Machines Hate Natural Language

Already there, and you got you a big microphone, and you happen to be sitting right between Lauren and I as we discuss uh recent movies that have come into the theaters. We frequently do that. Okay, And I say to Lauren, Hey, Lauren, did you see her? Interesting? Okay, I'm with you so far. Do you understand what the heck I am talking about? Jonathan? Know, I never understand what you and Lauren are talking about. Ever, you know I'm saying in your in your computer imagination world, Oh God,

got in this scenario? No, No, I don't understand what you're saying. That's perfectly simple to anybody who speaks English and in this sort of modern English lingo are familiar. What with what movies are recently came out. There's ambiguity in that sentence already, even if I weren't a computer, because unless I know for a fact you're speaking about a movie ahead of time, I just think you're using a a pronoun in a way where you could be

literally talking about anyone, Yeah, anyone who's female. But when we have everyday conversations like that, asking if somebody saw a movie or anything, that's simple, you don't think about how much incredibly complex figuring out you're doing in real time to make sense of their words shared so much context, the actual literal data is an entirely separate issue. Yeah, even if even if we were all to agree upon what we were talking about ahead of time, there's still

so many idiomatic ways of putting information in various languages. Uh, there are things like figures of speech, there's metaphors, all these other sarcasms. Sarcasm is a great example. There are all these ways that we can say sarcastically. Sarcasm is

a great example. But at any rate, there are all these different techniques we can use to communicate with one another that as you grow up in and around a language, you start to get a grasp on it naturally, right, you have natural affinity for that language, and you understand what the intent is, even if on the surface level it's different than what's the true underlying information. Right, so even if you say, great job at opening the pod

bay doors, how, he's probably not gonna get you. No, How's just gonna think, Wow, I thought I did a terrible job, but every one else didn't notice. Well, how how would probably understand? Because how is really clever like that? Yeah, well, there are a lot of these clever uh speech understanding

robots and computers in science fiction. You've got you've got all the cartoon ones like Rosie from the Jetsons, and you've got Hall nine thousand, who people just talked to and he responds in this calm voice and explains your doom sings Daisy right, uh, star Trek Star trek. Yeah, the computer's voice, which was voiced by Gene Roddenberry's wife, right, um, Margel Barrett yep so uh. She also played as I

recalled Diana Troy's mother. Yes, in the series. I'm still not sure whether it was sweet or creepy that that he had her playing. Yeah, let's not get into that. That's that's that's a podcast all on its own. I don't know, I mean more more the ship's computer. But but but anyway, um yeah, and so we have all of these terrific examples in science fiction, but it's not quite living up to that. Reality is not quite living

up to that. Yeah, Sirie isn't quite the inter prize computer. No, it's it's pretty good series, pretty good at voice recognition and responding. Google has its own Google Search that you can use with voice. Also can do things like transcribe a message you are speaking into a microphone as a text message or or whatever, or instant message. I mean, if you have an Android phone, you can pretty much do everything through voice. Then you've got uh the Google Glass.

Google Glass has voice control in it, although it also is very limited in what you can do. You can't just say anything. You have very specific command words. Wouldn't it be great if like to talk to people like you could not get their attention without saying okay Lauren. Yeah, that that's my idea of the perfect feature. Let's enact that, right.

I actually know some people that that's pretty much true anyway, but um yeah, and and I even had a joking little reference in here, like even even if you look back to Ferbie, Ferbie could respond to to voice commands, although Furbie spoken Furbish and would have gradually learned English, and it wasn't really responding in the way we're talking about here. Yeah, but one thing that I think is really important to make the distinction between is natural language

and voice commands. Right, being able to interpret sound as language is one thing, and and interpreting the context of words is something else entirely. Let's back up. Let's so we're going to talk about computers and robots understanding natural language. Let's step back and look at what is the essential difference between the language that computers understand and like to

use in the language that real that human beings used. Okay, so human language or natural language as we will refer to it in this podcast, and that's you know, generally accepted term natural language processing would mean using machines that could actually process uh commands are written out in whatever native language you happen to be working in human language.

