Interacting with chatbots

How chatbots have become an integral part of the fabric of the internet

A person is sitting down typing on a MacBook, with a book about Python Programming next to them

Contents

1. Introduction
2. Keyword recognition chatbots
3. Contextual chatbots
4. Conversing with bots
5. What problem is this solving?
6. When it goes wrong
7. Summary
8. Further resources

Introduction

Audio introduction for Top 5: Chatbots. A transcript is available for this podcast.

Student Team: Reflections on critical analysis and chatbot topics

A chatbot is a a computer program that simulates a conversation between a human and a computer. There are lots of different types of chatbots you might have come across on the internet, in the next couple of sections, we’ll explore two types you might have come into contact with.

Keyword recognition chatbots

These are the most popular bots in a customer service type situation and you’ve likely used one before. This kind of bot is able to recognise particular words and is trained to answer questions or fetch information. Kind of like an interactive FAQ, it provides a different way of accessing information you could likely find elsewhere on the website.

To illustrate this type of situation, I’ve made a video of my recent experience on the Hermes website, trying to figure out where my order was. This video shows me using the website to track a parcel the ‘manual’ way (using the search box) and then through the chatbot.

THIS VIDEO HAS NO SOUND — The video shows a parcel being traced. First with a reference number using the ‘Track Parcel’ tool on the Hermes website, and secondly with a reference number and the chatbot in the corner of the screen.

Transcript of the exchange I had with the Hermes chatbot
Chatbot: Hi there, to get started do you have a tracking number?
[a selection of buttons appears. I select ‘Yes I have a tracking number’
Chatbot: Go ahead and enter it please
Me: (writes tracking number)
Chatbot: Please enter the delivery postcode
Me: (enters the delivery postcode)
Chatbot: (replies saying that I need to contact the sender — very similar text was used when I tried out the Track parcel tool)

That video shows that it’s actually quicker to get to the same information by using the original tracking system versus the chatbot, in this situation.

That doesn’t necessarily mean that having a chatbot in this instance is pointless, but does it make the case for having one less compelling? What is the benefit of being in a simulated chat environment if the information is easily and readily accessible elsewhere on the website?

💬 Contribute

Read the following prompt then add your contribution in the box below. Responses from the same person are the same colour. All comments are anonymous.

This is just one example of the use of chatbots in a customer service environment. Use the discussion box below to talk about other situations where a chatbot might be useful. Can they be useful for accessibility, for example? Have you ever had a good experience with a chatbot?

If you can’t see the comment box, please write a response to this post instead

Contextual chatbots

These bots are able to learn about a user as they interact. Messages are saved and referred back to within the conversation. For example, if you’re ordering a pizza from a bot, the conversation might go something like this:

You: Hi, when is the PizzaTime shop on Wilmslow road open until?
Bot: Hi, PizzaTime on Wilmslow Road is open until midnight.
You: Thanks, can I order a Veggie Hot pizza for pick up please?
Bot: What size? You can choose small, medium or large?
You: Large please.
Bot: Great, I’ve ordered you a large Veggie Hot pizza for pick up at the PizzaTime shop on Wilmslow road

The context in this (made up) conversation is in the bold text. The Chatbot is programmed to look out for specific words and respond with prepared questions and statements, just like a keyword recognition chatbot, but critically, this bot remembers information you’ve previously given it. It can then use that information to provide a service for you.

Basically, any chatbot is only ever as good as its programming.

Conversing with bots

We’ve looked at some examples of how chatbots can be used in a business context, but what about non-service based interactions with chatbots? Maybe you just fancy having a chat with a bot, what then? There are lots of different reasons to talk to a bot: To ask it questions you can’t ask anyone else, because of loneliness, or just for curiosity’s sake. Sometimes companies provide versions of their AI chatbots as a way of advertising what they can do.

💬 Contribute

Read the following prompt then add your contribution in the box below. Responses from the same person are the same colour. All comments are anonymous.

