The newest generation of kids are growing up in a hyper-connected world where artificial intelligence is unavoidable. It’s shaping up to be a silent partner in their lives as companies race to develop AI assistants, and smart systems seamlessly blend into our surroundings. Until recently, the education sector did not give much thought to the technology. Then, along came ChatGPT, capable of generating essays, code, and other assignments with a human-like quality that prompted considerable debate over whether AI should even have a role in schools. Early discussions around using AI for the classroom have overwhelmingly been about cheating (AI-written essays are hard to detect), or helping teachers (AI can be an aide or a tutor, which causes its own anxieties). But AI is also a subject, one that future students — even today’s students — could benefit from diving into more deeply, so what might that look like? Writing for POLITICO Magazine this month, AI entrepreneur and skeptic Gary Marcus argued that as Congress considers AI legislation, it needs to mandate and fund AI-related curricula. His call conjured up past educational experiences, like elementary school computer classes and internet safety seminars, where students were taught how to engage with new technologies. One AI curriculum developer — Victor Lee, a professor at Stanford’s Graduate School of Education — says there’s a lot to learn from the shortcomings of earlier efforts, which seemed to inadequately prepare students to face cyberbullying and other issues that rose with the web. Lee studied AI back in the late '90s, and in early 2022, he helped launch the university’s CRAFT initiative, which designs free materials with high school teachers so they can teach AI concepts. He joins me for today’s DFD to discuss the concept of AI literacy, the considerations driving his work, and how it might evolve alongside future technological advances and the broader policy debates on AI regulation. Below is an edited and condensed version of our conversation: How would you define AI literacy? Is it more about preparing students if they want to seek careers in AI or equipping them to navigate a world where AI is omnipresent? When you say AI literacy, a lot of people are thinking about different things. So for some people, they're really thinking about, do you understand and have the foundations such that you could go on to be a developer of AI systems in the future? Those tend to emphasize things like the underlying principles behind machine learning, for instance, or classifiers, or transformers. There's another cluster of people who think AI literacy is understanding more fully, what are the potential risks and responsibilities that we have with respect to AI? That would involve issues of plagiarism and algorithmic bias. So it’s just trying to get people to slow down and think more about, where and how does this come about, and who is benefiting from it, and what are some unintended consequences? That's a bit more of a critical take, that a lot of the AI literacy community is doing. Then there's another one, and the one that we tend to lean into a bit more is how is this interfacing with the kinds of work that we do? What is the way in which we would be likely to be users not necessarily builders of the AI? But I would say those three are all representative. There's a lot of people talking past each other on what AI literacy means because it's a very seductive term. But when it comes down to what is it you make and emphasize as far as curriculum or lessons, it's a little bit all over the place, and there's some funny overlaps and some funny oversights. I don't think any one of them are wrong. We should just be explicit about what perspectives are being emphasized with any given AI literacy endeavor. Instead of offering dedicated courses on AI, you develop lesson plans and activities that teachers of different subject areas can add to their existing curriculum. Why does CRAFT take that approach? We subscribe to an idea, which I think is also taking off in the broader AI literacy community, that there’s a way to have AI be a topic that speaks to the different subject areas that different teachers are covering. So AI is impacting how we do science. AI is pertinent to a lot of the things we think about with social studies. AI is changing how we do art or composition or reading. And so we work in partnership to co-develop with the practicing high school teachers resources that would really fit and work within the constraints of school time and classroom time to make that bridge, rather than trying to create a separate course. Because it’s really hard for a school to free up an additional teacher, an additional class period etc. to have a dedicated AI course. There's a lot of sensible reasons to go at it from a multidisciplinary perspective. But you know, there's very much this pragmatic one: where are we going to find the time? There's not even enough time to cover the things that are mandated for the districts. I see the awareness really broadening on that. How much exposure do you assume students already have to AI by high school? I tend to be in a bit of a bubble where we're talking about AI all the time. But this is also why for me and in the CRAFT project, making sure we have people from around the country — from rural schools, from urban schools — is to get that reality check. There's still a very good subset of high schoolers who are just not familiar with all of these AI developments. They’re not aware that that's what those are, and how they're encountering it is pretty different. If you are thinking about the critical perspective of being able to create new AI, then you have to understand a bit about how data are really important to making the AI work. Kids are already now in the process of producing data because they use the family iPad, because they watch something on Netflix, and that is using AI as well. So I do think there are conversations to be had at younger ages. Certainly in high school, there's a bit more awareness and independence that's coming about, and some increasing focus on subjects that they're thinking may be part of their professional or post-high school future. To what extent do your lesson plans cover those more critical aspects of AI – the ones that are being raised in Washington’s current debates over regulation. I saw a CRAFT lesson about algorithmic bias. What’s your goal with resources like that? While I appreciate the work that our legislators try to do, there's sometimes a real disconnect with their understanding of what's going on or why things work the way that they do. If you take algorithmic bias, for instance, it's an interesting sounding term, but why does it come about? And is it because of company