Five Emerging Conversational AI Trends
A recent Gartner report highlighted intent driven design and development as key elements to delivering a successful chatbot, which is aligned to customer expectations. Another trend identified by Gartner is bootstrapping.
Introduction
Considering reports from Deloitte and Gartner, there are five emerging Conversational AI trends:
- Intent Driven Design & Development
- Bootstrapping
- Fragmentation
- Voice
- Large Langauge Models
Intent Driven Design & Development
One can argue that there are four general pillars constituting chatbot architecture, and to some degree voicebots. Although voicebots use the same core functionality, there are obviously more moving parts to voicebots. There have been efforts in the past to deprecate some these pillars.
The market has seen four emerging approaches to intent deprecation, but it has not been widely adopted.
However, on the contrary, the trend amongst the Gartner leaders are to merge or rather blur the lines between these pillars. Examples of this merging are:
- Assigning only certain portions of the flow to certain intents or sections of the NLU.
- Adding to an intent a confidence threshold, with low confidence scores triggering a confirmation prompt. All contained within the intent.
- Quick reply intents have also seen the light.
- The emergence of marketplaces…
- Adding structure to intents, like sub and nested intents.
- and much more…
Hence we are seeing an evolution of intents with functionality being added to intents.
According to Gartner, there is now the emergence of intent driven design and development. Too many organisations are starting their chatbot journey with attempting to select the right chatbot technology. And creating vendor comparison matrixes…
Gartner explains that an approach of intent driven design and development should be followed. I’ll refer to it as IDD henceforth…the starting point should be collecting user intents from existing customer conversations. These conversations can be customer agent conversations, email or live agent chat interactions and more.
There are various tools to cluster semantically similar utterances. By implication these clusters are intents. Closely related intents can constitute sub or nested intents.
Bootstrapping
The subject of bootstrapping has been raised quite often in recent times. The idea of bootstrapping is to fast-track the development of a chatbot. And especially the initial stages of the chatbot development process.
Bootstrapping does pose a few challenges…the first being the extent to which such an implementation is effective in addressing the complete ambit of customer needs. The degree to which fine-tuning is possible. And lastly, the fact that scaling the chatbot implementation in terms of functionality is often problematic.
Current avenues for bootstrapping a chatbot are:
- Bootstrapping a chatbot with search & semantic search
- Bootstrapping a chatbot with a Large Language Model
- Quick Reply Intents
- Knowledge Documents
- QnA
- Knowledge Intensive NLP (KI-NLP)
- Search and knowledge API’s
One criteria emphasised in the Gartner® Critical Capabilities for Enterprise Conversational AI Platforms Assessment is pre-sets. The Gartner report focuses on alternatives to bootstrapping; like predefined intents, entities and dialogue flows which are offered by the platform, suitable for specific use-cases and industries.
Fragmentation
I have alluded to this idea in my last few posts, but here are some arguments on platform fragmentation:
- A group of leading conversational AI development frameworks has been identified, these frameworks are self-contained and have extensive functionality.
- Emerging niche technologies are addressing vulnerabilities and blind-spots within the Conversational AI arena.
- This begs the question? Is this a zero-sum game where the Conversational AI Development Frameworks are in an arms race to develop in-house new features and functionality? With already high table stakes…
- Or is there an opportunity to become an app-store for Conversational AI development, a type of an aggregator, a market place?
- Also, is there not an opportunity for frameworks to fragment their framework and make it easy for customers to cherry-pick the functionality they require for a specific purpose?
- Is a framework like OneReach AI focussing on becoming a single Conversational AI portal to act as an orchestration engine / aggregator for Conversational AI experiences? Becoming the Twilio of Language Technology?
- And is this the direction in which Cognigy is moving with the Cognigy AI Marketplace?
Voice
There is a renewed focus on voicebots which are not associated with a dedicated device like Alexa, Google Home, etc. And there is a need from organisations to have their customers access voicebots via a phone call they place to a number.
In terms of customer care, incoming calls to a call centre is the preferred medium of communication of most customers, and the most expensive to field by companies. Automation has succeeded in other areas of Conversational AI, but voice calls are still lagging…for now.
Read more here.
Large Langauge Models
There is an emergence of Large Language Models (LLM), and these models address different conversational disciplines.
The conversational disciplines addressed by these LLM’s can be segmented into the five groups below.
- Embeddings: the clustering of utterances and sentences are analogous to intent detection, but in an unsupervised and automated fashion. An example of this, is the POC that HumanFirst & Cohere performed.
- Dialog Management with GODEL and Blender Bot are exploring avenues in determining the most probable next dialog turn. (Technology: GODEL (Microsoft), DialoGPT (Microsoft), Blender Bot (Meta AI)
- Generation not only generates bot responses, but maintains bot dialog, contextual awareness and session context. Technologies include BLOOM, Goose AI, EleutherAI, OpenAI, Cohere, AI21Labs.
- Question & Answer is being addressed by models like KI-NLP (Knowledge Intensive NLP). Where broad domain and general questions can be answered without querying an API or leveraging a traditional knowledge base. Technologies include Sphere (Meta AI), Commercial Search Engines, Wikipedia, etc.
- Language Translation is available on various platforms, with Meta AI’s NLLB.
Read more here.
Conclusion
Obviously there is no single solution to developing a Conversational AI solution. And a straight-up comparison of chatbot development frameworks will become harder as the landscape becomes more sophisticated and technologies develop.
