Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Lesson 1: Overreliance on AI
12 Plays21 days ago
As we offload our cognitive labor to machines, are we inadvertently erasing the 'road to expertise' and deskilling the very traits that define high-level thinking?
Topics discussed:
- The Hidden Cost of Deskilling: Why human proficiency and cognitive involvement decline when we outsource mental tasks to external systems.
- The "Levelling" Illusion: How GenAI allows users to perform at a high level without actual domain mastery, effectively breaking the ladder to expertise.
- The Multiplier Effect: Why the biggest performance boosts are reserved for those who already possess the skills to critically evaluate and edit machine output.
- AI as a Cognitive Amplifier: A three-stage framework—Foundational Fluency, Application, and Active Engagement—to ensure technology extends thinking rather than replacing it.
Recommended
Transcript
Introduction to AI in Learning
00:00:00
Speaker
What I want to do in this series of lessons is give you some ideas as to how to use artificial intelligence in ways that are constructive and disciplined.
00:00:14
Speaker
In other words, we're going to try to examine some possible strategies for learning with artificial intelligence. And maybe if I'm being a little grandiose, maybe even how to flourish with artificial intelligence.
Negative Impacts of AI on Cognition
00:00:32
Speaker
But the way I want to start this series of lessons is by by covering what's going to happen if you don't use AI in this disciplined way, as I mentioned, because a lot of people...
00:00:48
Speaker
the way they are currently using AI, not only is it maybe irresponsible, but I think it's also even detrimental to their cognitive abilities.
00:01:01
Speaker
In other words, they are undermining their own mental powers in the way that they are using the technology. And so that's why this first lesson is called over-reliance on artificial intelligence.
00:01:17
Speaker
Truth be told, there is actually already a term for this phenomenon. It's called de-skilling. Now, de-skilling has been in the literature for, i don't know, ah decades at least, 50 years, something like that.
00:01:31
Speaker
And one thing we can say about it is that it tends to crop up whenever there is a radical new technology. For example, when calculators came about, people were kind of afraid that you'd be unable to do calculations on your on your own because you're relying so much on the calculator. So let's begin here with a formal more definition.
Understanding De-skilling with AI
00:01:55
Speaker
Deskilling is the process by which human proficiency and cognitive involvement in a task decline as that task is outsourced to external tools or systems.
00:02:10
Speaker
Now, that is a very fancy way of saying it's when you offload you know your knowledge about how to do a task to a machine.
00:02:22
Speaker
Maybe the easiest example to get to our feet wet with this concept is what many of us who are roughly my age have done to themselves with their navigation apps on their smartphones.
00:02:37
Speaker
I used to be a pretty good ah navigator, I guess. I would know you know which direction Northeast Southwest is, right? And I can kind of guess also, you know oh we've been going for about a mile, two miles, five miles.
00:02:53
Speaker
All that is mostly gone for me now. i rely i over-rely, I should say. on my map app. And that is one example of de-skilling. I am no longer able to navigate the way I used to because I have offloaded those abilities, that task to a machine.
00:03:15
Speaker
right? A sort of set of algorithms. Maybe another example is that the other day I got into a car that didn't have a backup camera and boy, that I have a heck of a time parallel parking. I did it.
00:03:30
Speaker
So I'm not fully de-skilled in that domain yet, but it was considerably harder than it used to be. So these are examples of de-skilling and I hope you can see that's You know, it's not as necessarily bad, but there are some tasks that we want to be able to do without the aid of a machine.
00:03:54
Speaker
And maybe there's some tasks that we don't want a machine to at all, maybe ever do for us.
Smartphones and Cognitive Patience
00:04:01
Speaker
Okay, let's have an example now that's more so inside the realm of education.
00:04:08
Speaker
And it's a pre-generative generive AI example. That is this phenomenon that we're going to be talking about was present or seems to have been present even even even before the dawn of like chat GPT and stuff like that.
00:04:23
Speaker
And it comes by way of the researcher Marianne Wolfe and it introduces her concept of deep reading. What is deep reading? Deep reading is that slow kind of rich reading that so it's basically the opposite of skimming.
00:04:41
Speaker
You read a passage, your mind makes a bunch of associations, and then you read a little bit more and your mind makes more associations. It's the kind of reading basically that your literature teacher wants you to engage in all the time.
00:04:57
Speaker
And this is an ability that we have to cultivate. It isn't, you know, we don't come out of the box ready to read. We have to learn to read and then Deep reading is even more difficult because you have to know a lot of background knowledge so that when you're reading, you can make all these connections and it can be a richer experience.
