• Question: How is an AI build (such as Siri or Alexa) to know how to do stuff like turn you’re lights green? (if you connect it to you’re home as in) What does the machine or electronic need to have the knowledge of a human or smarter than?

    Asked by cookieowl on 19 Mar 2021.
    • Photo: Alessandra Mileo

      Alessandra Mileo answered on 19 Mar 2021:

      This is a very good question! Traditionally, AI systems used logic rules to represent human-like reasoning. Things like “IF this AND that happens, THEN do this action”. Imagine a lot of these, like “IF dark room AND it is Sunday, THEN turn lights green” and “IF somebody says “TV on” THEN switch on TV”. This is the KNOWLEDGE of the machine and and is written as a piece of (logic) code.

      What is after the IF in a rule is normally sensed by sensors and processed to see if it is true or not (in SIRI or ALEXA the device can understand the meaning of a word using dictionary of spoken words to compare it with, while in smart homes “the room is dark” is based on a light sensor that captures the amount of light in the room and check if the value is in the “light” or “dark” area).

      The action to be done (after the “THEN” in a rule) is normally translated into something mechanical and low level that the machine can activate (for example “turn lights green” can be translated into mechanically turning an electric switch in the green position).

      This rule or KNOWLEDGE is what scientists write to tell the system what to do. This knowledge is processed by a computer program (think about it as the BRAIN of the machine) that looks at all those rules and selects the actions to be done based on what has been sensed.

    • Photo: Alan Diaz

      Alan Diaz answered on 19 Mar 2021:

      That’s covered in a sub-area of AI that is called Natural Language Processing!
      What these AI applications attempt to do is to “understand” the context of phrases, i.e., “what’s the weather outside?” or “tell me a joke”. Each sentence has a number of words that the AI will separate and will try to identify what each of them means (part of the processing), then it will check the syntax rules that the AI has previously learned (i.e., when the word “what” is placed at the beginning of a sentence it is most likely to be a question). Last, the AI will come up with a few options of interpretation of what you have said, i.e.: option A) my human wants to know the weather right now (95% chances that this is correct), option B) my human wants to go outside (3% chances this is correct), etc., thus, it will reply to you according to option A.
      In reality it is a bit more complex than that, but that’s pretty much what you are trying to do.
      Keep in mind that if you get to learn how to process natural language, you’re already learning how to process a lot of data, like songs or images (technicalities may vary…).

      As for the “smarter than a human” part, if you build an AI like Alexa to control your home environment, it might be better than a human in say “turning on all the lights at the same time”, but if you ask Alexa to prepare a ham and cheese toastie it would be as useless as an umbrella in a windy day 😉 “Intelligence” is a very broad and controversial concept.