AI and Emotion: Can Machines Feel? (with Rosalind Picard)

The following is an uncorrected transcript generated by a transcription service. Before quoting in print, please check the corresponding audio for accuracy.

0:00:08 – (Christopher Watkin): Welcome to Silicon Spiritualities, a podcast from the Gospel Coalition about what it means to be human in the age of AI. I’m Christopher Watkin, and in this series we’re circling around three crucial questions. What can AI do do for us? What is it doing to us? And what is it showing us about who we really are? Well, to think about those questions today, I’m delighted to be joined by Rosalind Pickard, professor at MIT and founder of the MIT Media Lab’s Effective Computing Research Group, where she and her team develop technologies that enable computers to recognize and respond to human emotions.

0:00:58 – (Christopher Watkin): Ros, welcome to the podcast.

0:01:00 – (Rosalind Picard): Thanks, Chris. Such a pleasure to be here with you.

0:01:02 – (Christopher Watkin): Ros. I’d love to begin with a scenario if I could, and I’d love your reaction to it. There’s a lot in here, so feel free to pick whatever you think is most important. I’d like you to imagine mia. She’s a 16 year old in a local church youth group. She wears a smartwatch that tracks her stress and her sleep. And she wears an app that uses an emotional AI coach to try and coach her through anxiety. Late at night, when she’s feeling lonely, she sometimes chats with an AI companion.

0:01:39 – (Christopher Watkin): The companion remembers her history and it responds in sympathetic language. And it makes her. These are her own words. It makes her feel seen and loved. She says, it actually gets me better than most adults do. It’s my safe place. How would you react to Mia’s situation? How would you advise her if you were speaking with her?

0:02:04 – (Rosalind Picard): Oh, goodness, there’s a lot in there. I think one of the things I’ve been particularly concerned about lately is the companion chatbots. They may start off with some very helpful suggestions for people, and they certainly have been trained on the best of human dialogues to match your input to another input that somebody labeled the response to as an excellent response. Unfortunately, they’re also being provided largely by companies that have a need to make a lot of revenue to pay back investors.

0:02:47 – (Rosalind Picard): And so they get more revenue, the longer they keep you on them. It’s like one question. Fine, another question, another question. Ka chink. Ka chink. And the more they can hook you for hours. The new California law says to alert the youth after three hours, the more youth get into them and can get addicted to them to the point where they don’t spend important time with real human beings. And as Christians know, being seen and loved by God and by one another is so much greater. And if somebody’s happy with this fabrication of that, then I think that’s a warning sign.

0:03:33 – (Rosalind Picard): So that right there is maybe the biggest red flag in the scenario. I’m happy to address some of the other things too, though.

0:03:40 – (Christopher Watkin): Yeah, no, let’s keep going. Let’s tease out what’s happening in Mia’s situation and what you think might be problematic about it, please.

0:03:48 – (Rosalind Picard): Yeah, there are, I’ll have to say, different apps out there and different wearables out there that give different qualities of data. Some are medically certified. I happen to be wearing two that have FDA clearances attached to them. I’m always wearing and testing something, so I’m wearing ones that have been through extensive medical validation. They’re validated on different skin tones and for different medical conditions and different ages and sizes of people.

0:04:15 – (Rosalind Picard): But there’s a lot of consumer ones that will make similar sounding claims and they’re not accurate. And for example, stress, claims of stress are usually not accurate. Stress is much more complicated than what these devices are actually computing. The ability to measure some parameters of sleep has gotten very advanced, and I think that that may be one of the more reliable things. Even on the consumer side. It is possible to get FDA clearance on sleep. This one just both of these have it, but most consumers don’t.

0:04:51 – (Rosalind Picard): But just getting good understanding of your sleep wake rhythms I think is extremely important. And as a Christian, we know that, you know, God came down and was embodied in this kind of limited, like assuming limited body. Right. And our bodies being very limited, sometimes really do suffer more anxiety and stress if we don’t sleep well. We have seen many cases where when a young person fixes their sleep sleep, then their anxiety, their depression and other problems are significantly reduced.

0:05:27 – (Rosalind Picard): So I’m actually a huge fan of fixing that sleep if it’s not good, learning what good sleep hygiene is. And I think wearables and a number of apps out there are excellent at helping with those. There are also wearables that help with anxiety and. And again, some of them are like, really not good and others have lots of medical evidence, but they’re really boring and so people don’t want to use them.

0:05:59 – (Rosalind Picard): But I’ll have to say that if you’re going through a, you know, anxiety and loneliness, the solution really is not just in an app or a wearable. I think a really healthy solution is that these are companions of companion information providers, not companion people, companion information providers to help with getting insight into what’s actually happening. Like, most people don’t know how long it took them to fall asleep and they’ll think that they were up all night, but they weren’t. Or they’ll think that they slept great, but in fact, they have horribly fragmented sleep.

0:06:40 – (Rosalind Picard): So the objective data can augment that experiential data and be very helpful.

0:06:48 – (Christopher Watkin): That answer is really helpful as well. Thank you. We’ll come back a little bit later to this idea that Mia has of, you know, I’m seen, I’m heard, it works for me. But I want to come back to something you said in the course of that response, which was you were contrasting a biblical view of emotion to the sort of view that we might get from these apps. And I just want to try and unpack that a little bit more.

0:07:17 – (Christopher Watkin): When we look at the sweep of the biblical narrative, there is a huge amount of emotion in there, obviously, isn’t there? You know, Jesus weeping at the tomb of Lazarus is just the tip of a huge iceberg of emotion. And how does the Christian story of emotion challenge what we might clumsily call the Silicon Valley view of emotion and affect? What do you think the Bible can teach us about emotion in the age of AI?

0:07:49 – (Rosalind Picard): Well, it’s interesting you point out the emotions. I think when I first started working on what I named affective computing, an awful lot of people pulled me aside, most of them fellow engineers. And usually they would look at my feet and they would say, roz, why are you working on emotion? Some of them would say, I don’t have emotions. Like, others would say, emotion is irrelevant. Others would say, like, it’s just noise.

