Definition: A system is intelligent if is able to design an intelligent loss function.<p>From this recursive definition we obtain that there is only one type of intelligence (the root type), that is:<p>From any root type of intelligence there is only one extension type of intelligence given by the above recursive definition.<p>The main consequence of this definition: Root type of intelligence is of paramount importance because it defines the expansion type of intelligence.<p>Example of application: A startup could attempt to design a strong AI that employs this definition, if the startup sells several million agents then we would see an emergent landscape in which collaboration triumph to change the landscape.<p>So there would be a market in which AI system employ the recursive definition to construct, buy or sell other intelligent agents. So we arrive to the: Recursive AI agent market.<p>For those looking to downplay this definition: A projection into a stupid level is: This is just think from first principles, start by planning.<p>For those looking to enhance this definition: Apply concepts related to Shannon information theory to measure the channel capacity to apply this information into a concrete IA system.<p>For those looking just to tag and store concepts they would store this under: IA align, expand AI designing a loss function, variance of generative expansion of AI from a root type of intelligence.<p>I hope that our root human intelligence, the only one that we know, when expanded by the above definition allow us to survive.<p><pre><code> The new types of AI should be able to discuss political and economical decisions using the recursive definition of intelligence, that is explaining what is the loss function they are using to make decisions.</code></pre>