Latest / Dr Justin Coulson's Happy Families / AI Reduces Memory

Transcript

AI Reduces Memory · Dr Justin Coulson's Happy Families

• 0:05
  Speaker 1: What is AI actually doing to our brains? How is
• 0:09
  Speaker 1: it affecting our thinking? Are we doing too much cognitive offloading?
• 0:13
  Speaker 1: Is AI our autopilot or our co pilot? These are
• 0:17
  Speaker 1: the questions we're going to answer today in our Happy
• 0:20
  Speaker 1: Families podcast. It's a doctor's Desk episode where we have
• 0:23
  Speaker 1: a look at two brand new studies that are raising
• 0:27
  Speaker 1: flags of warning about how we're using AI. What is
• 0:30
  Speaker 1: doing to us, to our kids, to our ability to think.
• 0:34
  Speaker 1: So this stuff might sound like it comes from the
• 0:36
  Speaker 1: University of Duh. However, however this stuff matters because our
• 0:41
  Speaker 1: kids at school, at university, at life, and we as
• 0:46
  Speaker 1: well are using this stuff more and more every single day.
• 0:50
  Speaker 1: How is AI affecting our brains? The Doctor's Desk episode
• 0:54
  Speaker 1: of The Happy Family's podcast is next Stay with Us
• 0:59
  Speaker 1: gooda of the Happy Families podcast where you get real
• 1:01
  Speaker 1: parenting solutions every single day. This is Australia's most downloaded
• 1:05
  Speaker 1: parenting podcast where Justin and Kylie couson You Love Ai. Kylie.
• 1:08
  Speaker 1: This is a doctor's Desk episode especially for you. Are
• 1:11
  Speaker 1: you ready for you want to.
• 1:12
  Speaker 2: Throw me under the bus. You're hilarious, I am.
• 1:16
  Speaker 1: I've got two studies we've got we've gotta work cut
• 1:19
  Speaker 1: out for us today.
• 1:20
  Speaker 2: It's called cognitive off floating.
• 1:23
  Speaker 1: It's called cognitive offloading. That's right.
• 1:25
  Speaker 2: So, and since I carry the mental load of the family,
• 1:29
  Speaker 2: yes and all of that, I just need to offload, Okay,
• 1:34
  Speaker 2: sharing it, sharing it with someone?
• 1:36
  Speaker 1: Okay, do you feel like you've offloaded? Do you feel better?
• 1:40
  Speaker 2: So much better?
• 1:41
  Speaker 1: Gonna make feel worse, let's talk about it. Two studies,
• 1:43
  Speaker 1: most recent one comes out of Brazil, although it really
• 1:46
  Speaker 1: applies all around the world. There's a researcher there who
• 1:49
  Speaker 1: got one hundred and twenty research students together, sorry business
• 1:53
  Speaker 1: students together. They are undergrad students, and basically gave them
• 1:56
  Speaker 1: an assignment said here are some content that I want
• 1:59
  Speaker 1: you to work through or want you to figure it out.
• 2:01
  Speaker 1: Half of them were told that they could use AI
• 2:03
  Speaker 1: and should use AI. The other half were told not to,
• 2:06
  Speaker 1: and they had to prepare a presentation. There was a
• 2:08
  Speaker 1: whole lot of technical stuff, a bit of non technical stuff,
• 2:11
  Speaker 1: but they were given a couple of weeks to put
• 2:13
  Speaker 1: together a detailed presentation. They had to share the presentation
• 2:16
  Speaker 1: in small groups. They thought that was the research study,
• 2:20
  Speaker 1: but that sneaky psychology researcher pulled them back about forty
• 2:25
  Speaker 1: five days later, a month and a half later, and
• 2:27
  Speaker 1: gave them a twenty item multiple choice quiz on the
• 2:30
  Speaker 1: content of their presentations, on the content of what they'd
• 2:33
  Speaker 1: been asked to research, to see how the two groups
• 2:36
  Speaker 1: would go, one group using AI, the other using traditional
• 2:39
  Speaker 1: study skills and not being allowed to use AI. We
• 2:42
  Speaker 1: are presuming that they didn't cheat and use a bit
• 2:44
  Speaker 1: of AI.
