Latest / Azadeh and Guy on AI / Simulations, Sovereignty, and the Four Ds

Transcript

Simulations, Sovereignty, and the Four Ds · Azadeh and Guy on AI

• 0:00
  SPEAKER_00: Hello, everyone.
• 0:01
  SPEAKER_00: Welcome to another episode of Azadeh.
• 0:04
  SPEAKER_01: And Guy.
• 0:04
  SPEAKER_00: On AI.
• 0:06
  And today we're going to talk about AI fluency and also Guy
• 0:12
  attended a conference,
• 0:14
  AACD 2026.
• 0:16
  SPEAKER_01: And it was lovely.
• 0:17
  SPEAKER_01: It was New Orleans, two days after Mardi Gras.
• 0:20
  SPEAKER_01: So, you know, you got to take that under advisement.
• 0:23
  SPEAKER_01: Everybody's still recovering from
• 0:28
  SPEAKER_01: from Mardi Gras,
• 0:30
  but I had lots of fun and there were a few sessions about AI
• 0:36
  that were worth
• 0:36
  SPEAKER_01: listening to and many,
• 0:38
  SPEAKER_01: many conversations in the corridor.
• 0:41
  SPEAKER_01: So we had, since I'm on the Innovation and Technology Committee,
• 0:49
  SPEAKER_01: We had a special learning lab,
• 0:51
  which was delightful because we got to talk and it was more a
• 0:55
  conversation unless
• 0:58
  SPEAKER_01: we stand and speak at you,
• 1:00
  SPEAKER_01: we're speaking with.
• 1:01
  SPEAKER_01: We had four tables.
• 1:03
  SPEAKER_01: Each table focused on something else.
• 1:05
  We talked about international studies and what's happening
• 1:08
  across the world.
• 1:10
  SPEAKER_01: And that was John.
• 1:12
  SPEAKER_01: And then we had Ledi talking about prompting and...
• 1:17
  SPEAKER_01: I talked about data sovereignty, which is what we covered last time.
• 1:22
  SPEAKER_01: And so that was my practice run.
• 1:24
  SPEAKER_01: Thank you, everybody, for listening.
• 1:26
  SPEAKER_01: And if you missed this,
• 1:27
  we went a little bit deeper about data sovereignty and had
• 1:31
  different responses from
• 1:33
  SPEAKER_01: different participants.
• 1:34
  SPEAKER_01: So that was great.
• 1:37
  SPEAKER_01: And it was a lovely discussion,
• 1:40
  at least on data sovereignty and how we can awaken it and how
• 1:44
  early we can do this.
• 1:46
  SPEAKER_01: So that was great.
• 1:47
  SPEAKER_01: And the second one, I went to Punya Mishra and gave a talk about AI.
• 1:53
  SPEAKER_01: His thoughts are developing about AI every time that I listen to him.
• 1:59
  SPEAKER_01: He's done a little bit more thinking.
• 2:01
  SPEAKER_01: And we are very much on the same page.
• 2:03
  SPEAKER_01: We don't fully agree on everything, but we're very much on the same page.
• 2:06
  SPEAKER_01: And that is we're really excited about what we can do with AI.
• 2:10
  SPEAKER_01: We find it delightful,
• 2:11
  SPEAKER_01: but at the same time,
• 2:13
  SPEAKER_01: we find it somewhat scary and worrying,
• 2:16
  especially the way it might impact learning and social
• 2:21
  emotional growth in young
• 2:23
  SPEAKER_01: people.
• 2:25
  SPEAKER_01: And those are the things that we highlight,
• 2:28
  SPEAKER_01: of course,
• 2:28
  SPEAKER_01: environmental and social justice effects.
• 2:31
  SPEAKER_01: But on the flip side,
• 2:32
  we also worried about a world with AI where we're not preparing
• 2:37
  our students and
• 2:38
  SPEAKER_01: that's creating another problem with equity.
• 2:40
  SPEAKER_01: So it's really hard to escape from that.
