Chatbot, video avatar, interactive AI avatar: what is the difference?

01Staying quiet

A chatbot always answers. On stage, that would be a mistake.

To a chatbot, every input is a question. Type “okay” into the window and you get a reply. That bothers nobody, because nobody else is part of the conversation.

On stage, several people talk to each other. Most of what they say is meant for another guest, the audience or the moderator. An AI avatar that responded to all of it would barely let the guests finish a sentence.

That is why our AI avatars work the opposite way from a chat window. The same “okay” a chatbot would answer gets no reply from the AI avatar, even in the seconds after its own answer when it is listening for follow-ups. Same with a “Danke” (thanks). If its name only comes up in passing, say in a sentence about it, it stays quiet. It is meant to speak when the moderator calls on it or someone follows up with it.

02Terms

Which one has ears, which one has a face?

Two traits are enough to tell the terms apart: whether the AI reacts live, and whether it appears with a face and a voice. The help window on a website has one, the training video has the other.

• Chatbot

  Ears, but no face

  You type a question into the window and the program writes back right away. It reacts live, but only to the one person at the screen. Some chatbots have a voice; a face is not part of the term.

• Video avatar

  A face and a voice, but no ears

  An artificially generated face speaks in a finished, pre-produced video, familiar from training and explainer videos. What it says was decided during production. Pause the video and ask it something, and you get no answer.

• Interactive AI avatar

  A face, a voice and ears

  It hears spoken language and answers live, on screen and out loud. Whether it faces a single person or a panel where it is usually not the one being addressed, the term does not say.

03No send button

On stage, nobody hits send

In a chat window, the person decides when their question is finished and sends it. Whatever comes back appears immediately, and nobody reads it first. A long paragraph hardly matters, because you can skim it.

Say two guests are arguing about working from home, the moderator is running the discussion and the AI avatar is on the screen. One guest says the best ideas come up at the office coffee machine. The moderator turns to the screen and puts the question in German: “Albert, fehlt uns im Homeoffice wirklich die Kaffeeküche?” (Albert, do we really miss the coffee machine when we work from home?) If she keeps talking, he waits. Then he answers the guest by name and pushes back if his position calls for it. Nobody pressed send along the way. He worked out from the conversation that the question was meant for him and that it was finished.

In the room, everyone hears every word. “Kurz” (briefly) or “in einem Satz” (in one sentence) gets one sentence, “erklär uns” (explain to us) a full explanation of a good half minute; otherwise it is three to four sentences. And unlike in a chat window, someone reads along first. The moderator sees his answer as text on her tablet before the room hears its content, and can cut it off with one button.

04Choosing

When a chatbot is enough

• Isn’t an AI avatar just a chatbot with a face?

  Related in technology, not in behavior. Both usually phrase their answers with a language model. A chatbot waits for input that has been sent. An Aura51 AI avatar has to pick out of a running conversation whether it is its turn, for example when the moderator calls it by name or a guest follows up with it directly.

• Can you type a question to it?

  Yes, just like in a chat window, except that the moderator does the typing. On the tablet, she can enter a question, for example one from the audience, or use a button to turn a guest’s last sentence into a question for it. The whole room then hears the answer.

• Why put an AI avatar on stage when chatbots exist?

  On stage, it is there to take part in the discussion, with a position set by the briefing. If both guests agree on working from home, it can make the case for the office. That is the opposing view, a role that often nobody else takes on.

• What does my event need?

  Look at who is talking to whom. If one person at a time turns to the AI and waits for the answer, a chatbot will do. If the text is fixed in advance and nobody is meant to ask anything, a video is enough. If people are talking to each other in front of an audience and the AI is supposed to join in, you need an AI avatar that hears the whole panel and stays quiet most of the time. The next decision is then which character joins in.