• 2:45
  Speaker 2: I was going to ask if there was like a
• 2:47
  Speaker 2: control room, not a control group, to make sure that
• 2:50
  Speaker 2: they were doing it by the board.
• 2:52
  Speaker 1: Being supervised all the way through. Yeah, yeah, nice work.
• 2:55
  Speaker 1: So what would you anticipate were the results on the
• 2:59
  Speaker 1: ability to reac called content after forty five days following
• 3:03
  Speaker 1: the presentation.
• 3:04
  Speaker 2: Well, I'd be staffed regardless of whether I use AI
• 3:07
  Speaker 2: or not. Forty five days is.
• 3:08
  Speaker 1: A very long time, this is in our lives.
• 3:11
  Speaker 2: But I would suggest that if you are doing the
• 3:14
  Speaker 2: research and doing the heavy lifting, it's going to be
• 3:17
  Speaker 2: much easier for you to recall information than allowing somebody
• 3:21
  Speaker 2: else to do it for you, and you're just kind
• 3:22
  Speaker 2: of regurgitating what they've shared.
• 3:25
  Speaker 1: No surprise is here. Then sixty eight point five percent
• 3:27
  Speaker 1: of the questions were answered correctly by the people who
• 3:29
  Speaker 1: did the traditional learning. Fifty seven point five percent of
• 3:32
  Speaker 1: the questions were answered correctly by the students who use
• 3:34
  Speaker 1: the chatbot.
• 3:34
  Speaker 2: That's still pretty high. I'm actually pretty impressed.
• 3:37
  Speaker 1: They've barely passed. They got to fifty seven percent, whereas
• 3:40
  Speaker 1: the others were more than two Thursday, they were almost
• 3:42
  Speaker 1: at a distinction territory sixty eight point five. It's called
• 3:44
  Speaker 1: seventy percent. But here's what I would say. When you
• 3:48
  Speaker 1: have to stand up in front of people and give
• 3:50
  Speaker 1: a presentation, you still do have to learn the material
• 3:53
  Speaker 1: much more than if you're going to get a chatbot
• 3:55
  Speaker 1: to push out an essay for you, or do something
• 3:59
  Speaker 1: where you're just going to read it, over it, make
• 4:00
  Speaker 1: sure it sounds about right and submit it. So I
• 4:03
  Speaker 1: think that there are some limitations in this particular project
• 4:08
  Speaker 1: whereby these students still have to have a reasonable level
• 4:11
  Speaker 1: of investment in the project that they're doing. Because they
• 4:14
  Speaker 1: do have to stand up and present, they do have
• 4:16
  Speaker 1: to know.
• 4:16
  Speaker 2: As opposed to writing an essays that we mean precisely.
• 4:19
  Speaker 1: Because we talked about this in the pot a little
• 4:21
  Speaker 1: while ago, The Australian journalists from the Australian did a
• 4:25
  Speaker 1: big expose on how much cheating's going on in universities
• 4:28
  Speaker 1: and high schools today. Kids basically saying we just get
• 4:30
  Speaker 1: AI to write it. We don't even try to pretend
• 4:33
  Speaker 1: that AI didn't and they're getting high distinctions, but what's
• 4:36
  Speaker 1: their degree going to be worth in reality? So that's
• 4:40
  Speaker 1: the first study. It's pretty basic stuff. What it basically
• 4:43
  Speaker 1: shows though, is that the kids that did the traditional
• 4:46
  Speaker 1: learning they spent about six hours on the assignment. The
• 4:48
  Speaker 1: kids that use the technology assist only spent about three
• 4:51
  Speaker 1: and a half hours on the assignment. Even when they
• 4:53
  Speaker 1: did some statistical controls, they found that the people who
• 4:56
  Speaker 1: did the real learning rather than the AI learning, they
• 4:59
  Speaker 1: just had a deeper form of learning.
