Loading summary
Transcript84 lines
- [00:00]
Tara Tan
Think about 10 kids. Only three can read.
- [00:04]
Rei Nomoto
Oh, so 70% of the kids can't. They don't have reading classes at 12th grade.
- [00:07]
Tara Tan
12Th grade?
- [00:08]
Rei Nomoto
Wait, that's high school.
- [00:09]
Tara Tan
Yeah.
- [00:10]
Rei Nomoto
What? Welcome to Culture and Code, a podcast about the biggest shifts in culture and tech. I'm Rei Nomoto, a creative entrepreneur and founding partner of iancorp, a global innovation firm based in New York, Tokyo and Singapore.
- [00:26]
Tara Tan
And I'm Tara Tan, managing partner at Strange Ventures, an early stage fund focused on the future of computing.
- [00:38]
Rei Nomoto
So today's topic is Interfaces versus the mind. And this conversation is inspired by two separate anecdotes and headlines that I saw in the past couple months. So the first anecdote is more of an observation, and I'm kind of curious to hear what your observation might be on this topic. But I have a teenage daughter who's now 16 and she now is in a boarding school, so we don't live with her ongoing basis, but like a year or so she was at home and I would help her with her homework. And I'm watching her navigate her desktop on her PC and she doesn't quite understand the concept of folders or these nested folders. I'm pretty, I think, organized about trying to structure folders in a logical, easy way so that, you know, it's very clear, not only to me, but particularly like when I work with a team and team needs to find different files in different folders and more clearly labeled and more logically organized the files are, and then the folders are. It's easier to find, right? Yeah. But I've noticed that my daughter, our teenage daughter, doesn't organize files at all, and instead she exclusively relies on desktop search to find files. I then read an interview with a professor I teach at in graduate school in New York, but also I read an interview with a different professor who said that she notices her college students don't know how to organize files and they do a similar thing when finding files. They just exclusively use search to find files and they don't organize files at all. So bothered me thinking about how the interface. There are different types of interfaces, but particularly computer and screen interface and finding files and literally physically organizing files on a server or on a desktop and then cognitively organizing files that the influence of interface has on people's psyche and mind and just wanted to use that as a topic. So before I go to the next research topic that I was going to bring up, any reaction to that?
- [03:02]
Tara Tan
I mean, yeah, I guess I'm in between because I don't Overly organize, and I rely on search most of the time, but sometimes I get into a world and I start organizing. But, you know, it kind of reminds me of the library's, like, Dewey Decimal system, where previously, because it was sprawling, you had to be quite, you know, sort of specific and organized about where things went. And nowadays I think that the amount of content is just so high that it's impossible. And maybe it's impossible, but it's not, you know, it's also not very efficient because I have duplicates of stuff. I know I've downloaded stuff many times. Sometimes I can't quite find it. So it's there. Yeah, so it's interesting. I think it's just, you know, sort of digital habits. I wonder that there's a difference here.
- [03:48]
Rei Nomoto
Yeah, yeah. So the other point that I wanted to bring up was a research study by an MIT researcher. And this was a paper published maybe about two months ago. You know, we are recording the middle of September, and I think it was like in June or July of this year, so a couple of months ago. So the title of this research called your brain on ChatGPT, accumulation of cognitive debt when using an AI assistant for essay writing tasks. And you've seen either this headline or maybe a similar headline. Yeah, read the paper. Yeah. So just to give context to the people who haven't read this paper particularly, what this researcher did was take a group of people, divide them into three groups, and the researcher gave these individuals, which are students, a task, which is essay writing. And what they did was, for these three separate groups, they gave the same task to write an essay, but the process was deliberately different. So the first group was they were able to use ChatGPT or any AI tool freely to write their essay. The second group was that they had to start writing manually, but then they could use AI to improve it. And then the third group was they were not allowed to use any tools, but they just had to write an essay out of their brain. And the research found that. That the AI powered writing. The brain activity was way less than the third group, which didn't use any AI to write. And then the middle group, which is AI assisted writing, was somewhere in between. I happened to see an interview with this researcher, and then the interviewer studied the interview by asking the researcher, oh, does that mean that AI makes people dumb and dumber? And the researcher was quick to point out, no, no, no, that's not what this study means. And then she went on to explain how, like, you know, AI assisted writing makes the writing or the output of this AI prompted essays, quite vanilla and quite generic.
- [06:21]
Tara Tan
Right.
