Mint Primer | Why multiple chatbots are better than one
Summary
- Companies and users are getting better at deploying and interacting with virtual agents.
Now that users are getting comfortable with AI models and chatbots, the next step is to improve user-experience. But how? The answer lies in using many specialist bots, rather than getting one to do all the jobs—from banking to customer care to teaching. Mint explains:
What are multiple virtual agents?
Systems that rely on more than one chatbot or use multiple Large Language Models (LLMs) are called multi-agent systems (MAS). A chatbot on the website of a bank, auto, insurance, edtech or any other company has improved over the years, thanks to advances in AI and natural language processing. But it still falls short of answering questions that might require more ‘human-like’ capabilities or pulling data from different domains. The solution lies in using multiple bots or more than one AI (artificial intelligence) model—each with a different, specialized capability, and complementing each other to deliver a better response.
Read more: ‘AI will create job redundancies, but plenty of new roles too'
How are they better than a single AI model?
Companies and users are getting better at deploying and interacting with virtual agents. Open AI’s release of an upgraded GPT4 offers improved handling of text, vision and audio, boosts AI capabilities, and opens up new possibilities. Just as teams of humans are better in tackling complex problems, more than one chatbot will improve customer interactions. So, a Hindi language chatbot could combine with a math teaching bot and deliver math lessons in Hindi. Each is a separate LLM (which helps chatbots understand human input and offer answers) and is better at executing specialist tasks.
What will be the impact of having multiple bots?
If a chatbot gets stuck, the query is usually escalated to a human agent. Merging capabilities of multiple bots or making different bots work together is the way out. Microsoft, Meta, Exotel, Tokyo-based Sakana and others are trying multiple bots. A single bot can break down the task and give an answer but it may lag in response time and lack accuracy.
Read more: Where there is AI, there is data, in every industry
So, will we have teams of bots?
That’s the future. As machines get better with more computational power and ability to comprehend context and emotions, there will be many specialized bots. Some will take care of FAQs; others will offer deeper insights on, say, a holiday destination; another could plan the itinerary. But if you can’t go on the holiday and need someone to reschedule plans, it is beyond the capability of a current bot. But in future, even this will be managed by a virtual agent. It could be two or more bots managing such processes.
What are the risks of many chatbots?
Trust will be a big issue. The biggest risk is the virtual team messing up the outcome. You may not leave your entire mutual fund portfolio and money management to a bunch of bots. Single bots are getting better at customer care type of requests, like air con service or issuing cheque books. But in other use cases like content generation, there are doubts on accuracy and copyright issues. Even if teams of bots get it right, user acceptance will take longer. After all, it took over a decade to get comfortable with a single bot.
Read more: Mahindra Group Integrating AI and Gen AI to Improve Customer Experience, says Group CTO Mohit Kapoor
