AI chatbots living rent free in the brain (and why they have trouble with “off-color” jokes!)

AI chatbots living rent free in the brain (and why they have trouble with “off-color” jokes!)

living rent free in the brain

Artificial intelligence chatbots have rushed into our lives faster than our brains—or our institutions—know how to handle them. Some headlines still promise “supercharged learning” with tools like ChatGPT, while a newer wave of research raises a much more sobering possibility: that offloading thinking to AI may quietly undermine our mental edge over time. Recent experimental studies with students have found that when people lean heavily on AI for writing and problem solving, brain activity in regions linked to creativity, attention, and deep information processing can drop dramatically, and users often remember less of what they had “produced” with the tool’s help. 

Preliminary work in educational settings, however, suggests that chatbots can possibly improve short term learning outcomes—but mainly when they are designed to support metacognition, that is asking learners to explain, reflect or choose, rather than simply walking away with complete answers or drafts. Work with older adults likewise frames chatbots as potentially “helpful” companions for cognitive “upkeep” and training, prosthetic-like thinkers. Many of us are already navigating normal age related changes in processing speed and memory. The temptation to let AI “just do it for me” is understandable—but if we get the machine to work for us, promoting the kind of active engagement that helps protect brain health . . . 

Put simply, AI is a power tool. Used thoughtfully, it seems to have potential; used carelessly, it can (best case, just) encourage mental downshifting and dependency. But there is hope, actually . . . 

trouble with “off-color” jokes

In new research just summarized on Neurosciencenews.com, LLMs were seen to fail significantly on the classic SLOOP test protocol, where a subject has to quickly respond to words printed on a screen that are presented in colored letters. Sometimes the prompt may be, for example, the work "green," printed in blue fonts, etc. Humans are, generally pretty good at identifying the font color; AI was near incapable of correctly processing many prompts where the graphic "contradicted" the font color. That is, in principle, a potentially very important finding. In essence, it suggests that AI, at least the current generation of chatbots, does not have anything close to the executive function of the prefrontal cortext in humans. 

In other words, the human brain, when processing a sentence like, "This is a great party," is able to take into account not just the semantic meaning of the words and the grammar involved to come up with an interpretation, but also the "color" of the language, that is the special characteristics such as pitch, rhythm and volume, let alone all the context involved not present in the sentence or the culture or even the conversation that it occurred in. 

That certainly makes sense for words/color tasks, and it begins to explain why AI’s responses can so often be emotionally “random or more than slightly off.” Would the same principle not apply as well when confronted with all the other “coloring” of tasks as well? Let’s just assume so for now, ok? Future research will certainly unpack it. 

A few practical guardrails:

• Ask the AI to quiz you, not just tell you. Turn explanations into short recall questions you answer before revealing the response.

• Use it as a thought partner: “Here is my reasoning—critique it,” rather than, “What should I think?”

• Require yourself to rewrite or summarize what the chatbot says in your own words, away from the screen if possible.

• When studying Scripture, history, or philosophy, have the AI present opposing interpretations and then decide, in writing, which one you find more compelling and why.

• Finally, schedule some tasks—journaling, planning, prayer, or reflection—that you deliberately do without any AI assistance, to keep your native capacities in regular use.

My current work on the AI ROBICS project grows directly out of this emerging research landscape. Instead of treating AI as a magic pill for cognition, AI ROBICS explores structured, “exercise like” patterns of chatbot engagement designed to train attention, reflection, and discernment—especially for older adults who want to remain mentally and spiritually agile in an AI saturated world. 

In future research reports, I’ll be unpacking more concrete AI ROBICS routines you can try, so that your time with chatbots functions less like passive entertainment and more like a carefully planned workout for the mind and soul.

*And, of course, as AI engagement expands, so does the need for better scam, identity and deep web protection—and legal backup for us all. Connect with me when that makes sense for you! 

wracton@gmail.com

(423) 660-7400

www.williamacton.legalshieldassociate.com