They include things like syntax, grammar. They have these rules that structure that give you the framework you need in order to communicate an idea to someone else who speaks that language, and then a vocabulary of words for conveying specific ideas. Right, So, if you didn't have a vocabulary, you literally have nothing to say. If you didn't have the grammar and syntax, you wouldn't have a way of putting it together that would make sense to someone else. It would be a jumble of words with no no

real way of being sure what the underlying meaning is. Yeah, I think things like grammar and syntax give you a more precise grasp on meaning. So you could probably communicate some meaning with just words but no syntax, But syntax helps you put together complex site. We could all turn into Yoda and more or less understand what the other person is saying. But it gets increasingly difficult to uh

to express more complex thoughts like you were saying. It gets challenging to do that if you're not following a essentially a previously agreed upon framework. That's really what language is. But how or even though it's previously agreed upon, that does not mean it's static. Right, It changes all the time, which I think is one of the amazing and beautiful things about language that that new vocabulary enters our our

syntax all the time. Yeah, and the new structures, new structures enter enter our our understanding of language all the time because internet, right and text messaging. Uh. You know it's it's funny because you'll hear a lot of of linguistic purists kind of uh dismiss or or look down upon any kind of change to the languages. If language itself is a sacred thing that should never change. But that's if. If it doesn't change, it it stagnates. Well,

it becomes more difficult to communicate. I can understand the feeling, but it's just pure personal stubbornness. I mean, the language changed huge amounts before those people got to it in the form that they wanted to stick. Sure, we didn't even have grammatical rules or spelling rules before the printing press was invent to it was all kind of willy nilly, and it changes all the time. And the point we're

getting at is that language is very complex. It's something that as you grow up and as you are exposed to language and as you understand it better from either a practical point of view just because you're using it, or from a formally educated, you know, perspective. It's one of those things that you get more adept at over time. Machines have a very different way of communicating. It doesn't

have this this kind of level of linguistic complexity. It's really when you boil it down to the very basic level, a series of zeros and ones. It's binary code. It's essentially saying off or on, and these these little zeros and ones are instructions more or less. I mean, you

can have it represent something. Those zeros and ones could represent a once translated a a letter in a natural language, or it could represent a numeric figure, and then you would have other zeros and ones that essentially give instructions to a processor on how to push that other information through, and then you get an operation performed. Or you can use them to represent a photograph or a piece of music. Yeah. Yeah,

it really doesn't matter what the data is. It just means that you have to be able to represent it somehow in a quantifiable sense, to reduce it to that numerical component. I think one way that's interesting to think about the difference between natural language and machine languages. That machine language is very good at getting it right. Natural language is very good at making it work. If you

think about that, so machine language. Um, it's going to be very precise, but if a single thing is wrong in a long string, it's likely to come up the whole well, so that you know your entire meaning is lost. You can misspeak a sentence and people can still understand what you're saying. Luckily for us, because we do that

all time. Of course, if you're pedantic like I am, you cannot help but point out when other people make these little misteps and say I think you mean blah blah blah, which, by the way, it makes you sound just as obnoxious as I indicated, and I do it all the time. Well, it makes you as obnoxious as a computer because it seems like you're suggesting that you have a perfect grasp of meaning, which nobody who speaks

language does. Language is sort of natural. Language is necessarily approximate, and it's sort of adaptive and elastic as you use it. I might be as obnoxious as a computer, Joe, but I have feelings and words can hurt. Uh No, alright, So this brings us to programming language. So programming language is what a lot of maybe not a lot of some people who aren't really familiar with computers, they know about computers. They know in general what computers do, that

kind of thing. They've used computers, but they haven't ever gotten into any kind of programming or anything along those lines. Nothing and computer science a programming language to a lay person may seem like, oh, this is the stuff that