Try having a five minute conversation with each of these AI instances. Start with the same set of questions, and then see where you end up. Then use the below discussion box to talk about your experiences. You could answer any of these questions, or come up with your own: What do you notice about them? Do they answer in similar ways? What are your perceptions based on the look and feel of their presentation? What purpose do you think it is serving? How do you feel after the conversations?

Kuki: chat.kuki.ai

Evie: www.eviebot.com

If you can’t see the comment box, please write a response to this post instead

What problem is this solving?

When we’re thinking about technology, it’s important to ask ‘what problem is this fixing?’. That’s not to say that something should only exist if it’s solving an immediate problem, but it does form a large part of its overall usefulness. Something could be incredibly interesting and fun to use, think, or talk about, but if there are few real world applications, it’s unlikely to be very popular (Sorry Google Glass).

Something existing purely for the sake of it can be a wonderful thing, most of the time. But when we dig below the surface, at least in the tech world, it’s rarely the case. You will have likely heard the saying, “If you’re not paying for the product, then you are the product”. Whether it’s your data (for example on social media), or your time (for example, helping to train an AI by talking with it), you’re paying for what you use and consume on the internet. Then there’s another dimension if you’re paying to consume something that someone else has made on the internet. Looking at you, Spotify. Not forgetting the environmental impact that comes with technological developments.

When it goes wrong

There have been a few high profile cases of AI chatbots being compromised in recent years. The speed at which these chatbots have been trained by the public to spout racist and homophobic ‘opinions’ is alarming. Is this an upsetting reflection of society? Or is it a sign that there are thousands of bored people on 4chan willing to dedicate their energy to the next offensive project?

A South Korean Chatbot Scandal Shows the Threat A.I. Presents to Privacy

You might have heard about Microsoft’s Tay, who was released on the Twitter in order to be trained by public data (tweets), and learned to be a racist in less than 24 hours. This isn’t something that Microsoft will be allowed to forget, even though it happened back in 2016. Microsoft claimed to have ‘sorted and filtered’ the anonymised public data it used in the initial training of Tay, but clearly Tay was too trusting and naïve to be released on Twitter.

In the case of Tay, Microsoft wasn’t trying to hide that she was a bot. But some bots are set up to simulate human conversation. Do you think it should be made clear that you’re talking to a bot? I think that most of the time it’s obvious anyway, but this does raise the question, should companies be trying to pass their bots off as humans? Some say that the Turing test is outdated now, but it’s a good place to start from when thinking about human/robot interactions.

💬 Contribute

Read the following prompt then add your contribution in the box below. Responses from the same person are the same colour. All comments are anonymous.

Most chatbots we interact with are programmed to give responses to questions on specific topics and aren’t trained on data from Twitter. Aside from abuse, what else might be a concern if you are using a chatbot in a customer service situation, for example? Use the discussion box below to talk about any other vulnerabilities a chatbot might have.

If you can’t see the comment box, please write a response to this post instead

Summary

Thanks for taking part in this week’s topic. I hope you’ve had some productive and interesting conversations and enjoyed thinking about human-bot interactions. We’ve covered what chatbots are, how they have been introduced into our lives, as well as some of the drawbacks of companies using public data to train their bots.

Further resources

📺 How to deprogram homicidal sex robots
This is the CEO of Pandorabots, Lauren Kunze, giving a TED talk about AI, bots, gender and tech. It’s completely gender binary, but it’s definitely worth a watch, even though it feels like it was made 100 years ago (2019).

🔊 Reply All episode 178: I am not a bot
A podcast about when bots go bad (30 mins). This one isn’t about chat bots, but it’s fun and interesting.

#178 I Am Not a Bot | Reply All

👓 Singapore’s answer to COVID questions
Built to answer quick questions about non-life-threatening COVID situations.

👓 The Uncanny Valley but for audio?
This research paper from a student at Tilburg University, NL, discusses whether the Uncanny Valley applies also to audio.DigiLab Online (week 5)

^ back to contents

Any questions/comments, email digisoc@manchester.ac.uk.