• 5:02
  Speaker 2: So it would suggest that the using of AI and
• 5:06
  Speaker 2: chatbots and things like that is actually creating a learning deficit.
• 5:11
  Speaker 1: Yeah, I think that's exactly what it's doing. You just
• 5:14
  Speaker 1: don't have to process the information and if you don't
• 5:16
  Speaker 1: have to work as hard correct If you're not going
• 5:18
  Speaker 1: to work as hard, you're not going to get the
• 5:20
  Speaker 1: results over time.
• 5:21
  Speaker 2: So there's no point in going to the gym and
• 5:24
  Speaker 2: asking you to lift my weights.
• 5:26
  Speaker 1: Yeah, that's exactly right. Yeah, you can go to the
• 5:28
  Speaker 1: gym and a forklifts or a machine can lift the
• 5:30
  Speaker 1: weights for you. The weight still gets lifted, right, but
• 5:34
  Speaker 1: you don't get the benefit of it. And that's why
• 5:36
  Speaker 1: I think we need to be asking fundamental questions about education,
• 5:39
  Speaker 1: like what are we trying to do with education? We're
• 5:40
  Speaker 1: trying to get kids through school or are we trying
• 5:43
  Speaker 1: to get them to learn stuff? Because if we're just
• 5:46
  Speaker 1: trying to get kids through school and through university, that's
• 5:49
  Speaker 1: kind of what our system does right now, and I
• 5:51
  Speaker 1: think that's why it's doing such a bad job. Whereas
• 5:54
  Speaker 1: if we are really interested in helping them to learn stuff,
• 5:56
  Speaker 1: we've got to move away from this technology booster. After
• 6:00
  Speaker 1: the break, an even more interesting study, something that I
• 6:04
  Speaker 1: think is going to be a little bit more surprising,
• 6:06
  Speaker 1: that finds that most people just do what chat GPT
• 6:11
  Speaker 1: tells them to do, even when the AI is wrong.
• 6:16
  Speaker 3: Stay with us, We're back at a doctor's desk episode
• 6:26
  Speaker 3: of the Happy Families podcast.
• 6:27
  Speaker 1: Thanks so much for joining us, Kylie. Let's crack straight
• 6:30
  Speaker 1: into this one. First thing to highlight chat EPT Gemini
• 6:34
  Speaker 1: Claude Grock. They still make regular mistakes. They hallucinate, they
• 6:39
  Speaker 1: give you incorrect information. They make hallucinate. That's a technical
• 6:43
  Speaker 1: term for it. When an AI gives you incorrect information,
• 6:46
  Speaker 1: it's called a hallucination. You did another one.
• 6:48
  Speaker 2: I did not know that one. That's so funny.
• 6:50
  Speaker 1: So the BBC in the UK did a study last October.
• 6:54
  Speaker 1: They found that even the most advanced AI chatbots, and
• 6:57
  Speaker 1: I know that October last year. That's six months ago.
• 7:00
  Speaker 1: That's a long time in the AI world right now,
• 7:03
  Speaker 1: But in October last year, wrong answers forty five percent
• 7:07
  Speaker 1: of the time. That is not trivial.
• 7:11
  Speaker 2: It's almost half the time.
• 7:13
  Speaker 1: That's really concerning, and so why are we even using it.
