PUBLIC ANNOUNCEMENTS
Rethinking Conversational AI: Meet Digital Personas
The Promising Future of Human-Machine Interaction
Conversational AI is at a turning point. Human-machine interaction is not just about simple words anymore, it’s becoming much more complex. Companies are starting to combine different modalities and technological solutions to make the interactive experience more natural and immersive. Where’s all this headed? Will we soon move on from “stupid chatbots” to truly humanlike digital personas? What will they be like?
In recent decades, human-machine interaction has been evolving a lot. People have often had no choice but to learn how to interact with the machine; now it’s the machines that are, slowly but surely, learning how to interact with us, so that we can work with technological tools “without thinking about it too much”.
Using natural language to interact with machines is something extraordinary. But while it’s obvious that development in natural language processing (NLP) is crucial to reaching humanlike conversational AI (and NLP algorithms are indeed getting better and better each year), we must also acknowledge that truly natural interaction requires much more than just the words: the non-verbal, visual, or acoustic dimensions, among other things. Thus, the future of conversational AI must be based on a holistic approach — take into account multiple aspects of natural interaction. Such advanced interactive systems have been referred to as “digital personas”. So forget texting with a chatbot or talking to a voice bot, and prepare to interact with a persona!
Let’s have a look at what the first generation of digital personas should be able to do, how they should combine different communicational and technological skills.
Complex conversational skills
As mentioned above, purely conversational skills will of course remain the fundamental feature of all language-based interactive systems. This includes semantic analysis of words and sentences, the understanding of textual context, but also more autonomous dialogue management, which is related to detecting conversational patterns or handling intentions — those of the persona (managing the communicational goals it is pursuing) as well as those of the human interlocutor (estimating the user’s goals from what they are saying). Some conversational skills are subtle but equally important for the naturalness of the dialogue: choosing slightly different formulations in different contexts, detecting nuances in the meaning of synonyms, etc.
Wide scope of domains and topics
Digital personas won’t be restricted to a single domain. Using one chatbot to administrate your bank account, another to purchase flight tickets, and yet another for small talk and open conversation will no longer be necessary — you might have just one digital persona who knows all these things. The first step in this direction is probably the integration of small-talk skills into domain-specific goal-oriented conversational systems to make them more “friendly” and informal. The breadth of content is also the domain where generative models will gradually gain strength, probably combined with manual design into hybrid conversational AI systems.
Voice input & output
The growing importance of VUI (voice user interface) is undeniable; the voice-first approach is becoming the industry standard. Just like texting with a friend doesn’t feel quite the same as talking with them, talking with AI (both the user and the persona speaking aloud) can enhance the ease, naturalness, and perceived “cordiality” of human-machine interaction. Of course, VUIs have disadvantages, too: when people speak off-the-cuff, they often produce incoherent, interminable, or agrammatical sentences, they make pauses (which makes it hard to detect the end of the utterance)… That’s why for certain use cases, digital personas might continue to offer the written input option. However, according to current trends, the primary input and output signal will definitely be acoustic.
The body of a digital human
Apart from the digital persona’s abilities (the “mind”), its form (the “body”) will also undergo major changes and improve user experience significantly. While real-world physical hardware is in full development but still far from being humanlike, on-screen representations have probably surpassed the uncanny valley already: recently, the so-called digital humans — surprisingly realistic 3D avatars— have been gaining ground and they are starting to be combined with different conversational systems. Symmetrically to the previous paragraph: Just like talking with a friend over the phone doesn’t feel quite the same as talking with them in person, interacting with a 3D avatar takes user experience a big step further compared to just listening to the voice of the AI. These digital bodies and faces, albeit only virtual, really make a difference, especially because they enable the persona to make use of non-verbal elements of communication, like gestures and facial expressions.
Character & information profile
Two people could have the same abilities and knowledge, but interaction with each of them would still be different because of their different personalities, a.k.a. characters. That’s what digital personas will need to reflect if their goal is to be as humanlike as possible. Plus, when developing your own personas, you have the opportunity to promote certain morals and qualities that matter to you, e.g., by developing “AI that cares”. Apart from the character, another important feature is a coherent information profile: each persona should have a specific set of factual information about its identity, preferences, etc. (be it randomly generated or manually scripted, but always consistent). This, too, contributes to the humanlikeness of AI systems.
Advanced personalization
Advanced personalization doesn’t mean just remembering and reproducing extensive factual information about the user. It means adapting the flow and the overall style to fulfill communicational goals efficiently and appropriately. Achieving these abilities will be a high wall to climb, but definitely worth it since inflexibility and insufficient personalization count among the blemishes of current interactive AI systems.
Diverse sources of data
There may be limitations to technology in general, but there are almost none when it comes to combining existing technological solutions. Interactive AI systems are usually connected to services like ASR (automatic speech recognition), TTS (text-to-speech), or online information databases. But they can use almost anything as a source of data — location tracking to know where you are; eye tracking to know what you are looking at; facial/voice recognition to know who’s talking; facial expression analysis to guess how you are feeling… you name it. Naturally, connecting a persona to a new data source is one thing, and teaching it how to interpret the data and reflect it in its behavior is another. That will be the real challenge.
So, what do you think about the vision of digital personas? When will we achieve such high-quality AI systems — or do they already exist? Would you add some more features to my list? And… is it even desirable to have such advanced interactive AI? Where are the ethical boundaries? Let me know in the comments. (Seriously, I want to know what you think!)
And to conclude: it’s not a coincidence that you’re reading this on the PromethistAI blog. Although digital personas with all these amazing skills are mostly a vision, step by step, they are becoming a reality thanks to platforms like ours. Our multidisciplinary team is working hard to deliver all the necessary technological tools to create advanced digital personas; one already existing example is Poppy. Visit PromethistAI to learn more or the Promethist Platform to start designing right away!
