The AI Daily Brief: Artificial Intelligence News and Analysis - Are Educators Ready for a ChatGPT School Year?

Episode Date: September 1, 2023

As September and the Fall begins, NLW checks in on the conversation about ChatGPT among educators as they prepare to begin the first year where that technology will be available from day one. Before that on the Brief: Mustafa Suleyman says the US should use AI chip access to impose AI ethics on the world. Today's Sponsor: Supermanage - AI for 1-on-1's - https://supermanage.ai/breakdown ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, we're looking at how attitudes have changed among teachers about chat GPT. Before that on the brief, should AI chips be used as a choke point to determine how countries handle AI? The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our YouTube, our Discord, and our newsletter. Welcome back to the AI breakdown brief. All the AI headline news you need in around five minutes. Boy, you can tell that we are heading into a, long holiday weekend because the news is sparse. But where we start today is with a little bit of a follow-up to yesterday's discussion around the geopolitics of AI. Mustafa Sullyman is the former co-founder

of Google's Deep Mind and the new founder of Inflection, whose major product is, of course, Pi. Now, in addition to that, he is also promoting a new book that is coming out this month, and so has been all over the thought leader junket. He recently had a conversation with the Financial Times, in which he effectively validated the U.S.'s strategy of using access to AI chips as a geopolitical tool. In this discussion, he suggested that we should expand that policy and try to leverage it to enforce a new set of AI standards. In a conversation with F.T., he said, the U.S. should mandate that any consumer of NVIDIA chips signs up to at least the voluntary commitments, referring to the voluntary commitments that were won by the White House earlier this summer, and quote, more likely more than that.

That would be an incredibly practical choke point that would allow the U.S. to impose itself on all other actors in AI. Now, interestingly, it sounds like from this conversation that Silliman isn't so much concerned about the extinction and existential risk questions as relates to other users of AI, but on problems that are here and now. He said too much of the conversation is fixated on superintelligence, which is a huge distraction. We should be focused on the practical near-term capabilities which are going to arise in the next 10 years, and which I believe are reasonably predictable. He argues that fundamentally this technology is different from previous technologies in the ways that it will participate in the economy and in society. Specifically, whereas other

technologies are exclusively tools, AI agents will have a more direct role. In the future, he said, AI is going to be participating in the economy in a material way, unlike the way that Excel participates in the economy. It's going to be orchestrating actions using APIs. It's going to be booking and buying and planning and organizing. Now, when it comes to the critique of those voluntary standards, Suleiman argued that the critics were looking at it the wrong way. He said, practically speaking, the odds of passing primary legislation through the U.S. political process are very low. It's not very often you get the seven leading players in a new generation of tech to sign it voluntarily to a set of commitments. Now, in addition to that type of voluntary commitment

and using AI chips as a choke point to get other countries to sign on to those commitments, and of course needing new policy and regulation, Suleiman also said in this interview that there was the need for a new global institution that could bring transparency to AI models. Overall, he said, quote, inaction would be the worst of all possible worlds. Now, if you are like me and like interesting big picture power shift conversations, even if there is an air of just trying to promote the book, the fact that he is trying to promote this book leads to a lot of interesting discussions being prompted. For example, Suleiman also published in Time magazine an essay called How the AI Revolution Will Reshape the World. In it, he writes,

AI is different from previous waves of technology because of how it unleashes new powers and transforms existing powers. While all waves of technology create altered power structures in their wake, none have seen the raw proliferation of power like the one on its way. Maybe we will read that on a future Long Reeds episode. Another follow-up from the geopolitics story, Reuters reports that a number of companies are seeking approval for technology that can be used for image audio and video deepfakes. This is apparently different than the process by which LLMs got approved this week, and China so far has received 110 applications for approvals that relate to models that can be used to manipulate audio and visual data. Speaking of manipulating audio and visual data, one of the big concerns

that people in the U.S. have is tampering with the 2024 presidential elections. Already, media are having to deal with the reality that AI now exists. Insider publish a story this morning. A MAGA News network said its call with Trump was genuine after a report suggested it was an AI fake. Basically, this network, Real America's voice, streamed an audio-only interoperated. with Trump on Thursday, but because the audio was, quote, glitchy sounding with several breaks, a number claimed it was AI. One commentator wrote, doesn't sound like my president, that sounds nothing like Trump. He is slurring his words, cadence is wrong, diction is wrong, tone is wrong. Another wrote, his cadences off. It sounds like AI. Now, what's weird about this to me is that

representatives of the network have said it's real, but a representative for Trump himself declined to comment. Even if it does end up being real, it shows just how big of an issue this is going to be during the selection cycle. Moving on to our next story, one of the big questions right now is rights around data usage. We've discussed a number of times over the last couple weeks, how media companies are responding to chat GPT's new GPT bot and the ability to block it, and now apparently Facebook and Instagram have created a form that allows users to request to opt out of their data being used to train meta's AI.