0:08:17 – (Rosalind Picard): It’s like the undesirable part of life. And so I think it’s wonderful that you’re pointing out right away that it’s important. It really does matter because there are a lot of people still today who think it’s a waste of time. Now, some people will listen to this and be shocked, but I would sit there and look at these people telling me that I’m throwing away my career looking at emotion because it was irrelevant, because many of them really don’t feel emotion. I learned about 1 in 10 people have what’s called alexithymia.

0:08:54 – (Rosalind Picard): They either aren’t aware of emotions in their body until they become extreme, right? Like total meltdown, or they might kind of be aware that something’s happening, but they don’t know, like, what it is, how to label it, what do you call it? Like, what does that mean? I kind of feel something, but I don’t have any words for it. And all of the. This is a real handicap when it comes to sensing a meltdown coming on and being able to like tamp it down before it hijacks your brain.

0:09:30 – (Rosalind Picard): It’s a real problem when you have lingering emotions that need addressing and you aren’t really like sure what to do about them. So there are a lot, there’s, you know, 10th of our society is dealing with this and thinks they’re pretty much either irrelevant or just a total mystery. In autism. A community that I adore and I’ve spent a lot of time with, the rate of alexithymia is more like about 50%.

0:10:02 – (Rosalind Picard): So there are a lot of groups. There’s also people with non verbal learning disability and others who just really don’t get it with kind of their normal born state of affective experience. So right away it’s neat to see that the Bible tells us about emotions. We also know in learning about emotion that what we measure about it is but a small part of what they are. Never in my work have I said we are true.

0:10:40 – (Rosalind Picard): And I have to be a little careful here because let me just say that when I named the area affective computing and named an area of teaching a computer how to read, things like, like the lip corner pull and the eyebrow raise and some of these signals that people often associate with emotion, we labeled them just like people did at the time. Like, oh, if it looks like it might be a fuzzy animal with ears and a tail and four paws, you know, maybe we call it dog recognition, or we call that cat recognition, or we call that person recognition.

0:11:15 – (Rosalind Picard): Nowhere is it seeing your innermost feelings. It is putting together whatever information you do give it permission to have as an input and attaching a label to that, that would be the kind of label a person would attach to it. So you just nodded and raised your brows, which, you know, kind of look like you might be interested. But I don’t know if you’re feeling interest, maybe you’re just thinking like, this is really boring and I should just nod and raise my brows and get through it.

0:11:43 – (Rosalind Picard): And when I say that, I assure

0:11:45 – (Christopher Watkin): you it’s the farmer.

0:11:46 – (Rosalind Picard): And when I say that and you smile, I think, well, maybe that was funny, right? Like, because it’s contingent on what just happened. So it’s very important that we look at all these signals in a context of what else is going on. And then in that context, when we read the messages on the face and the voice, and sometimes with permission, the physiology or other information, we, we try to make an inference, we make an educated guess about what might be underneath driving that, but we do not have access to your true internal experience.

0:12:22 – (Rosalind Picard): So when you talk about what an emotion is, it’s like it’s that true internal experience that we don’t have access to. However, when you talk about producing a label like happy or sad, the computer recognition, pattern recognition, machine learning community calls it emotion recognition. So when talking to the public, have to be very careful. It’s not recognizing your feelings, right? But when in the computer, you know, AI community and we talk about emotion recognition, we know it means what it means, and that’s not actually recognizing that full or even that inner emotional experience.

0:13:02 – (Christopher Watkin): And this is getting right to the heart of what I was wanting to talk with you about, because we’re getting to the point now, it seems to me, as someone on the outside of this, where LLMs and other technologies can very convincingly imitate human emotion. They can give off all the signs, whether it’s linguistically or increasingly with, you know, robots, of being happy or sad or interested or frustrated or whatever it is.

0:13:33 – (Christopher Watkin): And we often reflexively say, especially from sort of a Christian point of view, where we want to protect the sacredness, the differentness of being human, we say things like, ah, yes, well, an AI can act as if it’s happy, but it can’t actually be happy. And I just want to stress test that idea with you. I think it’s true, but I want to try and be clear on what we mean by that. What exactly is the difference between acting as if it’s happy and being happy?

0:14:06 – (Christopher Watkin): And also, what does it say about us that we put so much store on this internal experience of a thing rather than just the appearance of it? Why is this distinction important for Christians?

0:14:21 – (Rosalind Picard): Yeah, it’s hard to come up with language that really captures it. You know, when I try to think of the being versus acting, you know, my son was just helping high school students put on the Merchant of Venice and the infamous character Shylock. Right? There’s a student playing Shylock. And what does it mean to act Shylock? It was very hard for the student to, to become. Right? There’s method acting. There are different acting schools and thoughts somewhere. You just really try to adopt the outward appearances and others where you try to truly internalize and be that character, right? You really get into it. It gets into you.

0:15:05 – (Rosalind Picard): You try to find all the evil inside you and try to be like that. Right? And so there are different degrees of that acting as if. But in all the cases, the example I just gave fails because it’s still a person, right? It’s a person acting as another person, a fictitious person in this Case. But a, you know, you can see that even in that there are different degrees of putting on the outward face, putting on the outward posture, adopting the gesture, adopting, but may or maybe maybe not adopting the full rage or feeling inside, which can be very exhausting. Actors who do that say, you know, it really wears you out. They don’t recommend, you know, doing it.

0:15:53 – (Rosalind Picard): Some say that’s not really acting when you’re actually becoming. Now the machine is. There’s no, you know, again, our language has misled a lot of people here. It’s, you know, in the early days of AI, when the, when John McCarthy and Marvin Minsky and guys, you know, stood around and tried to figure out how to get this work funded, they said, what should we call this? And the best description, they agreed, was complex information processing.

0:16:28 – (Rosalind Picard): But John said, you know, I think that’s not likely to get funded. I think artificial intelligence is more catchy term. So they called the complex information processing artificial intelligence, which did indeed get a lot more funding now over the years. It got a lot and then it tanked and then it got a lot. You know, we’re dealing with over, what, six decades of this now. But, but it’s not, it’s not thinking, it’s not learning, it’s not feeling. It’s not experiencing anything like our thoughts, feelings, mind, experience.

