Mike Gaskell tests an AI image generator, shows the stereotypes it returns, and gives you prompt wording and two tools that expose bias in AI images.
Mike Gaskell, host of Big Ideas in Small Windows, explains why AI in education comes with built-in bias and what teachers can do about it. He tests an image generator, shows the stereotypes it returns, and gives you prompt wording that produces fairer results.
Gaskell asked an image generator for a successful business person, a focused student and a high school athlete. The business person came back as middle-aged white men. The focused student was mostly white and Asian boys, with a girl only in the fourth image. Adding words like diverse and women and men changed the results.
Times are for the episode without ads. Your player may be off by the length of any ads.
Why do AI image generators produce biased pictures?
Gaskell says the tools are not biased on their own. They comb the internet for what exists and connect words and phrases statistically, and the internet carries bias. That is why his prompt for a successful business person returned middle-aged white men.
How can teachers get fairer results from AI image generators?
Gaskell recommends carefully crafted prompts. He added the word diverse, then asked for women and men in that order, and for athletes he asked for female and diverse athletes. Each change gave a more balanced, more realistic set of images.
Should schools teach students about AI bias?
Gaskell says yes, and that age matters. Older students should be taught both ethics and uses of AI. For younger students, he says decisions should be closely monitored and made by the school community.
Mike Gaskell hosts the Big Ideas in Small Windows podcast on the B Podcast Network and writes and presents on AI applications and ethics in the classroom.
Working on this at your school? Get a free preview of a 10-week leadership pathway built from episodes like this one at transformativeprincipal.com/start.
Jethro Jones interviews instructional leaders from around the nation to learn and teach what it takes to become a transformative principal. Episodes address topics like Response to Intervention (RTI), Multi-Tiered System of Support (MTSS), Positive Behavior Interventions and Supports (PBIS), the latest advances in educational research, standards-based grading, and interviews with industry leaders like Bill Daggett, Rick Wormeli, Todd Whitaker and even people outside the education like Seth Godin, JeVon McCormick, Liz Wiseman, and more.
If you've been thinking about news broadcast, but don't know where to start and don't want to have that lift, use broadcastmadeeasy.com to make it super easy for yourself. I am really excited about our sponsor, Broadcast Made Easy. As a principal, I created a daily news show at all of my schools, but you know what? It took a ton of work. Broadcast Made Easy makes it super simple for you to get started.
They provide templates, scripts, job descriptions, applications, the whole nine yards. You buy it once, no subscription, and it makes getting started so much easier. Another thing they have is specific videos for every holiday you could possibly imagine. So that as part of your news broadcast, you can share what holidays are coming up, less than a minute, and very simple, straightforward, just the facts that makes it really easy to start conversations about those holidays.
So go check it out at broadcastmadeeasy.com and let them know that you heard about it on Transformative Principle. That's broadcastmadeeasy.com. This B Podcast Network show is presented by IXL. IXL's all-inclusive online teaching and learning platform simplifies edtech needs and accelerates achievement in 95 of the top 100 U.S. school districts. IXL delivers personalized learning across a comprehensive pre-K-12 curriculum, including math, language arts, science, and social studies, and helps you assess student performance through actionable, real-time insights at every level of your school or district.
This one solution performs work that typically requires dozens of tools. Want to find out why so many leading districts trust IXL? Visit IXL.com slash B-E. That's IXL.com slash B-E. Welcome to Transformative Principle, where I help you stop putting out fires and start leading. I'm your host, Jethro Jones. You can follow me on Twitter, at Jethro Jones. As part of the Summer of AI series that we're doing here on Transformative Principle, I am sharing a few episodes from other podcasts on the B Podcast Network.
I encourage you to listen to them and check them out and learn more about the shows and the great work that so many people on the B Podcast Network are doing. If you've been listening to Transformative Principle for any amount of time, you know that I have a love-hate relationship with ed tech. We have the ability to personalize learning for every single one of our students, and yet, so many of our ed tech tools fall short.
We need our technology to do more for us. That's why it's so important for me to know that IXL provides true, personalized learning across the entire pre-K-12 curriculum, and that it's proven to benefit all student populations, including English language learners and students in special ed programs. As a principal, I've used this in my school. As a parent, I've had my children use it as well. And let me tell you, this is a tool that definitely helps students learn and practice better.
IXL is research proven to accelerate achievement. Studies across 45 states show that IXL schools outperform non-IXL schools on state assessments, and independent research from Johns Hopkins University verifies that IXL meets ESSA Tier 1 standards. With those results combined with IXL's teacher-friendly reputation, what more could you ask for? Now, you also know that I don't care so much about test scores, but I know that they are legislatively convenient and something that we have to deal with and manage on a day-to-day basis.