• 2:45
  SPEAKER_01: The third session was a panel discussion.
• 2:47
  SPEAKER_01: Yeah.
• 2:48
  SPEAKER_01: And in that panel discussion,
• 2:51
  there were five different scholars talking about the place of a
• 2:55
  technology in
• 2:56
  SPEAKER_01: general and digital technology more specifically.
• 2:59
  SPEAKER_01: And of course,
• 3:00
  SPEAKER_01: behind that discussion about digital technology,
• 3:03
  SPEAKER_01: it's really a discussion about AI.
• 3:04
  SPEAKER_01: We've had digital technology for a long time in teacher education,
• 3:07
  SPEAKER_01: but this is AI is just beginning.
• 3:10
  SPEAKER_01: And so there was a lot of reference to AI and the,
• 3:15
  overall agreement was that ai is an important development that
• 3:22
  it is a must for us
• 3:23
  in teacher education and whatever we think about ai whatever we
• 3:28
  think about whether
• 3:29
  it should or shouldn't happen we should prepare our teachers to
• 3:33
  deal in the world
• 3:34
  SPEAKER_01: with ai if ai doesn't come to be and
• 3:39
  SPEAKER_01: through regulation or something else,
• 3:41
  SPEAKER_01: it disappears,
• 3:42
  SPEAKER_01: then they've learned to be flexible with something.
• 3:44
  SPEAKER_01: But if it doesn't, they need the tools to deal with it.
• 3:48
  SPEAKER_01: And that's the role of teacher education.
• 3:50
  So this is something that is kind of at the forefront of many
• 3:55
  thinkers as the most
• 3:57
  SPEAKER_01: significant thing happening right now.
• 3:59
  SPEAKER_00: Interesting.
• 4:03
  SPEAKER_01: And...
• 4:04
  SPEAKER_01: Also a lot of side conversations,
• 4:06
  especially on the place it's taking in creating simulations for
• 4:10
  education,
• 4:11
  especially as part of teacher education as a way to make sure
• 4:15
  that all of our
• 4:15
  students have experience in trying to handle situations that
• 4:20
  don't see everyone.
• 4:23
  SPEAKER_01: every day in their practicum.
• 4:25
  SPEAKER_01: And so the idea is, at least for me, right?
• 4:30
  I don't want to replace the idea of placing students in
• 4:33
  practicum classrooms,
• 4:34
  SPEAKER_01: in real classrooms with real kids.
• 4:36
  SPEAKER_01: There's no replacement for that.
• 4:37
  And more to the point,
• 4:39
  that technology is cheaper than many other technologies,
• 4:42
  right?
• 4:43
  SPEAKER_01: What can I do in simulations?
• 4:45
  SPEAKER_01: In simulations,
• 4:46
  I could do something that I can't do or that our students will
• 4:50
  not get to see in a
• 4:50
  SPEAKER_01: regular classroom.
• 4:51
  So for example,
• 4:53
  an extreme event where a student is starting to become violent
• 4:57
  potentially.
• 4:58
  SPEAKER_01: That is not something that I want every student to experience.
• 5:02
  SPEAKER_01: And that is not something that if I just place them in classroom,
• 5:05
  SPEAKER_01: they will see because that's not very common,
• 5:07
  SPEAKER_01: at least in the classrooms in our immediate environment.
• 5:10
  SPEAKER_01: So if I want them to experience it and start thinking about what are some
• 5:13
  SPEAKER_01: solutions,
• 5:14
  SPEAKER_01: simulation is a great opportunity.
• 5:17
  SPEAKER_01: Or what do you do with students with not very common disabilities?
• 5:21
  SPEAKER_01: So learning disabilities are very common.
• 5:23
  Every classroom will have a few students that have some kind of
• 5:27
  a learning disability.
• 5:28
  SPEAKER_01: Same thing with a hyperactivity or attention disorder.
• 5:32
  But if you're talking about students who are blind or students
• 5:38
  who are deaf or
• 5:40
  SPEAKER_01: students who have a movement disability or things like that,
• 5:44
  SPEAKER_01: those are not very common in the schools,
• 5:46
  SPEAKER_01: thank God.
• 5:47
  SPEAKER_01: But you do have to experience them.
• 5:50
  SPEAKER_01: And that's a way to manage that kind of classroom is through simulation.
• 5:55
  SPEAKER_01: So in simulation, we've had simulations for a long time.
• 5:59
  SPEAKER_01: They were just not scalable because they were too expensive.
• 6:02
  SPEAKER_01: AI makes them scalable.
• 6:03
  SPEAKER_01: All right.
• 6:04
  SPEAKER_01: But you promised me to talk about AI fluency.
• 6:07
  SPEAKER_00: Yeah, yeah.
• 6:08
  SPEAKER_00: I saw a paper or a report, I would say, that was published by Anthropic.
• 6:12
  SPEAKER_00: I saw that on LinkedIn.
• 6:14
  SPEAKER_00: Mm-hmm.
• 6:15
  SPEAKER_00: And the main topic of that is AI fluency.
• 6:18
  So they have a very clear definition of AI fluency and they say
• 6:22
  it means engaging
• 6:23
  SPEAKER_00: with AI in ways that are effective,
• 6:26
  SPEAKER_00: efficient,
• 6:27
  SPEAKER_00: ethical,
• 6:28
  SPEAKER_00: and also safe.
• 6:29
  SPEAKER_01: And this is team out of Ireland, right?
• 6:32
  SPEAKER_00: Very true.
• 6:32
  SPEAKER_00: Yeah.
• 6:34
  SPEAKER_01: So we want to come visit you.
• 6:35
  SPEAKER_01: Please send us an invitation.
• 6:38
  They also talk about the AI fluency framework,
• 6:41
  which consists of four core competencies.
• 6:44
  SPEAKER_00: And they call them four Ds.
• 6:46
  SPEAKER_01: Okay.
• 6:47
  SPEAKER_00: Can you guess any of that?
• 6:50
  SPEAKER_01: I know about discernment, which is the one I'm most interested in.
• 6:55
  SPEAKER_01: But what are the other ones?
• 6:57
  SPEAKER_00: How do you define discernment?
• 6:58
  SPEAKER_01: For me,
• 6:59
  discernment is the ability to decide something about the
• 7:02
  quality of the output
• 7:04
  SPEAKER_00: you're getting.
• 7:05
  SPEAKER_00: Mm-hmm.
• 7:05
  SPEAKER_00: Mm-hmm.
• 7:06
  SPEAKER_00: So one of them is delegation,
• 7:10
  SPEAKER_00: deciding what work to do with AI versus yourself,
• 7:13
  SPEAKER_00: which is very important,
• 7:14
  SPEAKER_00: and description,
• 7:16
  SPEAKER_00: communicating effectively with AI systems,
• 7:20
  SPEAKER_00: discernment you mentioned already,
• 7:21
  SPEAKER_00: and also diligence,
• 7:22
  SPEAKER_00: ensuring responsible AI collaboration.
• 7:26
  SPEAKER_00: So,
• 7:27
  Anthropic is trying to answer the question that how people are
• 7:31
  using AI and how
• 7:35
  they are using AI and how they develop fluency with this
• 7:39
  technology over time.
• 7:41
  SPEAKER_01: And they have 24 different indicators of fluency?
• 7:44
  SPEAKER_00: Yeah.
• 7:45
  SPEAKER_01: Which sounds like too much to me.
• 7:47
  SPEAKER_00: Exactly, yeah.
• 7:49
  SPEAKER_00: But for this report, they only focus on 11.
• 7:52
  SPEAKER_01: Because they are the ones visible.
• 7:54
  So because this study is one of those studies that uses the
• 8:00
  data from the AI
• 8:02
  SPEAKER_01: provider,
• 8:04
  SPEAKER_01: it can only focus on the things that an AI provider can see.
• 8:09
  and one of the things i love about this is they have the
• 8:12
  technology to mask the
• 8:14
  users so you don't know anything about who the users are what
• 8:18
  are they doing and
• 8:19
  where are they going and that is really important because it
• 8:22
  allows us to do quick
• 8:24
  research on reasonable samples without a privacy infringement
• 8:30
  and that is magical
• 8:31
  yeah and their sample is fairly recent so they analyzed over 10
• 8:36
  000 anonymized
• 8:37
  SPEAKER_00: plot conversations from a week in January,
• 8:41
  SPEAKER_00: 2026 to see whether users show behavior that signal to AI fluency.
• 8:48
  SPEAKER_00: And they also, so they have part of that is very visible.
• 8:52
  SPEAKER_00: Part of that is not that much clear.
• 8:54
  SPEAKER_00: And by that one,
• 8:56
  SPEAKER_00: I mean,
• 8:57
  SPEAKER_00: when AI creates a kind of application code or artifact,
• 9:01
  and you cannot understand how people are analyzing those kinds
• 9:05
  of artifacts.
• 9:06
  SPEAKER_00: So for that part,
• 9:07
  SPEAKER_00: they said that,
• 9:08
  and I love how they explained about the limitations of
• 9:11
  everything because they said
• 9:12
  SPEAKER_00: they are going to do another qualitative study.
• 9:16
  SPEAKER_00: focus on that part and understand how people analyze AI products.
• 9:20
  SPEAKER_01: However,
• 9:21
  SPEAKER_01: just as a note,
• 9:23
  SPEAKER_01: everybody says that at the end of their study in limitations,
• 9:27
  SPEAKER_01: this is not qualitative.
• 9:29
  SPEAKER_01: So there's a lot we don't know.
• 9:31
  SPEAKER_01: And we suggest that the next step is qualitative.
• 9:34
  I wonder if they're doing it because everybody,
• 9:36
  I mean,
• 9:36
  it's kind of a standard thing to say.
• 9:38
  SPEAKER_01: It doesn't mean they're not doing it.
• 9:40
  SPEAKER_01: I don't know them personally.
• 9:41
  SPEAKER_01: Again, invite invitation to Ireland.
• 9:43
  SPEAKER_01: We would be happy to come and learn.
• 9:46
  SPEAKER_01: by sitting next to them.
• 9:47
  But at the same time,
• 9:51
  the question is always,
• 9:53
  when are you going to do it?
• 9:55
  Because you pointed out that this is very,
• 9:57
  very recent,
• 9:58
  which is great,
• 9:59
  right?
• 10:00
  Because one of the complaints that we have and that everybody
• 10:03
  has is by the time,
• 10:04
  SPEAKER_01: sometimes,
• 10:06
  SPEAKER_01: sometimes by the time
• 10:08
  SPEAKER_01: that the research is published,
• 10:09
  SPEAKER_01: it's like two years after you collected it,
• 10:11
  SPEAKER_01: and it's like five versions of different LLMs that have been published.
• 10:16
  SPEAKER_01: I mean,
• 10:16
  SPEAKER_01: just in the last month,
• 10:17
  SPEAKER_01: there were at least four versions of major LLMs that were published,
• 10:22
  SPEAKER_01: right?
• 10:23
  SPEAKER_01: Anthropic came out with one.
• 10:27
  SPEAKER_01: Google came out with one.
• 10:30
  SPEAKER_01: It keeps happening.
• 10:31
  SPEAKER_01: And so we need to be aware that we want results fast.
• 10:35
  SPEAKER_01: And that's the advantage of these reports.
• 10:37
  SPEAKER_01: They may be not vetted and they may not still yet be peer-reviewed,
• 10:43
  SPEAKER_01: but at least we're getting a window into what's happening.
• 10:47
  SPEAKER_00: Key patterns that emerge from their research.
• 10:50
  SPEAKER_00: Fluency is tied to isolation.
