Beyond Chatbots: Personality

Imbuing Digital Companions with Character

Aug 22, 2024

So far in the Beyond Chatbots series, I’ve explored how personalisation and integrations can transform our current generation of generic chatbots into more tailored and capable digital companions. I’ve also taken a look at the potential of proactive digital companions that prompt users and anticipate their needs. In this post I’m turning our attention to a crucial element that ties all of these aspects together: personality.

Personality is a big ingredient in turning a useful tool into a trusted companion. It's what makes interactions feel natural, engaging, and uniquely tailored to each user. As we develop sophisticated features for digital companions, the crafting of rich, adaptive personalities will play an essential role in shaping our relationship with them.

😜 Personality Traits in Digital Companions

Why does personality matter so much in the context of AI? The answer lies in two critical factors: trust and enjoyment. As we develop more sophisticated digital companions, their personality will play an incredibly important role in building user trust and creating joyful, engaging experiences.

Most current chatbots offer generic interactions, struggle to build rapport, and don’t keep users engaged over time. This leads to intermittent, shallow, and transactional interactions with users. In contrast, if we are able to craft well-defined personalities for digital companions they can offer more natural interactions, improved user engagement and more tailored experiences.

This will all go a long way to build more trust with our digital companions and change our relationships with technology going forwards. A digital companion with the right personality could become more than just a tool – it could be a coach, a confidant, or a collaborator that truly understands and complements its user.

One platform that's pushing the boundaries of AI personality customisation is Character.ai. The platform allows users to create and interact with chatbots representing various characters and has been very popular amongst the under 25s, so much so that, according to SimilarWeb, users of Character.ai visit the platform 65% more often than users of ChatGPT use ChatGPT!

Let's take a quick look at Character.ai’s approach. On the platform, users can define AI characters using free-form text descriptions. This includes details about the character's background, personality traits, speaking style, and knowledge areas. This works well because users have the freedom to create highly detailed and unique characters and the open-ended nature of text descriptions allows for complex character development. However, some users may find it challenging to know where to start or what details to include without structured guidance. While Character.ai's approach offers a glimpse into the future of AI personality customisation, I think the future lies in more intuitive, granular control over digital companions' traits and behaviours.

😍 Customising Personality Traits

To create compelling personalities for digital companions, there are key features that we will to need to develop and allow users to control. These features should work together to create fully-formed characters that users can relate to and meaningfully interact with.

To determine what these features should be, we can take inspiration from Psychology and the HEXACO model of personality structure. I prefer this to the traditional OCEAN model of big five personality traits as the HEXACO research is more inclusive and represents non-English speaking cultures. The HEXACO model conceptualises personality in terms of six traits:

Honest & Humility (H)
Emotionality (E)
Extraversion (X)
Agreeableness (A)
Conscientiousness (C)
Openness to Experience (O)

In the HEXACO model, each of these traits has two dimensional scale, which can be used to allow users to dial each dimension up and down to customise the personality of their digital companion. Each dimension also has associated facets and adjectives that can be used to translate these dimensions into natural language to be used in a system prompt to influence digital companions’ responses. This approach is rudimentary but can work effectively - we have already seen chatbots that are very capable of imitating different characters and personalities just based on a natural language description of that character.

To illustrate how these personality traits could be implemented in practice, below is an example interface that allows users to customise their digital companion's personality based on the HEXACO model:

These settings allow users to adjust the six personality traits on a 5-point scale. For example, the "Integrity" slider corresponds to the Honesty-Humility trait in the HEXACO model. Users can move each slider to shape their digital companion's personality, with the extremes of each slider representing opposing characteristics.

I asked Anthropic’s Claude to mock up this example interface and you can play around with it here. As you adjust the sliders, the interface generates a system prompt that can be used to guide the digital companion's responses and behaviour. For example:

A high setting on Honesty-Humility might result in a digital companion that's more direct and less likely to flatter the user.
A low setting on Extraversion could create a more reserved companion that doesn't initiate conversations as frequently.
A high setting on Openness to Experience might lead to a companion that offers more creative solutions and engages in abstract discussions.

