Middle of Somewhere Design

For an assignment in my Interactive Media 3 class, we were asked to experiment with the chatbot building tools Voiceflow and Tidio. I had never used either of these chatbot builders, but it was engaging and interesting to experiment with both and the experience helped provide me a better understanding of how chatbots operate and generate their responses.

Initially, we were asked to read more about Large Language Models, Prompting, and Hallucinations in the context of Voiceflow. The application seems very user-friendly as far as interface, allowing the user to produce a sort of chatbot that is able to take specific inputs and generate predictive answers. In order for the chatbot to answer questions, it requires an input of knowledge to base its responses off of. As part of the experimentation, I chose to provide the system with information about bicycles. I used Wikipedia pages about bicycles, gravel bikes, road bikes, mountain bikes, and bike racing. I was then able to preview how the chatbot would respond to different input questions I submitted that were related to bikes. With my own knowledge of bikes, I was able to gauge the responses to be very accurate, based on the information the bot had access to. I asked questions about components of bikes and what differentiates a road bike from a mountain bike. The answers were detailed and well structured. I was able to come up with a question about unicycles that I knew were not addressed in the source information and the chatbot had no response for me. The experimenting effectively demonstrated the strength and speed of the chatbot when provided with sufficient input data for the types of questions you would want it to provide answers to. It also was obvious how ineffective the chatbot would be if not provided with sufficient knowledge on whatever topic it is designed to provide support for.

After the experimentation with Voiceflow I moved onto the other building tool, Tidio. It was immediately clear that Tidio operates differently from Voiceflow, in the sense that the AI relies on manual dialogue response entry instead of pulling “knowledge” from data sets that have been provided. To set this up, a flow chart needs to be built, which provides different options, outcomes, and routes based on the users questions and answers. I found the interface to be far less user-friendly for a first-time user as the button choices and set up for the required user flows felt very complicated. Voiceflow was very self-explanatory and very easy to set up, whereas Tidio required much more trial and error and even some frustration with trying to create a testable flowchart for a chatbot. I stuck with the same theme of bikes for my experimentation and preloaded similar questions about the topic that I had used for the Voiceflow experimentation. The way this type of chatbot is set up creates very nice organization with the flowchart structure and very to-the-point responses for a user. The limitations seem to be that the chatbot is very confined to the exact questions created in the input flow and would likely more easily be unable to assist than the Voiceflow chatbot (assuming Voiceflow had sufficient knowledge sources provided).

Overall, I found this experimentation with Voiceflow and Tidio to be engaging and helpful in better understanding how chatbots can be set up and programmed to assist users in different ways. I personally was much more fond of Voiceflow’s capabilities and ease of set up than Tidio. I like that Voiceflow can provide such detailed responses if given enough input knowledge. As we read about, though, you need to be aware of possible “hallucinations,” where the sourced knowledge is combined inaccurately and generates incorrect responses. In contrast, I felt that the Tidio flowchart setup was tedious and frustrating. Even with a demo video it felt like it would take a lot more time to really form a polished flowchart that met all the needs of a successful product, in the form of a useful chatbot assistant. They clearly are both useful tools, though, and can each be implemented successfully if provided with appropriate inputs and flows for their intended application.

Voiceflow Demo:

Tidio Demo: