Why AI Won’t Replace Humans…It’s a Piece of Cake!

There’s been a lot of chatter online recently about whether or not AI will replace humans or create more jobs. In education circles, I often hear concerns about students using AI to cheat and skip ahead, essentially by-passing the learning process to get to their final product.

These are both valid concerns and in some areas, well justified. However, I don’t think AI will replace as many jobs as we think, nor will it create as many as we think. I also don’t believe it will reach true superintelligence for quite some time, if ever. The acceleration and development of AI platforms has been staggering, especially when it comes to video creation, but so far, I’ve found most large language models like Gemini and ChatGPT to still be pretty bland when it comes to outputs. And even after it has created whatever I have prompted (and re-prompted several times) there still is a LOT for the human to do when it comes to higher order thinking.

Making a cake

Here’s an example that I feel sums up the power (and limitations) of generative AI:

This past weekend, my niece hosted a wedding outside of Austin. As most in attendance lived elsewhere, we became the de facto hosting site to prep things like music playlists (I was the DJ), flower arrangements, and some of the food. My sister-in-law Traci (mother of the bride) came into town a couple of days early to help manage all of the last minute details and to get help with one of her biggest responsibilities…making the Groom’s cake.

We looked at various pictures on Pinterest, long considered the go-to source for such wedding inspirations, but didn’t quite find what she had in mind. She had made several “base cakes” in 8″ and 10″ circles because she wanted some extras in case mistakes were made. She knew she wanted it two-tiered and for the icing to be chocolate butter cream. Enter ChatGPT.

Despite my spelling and grammatical errors, I felt like it understood the assignment although I’m not sure how “easy” this design would be to do. Time for a re-prompt.

I’m not sure why I said “maybe” in the prompt, but it goes to show these tools are very conversational, so I talk to it like a human. The resulting cake is coming together, so with a few more prompts and some additional direction, we had reached what all agreed was the best model image as to what the cake could look like given the lack of cake-decorating experience in the house.

Now comes the interesting part…you still have to make the cake. AI helped by giving a model image and some instructions, but it cannot make, assemble and decorate the final cake. It can’t check the flavor profile or the consistency and moistness of the cake itself. So where would start if you were given this assignment? Would you just throw the cake together and try your best?

In our case, with all the extra cakes, we thought practice might be the best idea. Taking some wax paper out to try the icing patterns and making sure everyone in the house had a try to figure out who could nail the pattern best.

My daughter helped with some of the gold dusting of the strawberries and playing with the ganache to see if we could get the cool “drippy” pattern AI had suggested in one of it’s images.

The resulting mess and failure gave us some direction in terms of what we could and couldn’t do. You can also see from the image above that some of the cake was “taste-tested” by several in the house to taste the balance of chocolate cake and icing. Ultimately, after lots of practice and input, the cake was completed and ready for the wedding. Here’s the actual human result:

While not an exact replica, it turned out pretty close to the model image AI had given us and it tasted AMAZING!

Human in the loop

While AI played an influential role in the process of making the cake, ultimately it still came down to the humans. Let’s think about this in the context of school and learning. AI can help students brainstorm and it can help teachers come up with lessons. It can play the role of tutor or “thought partner” in various aspects of learning, but it can’t do the learning for you. Throughout this entire cake-making process there was a series of failures and what I like to call “productive struggles” that ultimately went into making the final cake.

My big takeaway from this is that we really need to encourage this productive struggle in our schools. We learn through failure and through trial and error. AI has the potential to take away some of that productive struggle when it comes to lower Bloom’s tasks like remembering and understanding. But given the current models out there and those in the foreseeable future, it can’t do the higher order thinking parts of Bloom’s like applying (practicing the icing), analyzing (figuring out where and how to apply it), evaluating (tasting for consistency), and ultimately creating (building the actual cake).

In our classrooms, we’ve needed to shift our focus from those lower-order thinking skills and AI might just be the impetus to make that happen. So the next time you prepare to embark on a project or paper and are worried about the students cheating or having AI do it all for them, think about which parts AI can’t really do and focus your evaluation on how they make their cake. Encourage trail and error by establishing it as part of the learning evaluation process.

Now, if you’ll forgive me…I have some delicious cake to “evaluate.”

Carl Hooker is an international speaker and educator. He works with schools and events across the country to thoughtfully integrate tools like AI into learning. He just released his second edition of the best-selling AI book Learning Evolution which shares several examples, strategies and ideas like this one. If you are interested in booking Carl for your next event or professional development day, fill out this speaking form to get more information.