They’re Promoting AI for the wrong reasons.

On a scale of creativity, I give ‘Chatbot ideation’ a solid C

I recently wrote to a data scientist in response to a Wharton School study she referenced (about AI’s idea generation) Give a large language model (LLM) the right prompts, she says, and it beats humans, hands down, when it comes to brainstorming.

Hmmm.

Well, I’ve done plenty of brainstorming in my past professional life, and lots of it now in education. We use Socratic discussions, and other methods every week, so we see how ideation works. In young children! My beef with the writer were the strategies she says are effective in getting a chatbot to spit out ideas. She recommended: ‘threats, tips, pleading, and emotional appeals’ as well as ‘chain-of-thought commands.’

Threats and emotional appeals? That’s like telling a parent to use the carrot and the stick to get a kid to do his homework.

Why the mad rush to bypass humans for original, well-thought-out ideas? Because it’s a shortcut? Is the alternative is so darn hard? It isn’t. I know because everytime I start a discussion in class I am surprised at the ways young minds come up with angles, nuanced ideas, and questions that provoke more questions. I end up with a white board chock-full of ideas.

So here’s the gist of my letter to the researcher:

My students are bursting with ideas, and a teacher’s role is to tap into these, and open the spigot, not introduce a bypass valve. I gave Poe (sometimes called the web browser for chatbots such as ChatGPT, Claude, Stable Diffusion etc.) a prompt to come up with a policy for cyber security in one paragraph. It was so bland, an 8th grader could have done better than that.

Just for fun, I then prompted it (using the ‘chain-of-thought’ method) to turn it into a rap song, and the results were…embarrassing. I could teach a 7th grader to write a rap in 30 minutes with better results.

Which brings me to my conclusion. LLMs are great at speed and variety, but poor in quality. They will dumb us down if we let them become our gold standards.

______________________________________

What, then are LLMs good for? I have seen some interesting graphics created in my class by students using Bing. Yes, even those embarrassing ones of people with a weird fingers. Like this one-if you look closely.

Or this one for a prompt for a ‘bull in a china shop.’

Perhaps there will be a place for AI — one day. I have read how some people use it to draft communication (emails and boilerplate gobbledygook for websites). In other words, they are great for people who are (a) incompetent (b) in a rush and don’t care about the output (c) just lazy.

As I have commented before, LLMs are solutions in search of problems.

_________________________________________________________

Note: As a teacher I have been testing a variety of chatbots over the past six months, and have been getting very mediocre to poor results. On the scale of creativity, I give it a C. Let’s not poison the well of creativity, or idea generation.