I really needed to hear that

There is a loneliness epidemic in our world and loneliness and depression are serious issues. According to the World Health Organisation, one in four older people experience social isolation and 5% to 15% of adolescents experience loneliness. So please don’t take offence to the following post. I don’t want to anger anyone but rather want to show how ridiculous some economic experiments can be.

A team of researchers from the US and Turkey wanted to know if chatbots can help reduce loneliness. To do this, they did a series of experiments, starting with the companion chatbot Cleverbot, which has been around since 2008. On two days, they were given access to the conversations, users in the US and Canada had with the chatbot. In about 5% of these conversations, users mentioned feelings of loneliness like in this conversation:

Chatbot: “Just let you know that you are not alone.”

User: “Thanks, I really needed to hear that.”

As a first step, they measured how long users interacted with Cleverbot depending on whether they expressed feelings of loneliness or not. There was a clear difference with the average length of the conversation being more than twice as long for lonely users.

Median length of conversation with Cleverbot

Source: de Freitas et al. (2024)

Then the researchers wanted to know if users felt better after talking to a companion chatbot. They did this in different ways. First, they looked at user ratings and user reviews for the three most popular companion chatbots: Replika, Chai, and iGirl (no, me either). The reviews that expressed feelings of loneliness rated the apps significantly higher than the average review suggesting that users find these chatbots help in alleviating loneliness.

A more direct result came with a different experiment where volunteers interacted with a companion chatbot the researchers created. This companion chatbot was based on chatGPT 3. They asked the volunteers to rate how lonely they felt before and after the interaction as well as before and after interacting with another human or watching YouTube videos for 15 minutes.

The charts below show that the reduction in loneliness scores was about the same whether the volunteers dealt with a human or with a chatbot. But when they interacted with a chatbot pretending to be a human, the reduction was much larger.

Reduction in feelings of loneliness from interactions

Source: de Freitas et al. (2024)

This is why I find these results so doubtful. Why would interacting with a computer interface that you know is a chatbot that just pretends to be human be more effective than personal interaction with another person? I think something is off here.

Besides, if we want to combat the loneliness epidemic, do these people think we need more software apps? In my view, the loneliness epidemic is a result of us spending too much time with software in the first place and not enough with other people.

To me, this is another one of these ‘Silicon Valley solutions’. A computer solution designed by a bunch of geeks for other geeks and completely missing the point of how real people live their lives and what makes them tick. It’s just another example of why we shouldn’t let Silicon Valley types run a country (just saying…).