Gizmodo writes, Meta introduced a new privacy setting Thursday that lets you ask pretty please for the company not to use your data to train its AI model. models. Buried in the nether regions of Facebook's privacy center, you'll find an entry called Generative AI Data Subject Rights. Here you'll find three options. You can tell Facebook you want to access, download, or correct any personal information, say you want to delete that personal information, or fill out a blank tech box if you, quote, have a different issue. Now, usually I would summarize this, but the Gizmodo author had a lot of fun with this piece and you can tell. Continuing, quote, the form then asks you for your name, email address, and country of

residents. You hit submit, the website tells you thank you for contacting Facebook. At this point, you'll probably want to do some occult ritual to ensure the data gods hear your plea. So there you have it. Good luck if you don't want meta to use your data for AI training. Now, closing out this brief with a trip into the world of science fiction, nature yesterday published in an article AI predicts chemical smells from their structures. Neural networks can provide descriptions such as grassy for a wide variety of molecules, including some that don't exist in nature. Nature writes, To explore the association between a chemical structure and its odor, the research designed an AI that can assign one or more of 55 descriptive words such as fishy or whiny to an odorant.

The team directed the AI to describe the aroma of roughly 5,000 odorants. The AI also analyzed each odorant's chemical structure to determine the relationship between structure and aroma. The system identified around 250 correlations between specific patterns in a chemical structure with a particular smell. The researchers combined these correlations. into a principal odor map, POM, that the AI could consult when asked to predict a new molecule scent. Now, apparently then, to test against this, researchers also trained 15 volunteers to associate specific smells with the same set of descriptive words, those 55 that were used by the AI. After asking humans to describe odors and then asking the AI to predict odors based on chemical structure,

the AI tended to guess very closely to the average response given by humans. In fact, it was generally closer than any individual's guess. Lastly, now that it's September, of course, we have to start thinking about holiday shopping, and if you are a gearhead, you will be interested to learn about Microsoft's new patent for a backpack that uses AI to superpower the wearer. ZDNet writes, Microsoft filed a patent for an AI backpack straight out of a sci-fi movie. Think Dora's backpack, but better. Now, basically, this backpack includes pressure sensors, a microphone, a camera, GPS, a compass, a barometer, biometric sensors, a speaker, a display for visual outputs, processor, and more. The example they gave in the patent is a skier who asked the backpack a question about the slopes,

which can then tell him which direction remains in bounds and which direction to go. And while this backpack will most certainly not be available anytime soon, Microsoft does have an event on September 21st in New York City, at which they promised to share a number of new AI innovations. That's going to do it for today's AI breakdown brief. I'll be back soon with the main AI breakdown. Before we get into the main AI breakdown, I want to tell you about today's sponsor, Supermanage.

If you work in a professional setting, you probably have some version of a one-on-one meeting, either with the people that work for you or the people that you work with. Unfortunately, all too often, those one-on-one meetings become glorified catch-up calls. Don't you wish you could jump right to the stuff that really matters? That's where SuperManage comes in. Supermanage AI magically distills your team's public Slack channels into a real-time brief on any employee, any time.

Catch-up on contributions, work-in-progress, challenges they're facing, sentiment, everything you need to show up ready for a truly meaningful conversation, and it's completely free. Visit supermanage.a.i forward slash breakdown today to start making the most of your one-on-ones. And thanks again to Supermanage for sponsoring the AI breakdown. Welcome back to the AI breakdown. Today is September 1st, and that means a few different things. First of all, it means we are firmly in pumpkin spice season. Second of all, it means beautiful fall foliage, rich family traditions.

