kirill_makarov - stock.adobe.com
Building a better conversational AI assistant requires emotion
Industry after industry is seeing benefits from chatbot implementation, but customers and developers are looking toward a future of more connected, intelligent conversational agents.
The first iterations of the conversational AI assistant saw success as simple chatbots that followed a pre-programmed set of instructions. They allowed for simple web surfing or augmented customer service.
Industries from retail to corporate product services saw near-immediate value in chatbots, but as consumers and businesses are demanding smarter, faster, more intuitive conversational assistants, we are witnessing the beginning of a potential chatbot revolution. Developers are focusing on conversational intuitiveness, intelligence and the ability to replicate human conversation.
As developers look toward what makes a well-rounded conversational AI assistant, it's imperative to denote the differences between utility and intelligence. Today's intelligent agents can be useful -- they can sell a product, direct a call or answer a question without human intervention. But the chatbots of the future will be more flexible and interactive and should be able to "offer products and services in a human way," said Sasha Caskey, CTO of digital platform Kasisto.
To train a computer to replicate human activity and speech, developers should start with contextual interactions, Karen Myers, lab director at SRI's International AI Center, said at the AI World Conference & Expo in Boston.
The first generations of chatbots were focused on one-off interactions -- book this event on a calendar, answer this specific question or direct a user to a webpage. The future goal of chatbot conversations is not a collection of sub-interactions, but a coherent multi-turn conversation that relies on context to build understanding.
Karen MyersLab director, SRI International AI Center
"You need for your conversational assistant to understand the history of your interactions and the overall goal of what the human is trying to achieve through the conversation," Myers said.
If intelligence is a buzzworthy goal of future generations of chatbots, evolution is another. With growing technology, there's a desire for chatbots to understand how the user's needs and priorities shift throughout a conversation, said William Mark, president of SRI International's information and computing sciences division, a nonprofit scientific research institute in Menlo Park, Ca.
Sentiment analysis and emotional conversation
A step up from mechanical, single-turn conversations will also require emotional analysis. The goal of an emotionally intuitive conversational AI assistant is for humans to not have to change their conversational style to speak with a computer assistant, Mark said. Human-to-human conversation is enabled by minute quirks including tone, emotional register and ensuring mutual comfort. Chatbots' current "personalities" don't take these things into consideration.
The development of neural networks alongside image and language recognition is pushing chatbots toward understanding tone, manner and diction and can help bots identify who is speaking, enabling them to interact with speakers in unique ways based on user profiles. Some users may want to receive short, informative directions, while others may prefer more conversational explanations.
Chatbots can help customers overcome the emotional hurdles that may make an interaction difficult. For example, Caskey noted that in the financial sector, customers sometimes feel uncomfortable discussing money, settlements and sensitive financial information with human counterparts. In order for chatbots to keep up an appropriate conversation, customers wanted understanding and responses that echoed their own feelings.
"When it came down to discussing finances, you don't want to broadcast and instead want a hushed chat conversation," Caskey said.
Once chatbots develop these kinds of conversational and emotional capabilities, it could allow for humans to be removed from the conversation completely and enable distributed coordination across multiple intelligent assistants.
While Myers and Mark agreed that the development of coordinated chatbots and virtual assistants is a long-term goal, the idea is to have smart agents delegating on behalf of their users, from low-stakes decisions like schedule coordination to higher-stakes financial delegation.
As conversational AI becomes more prevalent, it also become more specialized. This niche specialization will force AI systems to interact with each other in order to maintain a connected, user-friendly experience, Mark said.
Users want pleasant, simple, helpful virtual assistants that fulfill request without requiring too much information or follow-up, thus seamless information sharing between platforms will heighten the user experience and provide faster resolutions.
In addition to consumer-facing chatbots, Wai Wong, CEO of Serviceaide, a California-based software company, said chatbots can augment the internal business processes of employees.
If employees can use virtual assistants as digital personal assistants, the features can range from proactive individualized information retrieval and data sharing to support for collaborative activities with other employees.
"Imagine if every employee had a digital colleague available to help them, and these digital colleagues could also rely on increasingly domain-specific AIs to create a virtual empowerment team [across the organization]," Wong said.