- [06:21]
Rei Nomoto
And then the human written essays had more originality, maybe they had more mistakes, but the individuals in that group, each person has a different style and each writing was quite different. Whereas the first group was more homogeneous and then the middle group was somewhere in between. So the researcher went great lengths to explain that, oh no, no, no, AI is not making people dumb. But then I kept thinking, but that it actually may be.
- [06:50]
Tara Tan
I mean, I don't think it's dumb, I think it's just rewiring our brains. I mean, the Google effect is well studied. The research around the Google effect is that the search engine has actually rewired neural pathways for us over the last couple of decades.
- [07:03]
Rei Nomoto
Okay, CSMO on that.
- [07:05]
Tara Tan
Yeah. So that means that we actually have lost a lot of our memorization skills. So you remember when you were kids, you were able to remember like your house number, your friend's like landline number and all of that.
- [07:17]
Rei Nomoto
Yeah, landline numbers, yes.
- [07:19]
Tara Tan
The school actually has been lost and they actually studied it and found out that using Google over a couple of decades has actually rewired our neural pathways so that now we actually send less energy towards memorization and more energy towards referencing. So basically like, oh yeah, I went to this place and this reminds me of that. So it actually biologically changed our brains. So I will not be surprised if in a couple of decades we'll see the same thing happen with LLMs.
- [07:49]
Rei Nomoto
Right, right, right, right, right.
- [07:51]
Tara Tan
Like just use of them. And how does that reshape our neural pathways and where we send energy to and what where our brain allocates energy to. Right, and that's a real, real question.
- [08:01]
Rei Nomoto
Yeah. Do you find yourself thinking differently now that we all use AI one way or another way more frequently than we used to say three years ago, do you find yourself cognitively different?
- [08:12]
Tara Tan
Yeah, I think that's a couple of things. I think that many people tend to, especially if you're starting to first use it, rely extensively on LLMs to think first, if you know what I mean. So like I have a problem. Let me ask versus think and then, you know, sort of research or verify with the LLM. And I think that's a dangerous place to be because then you're not putting in original thought. And it's interesting how the different models are actually designed to respond to you. I noticed that clients like anthropics models are actually designed. They actually, you know, sort of design the experience to when you ask it something, it'll ask you Back So ask for human input. Whereas a lot of the other models would be like you ask it, it gives you back straight up. So this sort of encouragement or this banter model I think is actually quite different in my experience between using an anthropic model and say a gemini or an OpenAI model.
- [09:05]
Rei Nomoto
Do you have a preference?
- [09:06]
Tara Tan
I use it for different things. I hop between them.
- [09:08]
Rei Nomoto
Yeah. Okay.
- [09:09]
Tara Tan
For different things and sometimes I exhaust one and then I go to the other.
- [09:12]
Rei Nomoto
Give me a use case scenario of you switching between different models. Yeah.
- [09:17]
Tara Tan
If I'm researching something, I might use Gemini for like deep research through like, you know, it's amazing. It trawls through like 300 pages in minutes.
- [09:26]
Rei Nomoto
Yeah.
- [09:26]
Tara Tan
Like web pages. It's incredible resource. Whereas if I'm trying to stress test an idea or just have a sounding something to bounce back and forth, I would use Claude a lot. And an OpenAI I use for kind of general more like it's really good for unstructured into structured. That's my experience. So I use it for different things and I often hop between even with one problem space. But yeah, go back to your original topic around. Does it reshape our brains? Absolutely. I think that the temptation to rely on it to think is high.