and all of the greatness that is autumn. And third, of course, it means back to school. Now, this will be the first full school year in which generative AI in the post-chatCHAPT era is available. And it is 100% the case that because of that, this year at school will look very different from any other school year that has ever been. At the same time, it's highly likely that this will be the first year of what every year of school looks like going forward. So what we're going to do today is look at some of the posts out there, around how schools and educators are thinking about adapting to this new era of generative AI, including a post from OpenAI themselves, about how teachers might use chat GPT in their classrooms. Now, just to throw it back for a little bit of background, because, of course,

it's the AI breakdown and that's what we do here. This summer has been dominated by the narrative discourse around declining usage in chat GPT, even though it seemed like the obvious candidate for that was students not using it anymore because they weren't in school. One of the things that I found interesting about the discussion around this is how much media tried to make this some sort of threat to chat GPT. Insider writes, schools out for summer, and if chat GPT usage is falling because students are off, that's a bad sign for open AI. An analyst at Bernstein told this insider author, quote, if it's school kids, that's a real yellow red flag on the size of the prize. This idea that if chat GPT drop off is due to students on summer break, that implies a narrower audience and fewer use

cases. Far be it for me to critique a, quote, top internet analyst at Bernstein, but I don't know what the hell this guy is talking about. First of all, education and students are always the first users and the most robust early adopters of new technology. I don't know why we would think that chat GPT or generative AI more generally wouldn't be the same. Second, I'm not sure how the very obvious and clear uses for LLMs inside the educational environment says anything one way or another about whether LLMs are going to be useful for other areas of business or personal life. It reeked back then and it still reeks now of simply another new thing to say about chat GPT because it was getting boring at the time to talk about how amazing it was. With that little mini rant out of the way,

let's turn to how OpenAI is thinking about and talking about teaching with chat GPT coming into this new school year. The company writes, we're releasing a guide for teachers using chat GBT in their classrooms, including suggested prompts, an explanation of how chat GPD works and its limitations, the efficacy of AI detectors, and bias. Now alongside this blog post, they also released an educator FAQ. The questions they answer here in this FAQ include, how can educators respond to students presenting AI generated content as their own? Is chat GBT biased? How can chat GPD be used for assessment and feedback? Is chat Chbt safe for all ages? Does chat GPT tell the truth? Are there any resources for educators to learn more about AI. How can educators get started with chat GPT? And can I ask chat

GPT if it wrote something? Now, let's look at what they say about the big hot button issues. And let's start with these students passing off AI generated work as their own. The cynical take about LLMs in schools is that students are just simply not going to do their own writing anymore. They're just going to plug prompts into chat GPT or clod or whatever else they have access to, maybe tweak it here or there, and pass it off as something that they created from themselves. Last year, we had numerous stories about educators trying to use chat GPT itself to determine if students' work was created with AI, leading in many cases to false positives where chat GPT identified that it was the source of a student's work, when in fact, that wasn't the case.

OpenAI tries to address that one right on its nose, saying, chat GPT has no knowledge of what content could be AI generated or what it generated. It will sometimes make up responses to questions like, did you write this essay, or could this have been written by AI. These responses are random and have no basis in fact. Simply put, you cannot copy an essay into chat GPT and ask it if it was written by ChadCGPT. Hopefully that message will get through to educators this year. But what about other approaches to understanding if a student's work is AI generated? Outside of just updating academic honesty policies, what are things that schools can do? Specifically, are there third-party tools even if ChatGPT can't identify work that was created by

AI, might other solutions help in that area? Again, OpenAI is pretty blunt. Do AI detectors work they write? In short, no. While some, including OpenAI, have released tools that purport to detect AI-generated content, none of these have proven to reliably distinguish between AI-generated and human-generated content. They continued to elaborate on our research into the shortcomings of detectors. One of our key findings was that these tools sometimes suggest that human-written content was generated by AI. When we at OpenAI tried to train an AI-generated content detector, we found that it labeled human written texts like Shakespeare in the Declaration of Independence as AI generated. Now, importantly, they say this also gets into questions of bias and

fairness. They say that there were indications that this could disproportionately impact students who were learning or had learned English as a second language, or simply students who weren't necessarily as good at writing. This gets into that other question, is chat GPT biased? Rights OpenAI, chat GPT is not free from bias and stereotypes, so users and educators should care review its content. The model, they say, is skewed towards Western views and performs best in English. Some steps to prevent harmful content have only been tested in English. What's more, they say, the model's dialogue nature can reinforce a user's biases over the course of interaction. For example, the model may agree with a user's strong opinion on a political issue, reinforcing their belief.

