Chatbot for talk therapy — A Journey part 2

augmentedrobot
5 min readJul 2, 2020

--

Welcome to part 2 of this journey, the workshop and the digital survey.

A few weeks ago I posted a survey via PsyToolKit asking people what they would like in a chatbot. I made it in english to broaden the availability.

Here is a summary of 80 respondents (quite few, but I am still happy and grateful)

When it came to gender distrubution most people who answered identified as female.

Majority of those who answered were between aged 26–56, which is understandable considering it was maily spread on Twitter and LinkedIn. Had I targeted youth, TikTok or Instagram would have been a better target.

7% didn’t know if they talked to a chatbot or a human!

According to the results, a name was good to have, but a gender was irrelevant. The bot should be personal, but more like an aquaintance and not a friend.

The majority could imagine or are open to talking to a chatbot anonymously about their mental health and 7(!) percent didn’t know if they had talked to a chatbot or a human!

This bodes well for an eventual Turing test, which according to this result may already have been passed. OR, just maybe the human at the end of that chat conversation was really bad at communicating.

The main topics of choice for conversation were Work, Stress, Relationships and Conflicts.

The least popular were Celebrities.

Sex, Crime and Fitness — How to handle whatsubjects outside the scope of one’s bot.

One of the things me and my co-workers have to think about, and we discussed this during our workshop was how to handle topics outside the purpose of the chatbot.

I’ve already written a bit about the “Underwear Law” and flirting that can occur when talking to a chatbot.

There is a childish side to all of us, and chatbots trigger it. We humans are happy to test these systems with snuff, and sometimes threats.

When it comes to sex, it’s very easy to handle. The system can flirt back, reject or pretend that it doesn’t understand. I always think the system should write back, but not encourage snuff

An excellent answer to “what are you wearing” is simply to say “I don’t have a body so I do not need clothes, I am a digital system, but thank you for asking!”

When it comes to threats, harassment and admitting a crime, the law is a little complicated when it comes to automated systems. Soon, hopefully, an interview with a regional lawyer will clear some of the aspects of this out! I’ll translate and share the interview.

Words, don’t come easy to me, how can I find a way to make you see that words they don’t come easy….

I hope you’re old or versed enough in music that you’re hearing a tune while reading this header.

Finding words is hard for a chatbot. It soley relies on our ability to create a context and its content.

How chatbot technology works, is that the system stems common words. Stem means taking the stem of the word.

To get as many variations of the word as possible without having to enter it manually you can stem the word disaster. This becomes disa(ster) (sters)(strous). Pretty easy.

But if we take a word like catastrophic. It becomes cata(strophic)(strophy)

But there will also be cata(pult)(maran) etc.

Therefore, you have to manually segregate these words, and make sure that the system can make a differentiation between catastrophy and catamaran.

Finding words, the right words and ensuring that they are distinguished in the system is a challenge. A challenge we have now accepted and taken upon us. How many words will we miss? How many words will the system misinterpret? It will be interesting to see!

I apologiez, I did not understand what you said.

When a chatbot system does not understand what word you were referring to, or stems incorrectly, the bot is sent into the “I don’t know what you said” mode.

One of the big questions we (the team behind the chatbot) discussed was exactly how to solve this aspect and what it should say instead of “I don’t understand”

It’s frustrating because

1) It offends the user, dumbs it down and calls it incoherent

2) Reminds the person conversing that they are talking to a system.

An example of how this can go.

Please take care of yourselves. Wash your hands, keep distance and avoid crowded places.

While I’m at it, hydrate, don’t forget sun screen and don’t go swimming too far out in the ocean.

Have an amazing summer, part 3 coming up in August!

--

--

augmentedrobot
augmentedrobot

Written by augmentedrobot

I’m like an open book. Full of numbers.

No responses yet