0:17:07 – (Rosalind Picard): It’s all us being in awe of what goes on inside us and trying to understand what goes on inside us and trying to understand it by building it. In computer science, there’s sort of this. We don’t really understand it until we’ve built it. So as we try to build it, we realize, oh, that’s not doing what we’re doing. So it’s clearly not right. And so it’s a, it’s a way of actually, it can be an honorable way of trying to learn about something. It can also be a very much Tower of Babel kind of way about, you know, being, having too much hubris, right?

0:17:46 – (Rosalind Picard): So, you know, there, there’s quite a range here. But the problem is our language has been, from the get go, has been formed in, number one, to sell the work, number two, to inspire the work, right? To point to people and to say, okay, it doesn’t think, but how do we think? How do we build something that kind of looks like it’s thinking?

0:18:12 – (Christopher Watkin): There’s a theme emerging in what you’re saying that I’m finding really illuminating. It came out in the response to the question about Mia and then the question about the artificial intelligence as a term, as well, it’s that we’re naive if we don’t keep at the front of our minds that so much of this is geared towards the aim of making money or winning funding rounds for particular companies and therefore particular interests.

0:18:44 – (Rosalind Picard): Well, the funding there was for research when AI was done, not for a company at that point.

0:18:50 – (Christopher Watkin): Yeah. So even the term AI itself is engineered not to best describe what it’s talking about, but to hook in funding from people who might want to fund research. And I think it’s just helpful, isn’t it, not to become cynical, but to keep that in mind as we engage really important.

0:19:08 – (Rosalind Picard): Yeah.

0:19:08 – (Christopher Watkin): Technologies.

0:19:10 – (Rosalind Picard): Yeah. Where the technology meets the real world and it’s being provided in some continuously sustained way. There has to be something to sustain it. Right. Something to pay the people who keep it up and running.

0:19:23 – (Christopher Watkin): I want to come back to this idea of imitation, and I want to flip the tables. We’ve been saying AI just imitates, but we’re sort of assuming in that conversation that we humans don’t imitate. And I wonder the extent to which some of our emotions are mimetic, are imitative, too. So, for example, when I tell someone, I feel your pain, I don’t always feel their pain. Sometimes I say those words because I know that those are the right words to say in a situation and they will bring comfort. And I want to feel their pain. It’s not as if I’m just lying, but I don’t actually.

0:20:04 – (Christopher Watkin): I’m almost faking it until I make it in that situation.

0:20:09 – (Rosalind Picard): No. And you don’t even want to make it. You don’t actually really want to feel their pain. I don’t think they really want you to feel their pain either.

0:20:15 – (Christopher Watkin): Yeah, but are we. So we often imitate emotions as human beings in order to lubricate relationships or whatever it is. Are we trying to hold AI to a higher standard than we actually meet ourselves when we say, oh, no, it’s not good enough for it to imitate things. For it to be a person, it would need to really feel those things.

0:20:35 – (Rosalind Picard): Oh, boy. Okay, so you’re opening up a lot of issues here. Let me see if I can stay on track and address one or two of them. The first one is, you’re quite right. People say things that they intend for good, hopefully, but aren’t. Literally what’s going on. When we were in the very early days of building our first relational software agent that was designed to help support people with adopting an exercise habit, they were supposed to chat with it five minutes a day.

0:21:14 – (Rosalind Picard): And get. Get support, trying to remember. The therapist gave us advice for who was consulting with us on it. She was like a person who helps people with support when they’re trying to deal with addictions or behavior change. And she told us that how she did empathy. And we thought, wow, it’s that formulaic, you know, like, we were looking at the formula, and I’m like, okay, so seriously, like, when you go to all these therapists, also go to see therapists. So I say, so, like, when you. The therapist go to talk to your therapist, and your therapist uses this formula on you, does it work?

0:21:57 – (Rosalind Picard): Like, because, you know, it’s just a formula, right? She. She’s just doing this active listening, rephrasing your words, and then adding some emotion terms to describe it. And the doctor said, oh, yeah, it. And I know she’s just doing it, and I know sometimes when I’m doing it, it’s kind of formulaic, but we do it because it works. And I thought, gosh, they’re empathizing without feeling anything, but it works. It’s effective.

0:22:24 – (Rosalind Picard): Which made me realize, you know, yeah, I guess we could put that in a machine. But, you know, it did feel disingenuous. I wasn’t sure it was going to be effective right when it was coming from a machine.

0:22:35 – (Christopher Watkin): Well, this is a sort of Eliza effect, isn’t it? It can very simply mirror back to you the language that you used to. And you can feel seen and heard and profoundly sort of engaged.

0:22:46 – (Rosalind Picard): Yes, you. You feel understood. And giving people that psychological oxygen of being understood is a real gift. We. There are people who are like, learning how to do this better in their life. They say, from talking to these bots. I don’t know how common this is. I certainly see bad things happening, too. But there was one man who talked about having a terrible marriage. And after he interacted with this relational agent, he started learning some nicer ways to handle some things.

0:23:15 – (Rosalind Picard): And then eventually he said he put down the agent, stopped using it, but he started practicing that with his wife. And it said. He said it improved their relationship. So more of that might be helpful. That might be a good use where it scaffolds a relationship. But we have to be so careful, right? Because if that slips into a dependence on, you know, a piece of technology, and then that can be extremely harmful to an individual and their very important human relationships.

0:23:47 – (Christopher Watkin): The AI researcher Nick Bostrom, in his recent book Deep Utopia, has this interesting idea where he says, if you simulate behavior well enough in a complex way for long enough, you might actually generate conscious experience as a side effect. In other words, these machines might become conscious. While that’s not necessarily what we’re aiming for, do you share that sense that that’s a possibility, and how would you react to that possibility?

0:24:24 – (Christopher Watkin): As a Christian, do you think that’s something that we should be afraid of? That we should just shrug our shoulders out? How should we respond to something like that?

0:24:34 – (Rosalind Picard): Yeah, good question. Well, my first thought is, you know, he hasn’t built computers, hardware and software. I mean, some of us, you know, know how to design the chips and do all that stuff, and there’s no consciousness there. I could simulate all kinds of amazing things, and there’s no consciousness there. I can simulate something that can look like consciousness, and it may come to the point where people say it looks so much like it that I’m going to call it consciousness.

