Defining AI: The 8 Kinds of AI You Should Be Familiar With
By Scott Klososky, Founding Partner at Future Point of View
In the technology arena, we often fail to use language in beneficial ways. We name categories with monikers like Social, Cloud and Big Data and they soon after losing any valuable meaning. There is already complexity with digital systems and poor definitions make even more difficult. The latest unhelpful moniker may be the worst – Artificial Intelligence (AI).
This term has existed for many years and is applied loosely to a software system that "thinks" like a human. As long as it was mere concept and science fiction fodder, the name did not confuse people, nor bother me. This has now changed. People use the term "AI" as if it is something specific.
As an analogy, if you ask me what I am doing on Saturday and I say "sports," I would not have helpfully answered your question. If I said instead, "playing soccer or golf", the conversation advances. Sports is a category, not a helpful description. It has become vogue to state that every piece of software being marketed today is an AI, or is AI-powered.
Once this is stated, the conversation is about over because we are becoming numb to AI as a category.
Thankfully, there is a solution to this, adding additional words and meaning that creates the level of specificity needed for the conversation to go on. The following list defines a possible vernacular to help us move forward. These are in a loose order of sophistication level.
Decision Support Systems
This is an AI that helps humans to make wiser or more informed decisions. The intelligence of the software is meant to augment parts or all of a decision that also calls for some human intuition. The algorithm within the AI is never meant to make a decision completely on its own.
Reactive AI
This is an AI that waits for a human to make a request or provide a variable, then the AI reacts with a decision of its own. An easy example of this would be a game-playing AI. It does nothing on its own. It makes each decision based on a move by a human. This could be a support AI that does not engage to help unless a specific condition arises – then it reacts.
Assistive AI
This AI assists other machines – not humans. It exists wholly to watch over the performance of other systems and to adjust them, guide them, or report on them when necessary. An example would be an AI that oversees the robots at a factory by overseeing the whole manufacturing process and adjusting specific steps as needed.
Surveillance AI
This AI has a specific purpose, to watch over an entity, process or activity. It has rules that define the actions to take based on the variables that may come up while it is doing surveillance. This could be a watching over activities to assure safety and security or could be watching over customer behavior looking for a sign that action should be taken. Although other AI's might have a surveillance component, this AI's strength is its ability to watch over many things at once in a way a person could not.
Co-Working AI
This is a system that actively co-works with a human being to complete a process. They work together with the AI providing capabilities that are difficult for people and the person providing abilities that are not fit for an AI. In most cases, this type of AI is in use side by side with human activity. An example would be stock trading because a trader has AI's constantly making recommendations, researching, and providing alerts while the human is approving AI activities and learning from the AI's work.
Semi-Autonomous AI
This describes an AI that can do most of a task on its own with little to no involvement from a human. However, it is designed for human input in a few situations or to default to a human if it fails to have 100% confidence in a decision it needs to make.
Fully Autonomous AI
This AI system is built from the ground up to do a task or process with no human involvement. It knows its mission and will complete the mission to the best of its capabilities, over and over.
Self-Aware Cognitive Systems
This is the most sophisticated level of AI. It is a system that understands its mission, constantly learns (creates new rules) to provide an improved outcome in an autonomous profile. This system is self-learning enough to constantly improve on mistakes it might have made in the past and can make wise decisions even in a complicated environment.
Having studied technology vocabulary for years as a public speaker, I have seen this dynamic before. Very soon, people will move to more precise vocabulary to describe AI capabilities and this cannot happen soon enough.
The year 2020 needs to be when we start asking the question "what kind of AI" to anyone who claims their systems are AI-based.
Scott Klososky is Founding Partner at Future Point of View, and is an internationally recognized speaker, thought leader and technology expert. He and his team at Future Point of View, a boutique consulting firm based in Oklahoma City, architect world-class technology strategies for organizations ranging from Fortune 500 to universities, nonprofits, and countless professional associations and coalitions.