Beyond Chatbots: Proactivity

Enabling Digital Companions to Take the Initiative

Aug 08, 2024

In the Beyond Chatbots series so far, I’ve explored how personalisation and integrations can start to transform our current generation of generic chatbots into digital companions that truly understand and adapt to our needs. Next we’re looking at proactivity, which will turn our interactions with digital companions on its head.

In the context of digital companions, proactivity refers to the ability of AI to initiate interactions, anticipate user needs, and take action without explicit user prompts. It's about moving from a reactive model where the AI waits for commands, to a proactive one where it becomes a true companion, offering help before it's asked for.

Imagine opening your digital companion app and being greeted with a simple message: "Good morning! It's 7:30 AM. Would you like to see your schedule for today?" This small but significant shift — your AI initiating the conversation — is the first small step towards realising the potential of proactive digital companions.

As digital companions evolve over time, their proactive capabilities will become increasingly sophisticated. In the near future, you might wake up to a more useful prompt: "Good morning! I see you have an important presentation at 2PM. Would you like me to summarise the key points from your preparation notes?" Further down the line, your digital companion might even say, "Good morning! I've noticed you have an important presentation today on the product development roadmap for digital companions. I've prepared a summary of key points and some relevant competitive information. Would you like to review it over breakfast?"

This progression from simple, timely greetings to complex, anticipatory assistance isn't science fiction — it's where today’s chatbots will evolve to, and I don’t think that future is far off.

💬 Reactive Chatbots

Today's chatbots are impressive in many ways but fundamentally only ever react to the user. They wait for us to initiate every interaction, responding to our queries but never taking the initiative. This reactive approach restricts the potential of today’s chatbots in a few important ways:

The Blank Slate Problem: Users are typically met with a blank chat interface, offering very few indications of how the chatbot could help them or what it's capable of. This can be intimidating and limits the user's ability to fully understand a chatbot's capabilities.
Product-Market Fit Issues: The reactive nature of our current chatbots contributes to a significant problem with product-market fit. While hundreds of millions of people have tried today’s chatbots, most don't return regularly. This is something that proactive engagement could address.
Limited Learning and Adaptation: With fewer interactions, reactive chatbots have limited opportunities to learn from user behaviour and adapt their responses over time.
Lack of Trust Building: Purely reactive systems struggle to build a sense of rapport and trust between the user and the AI, which proactive interactions could encourage.
Limited Ability to Surprise and Delight: Proactive features can occasionally provide unexpected but valuable assistance, creating moments of delight that purely reactive systems could never match.
Missed Opportunities for Timely Assistance and Reminders: Without the ability to initiate interactions, chatbots can't offer help at the most opportune moments, or remind users of important tasks, events, or follow-ups without being explicitly asked.

A key issue underlying many of these limitations is the lack of a proper onboarding process. An onboarding experience could serve as the first proactive interaction between a digital companion and a user. The information gathered during onboarding, combined with contextual information (as discussed in our previous post on integrations), would create numerous opportunities for a digital companion to be more proactive in the future.

By addressing these limitations and moving towards a more proactive digital companion model, we can transform chatbots so that they not only respond to our needs but anticipate them, providing timely, relevant, and personalised assistance. This shift has the potential to significantly enhance user engagement, satisfaction, and the overall value proposition of our current chatbots.

🔀 Redefining the Interaction Model

The first step towards developing proactive digital companions requires a fundamental shift in how they interact with us. Instead of always waiting for user prompts, digital companions should be able to initiate conversations when appropriate. Proactivity also goes beyond merely initiating conversations. It encompasses a variety of behaviours, from simple prompts to autonomous actions. At its core, proactivity is about anticipating needs and offering assistance before it's explicitly requested.

This shift in interaction model could significantly change user expectations and experiences. Users might come to rely on their digital companions for proactive reminders, suggestions, and assistance, much like they would a human personal assistant. This could lead to increased productivity and reduced cognitive load, as users offload certain mental tasks to their AI companions. However, it also raises questions about dependency and the balance between helpful assistance and potential intrusiveness.

The key to effective proactivity lies in understanding context and user preferences. A digital companion should know when to offer help and when to stay quiet, adapting its proactivity to each user's personal needs and comfort levels.

Similar to how I suggested we introduce integrations to digital companions in my previous post, introducing proactivity should also be a gradual process. This approach allows users to become comfortable with increasing levels of their digital companion taking initiative. Below I outline a five-step process for implementing proactivity:

Step 1: Basic Greetings

Initially, a digital companion should proactively initiate a chat with a simple greeting. This small step begins to shift the dynamic from purely reactive to slightly proactive, establishing a friendly presence and gently introducing users to the idea of companion-initiated interactions. It sets the stage for more complex proactive behaviours without being overwhelming. For example, when you open the app, your digital companion might greet you with a simple:

"Good morning, [Name]! I hope you're having a great day."

This basic interaction might only be for the first ten interactions with a user, but starts building a rapport and prepares users for more advanced proactive features. This would be very simple to implement by coding a digital companion to speak first whenever a new chat is initiated by the user.

Step 2: Generic Offers of Assistance

After a short while, digital companions should start offering general assistance. This introduces the idea that the companion is ready and willing to help, but still leaves the specifics up to the user. It encourages users to engage more with the digital companion and helps them discover its capabilities while maintaining a low-pressure interaction. After the greeting, the digital companion might add:

"Is there anything I can help you with today?"

