When does AI become more than just complicated predefined logic?_问答_开发者

I cannot pretend to begin to understand how AI software is created, but while reading some news articles today the thought occurred to me: When does AI become actual AI and not just complicated IF s开发者_如何学Ctatements in the background? If everything software does comes down to determinable IF statements with some degree of randomness, how does it have any more or less AI than any other program?

AI doesn't "become" actual AI... It's the other way around, it stops being AI when you figure out how it works.

Recommended reading: PoAIP

When does AI become actual AI and not just complicated IF statements in the background?

Who says there's a difference? As far as we know, out brains are really just "determinable IF statements with some degree of randomness" as well. Physics is just applied math, chemistry is just phsyics with some details simplified to allow working at a larger scale, biology is just simplified large-scale chemistry, and so far we don't have any proof that psychology is not just simplified large-scale biology.

If everything software does comes down to determinable IF statements with some degree of randomness, how does it have any more or less AI than any other program?

Emergent complexity - the whole is more than the sum of its parts and, as it gets more complex, often shows behaviour that cannot be tracked down to any one of the parts, or even specific groups of them.

Everything in the universe is if statements. The difference between AI and non-AI (a.k.a. preprogrammed complex if statements) is all about who is making the decisions. The programmer or the computer?

If a programmer forces a Yes/No answer to a predetermined question, then it's not AI. For example:

if (question == "Is the sky blue?")
   return true
else
   return false

True AI would get the question externally (from a keyboard input, voice input, OCR input, or something else) and then figure out the answer on its own. A true dynamic if would be generated, but preprogrammed steps could still be taken to deliver the answer:

Use an attached wavelength camera to determine the answer.
Look up on Google or some other source for the answer.
Ask others in the real world who may know the answer based on trust.

But it goes deeper than that. Do we preload the wavelength answers for #1 above? Do we predetermine trusted sources for the computer in #3?

We (humans) are all preprogrammed from birth, via DNA, with a certain set of if statement answers built in, and then grow and learn from there. We know how to breath, how to breast feed, how to pump our heart, etc. the moment we become "alive". We also build a database of information from Day 1 of birth within our brains using senses.... sight, sound, touch, taste, and smell. We use this database for our "answers" to if statements.

So how many if statements and answers do we need to preprogram in a computer before humans determine that it's a true AI? At least some to start...

But ultimately, to be called AI, it needs to be able to dynamically build and answer if statements on its own and adjust the answers it gives as it learns from an internal database that it grows over time.

For starters, when you can 'train' your AI to behave a certain way. The AI can then make decisions based up on what it has learned from the training data, instead of from "hardcoded if statements". Obviously this is just scratching the surface, but you can see how AI using (for example) a neural network could evolve over time, as data in the network is changed.

I'd say that it's pretty close to AI when it can drive a real car in urban conditions.

Can you trivialize this by saying "it's just a bunch of 'if' statements"? I can't.

"Any sufficiently advanced technology is indistinguishable from magic." - Arthur C. Clarke.

Personally, I think auto pilots for planes and cars are quite advanced and impressive.

Although most of the approaches to AI is based on if-else structures, in fact the idea behind the AI is established on the computational models. For example, most of the recommender systems are based on the if-else structures, however, these are never AI. For AI, there must be a computational model in your algorithm which does nothing to do with if-else structures. Thus, it is mathematics! Another example is avatars speaking like SIRI. Although there are lots of ontology based approaches to language, I don't think speaking of SIRI is AI since it doesnot contain a computational model, although it may have computational model to help the if-else structures. But, computational models doesnot occupy the central stage. So, there is no newness, no creativity, no surprise if you approach the problem as an if-else structure.

a way to separate AI from complex "if-else" statements would be if it is able to comprehend what it is, as well as contemplate it's own existence.