The trouble with Chatbots is that they are either too dumb to be useful or so deceptively intelligent that their errors are cleverly disguised. Despite my favorable experience with my limited use of ChatGPT, most of the time I find service chatbots get everything wrong. Or the limited results are not so useful. Or my problem is too complex for the chatbot to understand. Even basic things get complicated with a chatbot.
When the weather finally seemed like it was staying nice, I contacted my dealer for a tire swap. By default I ended up with the automated system. A tire swap is a relatively straightforward request, so i thought it would be more efficient.
When prompted by the chatbot, I said something like “I want to change my winter tires.” The bot replied with something about scheduling an oil change. I probably used the word “change”, but never mentioned oil. I tried again, using different words such as “tire swap.” This time the bot replied with a service about a brake replacement. It was all very baffling.
I started requesting to speak with a person, but the chatbot kept offering to book me the next available appointment, without even confirming the service! Finally I got through to a real human who booked me in for something a lot sooner than the chatbot’s earliest date. It was all very weird.
On the other end of the chatbot spectrum are the powerful AI-driven chatbots such as ChatGPT. These chatbots can handle a range of highly sophisticated tasks including:
- research
- summarizing long complex works
- imitating styles
- creating deepfakes
- performing analysis
- and more!
And yet, they are prone to making up information, also called “hallucinating.” Sometimes they are accurate and factual. Other times they are believable, but made up. Discerning the difference is challenging. As things get more tangled using AI to perform tasks, the line between reality and Chatbot invented “reality” will becoming increasingly difficult. Soon people won’t know what to believe.
The promise of the chatbot is alluring. For some, it is already proving beneficial and helpful. But in some scenarios it is still lacking. The technology is either too sophisticated, or not sophisticated enough. In the near future, who knows? The service chatbots will likely respond accurately to inquiries. But can we ever trust the results from the more intelligent chatbots?
2 comments for “The Trouble with Chatbots”