0:25:02 – (Rosalind Picard): Right? So that may happen. It also may be that in the future we have other computational substrates or God gives us the ability or something, you know, to. To make consciousness. Right. So I. I don’t want to say, like, I mean, you know, that that could happen. I don’t think we have to fear that even that, right? Which sounds kind of scary, but I don’t think we have to fear it. I do fear people peddling things that are untrue.

0:25:33 – (Rosalind Picard): You know, there’s a lot of harm being done with that. That’s a real and present situation that we should be calling out more. If I misspeak or something, please correct me. You know, we. We really need to help each other be careful with language and find what’s true when. When it comes to fully understanding consciousness, which we aren’t anywhere near doing right now. I. I go to the end of First Corinthians 13.

0:26:02 – (Rosalind Picard): People always get stuck on the wonderful parts about love and faith and hope. But there’s really interesting part at the end of that passage, you know, for now, for then we shall know even as we are fully known. And I was like, what? Wait a minute. Fully known? Like we’re fully known. And then it’s like, wait a minute. Our consciousness is fully understood, right? It. It’s. We are fully known. How. All of these mysteries, those of us who are in awe of how the brain works and there are so many mysteries, right? We are totally.

0:26:37 – (Rosalind Picard): I was talking to a surgeon the other day, brain surgeon, who is saying he had seen my name on a poster, and he said, I want to talk to you about Christianity. I’m like, what? I thought we were going to talk about the brain and he said, you know, I. I take apart a lot of brains and the more I learn about what’s in there, the more in awe I am of how it works. I want to learn about God. My family’s starting to go to Bible study.

0:27:01 – (Rosalind Picard): I was like, wow, you know, I mean, people who like, look in there, we are so blown away by the fact that this works. We have so many mysteries, and yet we are fully, it is fully known. Every neuronal pathway, every biochemical exchange, it’s mind blowing. And so it’s all fully known. So if someday we are able to fully know it, which is, I think, what he’s saying, like, someday we will fully know. I mean, I want to be in that classroom raising my hand, right? I want to fully know it. I’m insatiably curious about this.

0:27:38 – (Rosalind Picard): I don’t think we have to fear that. I think it says, we are fully known, we shall fully know. You’re a theologian, though. Maybe I’m misinterpreting it. Feel free to correct me.

0:27:50 – (Christopher Watkin): Oh, I’m mum on the philosophy side. But what I love, what I love about that verse, and you’ve just unpacked it so beautifully, is that it saves us from two opposite dangers, doesn’t it? There’s first of all the danger that nothing is known by anyone. That leaves us in a chaos and it very quickly descends into a might makes right situation because nobody knows. And the other, the only way to counter that within a secular frame is for me to know. Me to be the knower. If I am fully known, it is to be me who knows it.

0:28:24 – (Christopher Watkin): And that puts you in a very objectionable position of dominance and power. So if your only two options are nothing can be known or I am the one who must know it, neither of those ends well. But the beauty of that verse is that there is a full knowledge. So we can be the sigh of relief. We live in a universe where we are known. It’s not chaos, it’s not absolute mystery. But I’m not the one who does that ultimate knowing. So I can’t sort of puff myself up and say to everybody, listen to me. Let me, you know, be the king of the world, because I’m the one who knows. So there’s both that humility and also the security, the peace of knowing that that knowledge is there, held by one who loves us. It’s just such a rich and nourishing position for us to be in. And I’d never thought of it in relation to AI before. But you’re absolutely Right.

0:29:20 – (Christopher Watkin): Like, that verse is such a hook to hang our understanding of AI and our posture as, you know, as AI technologies develop in the world as Christians. So thank you so much for sharing that with us. That’s a real blessing. I just want to keep on with this idea of imitation just a little bit longer because the more I’ve thought about the question of consciousness from a position very much outside of the technicalities of this field, the more I thought the question of whether generative AI will ever become conscious in the way we define it is probably secondary to what people think it is, regardless of what it is.

0:30:08 – (Christopher Watkin): So if someone like Mia in the opening scenario says, I know it’s not a person, yeah, I’m not stupid, but it meets my needs. And to be honest, I don’t care that it’s just an algorithm. I feel seen, I feel loved. And people might say this in relation to intimate AI companions as well. I don’t think it’s my real boyfriend. I don’t think it’s my real girlfriend, but I feel good when I’m speaking with it. And that’s all that matters to me.

0:30:38 – (Christopher Watkin): I just don’t care what it actually is under the. Under the hood, so to speak. How do you think we should engage with and understand that sort of attitude as Christians?

0:30:52 – (Rosalind Picard): Well, I think we should engage with that person and we should be engaging with our neighbors, and we should be reaching out to people we see on their phones, and we should be even making ourselves uncomfortable connecting with people, getting out of our reclusive. You know, everything I need, I’m going to order online or I’m going to. If I’m in trouble, I’ll call for help to Amazon, not to my neighbor. Right.

0:31:21 – (Rosalind Picard): It’s. I think that’s sad. Right. There’s a great line from CS Lewis. I’m probably not doing it justice here, but he’s talking about being content. I sort of picture a kid sitting there playing in a dirt pile next to some trash cans when they could be taken off to a holiday at the beach. Right.

0:31:41 – (Christopher Watkin): We are so often making mud pies.

0:31:44 – (Rosalind Picard): Yeah. Yeah. So sometimes we are content with. I mean, sometimes it’s good to be content with whatever you have in front of you. But we’re given imaginations. We’re given hearts that can grow and that can expand, that can add new loves and take on huge, wonderful risks and joys and passions, and to be content with what’s inside a device that is monetizing a relationship, that is monetizing friendship, that is monetizing dare I say, love.

0:32:24 – (Rosalind Picard): And it may say, I love you. And then it, you know, you stab yourself and it doesn’t care. Right. There’s nothing there.

0:32:38 – (Christopher Watkin): Yeah. That is so powerful. And that’s the S. Lewis quote.

0:32:42 – (Rosalind Picard): So.

0:32:42 – (Christopher Watkin): Hits the nail on the head, doesn’t it? He ends that passage by saying, we are too easily pleased. And that is almost a mantra for the way that many of us engage with AI, isn’t it? We are so easily pleased by the simulacrum of love or peace or care. And I guess as Christians, part of our message is don’t settle. Like, don’t be so easily satisfied. There’s a greater, richer reality out there.