- [09:58]
Rei Nomoto
Yeah. I wonder in terms of these tools influencing the way we think or don't think, you know, going to an LLM to think for us first, I wonder if it's also giving us an encouragement to be better editors. So for instance, when I use different AI tools and LLMs, one of the things writing and I've found whenever I used ChatGPT or Claude for idea generation, I find those ideas like, you know, let's say if I want to write a newspaper, a newsletter or something that I find their topic suggestion to be very boring or very like mundane, generic, really generic. Yeah. So I spend way more time on thinking of an angle, thinking of a perspective, thinking of an idea that I can only come up with or I bring my personal perspective to it so they become more interesting and a different point of view as opposed to something that anybody can write. So the initial like 0 to 1. I've tried to use AI to do it, but I've found it to be inadequate. Where I use and I found LLMs to be super useful is editing. So I would write and I'd ask for feedback and he would point out. And I don't necessarily agree with every point that it makes, but it makes me, oh, you know what? This is a good point. Oh, you know what? This is I'm keeping the original writing that I did. Oh, this is, you know, so I kind of go back and forth, go back and forth. Speaking of which, LLM, I happen to use this writing tool called Lex that allows you to switch between different tools, different LLMs, and depending on every week as a new model or updated version of the existing model. So it changes fairly frequently and it recommends the best model at the that given time. You know, let's say the clause, this version is better than Chat, GPT or Gemini or what have you. But anyways, so I use it for editing purposes. I used it to expand my writing, hey, add 200 words just to see one additional case. I found most LLMs to be kind of inadequate. And I switch out of that writing tool to do research to find other points to add to it and then write it, edit it. And then so like 0 to 1, I come up with my own 1 to 8. I go back and forth between myself and then the LLMs and then like 8, 9, 10 of 10 steps, exclusively me. So I find myself thinking harder to edit, like generating. I can rely on AI quite a bit, but I find myself editing more than I used to.
- [12:46]
Tara Tan
I started really trying to use it to challenge me, to challenge my. Okay, yeah, yeah. So I actually tried to move it into a, an opinion mode, which is actually quite interesting because I feel like that actually pushes both me and the LLM more because the LLM would basically prompt me be like, what do you really think? Or this isn't interesting enough.
- [13:09]
Rei Nomoto
So when you say open your mode, what do you mean by that? Exactly.
- [13:12]
Tara Tan
So like it kind of moves into more of an opinionated model as opposed.
- [13:16]
Rei Nomoto
To a fact based mode.
- [13:17]
Tara Tan
Yeah. Or moving away from more fact based or more just like pacifying when it's like, oh yes, great thoughts here, move it into an opinionated mode. And I find that that's actually more interesting for me.
- [13:32]
Rei Nomoto
Yeah.
- [13:33]
Tara Tan
So I should try to move it there.
- [13:36]
Rei Nomoto
Do you find yourself agreeing it or disagreeing with it? Quite a bit or even like offended?
- [13:42]
Tara Tan
Sometimes it's been a little rude. But you know, there was one when I was just kind of like. I mean, I was extremely tired so I kept pushing it and then I said I was like, no, I don't like any of this. And then at the third try, I think it got really frustrated or the pattern was like it told me, you don't want anything, just take a break now or just don't do anything at all, just go take a walk. It told me to go touch some grass no way. This is a parasocial relationship forming with the AI. But it's true. Yeah, I got it too. It's interesting. I find it now like I try to flip it into a. Either an opinionated mode or a mode where a little bit of personality comes out. And I find that challenges me more too. So it makes me actually get better at my writing and my thinking.
- [14:30]
Rei Nomoto
Yeah. This is a slightly different topic, but related to the initial topic of interface versus the mind when it comes to these tools for our kids. So our oldest one is 16 teenager. Our youngest one is just about to turn five. So there's a pretty big gap between the two. Oldest one, like I said at the beginning, she goes to boarding school, so she's out of the house. We can't quite control how she's using these, her tools. And I know for a fact that she uses ChatGPT. We don't let her use a paid version, we don't pay for her account. But I'm pretty sure she uses it to do her homework and all that. Right. And as far as we know that I like to think that she has good judgment and she's not having like strange thought or dangerous thoughts and having those conversations with AI and you know, having like thoughts that we don't want her to have. As far as Mino. But it's something that I think we have to watch out so that she doesn't go in a dark rabbit hole. Yeah. But for the time being I think she's got a pretty decent, I think judgment. Now our youngest one, I mean he has absolutely no concept of AI or what have you. Right. One thing that we've done pretty deliberately, about two years ago, when he was two, we stopped letting him use any screen. So we took away his iPad. We don't have a TV screen at home. We have device, you know, my wife and I, we have our individual devices, you know, iPhones. And sometimes he uses it to see photos or videos that we took. But rarely we give him that device to, to distract him because a couple years ago we noticed that when we would try to take it away from him, he would get really frustrated and really cranky that we felt that these devices and these screens were cognitively making him short tempered or at least whenever we would try to where we tried to take it.
- [16:37]
Tara Tan
Right.
- [16:38]
Rei Nomoto
Yeah. Screen dependent. Yeah, exactly. And in the past couple years we might be biased, but I think it was probably one of the best decisions that we've made made for him. So, you know, we read to him quite a Bit before he goes to bed, we don't show him screens at all. Yeah. And then the reason why I brought up is that when would I, or should I introduce AI to him?
