Impersonal personalities

Why chatbots need some style

It’s a new world for technology users - which means all of us! Gone are the days when our interactions with machines were confined to cold, robotic commands. Now, we can converse with our devices in a more natural, relaxed manner with the aid of chatbots and conversational interfaces. These encounters include voice assistants like Siri, Alexa, and Google Assistant, chatbots integrated into websites, and new Large Language Models such as ChatGPT that have recently set the online world ablaze. What, then, is the secret to introducing a successful chatbot? What makes one chatbot more competent and efficient than another? The answer perhaps lies in its personality.

Illustrated, of course, by the MidJourney Bot

Even if chatbots lack the inner depths of human personality, they do still have a certain personality-like quality if we consider it in terms of the patterns of behavior that make up their unique way of interacting with the world. We may not be able to relate to them on an emotional or intellectual level, but it’s possible to think of them as having a personality nonetheless. After all, don’t we all have a tendency to anthropomorphise, to give almost-human characteristics to animals, objects and even machines? It helps us to make sense of our environment, and it’s through this anthropomorphising that we can create stories, which are important for learning and memory. So although chatbots may not have a true human personality, they do have a personality-like quality that helps us to relate to them and make sense of the world.

In this way, a chatbot with a well-defined personality can make the user experience much more enjoyable, providing a sense of connection with the user. On the other hand, a chatbot without a clear personality can be robotic and impersonal, leading to frustration and a lack of trust in the technology.

Chatbots and Large Language Models

Although the term chatbot is now loosely used for any conversational interface, it is useful to understand that there are significantly different approaches at work behind the scenes.

Traditional chatbots are distinct from human conversations, being designed to answer specific input and holding to pre-established answers or activities. Such bots are generally rule-based, deciphering the user’s input and determining a suitable response. In contrast, human discourse is more open-ended, allowing for a greater range of flexibility and imagination.

ChatGPT, in contrast, is a type of natural language processing (NLP) technology that employs a wealth of text data to generate responses that are more human-like to user input. It is often used to create a conversation that is more natural and open-ended than a regular chatbot. Unlike a chatbot which is programmed to react to certain phrases or keywords, ChatGPT relies on machine learning to be able to understand context and answer accordingly with fluency and a wider range of options. Furthermore, ChatGPT is not intended to be integrated into a particular device or application, whereas the likes of Siri and Alexa are designed to be used through verbal interactions on certain devices.

Is personality more than just personalization?

Many conversational interfaces have been designed with a conversational style and a personality profile to give them a more lifelike and engaging feel. Depending on its implementation, the personality of a conversational interface can range from a basic set of likes and dislikes to a more intricate personality with a unique tone and style. Apple’s Siri is endowed with a helpful and sometimes amusing character. Alexa’s personality is not fixed, as it can be customised to fit the preferences of its user. Alexa’s default personality is friendly and helpful.

Some chatbots are designed to have an empathetic personality. This can be done by programming the chatbot to recognise and respond to user emotions. For instance, the chatbot could detect the tone of the user’s message, and then respond with an appropriate and empathetic response. Additionally, the chatbot could be programmed to ask follow up questions that show understanding and empathy, such as asking how the user is feeling or what their thoughts are about a particular subject.

One example of a chatbot with a successful personality is Siri. With her sassy and witty responses, she has become a beloved feature on Apple devices. This type of personality not only adds a sense of humor to the interaction, but it also humanises the technology and makes it more approachable for users.

But another example - and not so happy - is the chatbot Tay, created by Microsoft. While the initial intention was for Tay to learn and improve through user interaction, the chatbot quickly devolved into a biased and offensive personality due to malicious user input. This highlights the importance of carefully designing and monitoring a chatbot’s personality to avoid potential negative consequences.