computers communicate in, and it's not. A programming language. Is really just a set of rules that exist for our benefit for humans, benefits for for human computer scientists and computer programmers, so that they can follow these rules and create a program that a computer can then execute in

some future operations. Compromise. Yeah, it's it's really just like, this is the set of rules you follow, and as long as you follow these rules and you're really careful and you do them, do the steps properly, then the computer is gonna understand what you want it to do

whenever whatever input comes into that computer. This is all dependent on something called a compiler, which is basically a translation mechanism from from something that humans can process easily, a language that humans can process easily, to the code that lets the computer know what it's supposed to do exactly. So a compiler is that that intermediary step. It's what takes that that code, which to to a lay person

again doesn't look simple. Right someone who is never coded ever, if they were to look at just a page of raw code that someone, some programmer has created, to them, it looks like gobbledygook. It's just incomprehensible. But or you know, you might recognize some words here and there, but there's gonna be a lot of stuff where you're like, I don't know what that's a string of symbols. Yeah, but

but that's poetry. It's it's still it's still a format that humans find much easier to understand than just a bunch of zeros and ones, which if you were to look at a full page of zeros and ones, that would seem to be meaningless to to most people. I would say, there might be some people out there who could look at a page of zeros and ones and because their brains are wired a different way than mine,

is it totally makes sense. But for most of us, beyond like Neo, it's it's pretty difficult to so in that that sense, this compiler changes that programming languages into that that object code, those zeros and ones for the computer to understand what. Then it goes and runs whatever

the operations are in the program. This technology, by the way, was pioneered in the nineteen fifties by someone that you might have heard about, a Rear Admiral Dr. Grace Hopper, who was a mathematician and computer scientist who who was working for a number of decades, like the forties through

the nineties, and largely on compilers and computer languages. One of her nicknames, in fact, was Grandma Coble because she she helped create that particular language, not the planet and not the character in Mortal Kombat, but the common business oriented language. And and one one little piece of trivia

before we move on. She's also known for creating the term computer bug because, as the story goes, a moth flew into a circuit and short circuited the machine, and she referred to it as a bug in the system, and thus we have computer bug. There you go. Before this kind of work was was being done with compilers, computers were very different, and she she happened to work on one of those as well. She was part of

the original team working on the Harvard IBM Mark one. Uh. The kind of computer that that was was it was a punch tape computer, right and even even the later punch card computers used compilers. Here's the steps. I mean, it's crazy if you think about what what computer programs had to go through. If you think that those lines of code are difficult, if you think that's laborious, just imagine what would happen if you were programming computers in

the fifties and sixties. So you would get a probably you'd start with with some sheets of paper that are lined up so that you can actually write your program by hand using a pencil, and these sheets of paper you might even have to erase stuff over and over again. You then would either hand those that stack of paper that you've created that has your program on it to someone who is an expert using a key punch machine,

or you do it yourself. But you would then uh coordinate with the key punch machine to key in your program. It would punch holes into a series of cards. A program might be uh, you know, a few dozen cards, amount several hundred, it could be several thousand physical punch cards. Generally speaking, programs were about as heavy bell as long as the number of punch cards that could fit into

the hopper of a computer. So if you made a program that was longer than that, you pretty much gave up because it just it was too hard to feed it into the computer. Now that you don't just take your program straight to the computer. That's actually not the way it works. It's kind of similar to what we're talking about with that programming language. This is your programming language, your punch cards. You would take that to a compiler. So the punch cards you had created are your source deck.