- [17:00]
Tara Tan
It's such a timely topic and it's very polarizing. So I almost took the opposite approach.
- [17:05]
Rei Nomoto
You did?
- [17:06]
Tara Tan
I work fairly, like, you know, my kids five and four, we're not super. They don't have their iPads. You know, they have maybe like 30 minutes of TV a day, you know. But my kid right now, he's doing like Khan Academy math. He's in kindergarten. And actually I took the almost the opposite approach where I actually exposed him to AI this year, where I let him talk to the LLM directly to ask different questions. So, like, right now he's like super into Pokemon. So he's asking about, like, different types of Pokemon. And it's interesting. I mean, you know, I think Gen Alpha, which is your younger kid, I mean, they're growing up in a totally different world, right. And the way they learn is going to be supremely different from how we learned in school, how we step in, and I don't think that's the right answer. There are schools like the Alpha school, which has a very interesting model. The only way they learn is the kids. I think it's like, you know, sort of middle school. I believe they only learn for two hours a day with an AI tutorial. There are adults in school. They're not allowed to teach. So they're not allowed to teach. They're only allowed to make sure things happen, go haywire. But the only learning that kids do is with AI tutor for two hours and then the rest of the time is project time. So they build stuff with their friends and peers. They have different build projects, but that's a very controversial school. But apparently great scores have just like, you know, skyrocketed over the last six months. I just saw a report today that American kids, our test scores are going down the drain.
- [18:40]
Rei Nomoto
I heard that. Yeah. Yeah.
- [18:41]
Tara Tan
Guess how many. What percentage of kids have reading and Math proficiency at 12th grade right now?
- [18:48]
Rei Nomoto
30. 40%.
- [18:49]
Tara Tan
Which is crazy. 30%? Yeah. In most markets or most countries, it's about 90%. In Japan, it's like 99. 30%.
- [18:59]
Rei Nomoto
Wow.
- [19:00]
Tara Tan
30%. So think about 10 kids, only three can read.
- [19:06]
Rei Nomoto
So 70% of the kids can. They don't have reading at 12th grade. 12th grade.
- [19:11]
Tara Tan
12Th grade.
- [19:12]
Rei Nomoto
Wait, that's high school.
- [19:14]
Tara Tan
Yeah. What? In the U.S. something is not right. Yep. It just came out today, and I was looking at it. Something's not right. No.
- [19:22]
Rei Nomoto
You know what I meant when I say 30%. I thought 30% are the ones who can read and 70% can't read. It's the opposite.
- [19:29]
Tara Tan
You're saying 35% proficient in reading, 22% proficient in math. So 80%, 70 to 80% are basically illiterate.
- [19:38]
Rei Nomoto
And this is 12th grade, high school, 12th grade. That is crazy.
- [19:42]
Tara Tan
Like, this is not good. Like, this is so, so incredibly bad.
- [19:48]
Rei Nomoto
But is that because of the use of computers and devices?
- [19:52]
Tara Tan
I think there's a whole bunch of reasons. I don't think you can just blame it on computers and devices. You know, like, could be Covid. I don't know, a bunch of stuff.
- [20:01]
Rei Nomoto
Right.
- [20:01]
Tara Tan
But it's bad.
- [20:02]
Rei Nomoto
Wow. Like, wow, America.
- [20:07]
Tara Tan
It's like we're literally illiterate. Like, it's so bad. So. And it makes me wonder, like, I'm like, is it screens? Is it right to blame screens? I don't think so. Because if you look at other.
- [20:16]
Rei Nomoto
Yeah, it's probably not. Yeah.
- [20:17]
Tara Tan
Other countries, like, you know, even developing countries have like at least 80 or 90% proficiency.
- [20:23]
Rei Nomoto
Right.
- [20:24]
Tara Tan
They're like 30%. It's horrible. And it's not screens. Like kids are exposed to screens all around the world, you know, so like, something not right. Something's not right here. And I hear very different opinions from parents. Some parents are like, I don't like my kids touching AI. I don't want them touch it, to touch it until like at least college. But I do think that there is some sort of skill with working with AI. And so like, but there's a balance, obviously. So I don't think there's a right or wrong answer. But for me, I'm actually like slowly introducing them because I want to see how they interact with. I mean, it's partly selfish. I'm like so curious how they interact with it.