Eliza did little

Eliza was one of the earliest chatbot programs, created in 1966 by computer scientist Joseph Weizenbaum. It was designed to simulate conversation with a psychotherapist by using a pattern matching technique that included responses based on keywords. Eliza was capable of carrying on a conversation but was not designed to be particularly intelligent, the program simply used open-ended questions and often repeated back the user’s own words …

User: I’m feeling really anxious.

Eliza: Why do you think you’re feeling really anxious?

User: I don’t know, it just comes over me sometimes.

Eliza: It sounds like you’re not sure why you’re feeling really anxious. Can you tell me more about what’s going on?

You can see how that works straight away, can’t you? Eliza was designed to use a technique called reflective listening, which is a common technique used by therapists. Reflective listening involves repeating back to the speaker what they have said, in order to clarify their thoughts and feelings and help them to better understand themselves.

In the case of Eliza, if the program couldn’t parse (or understand) a user input, it could respond, slightly creepily, with something like this: What do you think your mother would say if she heard you talking like this? This was intended to help the user to reflect on their own thoughts and feelings, and to gain a better understanding of why they were feeling the way they did.

While Eliza was not truly able to provide therapy, of course, she was able to have conversations with users that were relatively natural-sounding for the time. However, some of the ethical problems that arose with Eliza included the potential for users to become overly dependent on the chatbot for emotional support, as well as the potential for users to divulge sensitive personal information to a machine that was not able to truly understand or help them.

Is ChatGPT any better?

As a language model, ChatGPT does not have a personality in the same way that a human or even a chatbot might. Instead, it is trained to generate text that is similar to human writing, but it does not have its own beliefs, desires, or emotions. Nor does it have a pre-programmed style of response It is up to the person using ChatGPT to give it a personality by providing it with specific input and guiding its responses in a particular direction.

One way to give ChatGPT a personality would be to provide it with a specific set of prompts or examples of writing in a particular style, such as a humorous or sarcastic tone. For example, if you wanted ChatGPT to have a humorous personality, you could provide it with a series of jokes or funny statements as input, and then ask it to generate responses in a similar style. This would allow ChatGPT to learn the patterns and characteristics of humorous writing, and it would generate responses that are similar in style. Of course, the specific personality of ChatGPT would depend on the input that it is given, so you would need to provide it with a wide range of examples in order to create a well-rounded and consistent personality.

Nevertheless, trying to have a humorous conversation with ChatGPT can be hard work …

What’s the difference between a hippo and a Zippo?

The difference between a hippo and a Zippo is that a hippo is a large, herbivorous mammal that is native to sub-Saharan Africa, while a Zippo is a brand of cigarette lighter. Hippos are some of the largest and heaviest land animals in the world, with adults weighing up to 5,000 pounds (2,268 kilograms) or more. In contrast, Zippo lighters are small, portable devices that use a wick and flint to produce a flame for lighting cigarettes. While hippos are known for their size and weight, Zippo lighters are known for their durability and reliability. The joke you mentioned about the difference between a hippo and a Zippo is based on the fact that hippos are heavy and Zippo lighters are relatively light, but the two are otherwise unrelated.

Notice that even this lengthy and leaden reply does not recognise that the joke is really based on the pun lighter and is enhanced by the rhyme between hippo and Zippo. ChatGPT analyses it, but doesn’t get it.

And yes, there are ethical concerns with ChatGPT. Since it is a form of artificial intelligence, it can learn from its interactions with humans, which means that it could potentially be used to manipulate people by providing false information or biased opinions. Additionally, because ChatGPT is a form of automated communication, it could be used to spam people with unwanted messages. There is also the potential for misuse of data collected through the interface, as it could be used to track user behavior and target advertisements. And of course there is also the potential for ChatGPT to create biased or discriminatory systems, whether accidentally or by manipulation, that could lead to unfair outcomes.

As you can see, a well-defined and appropriate personality can enhance the interaction and build trust with the user, while a lack of personality or a negative personality can hinder the technology’s effectiveness. As the use of chatbots continues to grow, it is important for developers to consider the personality of their chatbot in order to create a positive and successful user experience.