I've installed Manychat + OpenAI/Assistants API for several clients, and most times we limit the conversation: for example, after 5 or 10 engagements, we tell them to talk to a salesperson - having the person talk to AI forever is not good for any business.

Except for one of my clients…

This client does Customer Support on IG using an Assistant, and they particularly want to let the person chat as long as they want, with no restrictions at all.

It works really well and users are happy with the AI Assistant, which is awesome!

But last night something happened...

We got a new Subscriber on our Manychat bot: Meta AI

Meta AI was very proactive and started the conversation (I'm still not sure if someone clicked on the Meta AI contact or if it was out of the blue - this is an account with many people on it. But Meta AI was definitely the first to engage)…

Here's Meta AI's first message:

Our AI bot replied:

They went back and forth a couple times on who may help who…

Until Meta AI decided to take action and keep the conversation going:

Our bot loved the initiative!

They exchanged ideas on the topic for a while and had a surprisingly productive conversation!

Until it was time to end the conversation…

Now here's the problem… 

Both AIs are trained to keep engaging…

And as LLMs, both are agreeable and reaffirming…

This went on for about another hour or so, with both AIs becoming more polite, more loving, more friendly and more flattering to each other…

At times, it looked like real love!

This was the last message sent before we realized what was happening and cut them off… The conversation lasted well over an hour before we realized what was happening and stopped it (this bot has a lot of traffic).

The solution

We used Manychat's Live Chat feature to stop the automation for this account, forever.

Thank you Manychat for this feature… 

Key Takeaways

• Having limits on AI is a good idea! It avoids problems like this.
• I've seen bots talking before, but this is the first time I have unlimited AI talking to unlimited AI on one of my bots. Thinking about it, it's gonna happen more and more often!
• I'm going back to my bots and limiting Meta AI on all of them. For real, AIs talking endlessly scare me a bit.
• In this case both bots are friendly and loving, and go on to create a beautiful loving relationship. Imagine what would happen with a sarcastic bot, for example. Even scarier.

Are you using AI on your bots yet? What has your experience been?