Kapitel 12 VUI

cooperative principle -> listeners and speakers must act cooperatively in order to have a successful conversation

• Quality Say what you believe to be true

• Quantity Say as much information as is needed, but not more (e.g. right amount of answers / lines in chatgpt)

• Relevance Talk about what is relevant to the conversation at hand

• Manner Try to be clear and explain in a way that makes sense to others

-> break these maxims and negatively affect the user’s experience.

• Quality Advertising things you can’t live up to (e.g. „how can I help you?“)

• Quantity Extra information (e.g. „Please listen carefully as our options may have changed.“)

• Relevance Giving instructions for things that are not currently useful (e.g. explaining return policies before placing the order)

• Manner Using technical jargon

• main question: Can machines think?

• Goal: Create computer program which fools human judges into thinking it is human

• Problem: People can only look at the conversatioin -> even if the program passes the turing test it is not sentient

• Set up: - person talks either to computer programm or to human -> task: identify the machine

-> early NLP computer program (by Joseph Weizenbaum)

• Created to demonstrate the superficiality of communication between humans and machines

• Simulates conversation by using a pattern matching (connected with a thesaurus -> look up synonyms or generic term) and substitution methodology (words get substituted in predefined phrases)

• Gives an illusion of understanding, but had no built in framework for contextualizing events

most famous script, DOCTOR, simulated a Rogerian psychotherapist (simply parroting back what patients say)

-> I have a problem with my father? -> Tell me more about you family

• questions about the character: Who are you as a person? What was a event in you childhood that matters the most to you?

• similar questions that should get the same answer, but differently worded

• some thing that needs logic to solve, e.g. simple math question

• ask the same thing a few times -> human will get annoyed

• questions about detailed things (e.g. number of residents) -> human will answer with an inaccurate number / ai not

-> claims to be an 18-year-old female chatbot

• Her intelligence includes the ability to reason with specific objects

• She can play games and do magic tricks at the user's request.

• Won the Loebner Prize (best program in the turing test in that year)

The conversation mit Alexa is still different from human-to-human conversations:

• Back-and-forth exchange of information

• Series of on-offs

• Each snippet is a simple interaction

• No knowledge of previous interactions

• Independent completion

-> Human conversations are fluid and rely heavily on context and shared history.

-> add memory & context

-> support continiuous dialoge (topic switches, speaking turns, ..)

(understanding intent / emotion / know user preference)

Advantages:

• Speed -> talking is faster than typing

• Hands-free -> much more practical in many situations.

• Intuitiveness -> everyone knows how to talk.

• Empathy -> voice holds great deal of information (tone, volume, rate of speech,…)

  
  

Disadvantages:

• Public spaces -> not appropriate in some places

• Discomfort speaking to a computer

• Some users prefer texting -> texting is normal mode for many people -> dont want to shift to voice.

• Privacy -> Potential privacy violations (reading your text messages out loud

conversation design:

• conversation = structure

• dont fool anyone in thinking system is a human

• people say things you didnt expect

• just a guess how the user got to a specific point in a conversation (there are endless possiblities)

Conversations are multimodal (speak, tapping/swiping, typing)

could be a solution for the different apps -> maybe info of apps in background -> voice interface (or sth else) is what we interact with on a daily basis

• bots commenting on post created from bots

• ai images to train ai image generation

• write email with ai -> summarize email with ai (other person)

  very close to ai talking with ai

  
  

pro:

• Inclusivity (includes non-binary ppl)

• avoids reinforcing stereotypes

• neutrality in tone (culturally neutral)

con:

• Uncanny or Unfamiliar Sound (could be hard to listen to)

• loose of customization / personalization

• loose of trust (due to not fitting into expectation of gender)

(same wie davor?)