Now, what this all leads to, both in the bias section and in terms of how educators can deal with students passing off AI as their own work, is basically the need for a new pedagogical approach that incorporates ChatGPT as a thing that students are simply going to use. Now, this echoes an essay from last week by the New York Times, Kevin Ruse. The piece was called how schools can survive and maybe even thrive with AI this fall. Step one, Kevin writes, assume all students are going to use the technology. And I think Kevin is dead on here. Now, Kevin talks about how the first instinct last year when ChatGPT was released was for educational institutions to ban it.

Subsequent to that, there has, of course, been a lot of evolution in that conversation. conversation, if for no other reason than sheer practicality. Kevin writes, I encourage educators, especially in high schools and colleges, to assume that 100% of their students are using chat GPT and other generative AI tools on every assignment in every subject, unless they're being physically supervised inside a school building. Kevin points to a recent op-ed in the Chronicle of Higher Education called, I'm a student. You have no idea how much we're using chat cheptie. Now, Kevin's second two pieces of advice are that one school should stop relying on AI detectors, echoing what OpenAI said. And finally, that, quote, teachers should focus less on

warning students about the shortcomings of generative AI than on figuring out what the technology does well. Now, of course, a cynical take would be that OpenAI is incentivized to say something very similar, but when you look at their suggestions, it's hard not to see it as quite obviously the right approach. OpenAI's blog post gives a number of examples about how teachers are using chat chip-t. One example is role-playing challenging conversations. A professor, for example, at Old Dominion, has their students use ChachyPT as a stand-in for a debate partner to point out weaknesses in their arguments. Another idea is to build quizzes, test, and lesson plans from curriculum materials. A professor in Spain they point to, shares their curriculum to Chachyptee, and then

asks for things like, quote, fresh quiz and lesson plans to give themselves new ideas and help them in their own planning work for teaching. Another use, they say, is reducing friction for non-English speakers. A professor at the University of Johannesburg has been encouraging his students to use chat GPT for translation assistance to help them improve their English writing and to practice conversations. Finally, there is the vague, but ultimately probably most important idea, teaching students about critical thinking. This in some ways is turning the weaknesses of chat GPT on its head by asking students to recognize that chat GPT gives answers that are not necessarily credible or accurate, that potentially helps them think critically about whether they should trust

the answer, how to confirm information from other sources, and even how to push the system to produce better work. One of the things that run throughout OpenAI's recommendations, even if it isn't stated quite this crisply, is that just as students can use ChatCHEPT to improve their productivity, potentially to the extent that they don't even do the work, so too can teachers use ChatGBT BT as a tool to help them be the best possible teachers. Lesson planning is a time-consuming and creativity-draining endeavor. Chat-GPT is an incredible assistant for that type of work. Now, interestingly, if you look across a number of different sources and kind of review the articles that are coming out right now about how educators and students are thinking about AI heading into the next year, there would seem to be some pretty broad agreement that the understand that it's here and inevitable and figure out how to maximize it, point of view, is winning converts left and right. And one of the interesting things is that you're starting to see as educators embrace the technology, they're creating more space to figure out what it does well versus what it doesn't do well.

Katie Pierce, an associate professor at the University of Washington says, I love AI chatbots. I use them to make variations on quiz questions. I have them check my instructions for clarity. I have them brainstormed activity and assignment ideas. I've tried using them to evaluate student essays, but it isn't great at that. CBS News writes, educators say they are working with, not against AI in the classroom. One of the things that the professors quoted in this piece recognize is that even if they try to not expose their students to or block their students from using these types of tools, As soon as they get into the real professional world, they will need to be literate in exactly these

types of tools. Dan Wang, a sociology professor at the Columbia Business School, said, my opinion is that it is my obligation and responsibility to expose and immerse students in these generative AI tools. The reason why is because the MBA students I teach are going to be entering the workforce in about 10 months, and they'll often be working within companies and organizations that encourage employees to make use of generative AI tools. Indeed, putting a fine stamp on this, the Teachers College at Columbia University, one of the most respected teachers' colleges in the world, published a piece last week called Generative AI is here to stay. We have to learn from it. The gist is the same as all of these other pieces. This is simply not a genie that can be put back in

the bottle. Students are going to use it. The only question is whether educators can help them use it well. I think in the context of having had more time to start thinking about these questions and actually finding ways to use it effectively, we could start to hear some pretty amazing stories of how chat GPT and other tools like it are actually being used to significantly improve student outcomes over the course of the next year. So I, for one, am certainly excited and optimistic about these possibilities. So happy September 1st, happy Labor Day weekend, and until next time, peace.