00:05:18
Speaker
Well, Marianne Wolfe argues that smartphones have predisposed heavy users toward entertainment-oriented doom scrolling such that they can no longer engage in what she calls deep reading and now display a decline in cognitive patience.
00:05:38
Speaker
In other words, in your mind, you no longer have... that ability to let things take the time that they need to take.
00:05:49
Speaker
So to get some, you know, a rich experience from deep reading, that simply takes time. It's not something that you can really rush. And we have essentially trained ourselves, those of us who are heavy users of digital devices, to not be able to do that.
Digital Distractions and Memory
00:06:09
Speaker
And according to Marianne Wolfe, There are many negative results from this. First of all, we have shrinking attention spans. We have atrophying memory and we have a low boredom threshold.
00:06:24
Speaker
In other words, we get bored much more easily now. This is scary stuff, right? And you can tell quite easily that if you get bored really easily, then that's going to be bad news for your ability to learn. Sometimes when you're learning something, it actually takes a little bit of time, like a little bit of ah a time investment for you to see how interesting it is.
00:06:50
Speaker
But if you get bored too quickly and literally can't handle learning this thing anymore, well, the result is that you're probably not going to want to learn that thing.
00:07:01
Speaker
So scary stuff. But I did specifically choose Marianne Wolfe's research here for a strategic reason. Not everyone agrees that the reason for the you know decline in cognitive patients is digital device use.
00:07:22
Speaker
Some people say it might be other things. We see the decline in reading before smartphones were really prevalent. And so she's actually, her research is actually a great case study for the thing that I want to talk about.
00:07:38
Speaker
We are talking about the age of generative AI now.
Risks of Cognitive Offloading to AI
00:07:42
Speaker
That's the age that we live in. And I'm assuming that anyone watching this has already messed around with Cloud or you know Google's Gemini or of course, ChatGPT.
00:07:54
Speaker
And you realize the immense power of these large language models. they can do many very high level cognitive tasks, as I will demonstrate actually in many of these videos that we be I'll be making.
00:08:12
Speaker
And so what we're worried about, what the threat is with large language models, is that maybe we are outsourcing our thinking and even our creativity to large language models.
00:08:27
Speaker
So if we are you know getting de-skilled in anything, maybe what... what generative AI is de-skilling us in are these very high level cognitive tasks, right? So if you are a heavy user of this technology, if you use it for, you know, everything from writing emails and text messages to doing homework assignments to, you know, some people even use it to to get advice and and as a replacement friend, we'll talk about that. um
00:09:00
Speaker
But if you're using it in these ways, then maybe you're not working out the sort of mental muscles that you need to do a lot of things that are, well, to be honest, things that make us human, right? Things like logical structuring, creativity, problem solving, metacognition, empathy, maybe. Some people are just letting the machine come up with their empathetic responses for them instead of they personally working to you know figure out how to be empathetic to another person.
00:09:33
Speaker
Let's talk about just one of these right now. metacognition that's thinking about your own thoughts that is a very valuable skill when you are for example writing an essay you want to make it persuasive right and so you think about your conclusion and you have to think about why you find that conclusion so intuitive That exercise just is metacognition. You're thinking about your own thoughts. What thoughts lend support to this other thought, which I'm calling the conclusion?
00:10:13
Speaker
All that really is, iss metacognition. Now, if you offload this task to the machine so often that you can't, you know, really...
00:10:25
Speaker
flex that mental muscle anymore to think about your own thoughts, you might be in a lot more trouble than just not being able to write an essay, right? Thinking about your thoughts is absolutely essential for getting a lot of just daily, you know,
00:10:42
Speaker
high ah high cognitively demanding tasks done. So let me give you a little bit of food for thought here. We're gonna look at a research paper from k Croston and Bolici.
AI's Impact on Worker Skills
00:10:56
Speaker
Now, ah what they did for the record is they reviewed the existing literature on workplace applications of generative AI. In other words, this is not in classrooms.
00:11:11
Speaker
this it This isn't in in the workplace. But as it turns out, getting good data, at least as of now, about the effect of AI in classrooms is hard to come by. i will eventually share some of that with you. But there's a lot of good research on generative ai being used in the workplace.
00:11:31
Speaker
So. Here is the general finding. Low-skilled workers tend to see a massive boost in performance. In other words, you give someone who's kind of a newbie in a particular field, you give that person an AI for a crutch and they will do amazingly well. They sometimes do as well as professionals, right? Or high-skilled workers.
00:11:59
Speaker
What about these high skilled workers? They sometimes saw little to no improvement and sometimes they actually saw a decline in their quality of work.