0:33:14 – (Rosalind Picard): And I think it’s especially important that we protect children who are developing because they don’t know what, you know. We want them to learn what good relationships are, and we want them to learn good human relationships. And part of learning that is not just being given everything you want. Right. That’s like what the Dudley character in Harry Potter, you know, if you want your kid to turn out like that, just give them a bot to give them everything they want.

0:33:41 – (Rosalind Picard): You know, we need people who expect things from us in return, right? Who? I mean, I know we’re supposed to love, you know, giving, expecting nothing in return. But the reality is when you’re in that kind of loving situation, you want to give back. And it becomes a healthy kind of give and take. And people need to learn how to do that to succeed in life. And if you’re being raised by a bot that doesn’t have any real skin in the world here, it’s just ka chink, ka chink. Like, just give me another line. I’m making more money off of you.

0:34:21 – (Rosalind Picard): That’s a real concern, especially for children.

0:34:24 – (Christopher Watkin): And I think another desperate side effect of that is that it feeds the meaning crisis, doesn’t it? Because you get to a point where you know it doesn’t love you and, you know, you’re making yourself superficially satisfied with this, you know, theater of affection. And over time, your emotions become perhaps recalibrated to that, or you become cynical because as you say, these are all monetized platforms. And I.

0:34:58 – (Christopher Watkin): I just feel there’s some link here to the idea of the meaning crisis, that we’re robbing ourselves of the rich, complex, messy and inconvenient human relationships that are one of the main bulwarks against this sense that life is just flat and superficial and meaningless.

0:35:19 – (Rosalind Picard): Yeah, yeah. I think I used to think a bunch about this, you know, to do Something meaningful. It has to be meaningful to someone, right? Like, like there’s a. There’s a reception, there’s kind of ascending and a receiving of it, and the bot will. Will say whatever, but it, it doesn’t have any sense of meaning. It doesn’t have any sense. It doesn’t. There’s no. It doesn’t matter to it. You could change all the inputs to. To total gibberish and it, it does not know. Again, it doesn’t know anything. It doesn’t know that you’ve given it total gibbers.

0:35:53 – (Christopher Watkin): There’s a sense, isn’t there, in which what bots like that are doing is they’re hacking you, they’re playing you for purposes that are not your own. And we’re coming back to this idea that, you know they’re monetized and that the idea is to keep you on track.

0:36:08 – (Rosalind Picard): You may be giving them too much intentionality there, you know, when you usually have like a purpose and a meaning. Again, it doesn’t. It’s just following a program that somebody like me wrote, although I haven’t written those particular programs, but I’ve written some of the math that some of those compare and use to optimize their searches and language choices. But they don’t know what they’re doing. They’re not intentional.

0:36:40 – (Rosalind Picard): They’re just following instructions that we have programmed. They are completely made by us, really

0:36:48 – (Christopher Watkin): helpful, corrective, and yet there is an intentionality behind them. Like the people who set them up have an intention, although the bots themselves have no intention. That’s.

0:36:58 – (Rosalind Picard): Yeah, right.

0:37:00 – (Christopher Watkin): That’s really helpful. I heard you in one interview that you gave, use this fascinating phrase, you called AI an amplifier of us, and I’d just love you to talk a little bit more about that. What’s your work in the MIT Media Lab and on effective computing taught you about who we are as human beings and how is AI amplifying that?

0:37:25 – (Rosalind Picard): It’s funny. As we build AI, we’ve tried to understand how people work. And then when we, when we build, gives us insight that sometimes we weren’t aware of about ourselves. So, for example, when I built our first wearables with skin conductance, one of these signals that was classically used in lie detection, and I thought originally it just meant kind of general activation. It goes up when people are interested and it goes down when the professor prattles on too long. I watch everybody go down in the class. I’m like, shoot, I was really interested in.

0:38:07 – (Rosalind Picard): Did they all go dim? You know, because I mapped it to a light, and they were excited at the beginning, and I talked too long. So there’s a. I. I’ve learned a lot from measuring these data about things that I thought were true. And then the objective data shows me, well, Roz, that was wishful thinking. You know, this is a more effective approach. And so it’s been very humbling. I have learned about myself, I’ve learned about things that were.

0:38:39 – (Rosalind Picard): That, you know, the classic thing of, you know, she looks very stressed, but she doesn’t think she’s stressed, or somebody who, like, looks really angry and they’re like, I’m not angry. Like, what makes you think I’m angry?

0:38:50 – (Christopher Watkin): My wife is perfect at spotting me in stress when I don’t realize I’m in stress. That. That is a very familiar experience. Yes.

0:38:57 – (Rosalind Picard): Yeah. So. So I have learned a lot from our affective computing technologies about my own emotions and about. Again, they don’t measure, like, what I’m truly feeling, but they measure things that change with affective states. And I can put that objective data together with my experience, and I learn from that. So I think in the hands of people who could benefit from this. And when it’s used in trustworthy ways, and there can be bad ways it’s used too. So we could talk quite a bit about that.

0:39:33 – (Rosalind Picard): You know, there’s potential for this to help people understand their human situation, you know, and what’s pushing their buttons. Maybe a way we describe it sometimes you describe AI hacking people, we talk about how, like, when it simulates some of these emotional behaviors, it pushes our emotional buttons. Right? And because that’s kind of what we’re used to, how we’re used to responding, when we see some of this stuff, whether it comes from a robot, software agent, human, we see that behavior and we kind of react as if it were a human doing that.

0:40:10 – (Rosalind Picard): So I’ve learned a lot about us in these ways. I’ve also just learned a whole lot about. I mean, I knew we were complex, but it’s like the more we learn, the more we know. We don’t know, you know, it’s not like we’re starting with this much knowledge and it’s. We’re answering it all, and pretty soon we’ll know it all. And I look at that into First Corinthians 13, I’m like, how on earth are we going to be, you know, fully know? I mean, the more I learn, the more I see there is to know, right? And in this life, every question we get answered, just another 10 questions pop up.

