AI, a RoadMap for Context
Over-hype and AI seem to go hand in hand. Full General Artificial Intelligence, where a machine can have an intelligent conversation with a human is a dream which is not yet achievable and yet we are continuously bombarded by claims that such problems have been already been solved. The reality is we need to be honest about this — they will not be solved for another 10–15 years and the route to solving it needs to be defined so that everyone can understand where we are with this technology.
One aspect of a conversation is called context. But context is itself an aggregation and simplification of several totally different things. The implication is if we solve context we will be able to have an intelligent conversation. The reality is that context needs to be broken into its constituent parts and solved part by part over time. The problem is some people solve the simplest forms of context and then say they have solved context — this is misleading.
Types of context
Subject Context — remembering the subject of the conversion so that if you answer “yes” to a question or use the word “it” or “he/she” the system knows what you are talking about. This problem is being solved currently but for specific situations — subject by subject by Narrow Intelligences.
Environmental Context — (an optional context) sometimes the physical location or current weather at that location can become relevant.
External Context — (an optional context) world events such as a pop star dying can become the subject of a conversation, so the bot needs to be aware of external news events in case they become the subject of the conversation.
Social Context — this is changing the way things are expressed depending on the social state with the individual you are talking with. This context can itself be broken down in several types:
Formality Social Context. This is the level of formality or friendliness to be used in the conversation. Some languages such as French and German have this in the different forms of the word “you” being used. In German formally the word “Sie” is used and the informal singular word is “du”. Initially the machine could switch between the two formats dependent on the history with the individual being talked to, but in the longer term there is a wide spectrum of friendliness versus formality which needs to be dealt with. If you speak Geman try this tool which converts the “Sie” format to “du” format.
Emotional state social context (empathy level/sentiment analysis). Expressions can be used to transmit an emotion state. In its most basic form profanity can be used to convey a high emotional state. But the rules of when to use profanity are more complex than just emotional state. A child may not use profanity with its grandparent when it regularly does for other children in the same emotional state. At lower levels of emotional state these problems may be more easily solved. For example annoyance at a machine misunderstanding a question should be responded to which an apologetic language and an attempt to clarify the misunderstanding.
Jargon social context. An engineer will use different language with other engineers than he does with non-engineers. The reason is people create jargons to shorten conversations with like thinking individuals. The converse of jargon is simple language which would be boring to like minded people if used.
Humour social context. Humour is often treated as a measure of intelligence and humour can be used to entertain the human in the conversation. But humour itself is composed of types of humour. Simple humor may be simple puns, more complex forms involve sarcasm/irony or playing practical jokes or lieing to get an emotional response. One of the problems with humour is how often to use/reuse it. For example a joke should only be told once.
Conversational History social context. Needs to remember previous conversations by subject and know how successful they were. This is difficult to achieve because it needs to categorise each conversation and also understand something about the other person’s reaction to each old conversation.
Honesty social context. Sometimes depending on the circumstances the truth may be inappropriate. The concept of the “white lie” to spare someone’s feelings will be needed by an AI. Detecting this scenario will of course be difficult.
RoadMap for Context
So don’t be fooled by marketing hype from the software industry saying these issues have been solved by their 18 PhD Stanford graduates. They have not been solved yet but they will over the next 10–15 years. Below is a complete guess at the roadmap timescale. The Jargon social context will be solved by each subject area.
Discover Latest Jobs in AI, ML, Big Data & Computer Vision:
