Latest / Chief in Tech Deep Dive Shortcast / Chief in Tech S1E45 - Stephany Oliveros & Kristina Talova : Is AI Making Us Dumber? Understanding the Impact on Human Intelligence

Transcript

Chief in Tech S1E45 - Stephany Oliveros & Kristina Talova : Is AI Making Us Dumber? Understanding the Impact on Human Intelligence · Chief in Tech Deep Dive Shortcast

• 0:00
  So think about the last time you drove somewhere
• 0:03
  completely new without using your GPS. Oh, wow.
• 0:07
  I mean, I honestly can't even remember. Right.
• 0:09
  You probably can't. Over the last decade or so,
• 0:12
  we've just happily outsourced our spatial awareness
• 0:14
  to our phones. Yeah, completely. And, well, as
• 0:17
  a result, our natural sense of direction has
• 0:20
  severely atrophied. Like, we just don't know
• 0:22
  how to navigate anymore. And today, we are exploring
• 0:25
  whether large language models are doing the exact
• 0:28
  same thing to our minds. Yeah, it's a really
• 0:30
  important question. Are we just blindly following
• 0:32
  a cognitive blue dot? I have to say it is a phenomenal
• 0:35
  analogy because the psychological mechanics behind
• 0:38
  both of those phenomena are remarkably similar.
• 0:41
  We're really facing a massive cognitive shift
• 0:44
  right now. And that shift is exactly what we're
• 0:46
  focusing on today. We are pulling from the research
• 0:49
  of Kristina Tullova, specifically her work titled,
• 0:53
  The Cognitive Impact of Generative AI in Creative
• 0:56
  Problem Solving. It's a great paper. It really
• 0:58
  is. And our mission for this deep dive is to
• 1:02
  basically bypass all the surface level chatter
• 1:04
  about productivity and really explore the psychological
• 1:07
  impact of AI adoption. Right. Going beyond just
• 1:10
  the corporate buzzwords. Exactly. We want to
• 1:13
  understand how it alters the way you learn, and
• 1:15
  most importantly, how you can consciously use
• 1:18
  this technology to enhance rather than accidentally
• 1:22
  destroy your own creative problem solving skills.
• 1:25
  So, OK, let's unpack this. Yeah, to really understand
• 1:29
  what is at stake here, we have to look at this
• 1:31
  concept called cognitive economy. Cognitive economy.
• 1:34
  So the human brain is an incredible machine,
• 1:36
  obviously, but it is also an absolute energy
• 1:39
  hog. I mean, it consumes about 20 % of your body's
• 1:42
  energy. Wait, 20 %? Just the brain? Just the
• 1:44
  brain. And it only makes up like 2 % of your
• 1:47
  body weight. Oh, wow. That's a lot of energy.
• 1:50
  It is. Because of that. your brain is biologically
• 1:54
  hardwired to seek the path of least resistance.
• 1:57
  It wants to conserve glucose at all costs. And
• 1:59
  that's to be lazy. It wants to be efficient.
• 2:01
  Let's call it efficient. So when you introduce
• 2:03
  a tool that instantly synthesizes complex information
• 2:07
  and just, you know, hands you a perfectly formatted
• 2:10
  answer. Your brain just says, yes, please, I'll
• 2:11
  take that. Exactly. Your brain naturally wants
• 2:14
  to accept it and just power down. And to live
• 2:17
  as research points out that this leads to a really
• 2:19
  dangerous over -reliance on AI. Because it's
• 2:23
  doing the thinking for you. Right. When a tool
• 2:25
  doesn't just assist you. but actually processes
• 2:28
  information and generates solutions for you,
• 2:31
  it completely removes your cognitive load. But
• 2:34
  hold on, let me play devil's advocate here. Isn't
• 2:36
  that the entire point of technology? I mean,
• 2:38
  to offload the tedious heavy lifting so we can
• 2:41
  be more creative? Well, that's the pitch, right?
• 2:43
  Right. If the AI writes the boring email or,
• 2:47
  I don't know, drafts the standard weekly report,
• 2:49
  my brain has more energy to do the high -level
• 2:52
  innovative thinking. That is the ideal scenario,
• 2:54
  definitely. But what's fascinating here is that
• 2:57
  the reality is often the exact opposite. Wait,
• 2:59
  really? How so? When you remove the friction
• 3:02
  of the quote unquote, boring work. You often
• 3:06
  remove the deep engagement required to understand
• 3:08
  a problem holistically. Oh, I see. Think back
• 3:11
  to your GPS analogy. When we used to navigate
• 3:14
  without a map, we had to build complex spatial
• 3:17
  networks in our hippocampus. Yeah, you had to
• 3:20
  actually look around. Exactly. We noticed landmarks.