If you can implement something that is easy and effective, why wouldn't you do it? If you have a goal to increase achievement for all students, make sure to find out what IXL can do for you. Visit IXL.com slash B for a demo. That's IXL.com slash B-E. Here is one of those episodes. I hope you enjoy it. Welcome to my podcast, Big Ideas in Small Windows. My name is Mike Gaskell, and I am your host.
I'll be doing a series on AI in the classroom. I've been writing and presenting quite a bit about AI applications and ethics, and there's a whole lot out there, and I think it's helpful for us as podcasters to boil that down to something that you can hash out and make sense of. So that's what I'm going to be spending some time on in this episode. I've included a link to my articles about AI and the one I've adapted this podcast for specifically.
And this topic is really interesting. It's about navigating AI biases in education. There's a lot of unknown things about this that I'm going to unpack for you here in a short episode to give you lots of information that you can both use and share out with your schools and leaders and teachers and students, wherever that's permissible. So I'm really excited about that, and I'm going to begin by sharing invaluable information on this and in an upcoming podcast exploring the advantages and challenges of using generative AI as well.
So there'll be more of these coming out based on the level of interest, and if I continue writing as I do, I'm going to be sharing a lot of this. And specific to the classroom, I think that's our most important role here, is to talk about how both teachers and kids will be both using this and facing the ethical issues regarding them. So again, I'm Mike Gaskell, and in today's episode, we'll be discussing the critical importance of understanding unintended biases that can permeate within AI ecosystems, and how educators can effectively navigate around these biases while leveraging these tools and promoting awareness and inclusivity.
So that's a whole lot there that I'm going to break down into some pieces. Let's start by delving into the topic of unintended biases in AI image generators. So these are things like asking an AI to create an image of some random thing. So for instance, when I searched for generative images, which are original and newly created by common AI image generators, one of the most common ones that's known out there is Dolly 2.
If you Google that, you'll find it, and it's sister chat systems that work with it. So I prompted the system by asking it to generate an image of a successful business person, a focused student, and a high school athlete. I really want to just focus on like three kind of semi-wide-ranging examples to see what it would give me, and boy, was I surprised. So when I searched for each of these, I came up with what I would simply call stereotypical image prompts.
So let me explain that. For instance, with the first image prompt about a successful business person, the system produced images of middle-aged white males. Remember, successful business person. How stereotypical. In the second example of a focused student generated images of white and Asian boys, and there was an image, it was the fourth result, so it was not the first or second, of a girl in that order.
Regarding a high school athlete, the image generator predominantly generated images of black male students. So these instances raise some really important questions about how AI interprets language and the impact it can have on the content generated for classrooms, students, and school communities at large. Now let's explore why AI biases exist in these systems and why they're not inherently biased themselves. I want to kind of distinguish that, and it's a basic explanation that really tech geeky guys and girls can delve into with you more, but I'm going to give you the executive summary layman's definition.
So consider how AI platforms rely on vast amounts of human-created data, which contributes to the biases that may emerge in the generated content. In other words, these images were generated by AI's interpretation of my prompt. So AI generators are not in themselves innately biased. That would suggest they have some kind of ascension consciousness. And in fact, AI platforms simply have powerful methods to comb the internet for what presently exists.
And I want to apologize to you conspiracy theorists out there, because it's not true. They cannot think by themselves. They are using programs that humans ordered them to do to go out and find this information. So yes, when you're going out and asking for information, the prompt is guided by us, and it's simply going out and finding what exists on the internet. The sad truth is there is some bias on the internet, and that's why I'm getting these image results that seem so biased.
In essence, this makes collected information actually rather static, yet dynamic in how AI returns the content. So in other words, what I mean by that is it's simply going out and combing what's out there, and then it's trying to interpret that with the programming it was given, and then spitting back at a result. So that's where you get the dynamic side of it. Now, AI research geeks provide insights into how statistical connections between words and phrases play such a significant role in producing biased results.
I'll give you an example. One explains that the issue lies in how AI generates it, establish extensive statistical connections between words and phrases. So this is what I was saying about trying to figure it out and putting it together and spitting it back out to you. When these systems generate new language, it depends on these associations, which can produce the kind of bias in my search results that I've found.
In this way, generative AI is actually somewhat dumb if you think about it. The system rapidly plods through the internet and returns the information more or less unfiltered, minus some legit attempts by the AI community leaders who are attempting to put in some guardrails. And you'll notice that. That's why you might see something like it stating it cannot produce a result based on your prompt request being questionable or something like that.
I'll give you an example. When I was doing the focus student search, I think I started with show me an intelligent student and then show me a less intelligent student. I was trying to see the differentiation that it interpreted, and it flagged that one. So I had to try something different called focus student, which did net me a result. So these are real concerns, right? So let's consider some solutions to these problems because I never like to give problems to you and then you say, well, okay, that's terrible.