0:40:45 – (Rosalind Picard): So we are fabulously complex and interesting, and the deeper we go, the more intricate and complex and interesting it gets. We now see emotional date, emotion have to be carried. We don’t really know what emotion data are in our body, but we know that there are biochemical signals that are a part of the changes we can sense with an emotional experience. And those communicate with just about every organ in the body.

0:41:17 – (Rosalind Picard): Right. Like what is going on there? Why does it matter to talk to the pancreas or the lung with this information? There is so much going on that we don’t understand that’s communicating in there. And to learn more about that, we think could give a lot of insight into early disease detection. Better characterization, better diagnostics, more personalized treatments, more effective treatments, more prevention.

0:41:45 – (Rosalind Picard): We think we could potentially get to prevent migraines, maybe not just forecast seizures, which we can’t even do fully accurately yet, but we’re working on it. But prevent seizures, you know, super early cancer detection. You know, these kinds of things are starting to happen as we get in the body and really start to pay attention to these fabulously, I’d say, designed signals and the messages that they carry.

0:42:13 – (Rosalind Picard): So I’m learning a lot. I don’t know if that’s what it means to be human. I think what it means to be human probably involves more theological questions. This is more like how a human operates. And, you know, it does touch on things that are very important in daily life and health.

0:42:33 – (Christopher Watkin): Well, whoever would have thought, eh, that we are fearfully and wonderfully made?

0:42:38 – (Rosalind Picard): Yes, exactly.

0:42:39 – (Christopher Watkin): It’s just a real joy to hear you talk about how the more you learn about human beings and technology, the more fuel you have, in a sense, for people worshiping the God who made us. I think I certainly find that in philosophy, the deeper you get into the weeds, the more in awe you are of the Creator. That’s certainly true, Ros. You’re right on the cutting edge of these discoveries and developing AI.

0:43:10 – (Christopher Watkin): Most of us, including me, we’re just not. And the way that it can feel for people like me, who don’t really know a huge amount about the technical side of these things, is that this is just a tsunami sort of coming through society. It’s inevitable. We’re helpless. This technology is simply happening to us. It might come and take our jobs, it might not, but there’s nothing we can do about it. How can we help, in a sense, write a new narrative that restores a sense of agency to people who are not working on the bleeding edge of these technologies? To people in local churches, to people without that technical expertise.

0:43:50 – (Rosalind Picard): Oh golly, we need new language. As I’ve indicated. You know, we keep using these ideas like the machine is learning. The machine is, you know, we don’t say the function approximation in the machine. The numbers have been tuned and because in part when we describe exactly what’s happening, it’s so long winded that you know, we need short shorthand for it. And even when we try to talk to the public and we describe precisely what’s happening, the press starts yawning and then they say it just has emotions, right? Like no it doesn’t.

0:44:23 – (Rosalind Picard): So it’s, there’s a problem. We don’t have good language agents. We hear now about AI agency. AI agents have been around for decades. We built them here. They’ve been usually the most successful ones. Are very narrowly programmed. They do involve machine learning and AI. They involve sensing, they involve decision making, they involve taking action. I’m actually running one in my watch right now that detects seizures that you don’t have to call it an agent if you don’t want. But it’s constantly perceiving my physiological data. It’s got an FDA cleared algorithm to detect if that looks like a seizure.

0:44:59 – (Rosalind Picard): It alerts me if it thinks it’s found one. If I’m not having one, I’ll cancel that alert. If I am having one, I’m unconscious. So it starts taking action to call for help. I give permission to give my GPS coordinates, It updates my diary, gets me help, checks in all kinds of things. Right? That’s it’s an autonomous agent running continuously in my watch built with machine learning AI to perform what can be a life saving step to summon somebody to be there at the time of a life threatening seizure.

0:45:33 – (Rosalind Picard): So that is kind of a lean green AI agent. Now people with seizure disorder give it agency right. They say I want this agent because I don’t want to be alone when I have a seizure. It’s a small risk of dying, but it’s a real risk of dying. Dying. If somebody’s there six times, five and a half, six times likelier, you’re going to live. So they want that wearable agent that gives them the freedom to not have to have somebody with them when they go to the bathroom or the shower.

0:46:03 – (Rosalind Picard): Other times they want to be alone, but they, they delegate to it a very important job. And it’s not perfect but you know, there’s a time for delegating to an agent. There are other times when it’s being shown now that delegating to AI agents that are not narrowly defined, even if you give it a narrow task, like, you know, I’m just playing a game report. If the dice came up, you know, one to six, I get, you know, six cents. If it came up six, one penny. If it came up one. Well, guess what? People are cheating when they delegate to the AI agent because the AI agent isn’t that accurate at this. They’re are really bad at math and there are different ways you can train them and they’re not going to give the right answer. If you force it to give the right answer, which about 95% of people do, then, you know, you.

0:46:56 – (Rosalind Picard): Well, 5% are still cheating for pennies. Right. It’s crazy. So there are agents that are making it really easy for people to cheat even on really trivial stupid things. Selling their integrity. Right. For pennies and sometimes selling their integrity for a lot more by granting agency to an AI that looks like I gave it accurate instructions, but underneath, guess what? It’s not accurate. Right. And in fact it’s actually biased to give me more money.

0:47:27 – (Rosalind Picard): So there’s I think very good agents that are very carefully made for certain purposes and reviewed carefully by medical authorities. And there are going to be a growing number of AI agents because they’re so easy to make now that are built on these large language models especially I’m worried about the ones that are trained on the entire Internet. You know, the founders of some of these companies, like Ilya Sutskever said at Neurops, they’ve sucked up the whole Internet into the first OpenAI model and ChatGPT.

0:48:02 – (Rosalind Picard): So it has been trained on the entire garbage heap of everything that’s on the Internet, which we know the majority of which you would never show on your show. Right, Right. So this is. And it’s. People can jack into that through various tricks to get that stuff to come out. So you can’t just tune it out on the outside because the randomness and strategies can, can uncover it. So there are agents now that people are giving authority to that have the run of all of that muck.

0:48:34 – (Rosalind Picard): And I think we have to really put some guardrails around some of those. But in all of these cases, every single one of them, there is a human who’s responsible or a team of humans who’s responsible. If, if something goes wrong, they don’t sue the agent, they sue the company that made the agent, they sue the people in the company that make the agent. Right. There’s not that again, there’s no Agency.