• 3:24
  We understood cardinal directions. We were deeply
• 3:26
  engaged with our environment. And now we just
• 3:28
  stare at the screen. Right. When we outsource
• 3:31
  that to a screen, those networks literally weakened.
• 3:34
  And Tolova argues that the exact same thing happens
• 3:37
  to our critical thinking when we over -rely on
• 3:40
  LLMs. So the cognitive muscles literally atrophy
• 3:44
  because we aren't creating that mental map of
• 3:46
  the information. We're just looking at the final
• 3:48
  destination the AI gave us. Which is dangerous.
• 3:51
  And that introduces a much more insidious problem
• 3:53
  that she highlights in the paper reinforcement
• 3:55
  bias. Right, the echo chamber effect. I really
• 3:58
  want to try and break down the mechanics of this
• 3:59
  because I think people hear the word bias and
• 4:01
  they just automatically think about politics.
• 4:03
  Oh, sure. Yeah. But reinforcement bias and problem
• 4:06
  solving is structural, right? Like, if your cognitive
• 4:09
  muscles are atrophying because you are just accepting
• 4:13
  the AI's easy answers, you stop challenging the
• 4:15
  outputs. Yes. And what does an LLM do? It predicts
• 4:19
  the most statistically likely response based
• 4:22
  on its training data and your prompt. It gives
• 4:24
  you the average. Right. It is essentially designed
• 4:27
  to give you what it thinks you want to hear.
• 4:30
  You've completely hit the nail on the head. If
• 4:32
  you don't bring your own rigorous analysis to
• 4:35
  the table, you end up in this massive feedback
• 4:37
  loop. Because you're just nodding along with
• 4:39
  the machine. Exactly. You ask a question, the
• 4:42
  AI gives you the consensus average of the internet.
• 4:45
  You accept it, because like we said, your brain
• 4:47
  wants to save energy, and you just move on. You
• 4:49
  aren't actually problem solving at all. No. You
• 4:51
  are just participating in a synthesized consensus.
• 4:54
  The friction that creates true, innovative, out
• 4:57
  -of -the -box thought. is completely lost. Okay,
• 5:00
  that is genuinely terrifying. It is a little
• 5:03
  bleak, yeah. The idea that we might just smooth
• 5:05
  out all the rough edges of human creativity because
• 5:08
  we are literally too mentally tired to argue
• 5:11
  with a chatbot. That's the risk. But obviously,
• 5:13
  we can't just throw our computers in the ocean.
• 5:16
  I mean, we have to use these tools. And Talova
• 5:18
  conducted research with the Open University specifically
• 5:21
  on how to use AI in creative problem solving
• 5:24
  without falling into that cognitive trap. Yes,
• 5:27
  she did. And her research emphasizes that casually
• 5:31
  typing a prompt into a box whenever you feel
• 5:34
  stuck That's a recipe for that exact cognitive
• 5:37
  decline. So we need a better way to talk to it.
• 5:39
  We have to interact with these models using deliberate,
• 5:42
  structured techniques. And here's where it gets
• 5:44
  really interesting because her research at the
• 5:47
  Open University surfaced this fascinating concept
• 5:50
  called the Scratch Method. Yes, the Scratch Method.
• 5:53
  Now I was looking through the source material
• 5:55
  and it highlights this method as a major breakthrough
• 5:57
  for keeping the human in the loop. It's very
• 6:00
  effective. But I have to push back on this entire
• 6:02
  premise. Okay, let's hear it. Wait a minute.
• 6:05
  If the whole appeal of an AI assistant is its
• 6:07
  intuitive, magical ease of use, why on earth
• 6:11
  do I need to invent a complex, formal framework
• 6:13
  just to talk to it? I get that. I mean, it sounds
• 6:16
  like you're telling me to buy a state of the
• 6:18
  art calculator, but then insisting I still do
• 6:21
  the long division in my head on a piece of scratch
• 6:23
  paper just to build character. I completely understand
• 6:26
  the frustration, but the calculator analogy actually
• 6:29
  proves the point perfectly. Really? How? Well...
• 6:32
  A calculator executes a deterministic mathematical
• 6:35
  function. It does the math. There's only one
• 6:37
  right answer. Sure. Generative AI doesn't just
• 6:39
  do math. It generates ideas. It mimics reasoning.
• 6:43
  If you treat it like a calculator, where you
• 6:45
  just punch in a problem and blindly trust the
• 6:47
  output, you are surrendering your role as the
• 6:50
  thinker. Oh, I see. Because there isn't just
• 6:52
  one right answer in creative work. Exactly. The
• 6:54
  scratch method isn't about doing the long division
• 6:56
  in your head just for the sake of it. It is about
• 6:59
  recognizing that the AI is not some oracle delivering
• 7:03
  the absolute final truth. Right. It is a dynamic
• 7:06
  collaborative scratch pad. OK, so let's make
• 7:08
  this concrete. Because the source material introduces
• 7:11
  the term, but leaves the actual step -by -step
• 7:14
  application a bit open to interpretation, let's
• 7:17
  deduce what this actually looks like in practice.
• 7:19
  Let's do it. Say I have to write a complex, high
• 7:22
  -stakes marketing strategy. OK, good example.
• 7:25
  The lazy way, the cognitive economy way, is to
• 7:28
  just type, write a marketing strategy for a new
• 7:30
  shoe brand, and then I just copy paste whatever
• 7:33
  it spits out. Which guarantees a generic reinforcement
• 7:36
  biased result. Right. Total echo chamber. So
• 7:39
  under the scratch method, instead of asking it
• 7:42
  to write the plan, I would have to generate the
• 7:44
  initial raw material myself. Yes. I might dump
• 7:47
  in a messy, chaotic list of my own bullet points,
• 7:50
  maybe my half -formed ideas, my specific insights
• 7:53
  about the target audience. All your own original
• 7:55
  thought. Exactly. And then I prompt the AI with
• 7:58
  something like, act as a critical executive,
• 8:01
  find the logical holes in this strategy, and
• 8:03
  tell me where my argument is weak. That is the
• 8:06
  perfect implementation of it. Notice what happens
• 8:08
  to the cognitive load in your scenario. I'm still
• 8:11
  doing the hard part. You maintain the heavy lifting
• 8:14
  of the creation and the ideation. You are forcing
• 8:18
  your hippocampus to map the territory. to use
• 8:21
  our earlier analogy. Right. I'm building the
• 8:23
  map. And then you use the AI to assume the load
• 8:26
  of critique and pressure testing. So I am still
• 8:29
  in the driver's seat. It's like it's the difference
• 8:32
  between paying someone to do your homework for
• 8:34
  you versus sitting down for a really intense
• 8:38
  argumentative study session with a peer. Exactly.
• 8:42
  You are engineering. cognitive friction back
• 8:45
  into the process. Cognitive friction, I like
• 8:47
  that. The scratch method forces you to use the
• 8:50
  AI to augment your creative problem solving rather
• 8:54
  than replacing your human effort entirely. You
• 8:57
  remain the editor and the final arbiter of truth.
• 8:59
  But making that shift, like going from AI as
• 9:02
  an oracle to AI as a study partner, that requires
• 9:05
  a totally different mindset. It really does.
• 9:07
  You can't approach this technology like a traditional
• 9:09
  software developer looking for maximum corporate
• 9:12
  efficiency. You have to approach it like a teacher,
• 9:14
  which explains why Talova's own background is
• 9:17
  so incredibly illuminating here. It is literally
• 9:20
  the missing puzzle piece to her entire philosophy.
• 9:23
  Let's talk about that. Well, before she became
• 9:26
  an AI product manager, she spent nearly a decade
• 9:28
  as an educator. A decade? Wow. Yeah, she taught
• 9:31
  languages, organized workshops, hosted webinars.
• 9:35
  She fundamentally understands human development
• 9:37
  and how we learn. The fact that she was specifically
• 9:40
  a language educator is what really stood out
• 9:42
  to me. Oh, absolutely. Because think about what
• 9:45
  learning a new language actually entails. You
• 9:49
  aren't just memorizing a dictionary. No, not
• 9:50
  at all. You are literally learning a brand new
• 9:53
  way to structure your thoughts, like a new syntax
• 9:55
  for viewing the world. It rewires your brain.
• 9:58
  Yeah. And you can easily argue that she is applying
• 10:00
  that exact same methodology today. She's basically
• 10:03
  teaching people how to, quote unquote, speak
• 10:06
  AI. I love that phrase. It completely reframes
• 10:09
  the technology. It really does. When you view
• 10:12
  large language models through the lens of an
• 10:14
  educator whose goal is to create meaningful learning
• 10:16
  experiences, AI stops being the sterile, slightly
• 10:20
  threatening corporate hack designed to eliminate
• 10:22
  jobs. Right. It's not just about cutting costs.