• 7:17
  Speaker 1: That's why I couldn't and wouldn't use it when I
• 7:19
  Speaker 1: wrote my book, because it was so untrustworthy. I tried
• 7:23
  Speaker 1: to lean on it a couple of times to help
• 7:25
  Speaker 1: me with a couple of things that I was struggling with,
• 7:28
  Speaker 1: and I found it made up references, it gave me
• 7:30
  Speaker 1: false reports, It wrongly analyzed data. Like after I'd used
• 7:36
  Speaker 1: AI for about two weeks, as I was trying to
• 7:39
  Speaker 1: form some ideas with the book and develop some stuff,
• 7:43
  Speaker 1: I was like, I actually can't do this, the book
• 7:45
  Speaker 1: will be wrong and it will be worse for the
• 7:47
  Speaker 1: use of AI. So I had to write the book myself.
• 7:49
  Speaker 1: It was it was wonderful. I mean, I learned so
• 7:52
  Speaker 1: much right, but I was really hoping that I could
• 7:55
  Speaker 1: just get through a little bit easier and I couldn't.
• 7:57
  Speaker 1: I had to do it the right way anyway. So
• 7:58
  Speaker 1: AI makes a lot of mistakes, but a lot of
• 8:01
  Speaker 1: people don't understand that reality. So brand new paper from
• 8:05
  Speaker 1: the University of Pennsylvania. Stephen Shaw is the researcher there,
• 8:09
  Speaker 1: and a marketing professor by the name of Gideon Nave.
• 8:11
  Speaker 1: They did a serious experiments and they found the following
• 8:13
  Speaker 1: users tended to take the output of chat gpt at
• 8:16
  Speaker 1: face value, even when it gave them an incorrect answer.
• 8:21
  Speaker 2: So I just got to query you on this. Yeah, like,
• 8:24
  Speaker 2: did they know it was the wrong answer?
• 8:26
  Speaker 1: No, they didn't, So then you can't problem.
• 8:29
  Speaker 2: I feel like you can't judge somebody for following something
• 8:33
  Speaker 2: that's being recommended to them if they don't know it's wrong.
• 8:37
  Speaker 1: No, no, no, I'm not blaming the people. Well I
• 8:39
  Speaker 1: am kind of, because this is what happens when you
• 8:40
  Speaker 1: rely on AI.
• 8:42
  Speaker 2: But no, I mean, but the AI is wrong. But
• 8:45
  Speaker 2: you just look at you just look at life in general.
• 8:48
  Speaker 2: How many times, I mean we would just at this
• 8:49
  Speaker 2: house retreat and within their own spheres, we had kind
• 8:54
  Speaker 2: of experts in their own fields contradicting each.
• 8:58
  Speaker 1: Other ceaselessly, endlessly right.
• 9:00
  Speaker 2: So who do I follow? Who's right and who's wrong?
• 9:04
  Speaker 2: In so, AI is no different to life in general.
• 9:07
  Speaker 1: In the most striking experiment involving three hundred and fifty
• 9:10
  Speaker 1: nine participants, participants followed AI's correct advice ninety three percent
• 9:15
  Speaker 1: of the time. So they've got the advice. You're like, yep, okay,
• 9:18
  Speaker 1: I'll do it ninety three percent of time. But even
• 9:21
  Speaker 1: when AI gave them the wrong answer to your point,
• 9:25
  Speaker 1: they followed that wrong advice eighty percent of the time.
• 9:29
  Speaker 1: It was seventy nine point eight I'm rounding up by
• 9:31
  Speaker 1: point two eighty percent of the time.
• 9:34
  Speaker 2: What did our forefathers do when they needed to make
• 9:37
  Speaker 2: a decision?
• 9:38
  Speaker 1: They just figured it out right, And then if they
• 9:41
  Speaker 1: made a decision and it didn't work out, they were like,
• 9:43
  Speaker 1: I'm not going to do that again.
• 9:44
  Speaker 2: Well, it's usually a life of death situation sometimes. But
• 9:50
  Speaker 2: I feel like with all of the progression we've made
• 9:55
  Speaker 2: in technology, we've actually stopped list to ourselves and really, yes,
• 10:03
  Speaker 2: using our own brain.