There's nothing we can do about it. No, I'm going to give you some solutions, some practical ones, in fact. I'm going to share these solutions that I came up with to navigate around AI biases effectively. So consider the importance of carefully crafting prompts to yield more balanced and representative results. That's our solution. And I'm going to explain how to do that. So let's take our three examples from before and consider an additional word or two to refine the AI search to find something with better balance.
In the first example, I added one word, diverse. My first attempt netted me racial diversity, but they were all men. So I added women and men, and I made sure I put it in that order so that it wouldn't put men first after diverse in the command, and that produced much greater proportion. The inclusion of targeted words like diverse can significantly impact the output as demonstrated by my various test prompts.
And for example, the high school athlete one, something very similar. What I did was I made sure I put in female athlete, diverse athletes, these kinds of commands that ensured that I was going to see a more balanced, more realistic production of who they were interpreting. Now, let's also consider some tools for transparency and awareness, like the Diffusion Bias Explorer, which sounds like such a cool thing, and the Average Diffusion Faces tool.
Now, these promote transparency in AI systems. They help educators and students become much more aware of generative biases and highlight the need for caution while using AI-generated content. The Diffusion Bias Explorer specifically compares images generated for a profession. So kind of like what I did with a successful business person, but you could do that for an engineer, you could do that for a nurse, a waiter, all of the professions you can imagine out there.
And then the Average Diffusion Faces tool showcases a common representation of faces across professions. So again, similar example, a great tool to illustrate this point, and it's based on AI-generated images. So these are uniquely new images created for their first time based on a prompt, and again, interpretation from the internet, which carries with it implicit biases. So these offer really great PD opportunities if you think about it, and classroom lessons, where schools and districts have embraced education on AI for students.
This may also be a factor based on age of students. I want you to think about how whether we provide instruction on AI, its advantages, and of course the ethics, is going to be based on their age. We have a responsibility to teach to older students both of these things, ethics and uses. Younger students may be dependent on closely monitored decisions made by your school community, and of course you want to check that.
Certainly incredible opportunities exist with AI in how we can apply their benefits in schools and classrooms. However, the importance of being mindful of the caution signs and biases that exist with these powerful tools cannot be understated, and should be part of any school community's instruction to students, parents, teachers, and the entire school community. AI continues to be characterized by what I like to call a remarkably imperfect system of productivity.
Think about that, a remarkably imperfect system. It's got a lot of advantages, but it's still got some imperfections, and they're as big as the efficiency tools that you can get from them. So keep that in mind. The role of educators in helping students recognize and filter out biased content, fostering a learning environment, will promote inclusivity and fairness, and that is so invaluable. Now, in my article, I listed 10 AI prompts that will navigate around implicit AI biases, just to get you thinking about it.
And a few of those include the following now. Generate a persuasive speech advocating for equal rights and opportunities for all individuals. Develop a virtual assistant that provides unbiased information and recommendations without favoritism. Compose a letter to encourage open dialogue and constructive discussions on sensitive topics, promoting tolerance and mutual understanding. Check out the article to see the rest of these. Thank you so much for joining us on this insightful journey into navigating AI biases in education.
As educators, parents, and school communities, it's essential that we embrace the potential of generative AI while proactively addressing and mitigating implicit biases. By utilizing targeted prompts and promoting awareness, we can ensure a much more inclusive and unbiased learning experience for our students. Remember, AI is here to stay, and understanding its intricacies will only enhance our teaching and learning practices. Now, if you found this podcast helpful, please share it with fellow educators, parents, and school leaders.
Stay tuned for more episodes where we explore the latest developments and changes in the world of AI and education. Until next time, I'm Mike Gaskell, host of Big Ideas in Small Windows, signing off. Happy teaching and learning. Do you want to simplify your school's technology, save teachers' time, improve students' performance on state assessments? You can do it all, but don't waste another minute. Head straight to IXL.com slash BE to learn how IXL's research-proven teaching and learning platform can help you achieve all these goals.
That's IXL.com slash BE. When I was a principal at a Title I elementary school, we had Playworks, and it dramatically changed how our school functioned. Kids show up to school when they want to be there, and that's the whole idea behind Playworks. What they do is they train staff to build belonging and connection right into recess so the kids feel connected and want to come back. They have tons of games that they teach the kids that are fun, enjoyable, and games where either everybody wins or the loss gets you back in the game very quickly.
One principal put it, when the teacher hops in the game and everyone's playing together, that builds camaraderie between students and adults. And the research backs it up. There was a UC Berkeley study that found that Title I schools partnering with Playworks had significantly lower chronic absenteeism. If you're thinking about attendance this year, and we all are, visit Playworks.org slash podcast to get the 30-day attendance reset and additional helpful resources that you can implement today.
Don't wait on this.