After a good conversation with AI, you can feel oddly knowledgeable. You started out talking about animals, and now you feel ready to explain sleep and how to get work done. It's a satisfying feeling.
Sometimes, though, you look away from the screen and find that you can't explain much at all. To turn the conversation into learning, make room along the way to offer your own predictions, check the evidence, and finally explain the idea yourself.
So, does talking to AI make you smarter? It depends on what you do during the conversation. Let's follow one example to explore how to ask better questions, check sources, and see whether that feeling of understanding holds up.
Start with “Do moles get the blues?” and separate the questions
You can start with a playful question: “Do moles get the blues?” If AI introduces research on stress in animals, there's more to ask than “What other animals does this happen to?”
- What did the researchers measure when they called a state “depression-like”?
- How did they distinguish reduced activity from tiredness or difficulty moving?
- How much, if anything, can those findings tell us about humans?
Questions like these turn an interesting story into something you can investigate. If the study involves flies, for example, check which behaviors the experiment actually observed before concluding that flies feel down in the same way people do.
From there, the conversation might move into hibernation, seasonal changes in mood, or thoughts that won't stop going around in your head. After drawing those connections, check the evidence for each one. Similar-sounding phenomena don't necessarily share a single mechanism.
Ask AI for different questions to pursue, rather than a longer answer
Repeating “Tell me more” can simply produce a longer explanation. Try pointing to the part you're unsure about instead.
“Explain reduced activity from three perspectives: physical capacity, the environment, and learned behavior. What observations could help test each explanation?”
That makes it easier to choose what to investigate next. When you encounter a technical term, you can also ask, “What distinctions does this term help us make?”
Try putting your own prediction on the table first. “I think being tired might be enough to explain why it isn't moving. What observations would show that this explanation isn't sufficient?” Whether your prediction is right or wrong, you'll have a clearer idea of what needs revising.
You don't need a brilliant question to begin. You can say, “I'm not even sure which part I don't understand,” and ask AI to work through it with you, one question at a time.
When the explanation fits together neatly, go back to the sources
In this example, the shared feature of “reduced activity” can easily become an explanation for almost anything. It helps to identify what each source actually contributes.
- The experiment by Ries and colleagues reports reduced self-initiated activity in flies after vibration stress.
- The review by Dantzer and colleagues examines the relationships between inflammation, sickness behavior, and depression.
- NIMH describes seasonal affective disorder as a type of depression with a recurring seasonal pattern.
- The 1997 paper by Suddendorf and Corballis discusses “mental time travel,” the ability to imagine the past or future.
Jumping from these sources to “Winter depression is mild hibernation” or “Depression evolved to help us cut our losses” mixes research findings with metaphors and hypotheses. “Low-power mode” may be a useful metaphor, but it cannot serve on its own as an explanation of causes or treatment. Nor does citing a discussion from 1997 establish the current scientific view of other animals' abilities.
You can ask AI:
“Separate that explanation into research findings, researchers' hypotheses, and explanatory metaphors. For each study, give the title, year, study subjects, what was measured, and a URL.”
Then open the links. Check the title and authors, and read the abstract or relevant passage. A real link doesn't necessarily support the claim attached to it. If you can't find support for a claim, leave it unverified. This extra step helps distinguish a satisfying story from knowledge grounded in research.
A useful model helps you see what to test next
A model, as we're using the word here, is a simple map for organizing a situation and deciding what to try next. It doesn't need to be a grand theory that explains everything.
Take an ordinary situation: you're sitting at your desk, but the work isn't moving forward. Rather than stopping at “I'm bad at concentrating,” consider three possible explanations.
- Notifications keep interrupting you and pulling your attention away.
- “Prepare the presentation” is too broad a task, and you don't know where to start.
- You're sleepy, and what you're reading isn't sinking in.
Each suggests a different experiment. Turn off notifications and work for a short stretch. Start by writing three headings. Take a break and review how you've been sleeping. These are everyday adjustments to how you work. If mood or sleep problems persist and interfere with daily life, talking to a health professional is also an option.
To return to the mole metaphor, before spiraling into “Who am I without soil to dig?”, try “Where's some dirt I can dig in?” You might even ask, “Do I need to be digging in the first place?”
A good metaphor earns its keep by helping you ask a different question. Change one condition and see whether getting started becomes easier. If it doesn't help, redraw the map.
Can you explain it without looking at the screen? Can you use it in a new example?
When an AI explanation makes you think, “That makes sense,” pause for a quick check.
First, look away from the screen and put the main idea into three sentences of your own. Next, think of an example that didn't come up in the conversation. Finally, name a situation where the explanation wouldn't apply.
Could the idea that “changing conditions can change behavior” help with tidying up? Before deciding whether to throw things away, you might check whether the problem is that they don't have assigned places. But if you treat every difficulty as a storage problem, the explanation becomes too loose to be useful.
If you take your explanation back to AI, try this instead of asking it to polish your wording:
“Point out gaps in my explanation and places where I'm making claims stronger than the evidence supports. Don't give me the correct answer straight away. Ask one question that will help me rethink it.”
That shifts the task from copying an answer to improving your own understanding. When this article talks about a “mental workout,” it's a metaphor for this way of using AI. You don't need to measure success by the length of the conversation or the number of technical terms you've collected.
End the conversation with three things to keep
It's fun to keep following an interesting topic wherever it leads. If you want to use what you've learned later, though, leave yourself a short note at the end.
- What I can explain now: An observation that activity decreased and an explanation of why it decreased need to be checked separately.
- What I still don't know: Would the experiments discussed here produce the same results under different conditions?
- What I'll do next: Read the abstracts of the cited papers and compare them with AI's explanation.
That gives you a starting point the next time you open a book or paper. Use the conversation to map your questions, check them against the sources, and revise the map after reading. Knowing which parts deserve a closer look can help you make better use of the time you spend with the material.
Take an interesting idea far enough to think it through yourself
One pleasure of talking with AI is following an ordinary question into a neighboring field you didn't expect to visit. You can start with moles, or with “Why is getting started on tidying up such a chore?”
Make a prediction. Open the source. Try another example. Explain the idea in your own words, and revise it if needed. When you can move back and forth like this, more of the conversation stays with you.
Your next question can still be “Tell me more.” But every so often, try “What haven't we checked in this explanation yet?” That may be where the conversation gets more interesting.
References (4)
- Suddendorf & Corballis (1997), Mental Time Travel and the Evolution of the Human Mind
- NIMH, Seasonal Affective Disorder
- Ries et al. (2017), Serotonin modulates a depression-like state in Drosophila responsive to lithium treatment
- Dantzer et al. (2008), From inflammation to sickness and depression: when the immune system subjugates the brain