• 10:53
  SPEAKER_00: So most fluency behaviors show up when users stick with the conversation,
• 10:59
  SPEAKER_00: refine their prompts,
• 11:00
  SPEAKER_00: and we have seen that in the classrooms.
• 11:02
  SPEAKER_00: I mean,
• 11:02
  if you are interacting with AI with a super simple prompt and
• 11:06
  just ask a very basic
• 11:07
  SPEAKER_00: question,
• 11:08
  SPEAKER_00: you are not getting a good output.
• 11:10
  SPEAKER_00: So let me see.
• 11:13
  SPEAKER_00: The second one is people treat AI more like a partner than a black box.
• 11:19
  And their data suggests that the dominant mode of interaction
• 11:22
  is augmentation using
• 11:26
  SPEAKER_00: AI as a kind of thought partner rather than offloading work entirely.
• 11:31
  SPEAKER_00: And the last one is that there is still a blind spot around evaluation.
• 11:36
  And that's the thing that I've mentioned about AI products,
• 11:39
  I mean artifacts.
• 11:41
  And we're getting a lot of data about that because we are
• 11:44
  working in my classroom
• 11:45
  and we're seeing our students produce products and try to
• 11:48
  evaluate these products.
• 11:50
  SPEAKER_01: So that is back to the discernment kind of thinking.
• 11:56
  And it's always on our mind because our students are being very
• 11:59
  critical in
• 12:00
  SPEAKER_01: productive ways,
• 12:01
  SPEAKER_01: I think,
• 12:01
  SPEAKER_01: for the most part.
• 12:03
  SPEAKER_01: of the AI products,
• 12:05
  SPEAKER_01: at least outwardly,
• 12:07
  SPEAKER_01: sometimes rejecting them,
• 12:09
  SPEAKER_01: sometimes accepting them with some caveats.
• 12:13
  SPEAKER_01: So people from Ireland, we'll invite you here.
• 12:15
  SPEAKER_00: Yes, sure.
• 12:17
  SPEAKER_01: And can we have the names of the authors?
• 12:20
  SPEAKER_00: Let me see the names of the authors.
• 12:23
  SPEAKER_00: So if you scroll down,
• 12:26
  SPEAKER_00: Kristen Swanson found the research,
• 12:30
  SPEAKER_00: led the analysis and wrote this report.
• 12:32
  SPEAKER_01: Excellent.
• 12:33
  SPEAKER_01: Thank you, Kristen.
• 12:34
  SPEAKER_01: Yeah.
• 12:36
  SPEAKER_01: There's a number of other researchers that are associated with it.
• 12:43
  SPEAKER_01: So if you are interested, go to Enthropic,
• 12:50
  find the research and find the AI fluency index and it's all
• 12:54
  there and we thank
• 12:55
  them for doing really important work and really thinking beyond
• 12:59
  the notion of
• 13:00
  SPEAKER_01: literacy into the realm of fluency.
• 13:03
  The other interesting part on their website is that they have a
• 13:07
  learn menu and they
• 13:08
  SPEAKER_00: have anthropic academy,
• 13:10
  SPEAKER_00: tutorials,
• 13:10
  SPEAKER_00: use cases,
• 13:11
  SPEAKER_00: all of them are very helpful if you wanted to just
• 13:14
  SPEAKER_00: understand how AI works.
• 13:15
  SPEAKER_01: And lately I've been clawed happy.
• 13:17
  So that works for me personally,
• 13:19
  but to be honest,
• 13:21
  I use them in parallel.
• 13:23
  SPEAKER_01: So I use multiples as we go along and,
• 13:27
  SPEAKER_01: um,
• 13:27
  SPEAKER_01: I have access to more of them on model,
• 13:29
  SPEAKER_01: which is great.
• 13:30
  SPEAKER_00: Yep.
• 13:31
  SPEAKER_01: All right.
• 13:31
  SPEAKER_00: So thank you so much for listening.
• 13:34
  SPEAKER_01: All right.
• 13:34
  SPEAKER_01: We'll see you next time on Azadeh and Guy on AI.