- [21:01]
Rei Nomoto
Yeah, yeah, yeah. I'd say recently, just a couple weeks ago, I did let my four year old interact with the voice mode of ChatGPT.
- [21:11]
Tara Tan
That's right.
- [21:11]
Rei Nomoto
Yeah. And what I thought was interesting and potentially promising is so, for instance, I speak Japanese to me and my wife, we're Japanese, so we speak Japanese at home to him. Before this summer, he would answer everything that we say in Japanese. So we speak 95% of time in Japanese to him. And then maybe 5% when he doesn't quite understand what we're saying, we speak in English. But 100% of the time his responses were in English because he goes to. He's now in kindergarten but he was in 3K and his friends are English speaking and obviously. Right. School is in English. We go to Japan this summer and within three weeks he switched from English to Japanese. Wow. And within three weeks he was responding to us in Japanese. He was staying with my wife's family, grandparents. So he was 100%, not just us, but then other people are Japanese speaking. So he was able to, within three weeks, pick up enough words, enough Japanese to be able to imperfect, very broken. But still his responses became Japanese. And then we started school, obviously this September, he's still speaking back to us in Japanese, but just so that he can keep up with Japanese, I let him use ChatGPT voice mode and started having a conversation in Japanese. And he was having a conversation and he was getting such a kick out of it. Yeah, yeah. And the thing. And this is like kind of a wh sword. He would never stop talking like the LLM. Right. He would say something and he would ask back a question and then my son would say something and he would respond. And he was just kind of never ending. So we had to take it away from him at some point. Yeah. But I thought that was a decent use of AI for a young kid. And it may be even for grownups too, like to teach a language or to learn a language. And having that kind of conversation partner, before you had to be a human being, but now LLMs are good enough to be able to be a conversation partner. And that I saw a lot of potential.
- [23:14]
Tara Tan
I mean, going back full circle to how we started this conversation. Right. Like, I was thinking about it the other day and I was like, I was talking to a friend and they were like, so crazy. Just the amount of new releases and innovation and invention that's happening in the last few years, two years, feels insane. And I'm like, I actually think this is going to be status quo. Like, this is a new normal. The amount of speed of new information, it's a new normal. So actually, I think we're humans are in an evolutionary moment where we're going to be able to have to learn to take in that amount of information at that speed all the time.
- [23:48]
Rei Nomoto
All the time.
- [23:49]
Tara Tan
All the time. I think this is the new normal. I actually don't think that it's going to slow down.
- [23:53]
Rei Nomoto
No, it's not going to slow down.
- [23:55]
Tara Tan
It's not going to slow down. I think this amount of information is the new normal. So the question is, do we need to build tools along the way to help us keep up? Is curation going to be more important or are the kids going to come in and just be able to absorb this amount of information that quickly, but the speed of learning is going to be exponential? Absolutely.
- [24:17]
Rei Nomoto
Yeah. Cool. So just to wrap up to two things before we wrap up conversation. Yeah. I want to do our takeaway, each of our takeaways, and I want to try this new segment idea that you and I talked about before recording, which is sharing a story. But so the takeaway for this conversation about the influence that interfaces have on our minds, I'll go first, which is borrowing what you said. I don't think that these new interfaces and these new tools will make us more stupid. I think they may make us more lazy. But to your point, I think they rewire our brains to think differently or to react differently to whatever that comes at us. So my hope for observation of the current state and the future state is that these tools, whether the prompt based tools or voice based tools or other tools, they may make us a little bit lazy in some ways, but I think it opens up other opportunities and I think they rewire our brains definitely to open up possibilities. That's my takeaway from this conversation.
- [25:25]
Tara Tan
Cool.
- [25:26]
Rei Nomoto
Yours.
- [25:26]
Tara Tan
My takeaway is maybe more of a parting thought, which is you talked about how your kid is able to like, after three weeks in Japan, pick up Japanese. And I would say embrace that child in you and keep an open mind and have some sort of neuroplasticity, that sort of framework. I would encourage, like everyone to kind of stay in that, stay in that kid mode where your mind is open and you're learning all the time. And I think that's the only way to move on.
- [25:59]
Rei Nomoto
Cool. Cool.
- [26:00]
Tara Tan
All right.
- [26:00]
Rei Nomoto
All right, all right. Yeah. Oh yeah. Oh, yeah. Oh, yeah.