00:12:11
Speaker
That's worrisome, right? So it's really good for low skilled people and either not good at all or actually negative for high-skilled people. Now, those aren't the actual the only options. I'll give you some more options, but let's talk about one study in particular first, and and then I'll explain to you in more detail.
00:12:33
Speaker
So, Croson and Bollici reviewed lots of studies. i'm gonna i chose this particular study because it's very illustrative of the phenomenon that's going on here. so Okay, in this study, professional programmers were compared to a group of 50 non-programmers using AI to complete a coding task.
00:12:56
Speaker
Actually, there's three buckets of people here. There's professional programmers not using artificial intelligence. That's one bucket. The second bucket is professional programmers, but they are allowed to use artificial intelligence.
00:13:12
Speaker
And then there's these non-programmers, right? So... Maybe they know a little bit about coding, but they're not professional programmers, but they're allowed to use AI. So we'll see what happens, right?
00:13:24
Speaker
Okay, let's go over the results then. The professionals were 30% faster with ai compared to professionals not using AI.
00:13:34
Speaker
In other words, if you compare it just the pros, those professionals that used AI were a lot faster than those that didn't. That's probably...
00:13:45
Speaker
Not terribly surprising, right? Just, you know, the AI does a lot of the work for you and you get to just kind of move along faster. Okay, awesome. Let's talk now about the non-programmers. 95% of the non-programmers finished the task in roughly the same time as professionals.
00:14:06
Speaker
It's also, by the way, roughly the same quality, not always, but um they basically got the job done at about the same time as the professionals, not using the AI, by the way.
00:14:20
Speaker
So that's actually very, um you know, ambiguous. What should we make of this finding? You might say to yourself, that's pretty awesome. You can basically not know how to code it very well and perform about as well as someone who is a professional in that job.
00:14:39
Speaker
Okay, maybe that sounds good to you. Let me tell you why it's probably a bad thing that these individuals perform so well despite being low skill in that profession.
00:14:54
Speaker
This is called the leveling effect. That's when people with low skill using AI kind of ah perform at the same level as professionals.
00:15:05
Speaker
And this is actually, at least in many cases, de-skilling, right? These low skill users are merely performing at a high level But without really being proficient in that domain, right? Like if you were to take away the AI from these non-professionals, well, then they wouldn't perform anywhere near as well.
00:15:30
Speaker
And certainly nowhere, you know, and not not nearly as quickly either. And so this is a case of, sure, the output is the same, but the process, especially the mental process, is not at all the same. These individuals cannot code without artificial intelligence. And here's the really worrisome thing about this.
00:15:55
Speaker
if you're using generative artificial intelligence in this way you are not creating a pathway for learning how to become proficient in that domain in other words If you're using generative AI like this to, you know, come up with some, some finish some coding task, you are essentially erasing the road to expertise.
00:16:21
Speaker
That is not how professional programmers got the ability to write code. They did it by, you know, kind of slogging through a lot. They would read a lot of people's codes. They would work on a lot of, you know, easier tasks and until they could work up to harder tasks. There's a whole lot going on when someone is becoming an expert coder. And by using AI, you are essentially following none of those steps toward expertise.
00:16:53
Speaker
So I hope you can see here that this use use of generative ai doesn't actually impart in you any knowledge or wisdom or you know any practical reasoning skills for how to be good at that job. Because you want to be good at these jobs, right? Otherwise, you are replaceable. If anyone with low skill and an AI can do a job,
00:17:19
Speaker
well then they'll just get someone else that can pay less than they're paying you, right? So it's pretty important to actually attain expertise.
AI as a Productivity Multiplier for Experts
00:17:28
Speaker
Speaking of expertise,
00:17:31
Speaker
There are some cases where the multiplier effect has been documented. What is the multiplier effect? That is when experts actually perform better with artificial intelligence. So i mentioned earlier, sometimes there is no you know,
00:17:51
Speaker
very good effect with AI if you are a pro. Sometimes it's actually a negative effect, right? There's actually a decline in quality. But sometimes there is this multiplier effect.
00:18:03
Speaker
So in essence, you are enhancing your cognitive powers. Maybe you perform the task at an even higher quality than you could have without the artificial intelligence. Or as the case that we discussed, maybe perform the task way faster than you could have without artificial intelligence. So Let's talk about this multiplier effect.
00:18:30
Speaker
um Why do where does it appear? First of all, it appears only in those that are highly skilled. You have to already know a whole lot about whatever task you're doing before you can see the multiplier effect.
00:18:47
Speaker
And I should also mention this, it's not because these people are really good at prompting, although they do make really good prompts. Let me just say this real quick.