policies? Or is it because of the datasets that were used to train it? Or is it because of the particular algorithm on which the training is done? Or is it based on the actions that the data and algorithms serve? Or because there's not a human in the loop in there? What my larger goal, and what I think is important for AI literacy, is to be able to participate in those conversations more accurately. With all this misinformation and AI hallucinations that have come out, people’s first response is, “well, just cite your sources.” That's not actually how these things work. So knowing that is important for thinking about what is the solution for dealing with inaccuracies that come out of generative AI. AI curriculum currently is not mandated in state academic standards. Do you think it should be, and are you advocating for its inclusion? There's definitely a push toward it. One of the challenges is that AI is definitely having its extended peak moment, and I do think AI is going to stay. But we go through hype cycles all the time. Clearly, there's things that are going on with new types of technology and computing that we were not as concerned about 20 years ago. Knowing that that's coming, what are the things that we need to know so that way we're not just creating an infinite list of standards that no school can participate? There's work underway for it with thoughtfulness and caution before saying, “well, this needs to be fully adopted as whole new standards” because we have a tendency to pile on the school systems. What are the sensitive topics that have come up in those discussions? One theme we hear from lawmakers is whether to focus on the more existential versus immediate risks of the technology. Has that trickled down into the classroom? There's been sensitivity around the safety and the risks. In California, there's some heated conversation going on that fundamentally raises the question of how much do we need people to be trained, to be equipped, to develop the technologies, to really build them out versus be able to use them. For AI literacy, it does distinguish itself in terms of not pushing a full computer science, engineering and programming curriculum. In rooms with state legislators, with policy leaders, with researchers, everyone has a broad nod of agreement about concern about ethics and safety. Whether they're thinking about the same thing — whether on the one hand, it's ‘will the machines rise and overthrow humanity’ versus ‘are we going to exacerbate social problems and inequities?’ - that's another question. When they say we have to think about AI ethics and bias and human in the loop — it’s an interesting time in the education system right now because there's a lot of stuff being worked out. What’s on the horizon? There are going to be these questions about inequities in society, how the ways in which those inequities are talked about given things that are happening in other parts of the education system across states and curriculum. I think there are also questions about, how much does this become support for democracy when we see a lot of consolidation of models and transformers that are limited to a few entities? This one has probably been more my skew — the green impact of AI. There's a bit more murmur, and it is a concern. There's also the day-to-day use cases: what it means to be a responsible person for schools, and there's more cases coming up of cyberbullying by putting someone's face on an inappropriate image and circulating it at school. Those I feel should be talked about in terms of what our expectations are for good conduct given these technologies. Say, we take some recordings of someone’s voice and then make it sound weird because we've trained AI to talk in a certain way. While that may be done out of jest or even for one's own enjoyment, think carefully about the risks that are involved, and what we deem acceptable or not acceptable. How have discussions about the role of AI in education evolved over time? When ChatGPT first came out, the concern in schools was heavily focused on plagiarism risks. With those new cases and technologies, has that been overshadowed? There's definitely a lot of people who are still very concerned about plagiarism. I do hear it as a huge ongoing concern. But I don't think it's as dominant to the degree it was before. There is some sense of, “well, there’s ways in which it might be useful, and it's going to be part of our world, so how do we do the work of using it?” But there's a lot of uncertainty because there's just so much chatter going on, and there hasn't been enough clarity in the U.S. with respect to what's okay, what's not okay. Deepfakes — I think there's some awareness. Right now, there's also “well, what is the right amount of AI to use?” There's some school districts like Los Angeles Unified School District that are building their own in-house chatbot technology, and the tech enthusiasts are super intrigued by “what would that mean, and what will look like?” And various other school districts are embarking on that, and others are just moving more cautiously. What lessons have we learned from previous ways technology use was discussed in schools, like safety and cyberbullying seminars for social media and the internet? We have an important opportunity that we can't afford to pass up right now. We saw what happened and were caught on our heels with misinformation and social media. Right now, we're at the precipice of the version of that for AI. It's already happening, but to the extent that it needs to be taken seriously, we have the opportunity now that we have to broaden. There are also some other fundamental things. A lot of kids may hear their parents or TV programs, where a voice assistant is demonstrating Siri or Alexa or whatnot. Something that I think we need to be on the lookout for is the over-personification of AI, and thinking about it as being like a person, or thinking about only the technologies that exude person-like qualities as being AI. So recommender systems, your phone being unlocked based on your face or your thumbprint all use AI. Spell checking and grammar correction on your devices use AI. Email filtering uses AI. Those don’t look like robots. We could get into in a few decades a way of relating to AI that I think is not consistent with how a lot of us understand what AI is. There are open questions about what's important to know in this world where data drives a lot of technology, our processing powers are massive, and we're so far removed from where a lot of these things are happening — whether it's these massive server farms or whether it's the moderators who are filtering the content. Just starting to plant those seeds of awareness — I would consider to be important.
| 
Comments
Post a Comment