¿Qué es IA?

Hello everyone. In this video, we're

going to try to address the question of

what is AI? In fact, there are many

definitions for AI systems. But these

various definitions can fall under one

of four different categories. The first

category says that AI systems are

systems that try to think like humans.

Of course, these are involved with

trying to understand how humans think.

The other set of definitions is involved

with or they define AI systems as

systems that act like humans and these

are involved with understanding human

behavior, how humans behave. The third

category of definitions focuses on how

to think rationally. They say AI systems

are systems that kind of think

rationally. And of course this is

involved with the question of okay how

do we make rational decisions and how do

we define rationality in the first

place. Now the last category of

definitions is focused on acting

rationally. They say that AI systems are

systems that are able to act rationally

meaning how to act rationally in order

to achieve a certain goal. Okay. Now

let's take a look at each of these set

of definitions. If we try to focus on

systems or if we define AI as systems

that try to think like humans then we

need to understand how humans think.

This knowledge is will take us to things

like neuroscience and cognitive science

that basically try to understand how the

human brain works and how the thinking

process takes place. As for definitions

that rely on acting like human or

behaving like human, well, in that case,

we need to understand human behavior and

that kind of depends on psychology.

But Alan Turan actually came out with a

test that allows us to evaluate an AI

system and whether it does act like

humans or not. His test is very simple.

He basically said, okay, let's have some

sort of an interrogator. First of all,

you can imagine that this AI system is

something like a chatbot. Okay? And now

a human interrogator is interacting with

this chatbot. If this human is unable to

to figure out whether this is a bot or

an actual human that is chatting with

him or her, then we would say that this

AI system has passed the Turing the

Turing test and we can say that it is

acting like humans.

So in order of course for an AI system

to be able to act like humans, it needs

to understand multiple things

to understand natural language

processing because it will interact with

humans, right? So it has to have a way

to process this natural language and use

it to represent knowledge and also has

to have a mechanism to reason about this

knowledge, process it somehow. And it

needs of course means for machine

learning, right?

>> [snorts]

>> in order to learn something and

eventually respond to this human.

Now, as for thinking or focusing on

thinking rationally, now we need to

define what is rational thinking. Of

course, this is a philosophical

question, right? and Aristotle was the

first to try to address it and that led

him to the development of logic and the

inference rules that we can use in order

to infer correct results whenever we are

given correct premises.

Okay. So let me give you an example. So

if we have two premises like this, if we

say you know what Socrates is a man and

all men are mortal, then from these two

premises we can infer or conclude that

Socrates is mortal. Okay.

Now, of course, trying to answer the

question of what is rational thinking is

something that usually philosophers or

people that belong to different moral

value systems would be concerned with.

But in AI systems, we try to use the

same terminology that is used in

philosophy,

but we give it very specific measurable

definitions. Okay? Okay. So, this is

just something to keep in mind as we go

forward because we will be using terms

that normally humans do it to represent

or to describe human behavior. We're

going to use it to describe the behavior

of these AI systems, but they don't

necessarily mean the same thing. It's

like a very watered down version of of

the same definition of the same word.

Okay. Now moving to the last set of

definitions that relies on acting

rationally. Well again we need to define

what is rational behavior. In fact we

define in AI we define rational behavior

with respect to or in terms of a given

goal. We say that rational acting leads

to achieving the best possible outcome.

And as you can see this definition, the

way we define rational acting over here

is very different from the way a philos

a philosopher would uh define the same

term. Okay. So again we are using very

specific definitions of something that

maybe in philosophy people keep fighting

about okay what is the correct

definition but over here in AI we try to

make everything measurable and

quantifiable. So again rational behavior

is defined in terms of trying to achieve

a certain goal. A rational acting is

going to lead to achieving this goal or

achieving the best possible outcome.

Sometimes the goals are very complex.

They have different components. So we

are at least trying to achieve the best

possible outcome and that would be what

we call rational acting or behavior.

Now, as you can see, all of these

definitions

borrow information from different

disciplines and sciences. Like trying to

think like humans, well, we have to

understand how humans think, how the

brain works. And this borrows

information from neuroscience.

Trying to have AI systems to act like

humans. This will need to borrow

information from psychology because this

is where we study how humans behave,

right? Thinking rationally borrows

information from logic and acting

rationally borrows information from

economics because how can you measure

whether you have reached a goal or not?

Especially if you have uncertainty in

the environment around you and if the

goal is complex and it contains multiple

different components that may kind of

oppose each other. Okay. Now, AI systems

also depend on math in order to try to

quantify the goals that we're trying to

reach and how to measure the extent by

which we have reached these goals. Of

course, it borrows or it relies on

computer engineering in order to build

the hardware of these AI systems. It

borrows information from linguistics in

order to enable these AI systems to some

somehow process natural language. I

don't want to use the word understand

human language because understanding is

a debatable term itself. What do we mean

by understanding? But we can say AI

systems should be able to process

natural language in some way. [snorts]

And AI systems also rely on control

theory because these AI systems or

agents should control themselves and

have some control over the environments

they're in.

So to wrap up, there are actually many

definitions of AI. All of them rely on

theories that come from other

disciplines as we just

saw and the most common approach to AI

actually leans off definitions that

relate it to rational acting. Okay. So

most

modern AI focuses on definitions that

relate AI systems to how to act

rationally in order to achieve a

specific goal or maximize the outcome.

Okay.

Now AI actually requires a lot of things

and it borrows information from a lot of

things but these are the main very basic

things that AI requires. Some way to

process natural language. Some methods

to represent human knowledge and to be

able to reason about it to infer new

information from the information and the

knowledge that we already possess.

should be able to reason under certainty

and uncertainty because sometimes we do

not have full knowledge actually most of

the time in most complex AI systems we

do not have full knowledge right so

there's a notion of uncertainty that is

going on AI systems requires an ability

to allow the machines to learn new

patterns and use them to enhance their

performance

and AI systems require a form of

perception and action or acting that

utilize izes computer vision and

robotics simply because these AI systems

they interact with the environment

around them. Again the term perception

is in AI systems is used to mean

something that is kind of relevant to

human perception of the environment but

it's not the same. So basically AI

systems requires means to take input

from the environment okay and process it

somehow through computer vision and it

should also be able to act upon the

environment through means of robotics

and things like that that allow it to

achieve its goal if its goal needs that.

So that's pretty much it. This was a

quick summary that tries to answer the

question, what is AI and what are the

different categories of definitions for

AI systems and the main set of

definitions that modern AI relies on

which is trying to create systems that

act or behave rationally. Thank you.