• 10:25
  It becomes a deeply human, transformative tool
• 10:29
  for empowerment. It's not about what the machine
• 10:32
  can do, you know. It's about what the machine
• 10:35
  can help you become. If we connect this to the
• 10:37
  bigger picture, you can really see how her background
• 10:41
  directly informs her advocacy today. How so?
• 10:45
  She is fiercely vocal about the need for ethical,
• 10:48
  unbiased, and transparent use of AI. But she
• 10:52
  understands that Ethical AI is not merely a technical
• 10:55
  glitch that a software engineer can just patch
• 10:57
  in a vacuum. It's not just a line of code you
• 10:59
  can fit. Exactly. You cannot have ethical, transparent
• 11:02
  technology if you do not have an educated, critically
• 11:05
  engaged populace interacting with it. The two
• 11:07
  are completely intertwined. Because if the tool
• 11:09
  is actively shaping how we think, as we establish
• 11:12
  with the cognitive economy and the scratch method,
• 11:14
  then the underlying ethics of how that tool is
• 11:17
  built will literally dictate the future of human
• 11:19
  thought. Which means we have to look very, very
• 11:21
  closely. Who was actually building the tools?
• 11:24
  Right. And it's one thing to have this beautiful
• 11:26
  psychological philosophy about education and
• 11:29
  unbiased AI, but it is entirely another thing
• 11:32
  to build the real -world infrastructure to achieve
• 11:34
  it. Yes, the execution is the hard part. Which
• 11:37
  brings us to SheAI, an initiative Talova actually
• 11:40
  co -founded. This is where her educational theories
• 11:42
  manifest into tangible global action. So SheAI
• 11:47
  is an educational platform explicitly backed
• 11:49
  by the United Nations and its focus is laser
• 11:52
  targeted on female AI literacy. It's a fantastic
• 11:56
  program. They are working to close the gender
• 11:58
  gap in the tech industry, empower women with
• 12:01
  the skills to harness AI in their careers and
• 12:03
  basically build a future where women are actively
• 12:06
  shaping this technology. It is a critical mission
• 12:08
  and it directly addresses the most glaring vulnerability
• 12:11
  in the ecosystem we've been discussing today.
• 12:13
  So what does this all mean? Like, in the context
• 12:15
  of our deep dive today, moving from cognitive
• 12:18
  overreliance to the scratch method to the mindset
• 12:20
  of an educator, why is female AI literacy specifically
• 12:24
  cited as the linchpin for achieving that unbiased
• 12:27
  AI development we need? This raises an important
• 12:30
  question because it forces us to look at the
• 12:32
  structural reality of the tech industry right
• 12:34
  now. Let's trace the logic back to the problem
• 12:37
  of reinforcement bias we unpacked earlier. OK,
• 12:40
  let's go back to that. We established that an
• 12:42
  AI model feeds you back the assumptions baked
• 12:45
  into its training data, right? Right. It gives
• 12:47
  you the consensus of the people who built it
• 12:49
  and the data they selected to feed it. Now ask
• 12:52
  yourself, whose assumptions are currently baked
• 12:54
  into those models? Oh, I see where you're going
• 12:56
  with this. The artificial intelligence field
• 12:59
  suffers from a severe lack of demographic diversity.
• 13:02
  It is largely a monoculture. Yeah, that's definitely
• 13:05
  true. If the teams writing the algorithms, selecting
• 13:08
  the training data, and defining the safety parameters
• 13:11
  all share a similar background, their unconscious
• 13:13
  biases inevitably become the AI's biases. Ah,
• 13:17
  I see. It's a massive data integrity issue. Precisely.
• 13:20
  You literally cannot... solve data bias or reinforcement
• 13:24
  bias if the industry building the tool is just
• 13:26
  a monolith. It is structurally impossible and
• 13:30
  LLM acts as a mirror to its creators. A mirror.
• 13:33
  If AI is going to process information and develop
• 13:36
  solutions that serve a diverse global humanity,
• 13:39
  it must be guided and scrutinized by a diverse
• 13:42
  representation of humanity. That makes total
• 13:44
  sense. Closing the gender gap and really Advocating
• 13:47
  for broad diversity across all spectrums in tech
• 13:50
  isn't just a nice societal bonus or PR talking
• 13:54
  point. It's not just a diversity quota. No, it
• 13:56
  is a practical, absolute necessity for creating
• 13:59
  technology that doesn't just reinforce the narrow
• 14:02
  worldview of a select few. It's the ultimate
• 14:05
  structural safeguard against the echo chamber.