00:18:58
Speaker
I looked at the prompts um that that the the professionals made and the non-professionals, and obviously the prompts from the professionals are way better.
00:19:09
Speaker
Because you can tell in as they're prompting the artificial intelligence, they they already kind of know what the code is going to look like. But the funny thing is, whether you're really good at prompting or not, the AI is getting so good that it kind of already knows what you want.
00:19:29
Speaker
And it'll actually give you, you know, something very similar, whether or not you prompt it very well. In other words, what I'm saying is that the people that had really good prompts, the professionals,
00:19:42
Speaker
their output with the AI gave back to them wasn't like that much better than the non-professionals because the AI is so powerful.
00:19:53
Speaker
The real difference here, the the reason why highly skilled people can have the multiplier effect relative to non-highly skilled people is that highly skilled people are better at evaluating and editing the output of the ai In other words, they already know sort of what to expect from the AI, what it's going to spit out, and they can look through it really quickly and fine tune it And essentially what they did is they automated most of the process, right? They just had the AI make whatever, you know, code.
00:20:31
Speaker
And then they just kind of fixed it up a little bit. And so that made it so that they could work really, really quickly and produce a very high quality product. And that's how you see the multiplier effect.
00:20:45
Speaker
Okay. Hopefully I'm beginning to convince you that you want to be an expert in whatever domain matters to you so that maybe you can see the multiplier effect and you want to avoid de-skilling as much as possible.
Mass De-skilling Risks with AI
00:21:04
Speaker
If I haven't convinced you of this yet, let me give you maybe one more example. That hopefully will get you shaking in your boots a little bit and we'll see we'll see what you think.
00:21:17
Speaker
Okay, one recent meta-analysis, give some real signs for concern. Okay, here's the question. Is mass de-skilling possible?
00:21:30
Speaker
Like, in other words, is it possible that we're moving toward a future where a whole lot of people just literally forget how to do a task because the machine is doing all the tasks for them?
00:21:46
Speaker
Kind of like in a in that movie WALL-E where people don't even know how things work anymore because the robots have been doing everything. Well, um this article does not say that we're heading toward the WALL-E future, but there are some signs for concern. Let me tell you the main finding. There's actually a lot of findings in here. I'm only going to mention one of them.
00:22:10
Speaker
There is a decline in mid-skill jobs and a growth in low-skill and high-skill jobs. What does this mean? Well, if you are sort of in between a professional and a newbie, you know, the mid-skill section of the job sector, there's not a lot of jobs for you. The jobs for you are declining.
00:22:33
Speaker
If you are a newbie, lots of jobs for you and if you ah ah have a lot of you know if you're an expert in some field there's also work for you although not as many as for the newbies right but if you are an expert there are still jobs for you obviously and this is not necessarily a good sign. You kind of want a lot of skills in the mid-skill section ah because that should be where most of us are.
00:23:04
Speaker
And you want the fewest amount of jobs sort of in the low-skill section, right? Because if you work as low-skill worker, your pay that a low-skill worker. your pay is like that of a low-kill worker And the fact that there's less mid-skill jobs means that even if you are a mid-skill employee,
00:23:28
Speaker
What job are you going to get? Well, there's more low skill jobs. So that's probably what you're going to have to work for. And that is, again, a low skill wage.
00:23:38
Speaker
Right. So essentially, we're going to get worse economic inequality. More people are going to work low skill jobs for worse pay.
00:23:50
Speaker
By the way, interestingly, you do need high-skill employees still because it is the high-skill employees that evaluate the work of the low-skill people using artificial intelligence.
00:24:04
Speaker
Because remember, the low-skill people using AI will perform about as well as someone with high skill, but you need someone with high skill to actually evaluate the work of the low skill worker.
00:24:17
Speaker
That's kind of funny, but that is that is a skills gap, right? There's gonna be a lot of people with low skills and a few people with high skills Not a whole lot of people in between.
00:24:30
Speaker
and that, you know, that's just kind of a bad sign, right? I put in here that this might make technological underemployment more prevalent. Underemployment, there's a lot of ways you can be underemployed.
00:24:44
Speaker
One of them is that you're basically overqualified for your job. And so, as i already mentioned, that's one thing that might happen and in the near future as AI is used more and more in the private sector.
00:25:01
Speaker
Okay, well, what does this mean for us?
Practical AI Use vs. Cognitive Laziness
00:25:06
Speaker
How can we safeguard ourselves against the, ah I don't know, I don't know if the mass de-skilling is going to happen, but how do we safeguard ourselves from de-skilling in general?