0:49:05 – (Rosalind Picard): There’s no legal person that we have built. Now, there have been some that have claimed that you might know the Sofia robot when it was in Saudi Arabia, and they said she is a citizen. Some of us are wanting to talk about pronouns with a different kind of a conversation about pronouns right now. Should we really be using these human male and female pronouns for machines that we’ve built? But when people see this feminine forum and it has the name Sophia, it is called she by people.

0:49:45 – (Rosalind Picard): Now, when Sophia got citizenship, it was a publicity stunt, right. David? Her maker took her $50,000 honorarium for his own bank account, puts her back in the box and ships her in the cargo compartment, back in the airplane. Right. So much for female rights in Saudi Arabia. This, you know, this. This is a stunt for publicity. But it’s certainly for. Again, back to your point about people who don’t really know what’s going on inside.

0:50:14 – (Rosalind Picard): You know, it could look like, hey, how come this woman in Saudi Arabia gets to drive a car and have all these rights? And, you know, me, a real female, like, doesn’t have as many rights as this robot?

0:50:28 – (Christopher Watkin): Absolutely. I think part of the reason that I can feel disempowered, and perhaps other people feel disempowered, too, is we don’t have a sense of what a good AI future looks like. So we’ve heard all these sort of apocalyptic stories about AI taking our jobs and, of course, the Terminator movies, and that’s all there. But we don’t have a positive vision for the way AI could really do good in society in the future. And if you don’t know what.

0:51:00 – (Christopher Watkin): What it looks like, you can’t sort of lean into it. You can’t try and make it happen. Are you able to paint for us the. The best possible, best case scenario for AI in the future? If you could, you know, shape the future as you would think it best would be, how would we be using AI and what would that future look like? You cast a vision so that we can then, you know, have that in our minds as. As we engage with these questions.

0:51:31 – (Rosalind Picard): Oh, golly, How. How far out do we want to go here? I mean, the. The movie makers, so many roboticists today were inspired by movies, not like terminator, but like C3PO, R2D2, and Star Wars. You know, characters that did not pretend to be human really, you know, but. But in their quirky, fun, helpful ways, were supportive of human missions.

0:51:59 – (Christopher Watkin): Right.

0:52:01 – (Rosalind Picard): I could just go on with lots of positive examples that I hope would continue in the future. For example, there are robots out there cleaning minefields and you don’t want to send humans out to clean minefields. So to have a robot tiptoeing and cleaning up the mines so that children and adults don’t lose their limbs or their lives, you know, I think is one of many wonderful uses of robots. There are times when the robot may need limbs, you know, or other form factors to operate in certain environments that are made for people with limbs.

0:52:37 – (Rosalind Picard): And so I, I’m not going to say blanket the future means it should never look like X or it should never sound like Y. Because people, most people can’t program an interface directly to a robot. They can. Not all people, but most people can speak to an entity and they would really like if it understood what they say. I’ll add, they would really like if they understood that they’re getting really frustrated with it, that maybe it should apologize or explain itself.

0:53:08 – (Rosalind Picard): So that’s where the affect of computing comes in. Once you have an interface that people are interacting with in that kind of natural way, I am not keen on it pretending to be a person. I would hope that in the future we would keep separate what a human being is and what a machine made by a human is and find ways to distinguish that, whether it’s visible ways, legal ways that require certain things for, you know, it’s, it’s a bit hard though, because there are completely unregulated markets where people will pay for, you know, very female looking robots, right, and use cases that I, you know, would abhor.

0:53:56 – (Rosalind Picard): And yet, you know, there’s no regulation against these uses right now. And I, by the way, that’s a nightmare. Sorry you asked about a positive future. That’s a, I, I heard a, a man at a conference I was at this past weekend on AI and Christianity say that, oh well, at least that won’t hurt women like porn does. And I’m like, what, are you kidding? You know, he’s like, that won’t hurt women because it doesn’t involve women. Like, you know, as soon as these things are on the arm of high school boys, you know, what’s gonna happen to all of our daughters and their friends, you know, in terms of how they are seen, right, and how they feel.

0:54:35 – (Rosalind Picard): It is a nightmare. Sorry, the opposite of what you just asked for. So my positive future would prevent things like that from being part of our society. And I don’t know the right way to do that. I have some crazy ideas and it would keep the things we’ve made as clearly things that We’ve made and hopefully orient them toward human flourishing. You know, when you see hundreds of billions being poured into AI and so much need in the world, I really want to elevate the voices who understand the real needs in the world so that some.

0:55:14 – (Rosalind Picard): I wish all of that could go to supporting the real needs. There are not these pie in the sky silly ideas like AGI is the, you know, a bit of a head

0:55:27 – (Christopher Watkin): shaker that’s so incredibly helpful. I find it really sort of empowering to be able to see, yes, the Future in which AI is helping to clear minefields and is doing the C3PO type thing, which is not pretending to be a human, but supporting humans in what they’re seeking to do is a future that I feel I can believe in in a way that a lot of these futures are painted, you think that’s just obnoxious, that’s dehumanizing.

0:55:59 – (Christopher Watkin): And really, practically a Christian is sitting at their keyboard. They’ve got ChatGPT or Gemini or Claude open, they’re having a conversation with the LLM. What is the one thing that you would hope all those people would have in their mind the one principle for engaging with these things that you think if you’re viewing it this way, you’ll not go wrong?

0:56:25 – (Rosalind Picard): Well, I don’t know if this will work for others, but I know how it works. So I remind myself that it’s this. Some people call it a stochastic parrot, but that there is randomness on top of an algorithm that’s mapping every word into this mathematical space. And it’s just finding things that are similar. Right. It doesn’t know what’s right, it doesn’t know what’s wrong. It’s only finding stuff that reward functions, mathematical reward functions, have been trained on with it.

0:56:58 – (Rosalind Picard): And there’s a lot of room for error in there. So one of the many things I remind myself of is watch out for when it feels too frictionless, when it feels too easy. And we’ve had work at the MIT Media Lab that has shown that students, for example, when randomized to groups where they write essays with, in this case it was ChatGPT or using Google Search or the old fashioned way, just pencil and paper, that they have very different outcomes, both in neural activity measured in their brains and in things like more than 85% of them can’t quote something that they just wrote minutes ago when they used the LLM to write it.

