Microsoft Thinks the Key to Making Chatbots More Reliable Is... More Chatbots
What could go wrong?
AI chatbots have been a popular tool for getting in-depth answers, but they haven’t come without flaws — Google Search’s AI Overview recommending that people add glue to their sauce to prevent cheese from sliding off their pizza, for example.
We would hope most people have the common sense to not do this in the first place, but it would be even better if there were a way to avoid nonsensical responses altogether. That’s where Microsoft’s AutoGen framework could come in, which uses multiple AI chatbots that can talk to and double-check each other to ensure a quicker, more reliable response.
Collaborative AI Chatbots
Unlike existing options like ChatGPT or Anthropic’s Claude that use just one AI chatbot, AutoGen’s approach can activate several AI agents to handle more complex tasks with a more collaborative mindset. This way, the idea is that one AI chatbot can overcome another’s weaknesses and prevent spewing misinformation. Currently, the AutoGen framework runs off OpenAI’s GPT-4 large language model, but you can use your own.
A diagram showing how AutoGen’s multi-agent approach can be customized.
Not only can you have several AI chatbots converse, but you can even control how the conversation goes. For example, you can set them to talk with each other freely, or you can create a hierarchal format where chatbots are only allowed to talk to a manager. Chi Wang, principal researcher at Microsoft Research, gave some unique examples of AutoGen’s multi-agent approach, including a chess game between two players or even a football game with multiple chatbots.
A recent Wired experiment also showed an example of one AI chatbot acting as a reporter, while another acted as an editor. However, as Wired points out, more AI chatbots working on one task could introduce more complexity than necessary, which could lead to more errors.
Not Quite Ready for Primetime
As promising as AutoGen sounds, it’s still just a framework. We’ve yet to see any major tech companies incorporate this multi-agent approach with their AI chatbots, with only OpenAI’s GPT-4o getting close in this demo where two of them are singing together. However, the team behind AutoGen is constantly looking to improve the framework and has plans to incorporate Google’s Gemini AI.
We’re only in the early stages of AI chatbots in general, even though there are already a lot of options out there. AI chatbots telling us to add glue to pizza is still a huge issue without a solution. Whether adding more AI chatbots to the equation is the answer is anyone’s guess, but I guess two heads are better than one... right?
