.webp)
For Conversational User Interfaces to improve, the reliance on social intelligence will need to be leveraged. Many now regards social intelligence is a prerequisite for human-like Artificial Intelligence.
Introduction
We have all heard people say, be nice to everyone even if they are rude to you because you don’t know what that person has been through today.
This speaks to us as humans that we should be having a sense of social intelligence.
Social intelligence refers to the ability to navigate and understand social situations effectively. It involves being aware of the emotions of others, as well as being able to interpret social cues, communicate effectively, and maintain positive relationships.
Socially intelligent individuals are attuned to social norms, cultural differences, and situational contexts.
They can read social cues and understand the unspoken rules of social interaction. This allows them to navigate social situations with ease and adapt their behaviour accordingly.
But how will a social intelligence system for AI look? This study looks at the taxonomy of such a system.
Multimodality
There has been keen interest in the introduction of multimodal features to Large Models, with image and video generation and detection. However, I see a huge significance in making use of multimodal features within a conversational UI, allowing for a more nuanced conversational experience.
Understand
We have long focussed on social intelligence by only considering the cognitive aspect of it, and our systems’ ability to understand what users’ intent is. And understanding the user, based on their input and any related customer data the system can glean from back-end lookups.
Behavioural
As conversational UI designers, we are often also fixated on the behavioural part of the bot; hence designing bots which behave in a certain way.
Both understanding and behaviour are important to building to good conversational UI, but not sufficient.
LLMs perform worse than best human performance on most tasks on cognitive and situational intelligence.
~ Source
Social intelligence data infrastructure
What this study attempts, is creating a taxonomy of what a Social Intelligence system would look like.
A seen in the image below, the study on social intelligence has identified three distinct types of social intelligence:
- Cognitive Intelligence
- Situational Intelligence
- Behavioural Intelligence
Cognitive intelligence
Under cognitive intelligence traditionally chatbots have covered items like intent and emotion quite well. Advances have been made on the emotional front, ensuring that Conversational UIs shows some level of empathy.
As most of our chatbots have been task orientated, much focus has been placed on intent recognition.
Future datasets should focus more on specific, nuanced, and long-tailed social situations. ~ Source
Situational intelligence
This aspect considers the social context, and social context informs cognitive and behavioural intelligence.
Cultural differences between high-context and low-context cultures refer to variations in how communication is conducted and interpreted within different societies.
These differences can have significant implications for interpersonal relationships, business transactions, and overall social interactions.
High-Context Cultures
In high-context cultures, much of the meaning in communication is conveyed through implicit cues, such as non-verbal gestures, tone of voice, and contextual clues.
People in high-context cultures rely heavily on shared cultural knowledge and understandings to interpret messages. Examples of high-context cultures include Japan, China, and many Middle Eastern countries.
Low-Context Cultures
In contrast, low-context cultures rely more on explicit verbal communication to convey meaning.
Messages are typically straightforward and direct, with less reliance on context or shared cultural knowledge. Examples of low-context cultures include the United States, Germany, and some Northern European countries.
One can argue that in the recent past, chatbots have been low-context communicators.
Behavioural intelligence
Behavioural intelligence encompasses the skills and abilities needed to effectively communicate and interact with others in order to achieve social objectives.
It goes beyond mere communication and encompasses a range of behaviours and actions aimed at navigating social situations with finesse and achieving desired outcomes.
At its core, behavioural intelligence involves understanding social dynamics, recognising social cues, and adapting one’s behaviour accordingly to build rapport, influence others, and foster positive relationships.
This includes not only verbal communication but also non-verbal cues such as body language, facial expressions, and tone of voice. This is something which in the past has been referred to as face speed. The cues we send to each other via expression, tone and posture.
Considering the image below, On the left side, you’ll find the distribution of three intelligence types.
While on the right side, you’ll see the frequency of various subcategories within cognitive, situational, and behavioural intelligence.
Multi-modality
To solve for challenges like face-speed, gestures and other human communication modalities beyond language…
Future datasets will have to be integrated to multiple modalities, which can aid in the development of AI systems with a deeper and more comprehensive grasp of social contexts and cues.
This will ultimately enhance social intelligence by providing a more accurate understanding of human interactions.
Conclusion
The study presents a social AI data infrastructure, featuring a well-founded taxonomy and a comprehensive data library containing 480 NLP datasets.
This infrastructure aims to standardise the concept of social intelligence in AI systems and streamline the organisation of existing NLP datasets.
Additionally, a perform an in-depth analysis of the data library and assess the performance of LLMs (Large Language Models), providing valuable insights into the current landscape of data and guiding future developments in dataset creation to enhance social intelligence in NLP systems.
This initiative facilitates the curation of high-quality datasets and fosters the holistic advancement of social intelligence within the NLP field.
Find the study here.
