I believe artificial intelligence (AI) term is overused nowadays. For example, people see that something is self-moving and they call it AI, even if it's on autopilot (like cars or planes) or there is some simple algorithm behind it.

What are the minimum general requirements so that we can say something is AI?


5 Answers 5


It's true that the term has become a buzzword, and is now widely used to a point of confusion - however if you look at the definition provided by Stuart Russell and Peter Norvig, they write it as follows:

We define AI as the study of agents that receive percepts from the environment and perform actions. Each such agent implements a function that maps percept sequences to actions, and we cover different ways to represent these functions, such as reactive agents, real-time planners, and decision-theoretic systems. We explain the role of learning as extending the reach of the designer into unknown environments, and we show how that role constrains agent design, favoring explicit knowledge representation and reasoning.

Artificial Intelligence: A Modern Approach - Stuart Russell and Peter Norvig

So the example you cite, "autopilot for cars/planes", is actually a (famous) form of AI as it has to use a form of knowledge representation to deal with unknown environments and circumstances. Ultimately, these systems also collect data so that the knowledge representation can be updated to deal with the new inputs that they have found. They do this with autopilot for cars all the time

So, directly to your question, for something to be considered as "having AI", it needs to be able to deal with unknown environments/circumstances in order to achieve its objective/goal, and render knowledge in a manner that provides for new learning/information to be added easily. There are many different types of well defined knowledge representation methods, ranging from the popular neural net, through to probabilistic models like bayesian networks (belief networks) - but fundamentally actions by the system must be derived from whichever representation of knowledge you choose for it to be considered as AI.

  • $\begingroup$ Given that "we" clearly means Russell and Norvig themselves, I doubt they overstepped ... $\endgroup$ Jan 6, 2017 at 7:07
  • 7
    $\begingroup$ Now define "percepts from the environment" and "actions". Any algorithm can fit this definition. $\endgroup$ Jan 17, 2017 at 10:05
  • 1
    $\begingroup$ @BenediktS.Vogler Agreed. In the most reduced definition, any decision making algorithm would qualify. Level of intelligence (degree of utility) is corollary. $\endgroup$
    – DukeZhou
    Dec 1, 2018 at 0:38

In addition to what has already been said about AI, I have the following to add. "AI" has had quite a history going all the way back to the original Perceptron. Marvin Minsky slammed the Perceptron in 1969 for not being able to solve the XOR problem and anything that was not linearly separable, so "Artifical Intelligence" became a dirty word for a while, only to regain interests in the 1980s. During that time, neural nets were revived, backpropagation used to train them was developed, and as computer technology continued its exponential growth, so did "AI" and what became possible.

Today, there are lots of things we take for granted which would've been considered "AI" 10 or 15 years ago, like speech recognition, for example. I got my starts in "AI" speech recognition back in the late 70s where you had to train the voice models to understand a single human speaker. Today, speech recognition is an afterthought with your Google apps, for example, and no a priori training is needed. Yet this technology is not, at least in general audiences, considered "AI" anymore.

And so, what would be "minimum requirements"? That would depend on whom you ask. And what time. It would appear that that term only applies to technology "on the bleeding edge". Once it becomes developed and commonplace, it is no longer referred to as AI. This is true even of Neural Nets, which are dominant in data science right now, but are referred to as "machine learning".

Also check out the lively discussion on Quora.


This is an "in human language" (non-technical;) synopsis of the core of Kaiesh's excellent answer.

In the most basic sense, any decision making algorithm can be regarded as a form of Artificial Intelligence.

The History of Artificial Intelligence wiki gives a pretty good overview. The roots of the field are generally ascribed to Symbolic Artificial Intelligence, but it might be said to go back as far as Babbage. The first functional game AI in the form of an "analytic engine" may be Nimatron (1940). More recently, Machine Learning in all of its various forms, including Neural Networks and Genetic Algorithms, have been delivering exciting results. Bayesian networks are another form of probabilistic AI.

Utility, the means by which we evaluate the degree of intelligence of algorithms, is separate from the mechanism.

AIs can be weak or strong. Strong means better performance at a task than a competing rational agent, typically humans. ("Man is the measure of all things." Protagoras) The term strong in relation to AI has traditionally been taken to mean Artificial General Intelligence [see also the Turing Test], but current algorithmic intelligences are only "narrowly strong".

Intelligence is a spectrum, therefore:

The minimum requirement for AI is that an algorithm make decisions based on data, irrespective of the quality of the decisions.


There is also the AI effect, that is, the tendency to not consider something an AI once it is well understood. For example, neural networks are not yet fully understood, so people still tend to call them AI. Once we know exactly all the details about neural networks and their inner workings, we might start to consider them just computation. This is an old philosophical topic that goes back at least to the famous Jacques de Vaucanson's defecating duck and automatic loom.


From "Artificial Intelligence And Life In 2030: One Hundred Year Study On Artificial Intelligence":

In fact, the field of AI is a continual endeavor to push forward the frontier of machine intelligence. Ironically, AI suffers the perennial fate of losing claim to its acquisitions, which eventually and inevitably get pulled inside the frontier, a repeating pattern known as the “AI effect” or the “odd paradox”—AI brings a new technology into the common fold, people become accustomed to this technology, it stops being considered AI, and newer technology emerges.

Consequentially, I believe we can not choose a fixed set of requirements for something to be considered AI; rather, at any given moment in history, AI is a set of programs which can achieve something that before was generally considered to be solvable by humans only. As technology evolves, the boundaries keep getting pushed and pushed, and the bar rises higher. Consider chess playing: once chess engines were considered one of the pinnacles of AI, while nowadays such programs are perceived as "blind search" and not truly intelligent.

To quote Larry Tesler, Intelligence is whatever machines haven't done yet.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .