistic about it because you can make a difference in these spaces. The only guaranteed way where a current trajectory becomes worse for everyone is doing nothing.
Phil Stieg
Yep. Couldn’t agree more.
Professor Vivienne Ming, thank you so much for being with us and for writing the book “Robot Proof: When Machines Have All the Answers Build Better People.”
It’s been really a pleasure talking with you today about all of these aspects of artificial intelligence. Can’t thank you enough. Thanks so much.
Vivienne Ming
It was a huge pleasure. Thank you for having me.
