When someone posts in an online support group, most of us would agree that getting a quick reply is a good thing. And it probably is, especially in the cases where a person has been sitting with something they haven't been able to tell anyone else in their real life. But there are other people in that group and those people’s comments (or lack thereof) are often impacted by the initial comments made before they offer theirs.
That is part of what I found interesting in this week's research. One study looked at what happened around posts that received replies from bots. Another moved people around into different parts of the writing process, to see where their ideas made a difference, when paired with AI writing. And in a third, AI helped researchers build a game that encouraged students to learn a specific scenario, and then compared their understanding to other students who received the same information in a standard lecture.
STUDY 02
Where a person's ideas make a difference
Researchers divided story creation into planning, writing, reviewing, and revising, then changed which jobs were done by people or AI. Stories produced entirely by AI received higher creativity ratings than all-human stories, but were more similar to one another. Letting a person make the plan or do the revisions produced a wider range of stories than leaving everything to AI. Simply having a person review the AI's work did not produce the same improvement in variety.
Read the study ↗
STUDY 03
Learning the law by playing through a case
Researchers used AI to help create characters and scripts for a game in which law students played explorers trapped in a cave with too little food. After the researchers checked and revised the material, the students played their roles and negotiated with one another about how to survive. Compared with students taught the same material through lectures, the game players scored higher on the later legal-knowledge and reasoning test. AI helped develop the game, but did not supply the students with answers during play.
Read the study ↗Also worth your time
Explaining the answer to other people. In a probability course, students who met in small groups and worked through unfamiliar versions of homework problems without notes scored an estimated 11.4 percentage points higher on the final than classmates given standard graded homework. Read the study ↗
Feeling better about words that feel less like yours. People who used AI to reframe an account of a stressful experience generally reported less negative emotion afterward, while also feeling less ownership of the text and less like it represented their current selves. Read the study ↗
Why some people would keep working. In a survey about work and its value, about one in five workers said they would keep their job even if offered a guaranteed income worth 120% of their current pay. Read the study ↗
SOMETHING TO PLAY
Try Odd Jobs
Place ramps to get a parcel through the workshop, run your machine, and see where it ends up. Once it works, see if you can collect all three gold bolts along the way. You're planning, testing, and changing your mind as you go, but mostly you're trying to get a slightly ridiculous machine to cooperate.
Play Odd Jobs →The Reckoning
I think we underestimate how much we get from having something to offer other people in our everyday lives. We talk about wanting support, and of course all of us do, but we also want to be the person someone comes to for help occasionally. We get something out of being asked for our help or our expertise, even if we feel like there are not enough hours in a day to do all the other things we have on our plates. Giving requires our time, but it also gives us a fulfillment that almost nothing else does.
A support group has two types of people that can benefit. Not only the person looking to solve the problem, but the person who can provide value by assisting in a way that few else can. Because that person has experience with that issue, or at least something very similar. And in a scenario where AI gives an immediate response, and it might be something more thorough and polished than something we might be able to offer, we might be happy to let AI handle more of it. But the data is beginning to show us that not only do we take away the gift of being able to support and offer guidance, but the person receiving the help doesn’t feel as much connection from receiving the advice, and that lack of connection just might lead to another case where they need some kind of emotional assistance in the future.
That is why I liked the game that I put in this week's research. AI helped create something people could work through together. The students still had to decide what they believed was fair, hear what the others wanted, and try to come to some kind of agreement. There was a use for the technology that left plenty for them to do, and most importantly, they were doing it together.
We are going to keep finding things that AI can do well. But that doesn’t mean that we should just fully hand things over to AI, so that we can get to the destination quicker and with less friction. Sometimes doing something for another person is how we become important to them, and I would like us to think about that before we decide to hand everything over to the machine.
Michael McNamara
Illustrations created with AI. They are editorial images, not depictions of research participants.
STUDY 01
Who replies after the bot?
Researchers compared posts in online mental health communities that received bot replies with posts answered only by actual people. The posts with bot replies tended to get faster responses, but their authors replied less afterward and fewer other people joined in those scenarios as well. In a separate experiment, participants felt more companionship from human responses. They also saw those responses as more authentic and empathetic, and felt closer to the person responding.
Read the study ↗