Love and Work Are Not the Same
I love pies. My favorite way to use data is with pie charts.
No one says this. At least, I haven’t heard much about it as a data scientist. If you said something like that in a meeting, you’d get polite laughs reserved for people who might be joking. Because we all understand, instinctively or by trade, that liking something has nothing to do with whether that thing helps you understand anything or do better. Or, to be more precise, the message or story you share determines the most effective way to communicate with your audience. Finding something you love can increase satisfaction and motivation. But that is different than reading.
A pie chart with 14 roughly equal slices tells you nothing. A pie chart that compares prices over time tells you the least (misleading). And no personal preference for cakes can change what the human eye cannot and cannot judge. We are bad at comparing angles. We can compare the length. Don’t say angry 3D charts!!
This is why the boring bar chart keeps winning. Your preference does not receive a vote. We, data scientists, choose visualizations based on data. Whether you’d like to see more pies or not, that’s entirely your call.
But I really enjoy pie charts.
Of course. You may also enjoy reading quarterly sums like a limerick. The question was not what you like. The question is what helps you see the details of the data.
Learning Pie Charts
Which leads me to the never-ending story of learning and learning styles. We’ve spent decades doing this with a straight face: someone says “I’m a visual learner,” and instead of a polite laugh, they get a redesigned lesson with lots of pictures. We survey people about their preferred learning style. We sort them into buckets: visual, auditory, emotional. We create content to match. We call it reader-centered design and we feel good about it.
When researchers looked for evidence that matching instruction to learning styles improves learning, they found little evidence. They found nothing, and several well-designed studies contradicted the idea (Pashler, McDaniel, Rohrer, and Bjork, 2008). Yet a systematic review found that approximately 89% of teachers still believe in matching instruction to learning styles, and most report that they actually do (Newton and Salvi, 2020). And now the AI is trained on all that fiction. No wonder AI agents are as confused as humans.
Confusion
We are confusing two different questions. “What do people prefer?” real question. It’s worth asking. Likes affect motivation, and motivational issues. “What exactly works?” a different question. It has a different answer. And when these two conflict (as they often do), the “practical” must prevail, because the student’s intention was never taken care of. It was about getting better at something.
A person who wants everything in pie charts doesn’t need other pie charts. They need someone to show them a bar chart and say: Look how fast you just got the answer. The self-proclaimed visual learner doesn’t need his safety training turned into infographics. They need to practice retrieval, space, feedback, worked examples, unpleasant things with real evidence behind them, regardless of which sensory channel they claim is their brand.
Measurement is Key
Data preference. But the data is about comfort, not efficiency. Design it, and you’re preparing how the learning feels instead of how it happened. Those two measures diverge more often than we would like. A smooth, pleasant experience often produces worse retention than a hard, less comfortable one. If you’ve ever seen glowing academic tests sitting next to flat performance numbers, you’ve watched a breakup happen.
So the next time someone asks for a training version of a pie chart: short, cute, tailored to their style, take the request seriously as a signal about motivation. Then ask a better question. Not “what do you like?”
“What would it take to get better at this in 90 days?” And measure it. No one ever answered that one with “many pies.”
Wait, It’s Getting Bad with AI
For decades, one thing has quietly shielded us from our negative outlook: cost. The cost of resources, time, and effort to produce the same “learning style” content. Creating three versions of the courses (visual, audio, and human), 3 styles out of 75 different ones, was expensive. Budgets forced trade-offs, trade-offs forced questions, and at some point one often asks, “wait, do we really need this?” Friction was our quality control by mistake.
That’s over. Get ready for personalized pie chart lessons.
AI can now generate a version of your lesson for every student. Not 3 styles, all 3000. A podcast version for “the listening reader.” An infographic for the “visual reader.” Imitation of anyone who has tested “hands on.” Each is produced in minutes, each is polished, each is so personalized that people can’t even keep up. Dashboards will light up. Students will report that they like it. Marketers will find funny AI stories they can play at twice the speed about the product experience. Win-win!
- Did I mention measurement issues?
From now on, it will be the only thing that matters. If the basic theory the AI was trained on is wrong, then we haven’t actually produced personalized learning. We made a pie chart industry. In 3D with rainbow colors that speak.
AI does not validate our design assumptions. It grows them. He feeds the myth, and will fulfill that myth flawlessly, at scale, with a confidence that looks like a lot of evidence. A bad idea often failed a little, in one lesson, when someone might notice. Now it can fail well in the whole company, wrapped in the word “adaptive.”
There is a second trap hidden within the first: measuring the wrong thing. When AI is trained to satisfy students by giving them what they want (rather than what they need), satisfaction scores will skyrocket. Smooth, engaging and fun. It feels good to read. The problem is that the struggle is what makes the habit stick. An unconventional format that commands real attention. A wrong answer to live with before being exposed. An improvement loop identified in satisfaction will promote learning with pleasure from learning.
None of this makes the AI a villain. The same machine that can produce an endless pie is a machine that can finally do things that we have never had the power to do: adapt to what the student knows rather than what he likes, produce a habit of returning to the edge of a certain power, put in place within weeks, change the format deliberately so that it does not match the style, but to break the comfort of another.
We, the people, are responsible for how we use learning AI. We, the people, are responsible for choosing and influencing the path of “AI solutions” for technology vendors. We, the people, have a responsibility to say no to lessons where they are not needed, and yes to solutions that may fall outside of traditional technology. We don’t need to write white papers to convert learning style believers. We need to put a measure that shows the impact on the work. Not just rote memorization or programmatic recall, but real, ongoing behavior change that occurs under real-world conditions.
“Feed content” is as bad as 3D pie charts of all data. The problem is not the AI. The problem is the person. We have an opportunity to stop hiding because of a lack of resources and technology to do the right thing. No more excuses. But, without programming thinking and focusing on impactful behavior change, this “content feed” may turn out to be just what it says: upload your PDF, and it creates amazing pie charts from it.
No pies were harmed during this article.
References:
- Pashler, H., M. McDaniel, D. Rohrer, and R. Bjork. 2008. “Learning Styles: Concepts and Evidence.” Psychology in the Public Interest 9 (3): 105–19.
- Newton, PM, and A. Salvi. 2020. “How common is the belief in the learning neuromyth, and does it matter? A pragmatic systematic review.” Limits in Education5, 602451.