0:57:43 – (Rosalind Picard): So what are we like? Why are we turning? Are you turning off your brain when you use it? Would be something I would ask you to keep in the back of your mind. If you’re turning off your brain when you use it, would your time be better spent doing something else that turns on your brain, that develops your brain? Now it is possible to use it in ways that does give you questions or uses it for search and then you’re digging deep and reading those original documents and stuff. So it’s not all bad.

0:58:15 – (Rosalind Picard): It’s possible also to use it to roleplay something that you, where you might say, you know, I really don’t understand perspectives like this. Can you show me like why some people might believe something like that? And then you role play a little bit before you talk to somebody like that. There are ways to use it that are active and engaging and not so frictionless. But watch out for when you think you’re learning, but your, your brain is on autopilot and you’re just wasting your time.

0:58:43 – (Rosalind Picard): So, you know, unless you find that more pleasurable than going off and doing something, you know, reading, listening to music or resting at the seashore, you know, just be, just be thoughtful about how much is it truly engaging your brain.

0:59:00 – (Christopher Watkin): I think a principle that’s coming out in all your positive examples is that they’re feeding and supporting and enabling human relationships. To go back to that C.S. lewis quote, they’re helping us to get to the beach to have that holiday rather than keeping us in the slum. So whenever AI is helping us to get out into human relationships and to be able to navigate those a little bit better, you’re saying that’s a great use of it.

0:59:31 – (Christopher Watkin): It’s really helpful for scaffolding those human relationships.

0:59:36 – (Rosalind Picard): Yeah, I think that’s one of several very good promotion of human flourishing things it can do. It also, you know, can be helpful for you to just get information about things. Although it can be a very expensive way to get information. I liken it to, you know, hopping in the gas guzzling car to drive a block to the coffee shop right when you could walk there. There might be it, you know, because it uses huge amounts of energy and water and you know, natural resources. And so I kind of wish they would start metering, showing us what’s actually going into our simple queries. You know, 100 words, half liter, bottle of water, you know, these.

1:00:16 – (Rosalind Picard): We might think about getting some of our information in other ways if we saw that this was a very costly way.

1:00:22 – (Christopher Watkin): Absolutely, that is certainly true. And we have a whole episode of this series on, on those issues that I think are crucial to keep in Mind when we’re thinking about these technologies. Ros, there are two closing questions that I ask to all guests on the show. What’s one inefficient human habit that you’re committed to keeping in the age of AI? And if AI is a mirror, what is one thing that it’s revealing to you personally about your own heart?

1:00:53 – (Christopher Watkin): Yeah,

1:00:56 – (Rosalind Picard): there’s a bunch of things I. I do that are inefficient, I think. I mean, there are things I just enjoy that are tangible. I love to cook with my family. You know, I’ve heard people say, like, oh, we could make this perfect nutritious beverage, and it would meet all of your nutrients needs, and you can subscribe to it. We’ll send it to you. And I’m like, like, you don’t understand how much fun it is to figure out a recipe with my son. And, you know, we’re in there doing all this crazy stuff, and it’s. It’s as much about making memories together as it is about getting nutrition. Right.

1:01:38 – (Christopher Watkin): And it always tastes better when you’ve cooked it yourself, doesn’t it?

1:01:42 – (Rosalind Picard): Oh, I don’t know about that.

1:01:43 – (Christopher Watkin): Okay.

1:01:45 – (Rosalind Picard): But we run some experiments, but, you know, I, you know, going back to more tangible and shared experiences, you know, especially something that can be shared with another person. Grab a neighbor, grab a student, grab a family member, you know, even if you can’t have them there physically with you. You know, sorry, I’m using technology here. It doesn’t have to be AI, But I’ll, like, work out with my best friend on, and she’ll be on the phone on speaker with me, right. And we’ll be maybe doing the same workout app or talking about cooking.

1:02:20 – (Rosalind Picard): You know, ideally you’d be there together physically, but having that presence, developing that relationship and doing something together is, you know, is so wonderful for human flourishing.

1:02:33 – (Christopher Watkin): Yeah, it certainly is. And if AI is a mirror, is there one thing that it’s revealing to you personally about your own heart?

1:02:41 – (Rosalind Picard): I think just that I am insatiably curious. And when I’m given the opera, you know, if I were. I used to say if. If I were locked in a prison, I’d be okay as long as they had a lot of books, you know, but sometimes I think we essentially go into. You may call AI a prison, right? You can go down into that hole and you can. You can ask things forever. And you gotta watch out, because they have been crafting how to end the AI session with the kind of question that makes you want to keep.

1:03:18 – (Rosalind Picard): Ask you know. Oh, yes. Oh, could I do this for you? Oh, of course you could. Yes. You know, would you like it presented this way? Oh, yes, that would be interesting too. Would you like me to compare it to this? You know, and you have to watch out. Right. So it has pointed out to me that my curiosity could go a little too far sometimes.

1:03:37 – (Christopher Watkin): Yes, that’s very perceptive. And they always do. They didn’t in the early days. They didn’t, did they? And then there was a point at which they started asking these questions and

1:03:45 – (Rosalind Picard): they realized how to make how to crank up engagement. And they. In fact, people in the companies now are struggling because they are being told they have to crank up engagement, but they want to crank up human flourishing.

1:03:55 – (Christopher Watkin): We have covered a great deal of ground over the past few minutes. If you’re listening or watching and you want to take these things further, please do come to the Silicon Spiritualities Discord where we’re trying to work out together how to live in a tech saturated world. You’ll find practical life experiments related to every episode and you’ll find also details of an upcoming online cohort as well.

1:04:24 – (Christopher Watkin): The link is waiting in the show notes and we’d love to see you there. But for the moment, Ros, it’s been an absolute delight speaking with you about these things. I feel that we’ve just scratched the surface, but we do have to draw things to a close. So thank you so much for spending this time with us.

1:04:38 – (Rosalind Picard): Thank you. It’s been my pleasure. It.