1
$\begingroup$

If I create a program which takes an input, gives an output and then requires a response to let it know whether the answer it gave was any good does it count as AI?

If not, what is the process of AI? Does it not always need specific parameters? For example, I ask it "Who is the president of the USA?", and I have programmed it to look for news articles in SEOs and remove the "Who" part, is that AI?

$\endgroup$
1
  • $\begingroup$ Excellent question. Welcome to AI! $\endgroup$
    – DukeZhou
    Jul 27, 2017 at 16:22

3 Answers 3

1
$\begingroup$

There is no "process of AI" as such. There are many, many different approaches to AI, different ones of which are used in specific applications.

As to whether a purely trial and error approach could be considered AI... I'd offer up a qualified "maybe". If you do nothing but an exhaustive scan of the solution space, for every trial, then I'd say "No, it's not really any kind of AI". OTOH, if you're using a knowledge-base of some sort and applying some kind of reasoning (even if it's a heuristic) , and if you have a system that somehow learns from the feedback from the user and gets "smarter" over time, then you're likely working on something that could be considered AI.

All of that said, the exact definition of what is and isn't AI is somewhat fuzzy. One popular definition is something like "any technology that allows a computer to do something well that currently only humans can do well". So if you're doing something that fits that descriptions, it's possibly an aspect of AI. And consider again that most people don't really consider "brute force" solutions to be AI.

$\endgroup$
0
$\begingroup$
  • If an algorithm is making decisions, it can be deemed AI.

The decisions don't have to be better than a human's, or even good. (AI can be limited or "dumb"--on my project we specifically make dumb AIs so children and new players can win;)

From this perspective, the process of the algorithm choosing an output based on an input qualifies it as basic AI. Asking for a response on the quality of the output is merely the validation procedure.

  • If the algorithm learns from the validation process, it constitutes a type of machine learning.

It may not be the "sexy" Machine Learning that has sparked renewed interest in the field of late, but it constitutes an automata that is learning nonetheless.

$\endgroup$
3
  • $\begingroup$ So an AI could be something as simple as in if/else or switch statement with access to a database where it can save answers? Thank you for the welcome. $\endgroup$
    – Alanay
    Jul 27, 2017 at 16:38
  • $\begingroup$ At the most fundamental level, absolutely. It's just that with the cutting edge AI, the if's and then's become extremely complex, and you get into ideas such as deterministic vs. non-deterministic algorithms. As @mindcrime notes, the formal definitions are sort of fuzzy and continually evolving, but decision-making seems to be at the root of algorithmic (artificial) intelligence from the standpoint of both purpose and function. $\endgroup$
    – DukeZhou
    Jul 27, 2017 at 16:45
  • $\begingroup$ I can say with absolute certainty that "any technology that allows a computer to do something well that currently only humans can do well" is merely a qualifier for AI strength, and is in no way a fundamental description of AI in general. (This is why mindcrime was careful to present it as a "popular definition", and make the point about non-universality of opinions regarding brute force approaches--when Euler started analyzing Latin Squares, he used brute force and I don't think anyone would say Euler wasn't intelligent!;) $\endgroup$
    – DukeZhou
    Jul 27, 2017 at 17:33
0
$\begingroup$

I would say that, while it does count, it only counts in certain circumstances, as trying to use trial and error with already predetermined data as its objective to output isn't AI, as it already has data that is fine as-is. For example:

Say your AI is trying to point at a precise location but is only given the current accuracy of its positioning. It could check every location and find the best of all of them, or, it could do this:

  1. Correct the up/down(vertical) positioning:
    1. Start moving up.
    2. If the accuracy gets better:
      1. Keep moving up.
      2. Else move down.
    3. Continue until the accuracy starts to get weaker.
    4. Move back just a tiny bit to increase again.
  2. Correct the left/right(horizontal) positioning:
    1. Start moving right.
    2. If the accuracy gets better:
      1. Keep moving right.
      2. Else move left.
    3. Continue until the accuracy starts to get weaker
    4. Move back just a tiny bit to increase again.
  3. You're done!

This is AI, as it learns what to and what not to do, and figures out a location based off of the accuracy of its positioning alone. So while the "trial and error" method is AI, it only counts when it has no predetermined calculation to figure out the result without said trial and error. Trying to find numbers of Pi, for example, while it is technically trial and error, it uses math functions and requires multiple inputs to calculate its output, and, in the end, it only partly uses trial and error, therefore, it is not AI in the end.

$\endgroup$

You must log in to answer this question.

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