This open-ended question invites users to explore the digital companion’s abilities at their own pace, gradually increasing their comfort with its proactive nature. This would be similarly straightforward to implement by making the digital companion’s initial prompt more sophisticated.

Step 3: Contextual Suggestions

In the next stage, the digital companion should start to use available contextual information such as time, device, location, and calendar events to offer more specific assistance. This approach provides more relevant and timely help, demonstrating the companion’s ability to understand context and increasing its perceived value. For instance, your digital companion might say:

"Good morning! I see you have a busy day ahead with three meetings scheduled. Would you like me to summarise your day's agenda?"

By offering contextual assistance, the digital companion shows it can anticipate needs based on readily available information, making it a more useful companion in the user's daily life. Implementing this level of proactivity will require the personalisation and integrations that I’ve discussed in previous posts and developing ranking algorithms to determine the most important suggestions in a given context.

Step 4: Predictive Assistance

As digital companions become more sophisticated, they could use patterns from past user behaviour and current data to anticipate needs and offer proactive solutions. This stage offers much more highly personalised assistance to the user, helping them manage complex tasks and schedules, and demonstrates an advanced understanding of user needs and preferences. For example:

"I've noticed that you usually start preparing for quarterly reports about two weeks in advance. Your Q2 report is due in 18 days, but your calendar is quite full next week. Would you like me to block out some time this week for report preparation and gather the key financial data you typically include?"

This level of proactivity would show the digital companion's ability to learn from user behaviour over time and offer increasingly valuable assistance. This level of proactivity will require an improvement in the planning and reasoning capabilities of large language models, so is probably at least 12 months away from being practical and reliable.

Step 5: Autonomous Actions

When digital companions have much more sophisticated capabilities, they should be able to take predefined actions on the user's behalf, always with clear consent and easy override options. This approach saves time, reduces cognitive load for users, and handles routine tasks automatically, representing the highest level of proactive assistance. An example might be:

"I've noticed a recurring conflict in your schedule. With your permission, I can automatically reschedule the lower-priority meeting to resolve this. Would you like me to do that?"

This demonstrates the full potential of a proactive digital companion, taking initiative to solve problems while still ensuring the user remains in control. There is obviously a lot of product development to do before we get here, but these are the types of interactions we should be aiming for with our digital companions in the future.

This gradual approach to introducing proactivity will allow users to acclimate to more proactive behaviours from their digital companions, building trust and comfort over time. It's important to note that users should always have control over the level of proactivity they're comfortable with, and be able to adjust these settings as their relationship with their digital companion evolves. By implementing proactivity in this gradual manner, we can transform chatbots into true digital companions that not only respond to our needs but in the future anticipates them, providing timely, relevant, and personalised assistance.

The evolution from reactive chatbots to proactive digital companions isn't just a technological leap - it will fundamentally shift how we interact with technology.

🚧 Challenges and Considerations

While the potential of proactive digital companions is exciting, it's crucial to address several challenges:

Privacy Concerns: Proactivity requires access to and analysis of user data. It's essential to implement robust privacy protections and give users granular control over what data is used for proactive features. For example, a proactive digital companion might need access to a user's email, calendar, and location data to provide timely and relevant assistance. This data shouldn’t leave a user’s device and only abstracted, non-personal data should be saved as memories for future use.
Avoiding Overreach: There's a fine line between helpful and intrusive. Digital companions need to learn and respect individual user preferences for different levels of proactivity. Users should be able to select how proactive they would like their digital companion to be during onboarding and to be able to adjust these settings whenever they please.
Handling Errors and Misinterpretations: Proactive suggestions based on misinterpreted data could be frustrating or even harmful. Implementing graceful error handling and continuously learning from user feedback is crucial. A potential solution could be to ask for clarification before acting on potentially sensitive interpretations.
Ethical Considerations: We must ensure that proactive features don't reinforce biases or manipulate users. Maintaining user autonomy in decision-making should be a priority. This is an area that needs to build on current efforts to reduce bias in models, expanded out to consider proactive use cases.

Addressing these challenges will be key to creating proactive digital companions that enhance our lives without introducing new frustrations or concerns. As the technology continues to advance, we can expect proactive digital companions to become increasingly sophisticated and helpful. One area I’m particularly excited about is emotional intelligence, allowing digital companions to better understand and respond to user moods and emotional states.

🏁 Conclusion

The shift from reactive chatbots to proactive digital companions represents a big change in how we interact with technology. By initiating interactions, anticipating needs, and offering timely assistance, digital companions have the potential to significantly enhance our productivity, well-being, and quality of life.

However, as with personalisation and integrations, it's essential that we develop these features thoughtfully, with a focus on user empowerment, privacy, and ethical considerations. The staged approach to implementing proactivity that I’ve outlined allows us to build trust and comfort gradually, ensuring that digital companions enhance our lives without overwhelming or intruding.

In my next post, I’ll explore another crucial aspect of digital companions: Personality. How can we craft AI assistants with distinct, customisable character traits that build rapport with their users? Stay tuned to find out.

What are your thoughts on proactive digital companions? How would you feel about a digital companion that initiates conversations with you? What kinds of prompts would you find most helpful? Share your thoughts in the comments below.

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