• 14:08
  Exactly. Talova notes that she believes technology,
• 14:11
  when used with purpose, can create unparalleled
• 14:14
  opportunities. But that purpose requires diverse
• 14:17
  hands on the steering wheel. You need different
• 14:19
  perspectives. Right. By empowering women with
• 14:22
  AI literacy, SheAI is ensuring that the people
• 14:24
  building, critiquing, and utilizing the technology
• 14:27
  actually represent the populations who will be
• 14:29
  affected by it. It ensures that the frameworks
• 14:31
  we develop, like the Scratch method, for example,
• 14:34
  are robust, inclusive, and fundamentally human.
• 14:37
  Man, it is incredibly satisfying to see how tightly
• 14:41
  all these concepts interlock. You really do all
• 14:43
  connect. We started this deep dive exploring
• 14:46
  a very personal internal fear, you know, that
• 14:50
  staring at a blinking cursor and letting a supercomputer
• 14:52
  do the work. was going to make us cognitively
• 14:56
  lazy. Yes. We worried about losing our mental
• 14:58
  map. A completely valid fear, by the way. It's
• 15:01
  rooted in the biological reality of how our brains
• 15:04
  seek to conserve energy. Right, the glucose thing.
• 15:06
  Yeah. But then we looked at how we can fight
• 15:09
  back against our own biology. We deduced how
• 15:12
  structured problem -solving techniques, like
• 15:14
  the scratch method, can artificially engineer
• 15:17
  cognitive friction, forcing us to stay in the
• 15:19
  driver's seat. We have to keep ourselves engaged.
• 15:21
  And we saw how adopting the mindset of a language
• 15:24
  educator transforms AI from just a, you know,
• 15:28
  productivity crutch into a dynamic study partner.
• 15:31
  It completely shifts the paradigm. And finally,
• 15:34
  we saw how that same educational drive translates
• 15:37
  into massive structural action through SheAI,
• 15:41
  proving that the only way to build truly unbiased
• 15:43
  AI is to ensure that a diverse spectrum of human
• 15:46
  voices are the ones shaping it. It's a profound
• 15:49
  journey. It moves from the microscopic architecture
• 15:52
  of the human hippocampus all the way up to the
• 15:54
  global architecture of the tech industry. And
• 15:56
  most importantly, it places agency firmly back
• 16:00
  into the hands of the user. Which brings it all
• 16:02
  right back to you listening to this deep dive
• 16:04
  right now. The next time you sit down at your
• 16:07
  keyboard, I want you to ask yourself a hard question
• 16:10
  about your own workflow. A very hard question.
• 16:12
  When you type in that prompt, are you just blindly
• 16:14
  following the cognitive blue dot? Are you letting
• 16:17
  the LLM do the heavy lifting just because it's
• 16:19
  easy? It's so tempting. It is. Or are you using
• 16:23
  it, purposefully feeding it your messy chaotic
• 16:25
  ideas, forcing it to critique you, and using
• 16:28
  it as a scratch pad to genuinely expand your
• 16:30
  own creative problem -solving skills? Because
• 16:32
  the inescapable reality is that this technology
• 16:34
  is only going to become more seamless, more integrated,
• 16:39
  and far more capable of doing the thinking for
• 16:41
  us. That's true. It's only getting better. Which
• 16:44
  leaves us with a final thought to mull over long
• 16:47
  after you finish your commute or your workout.
• 16:50
  Oh, lay it on us. Well, we have spent this entire
• 16:53
  discussion exploring how humans must constantly
• 16:56
  invent new methodologies, like the scratch method,
• 17:00
  just to prevent the machine from making us cognitively
• 17:02
  lazy. Right. We are continuously having to adapt
• 17:05
  our own problem -solving frameworks to match
• 17:08
  and counter the evolving capabilities of these
• 17:11
  language models. We have to keep up. So when
• 17:13
  you really step back and observe that dynamic,
• 17:15
  who is actually training who? Oh, wow. Are we
• 17:18
  just programming the AI to give us better answers?
• 17:22
  Or... by forcing us to constantly change how
• 17:24
  we learn and think to accommodate it is the AI
• 17:27
  subtly, permanently reprogramming the fundamental
• 17:30
  structure of human thought. It is the GPS navigation
• 17:33
  all over again. Exactly. Only this time, it's
• 17:36
  not our sense of direction at stake. It's the
• 17:38
  very map of our minds. Definitely something to
• 17:41
  think about the next time you hit generate. Thanks
• 17:44
  for joining us on this deep dive.