00:25:19
Speaker
Well, here's one thing you can do. i don't recommend it, but you can go all Nancy Reagan and just say no, right? Don't use the technology. I don't think that this will work. I'll give you two reasons. One of them is here on the slides, but here's the first reason why I don't think it'll work. Eventually, you're going to get tempted to use this technology. And that's because the human brain evolved to be, we can say, cognitively lazy.
00:25:51
Speaker
but maybe the more technical term here is energy efficient. If you don't have to think hard, your brain doesn't want you to think hard, right? Your brain wants you to take it easy, conserve your calories, right?
00:26:05
Speaker
That's sort of the environment that we evolved in. And that's sort of the way our brain still functions. And so for that reason, we will almost always choose convenient over convenience over effort, right? If something is convenient...
00:26:24
Speaker
yeah, we're gonna do that that's That's way more appealing to us than doing the hard thing just for the sake of doing hard stuff. Some people do that, but for the most part, most of us will go with convenience over effort. And so when it comes to these powerful technologies,
00:26:45
Speaker
you know, it's very likely that you're going to want to offload our your cognitive labor to the machine. If and when you can, you're going to use the technology in this sort of sloppy, undisciplined way that probably, maybe, leads to de-skilling.
00:27:07
Speaker
That's one thing I learned wanted to say. a second thing that I wanted to say, though, is that there really is... a better approach, right? It really is the case that it's possible to create the conditions under which you can see the multiplier effect, right?
Enhancing Thinking with AI
00:27:26
Speaker
You can use AI as a cognitive amplifier. In this case, in this sort of situation, the AI doesn't replace your thinking.
00:27:38
Speaker
It literally enables and extends your thinking, right? So in the lessons that follow, I'm going to give you some ideas as to how to do this with artificial intelligence. And let me just tell you now that even though there's many ways to break up the stages of learning i'm going to do this in three stages and you know roughly i think we can have ai exercises that align with those three stages and then you know i'll just give you some other ideas that i have that are maybe don't fit into this model but
00:28:17
Speaker
First of all, we're going to use artificial intelligence to establish foundational
AI for Foundational Fluency
00:28:24
Speaker
fluency, right? When you're learning new content, some new field, some new ah concept, some new whatever, right?
00:28:31
Speaker
The first step is just to learn the who, the what, and the where, the basic foundational terms, right? And you can use AI for this for in many ways, right? But more than anything, going to You can make sure that you keep your motivation high by using AI as a diagnostic. When am I ready to go to the next level? If you go to the next level too quickly, you're going to get frustrated and you're not going to want to study anymore. You're not going to want to learn anymore.
00:29:02
Speaker
But with AI, you can check yourself very efficiently. You can make sure that you've mastered everything that's relevant before moving forward. right So that is a very important tool in this stage.
00:29:15
Speaker
Stage two is application. um i really like, there's something called Bloom's taxonomy that's sort of like a hierarchy of when you're learning new things. And application is sort of right in the middle.
00:29:27
Speaker
And I think this is a very useful way to solidify knowledge of new material. And so I think we can use AI for substantive application exercises. So I'll show you how to do that.
00:29:43
Speaker
And then there's sort of the mastery level. So I call this stage active engagement.
Deeper Engagement and Analysis with AI
00:29:49
Speaker
And the way that I describe this is that you're making learning content more tactile. In in other words, you can literally sort of play with it.
00:30:00
Speaker
And these are things like evaluation, analysis, even creation. You can even sort of create new concepts. If you really understand something, you can make something with it.
00:30:12
Speaker
And so that's sort of the idea here behind stage three. So I will give you, um there'll basically be a video for each of these stages.
Cognitive Ladder Model for AI Use
00:30:22
Speaker
But I want you to sort of visualize this as a cognitive ladder. The way that I think of ah the way people are using ai right now is that they're climbing up a broken ladder. A lot of people are basically using it in a way that always keeps them as a novice, as a newbie, right?
00:30:43
Speaker
And the rungs up to expertise they've broken off, that's here on the left. And so in other words, I guess the technical way of saying this is that people are using ai in a way that de-skills them.
00:30:58
Speaker
But obviously, I think, hopefully at least, I've inspired you to want to become an expert in you know whatever domain matters to you. And for that, you need to climb up this ladder, right? So we need to essentially you know rebuild those ah broken rungs of the ladder.
00:31:18
Speaker
And as I mentioned, I think there's roughly three stages up that ascent. And i will go over those stages, but they are very basically foundational fluency, application, and active engagement, or what I sometimes call mastery.
00:31:37
Speaker
So we will go over all those, starting with foundational fluency in the next video.