• 10:06
  Speaker 1: Here are the keywords. The researchers in this study at
• 10:09
  Speaker 1: the University of Pennsylvania use the term cognitive surrender. Cognitive surrender,
• 10:15
  Speaker 1: and that's exactly what you're saying. It's like, somebody can
• 10:18
  Speaker 1: do it for me. I don't have to think about it,
• 10:20
  Speaker 1: I don't have to use my brain. I'm literally going
• 10:22
  Speaker 1: to surrender my cognitive capacity to the machine. I'm just
• 10:25
  Speaker 1: going to trust the machine. Would you do that?
• 10:27
  Speaker 2: When we hire an accountant, we surrender to them. Yeah.
• 10:32
  Speaker 1: We we do outsource a lot of stuff because we
• 10:35
  Speaker 1: can't do everything, Like we can't all be a business
• 10:37
  Speaker 1: manager and an accountant and a salesperson and a shoemaker,
• 10:41
  Speaker 1: and we can't make our clothes and make our food
• 10:44
  Speaker 1: and grow our crops and do all that. We have
• 10:45
  Speaker 1: to outsource. But we're doing it now on a scale
• 10:48
  Speaker 1: that we've never done it before, and we're actually giving
• 10:51
  Speaker 1: up knowledge that we have. I'll give you an example
• 10:53
  Speaker 1: or study that we talked about on a Doctor's Desk
• 10:55
  Speaker 1: episode a couple of months back, showing that doctors who
• 10:59
  Speaker 1: used AI to assist with diagnoses six months after they
• 11:04
  Speaker 1: began had less faith in their ability to diagnose patients
• 11:08
  Speaker 1: without AI subsequent to their time using it.
• 11:13
  Speaker 2: In other words, well, because every time they give a diagnosis,
• 11:16
  Speaker 2: they recognize it's not their diagnosis.
• 11:19
  Speaker 1: But they start second guessing themselves.
• 11:20
  Speaker 2: That's exactly right, though, but there's no faith in that diagnosis.
• 11:24
  Speaker 1: They were getting it right before as much as they
• 11:26
  Speaker 1: got it right with the AI, sometimes more because the
• 11:28
  Speaker 1: AI makes so many mistakes. Interesting, Oh my goodness. So
• 11:31
  Speaker 1: the take home message here everybody thinks.
• 11:35
  Speaker 2: Don't believe everything you read. That's the take home message.
• 11:38
  Speaker 1: Yeah. Yeah, everyone thinks that AI is getting better and
• 11:41
  Speaker 1: better and better, and we can surrender our thinking to AI.
• 11:44
  Speaker 1: But humans are becoming inappropriately reliant on AI and it
• 11:51
  Speaker 1: is not actually always helping us.
• 11:54
  Speaker 2: It's a whole native surrender.
• 11:56
  Speaker 1: Yeah, yeah, it sounds nice. We're losing a species something
• 12:01
  Speaker 1: critical to our existence, and that's the ability to think
• 12:05
  Speaker 1: things through for ourselves and make our own decisions and
• 12:07
  Speaker 1: have our own learning. The question is what are you
• 12:10
  Speaker 1: going for the outcome or the growth. If you're going
• 12:13
  Speaker 1: for the growth, do it the hard way. If you're
• 12:16
  Speaker 1: going for the outcome.
• 12:17
  Speaker 2: Wow, it doesn't matter.
• 12:19
  Speaker 1: It doesn't matter. That's our doctor's desperate today. Thank you
• 12:22
  Speaker 1: so much for listening. We hope that it gave you
• 12:23
  Speaker 1: some food for thought and we can't wait to be
• 12:26
  Speaker 1: with you again tomorrow for another episode of the Happy
• 12:28
  Speaker 1: Families podcast, which is produced by Justin ruland from Bridge
• 12:31
  Speaker 1: Media additional support from Mim Hammond's. For more and from
• 12:34
  Speaker 1: about making your family happier, visit happy families dot com
• 12:37
  Speaker 1: dot au