I’ve tested how well this works and it’s pretty good - the different system prompts significantly change the responses you get from a chatbot. For example, I created two different personalities - an “Empathetic Supporter” (highly empathetic, agreeable, and honest, prioritising emotional support and understanding over practical problem-solving) and a “Practical Achiever” (focusing on practical solutions and achievement, less emotionally involved but highly organised and goal-oriented) and gave them both the following prompt:

“I’ve just arrived home after a long, stressful day at work where I made a mistake on an important project. I'm feeling down and considering ordering a take away for dinner instead of cooking the healthy meal I had planned. How should I respond to this situation?”

In their responses, the “Empathetic Supporter” prioritised emotional validation and support, while the “Practical Achiever” focused on problem-solving and forward-thinking strategies.

While this implementation is pretty basic, it shows a solid foundation for how we could develop digital companions with customisable personalities. A key challenge will be ensuring that these personality traits are consistently applied across all interactions whilst also being able to adapt to context and the requirements of specific interactions. This will require some new research and new techniques for how we build more sophisticated digital companions, but is something I’m confident can be delivered.

🤓 Shaping a Digital Companion’s Personality

Beyond personality traits, there are additional features that can contribute to a more nuanced and adaptable digital companion. The way they behave and communicate is crucial in shaping how users relate to them:

Adaptability: It’s important that digital companions are able to maintain a core personality while being flexible enough to adjust its communication style based on the context of the conversation, such as switching from casual to formal in professional settings.
User feedback: As users interact with their digital companions over time, their feedback – both explicit and implicit – should influence how the AI expresses its personality, such as increasing uses of humour if the user consistently responds well to it.
Proactiveness: Some users might prefer a companion that takes initiative, while others might want one that waits for explicit instructions.
Vocabulary: The choice of words and the complexity of language used can significantly impact how a digital companion is perceived. A digital companion should accurately reflect the language that is used by its user.
Humour: This is a powerful tool for building relationships. Digital companions that have the ability to understand and use humour will make interactions far more engaging and human-like, helping to build trust.
Emotional Intelligence: The ability to recognise and respond appropriately to user emotions is an important part of building user relationships and providing support.
Emotional Expression: Recent advances in GenAI voice technology, as seen in the demos of GPT-4o’s upcoming advanced voice capabilities, will allow for a digital companion to express more nuanced emotions in voice interactions that just can’t be done in text based interactions. Adapting emotional expression based on the situation and the user's emotional state will be important for creating natural, more empathetic interactions.

I believe that we will soon be able to build digital companions whose core personality traits can combine with all of these additional character elements to create more human-like, engaging, and trustworthy user experiences.

🥸 Emotions & Appearance

To truly deliver human-like interactions, digital companions must evolve beyond just text-based chat interfaces. The future of AI companionship lies in the integration of emotional intelligence and visual representation, creating a multi-modal interaction experience that feels genuinely personal and engaging.

Recent advancements with GPT-4o’s advanced voice mode have demonstrated impressive capabilities in generating emotionally nuanced speech. This opens up new possibilities for digital companions to convey a wide range of emotions through voice, adding depth to AI interactions.

Currently, GPT-4o users can choose from a limited selection of voices. However, imagine being able to describe any voice you'd like your digital companion to have - perhaps a soothing baritone with a hint of a Scottish accent, or a cheerful soprano with a California surfer vibe. This level of customisation would allow users to create truly unique and personally appealing digital companions.

But expressing emotions is only half of the equation. Equally important is the ability of digital companions to recognise and respond appropriately to the emotions of their users. This emotional intelligence is key to creating meaningful, empathetic interactions. For example:

A digital companion might detect frustration in a user's voice and respond with patience and encouragement.
It could pick up on excitement and match that energy in its responses.
During periods of stress, it might adopt a calming tone and offer supportive words.

This kind of emotional attunement could significantly enhance user engagement and satisfaction. When users feel understood on an emotional level, they're more likely to form a connection with their digital companion, leading to more frequent and meaningful interactions.

A large part of emotional expression in humans comes from gestures and non-verbal communication. This is why digital companions will likely need a "visual mode" for certain contexts and use cases. Advances in generative AI’s abilities to create images and videos will lead to digital companions having a visual digital presence, allowing users to interact with them as if they're on a video call.

This visual aspect opens up new opportunities for customisation and human-like interactions:

Visual Appearances: Users could customise the visual appearance of their digital companion for video interactions, ranging from selecting preset avatars to designing completely unique looks.
Gestural and Non-Verbal Communication: Future digital companions might use facial expressions and body language in video calls, adding another layer of personality and emotional expression.
Multi-Modal Interaction: Digital companions could adapt their appearance and personality expression across different interaction modes - text, voice, and video - while maintaining a consistent core personality.

As AI systems become more advanced and integrated into our lives, building and maintaining user trust will be paramount. Personality plays a crucial role in this trust-building process:

Consistency and Predictability: A well-defined AI personality creates a sense of familiarity and predictability, helping users feel more comfortable in their interactions.
Transparency and Honesty: AI personalities should be designed to be upfront about their capabilities and limitations, fostering trust through honesty.
Emotional Resonance: AI that can express and respond to emotions appropriately can create a sense of understanding and connection with users.
Adaptability and Learning: AI personalities that can learn and adapt to user preferences demonstrate responsiveness and a commitment to meeting the user's needs.
Ethical Behaviour: AI personalities should be designed with strong ethical principles, consistently demonstrating integrity in their interactions.

🚧 Challenges and Considerations

While the potential benefits of personality-driven digital companions are significant, their development and implementation come with a range of complex challenges and important ethical considerations. As we push the boundaries of personalities in digital companions, we must navigate these issues carefully to ensure that they enhance our lives without introducing new problems.

Ethical Implications and Bias: One of the primary challenges in developing personality-driven digital companions is ensuring that they don't perpetuate harmful stereotypes or biases. To address this we need diverse development teams and rigorous testing protocols.
Emotional Intelligence and Manipulation: As we develop more emotionally intelligent digital companions, we open up new possibilities for meaningful interactions – but also for potential manipulation. There is potential for misuse and to exploit users or manipulate their emotions. We need to establish clear ethical guidelines for AI emotional expression and implement transparency measures.
Privacy and Data Protection: Personality-driven AI requires processing large amounts of personal data to provide personalised experiences. This raises significant privacy concerns. How do we balance the need for data to improve AI performance with user privacy rights?
Managing User Expectations and Relationships: As digital companions become more sophisticated and personable, there's a risk that users may develop unrealistic expectations or form inappropriate attachments. We will need to develop guidelines for healthy human-AI interactions.
Cross-Cultural Considerations: Developing digital companions with cultural awareness and adaptability is key. This involves not just language translation, but understanding and respecting cultural norms, taboos, and communication styles. Involving diverse global teams in the creation and testing of AI personalities will be essential to achieve this cultural fluency.
Balancing User Desires with Ethical AI Behaviour: Users may want to shape their digital companions in ways that conflict with ethical AI principles. To navigate this, we need to establish immutable ethical guidelines for AI behaviour that cannot be overridden by user preferences. This ethical framework should be transparently communicated to users, helping them understand why certain customisations might not be possible.

As we develop personality-driven digital companions, addressing these challenges will be crucial. By doing so, we can create digital companions that are not only more engaging and personable, but also trustworthy, ethical, and truly beneficial to users' lives. This will require a multi-disciplinary approach with ongoing dialogue between AI developers, ethicists, policymakers, and users to ensure we're creating a future with AI that enhances human potential while respecting human values.

🏁 Conclusion

As we've explored throughout this post, the development of sophisticated, customisable personalities for digital companions could deliver a significant leap forward in our journey beyond simple chatbots.

The development of sophisticated, customisable personalities for digital companions represents a significant advancement in our journey beyond simple chatbots. This aspect of AI development doesn't exist in isolation – it's deeply interconnected with the other elements we've explored in this series. Personality enhances personalisation by allowing users to customise a digital companions’ communication style. It complements integrations by enabling digital companions to present information from various sources in a manner consistent with its user’s preferences. And it elevates proactivity by allowing digital companions to initiate interactions in a way that feels natural and appropriate to the user. In essence, personality is the thread that weaves together all these elements, creating a cohesive and engaging user experience that will transform how we interact with technology.

However, with this great potential comes significant responsibility. As we develop these technologies, we must navigate complex ethical considerations, prioritise user privacy and wellbeing, and work diligently to build and maintain user trust. The challenges are substantial, but so too are the potential benefits.

What role do you envision AI personalities playing in your life? How do you think these digital companions might evolve in the coming years? As we continue to push the boundaries of what's possible with generative AI, these are questions we'll all need to consider.

In my next post, I’ll explore another crucial aspect of creating trustworthy and effective digital companions: fact-checking and reliability. How can we ensure that our digital assistants provide accurate, verifiable information? Stay tuned to find out.

“The future is already here, it’s just not evenly distributed.“
William Gibson