Starts to sound like I'm playing Magic Big Gatherers. You then run it through the compiler. The compiler then compiles your program and creates an object deck, which is a new deck of cards with holes punched in it, but it really kind of compresses down the number of cards you need that compiled deck. That object deck is your program, which you then can take to the appropriate computer and run through to see if you did it correctly or if the computer gives you an error. Therefore, you can

write much longer programs than you know. If it all has to fit into that one little stack, well, you could definitely, yeah, much you would have. You would have to. You would have to break it up into uh into sections, which might be somewhat challenging, but uh yeah, And I mean,

and here's the thing. You need to make sure you number those punch cards, y'all, because there's nothing like watching a computer engineer just weep uncontrollably after someone accidentally spills over a deck of cards that were not numbered and you have no idea what order they're supposed to go in. Yeah, it still happens a lot today, just kidding. Actually it's

a lot better today. Today we have these programming languages like like See and Java, you know, and and it's all and we have monitors and we have story you know, like before we're talking about you're you're creating these physical programs on physical pieces of paper because there was no other way of doing it, right, So these programming languages we have today and all the stuff like monitors and and everything that makes it a lot easier to write programs like this, But it's still a lot of work

on the front end. Is there a way that we could make computers as easy to talk to as our coworkers so that when we have an idea of what we wanted to do, we can just explain it to the computer, right, So this this is sort of the idea also behind things like the semantic web, where we're able to have the web not only understand exactly what we want, but begin to anticipate it based upon that.

That obviously has other elements of artificial intelligence on top of natural language processing, but it's natural language processing as a as an integral part of that. So this is a really difficult problem, and it's a problem to the problem on multiple levels. We probably started just the absolute simplest thing, like just getting it to understand regular words

and stuff. I mean, it's you know, the components. You can break it down to include things like understanding sentences, so actually understanding the structure and meaning of a sentence, which is not a trivial task. It's really complicated. Uh. There's also the machine translation, which is the whole idea of translating one language into another language, which is really what we talked about in the video, is that idea

of real time translation. Again, very difficult to do if you want to preserve the intended meaning of the original message. There's parsing and tagging, which is all about examining the parts of a statement to determine the relationship to one another. So being able to identify what part of a sentence

is what and how they interact. So if I say a simple sentence like Joe, go get me a cup of coffee, that it's able to figure out all of those things, like what what all of those things mean, and what what I'm actually what the command actually is well, and even that it's a command, I mean, it needs to understand that you are giving an imperative sentence, Joe, you do this, not a descriptive sentence some subject called Joe should got went and got me a cup of corps,

right right, some something that already happened. It clearly has not happened because I sit here with no cup of coffee in front of me. Joe, you just don't take a hint anyway. There's also a grammar induction. So this is the idea of mapping out the hierarchical structure of

a natural language statement. Um. And this is the way I try and explain to other people is it's like your diagramming a sentence, where you're identifying what each part of the sentence actually is, what it what its purpose is within a sentence, and these as we know, like these things can change depending upon the structure of the sentence. An object can become the subject in the next sentence, right,

But a computer doesn't know that. So if you were to say, take one sentence and you define all the different parts of that sentence, for for the the the what they're doing within that sentence, and then you create a brand new sentence using those same words, the computer wouldn't necessarily be able to tell you what was what. It may be basing its decisions on what the previous

s and said, and then everything's wrong and nothing makes sense. Um. On top of that, there's the word sense, disambigulation, ambiguation rather and it's um, I know, I know, I like to add ls into words where they don't belong. Disambiguation, I was being ambiguous. Uh yeah, so yeah, these are words that can have multiple meetings, sometimes contradictory meetings. If I say that something is cool, I could mean one of two things, you know, at least one of two things.

If I'm being really ironic, I could be meaning multiple things. So uh, it's it's one of those deals where it's again something that we humans can figure out, usually through context and tone, but a computer may have real problems with that. And none of these are trivial right. All of these are are challenging issues and U and it's interesting to me how different people have gone about trying

to address these oallenges. Uh. For example, just trying to teach a computer all these pieces, even if you're working with a very limited vocabulary. It's exhaustive to try and get all the different variations on there of all the ways a human could use those uh, those words and those commands to mean different things. And just teaching a computer that just using brute strength, it requires a lot of processing power and it it just takes a lot

of time. Sure, there's a lot of parallel processing that goes on in the human brain and this is this is what allows babies and foreign language learners to pick up on this kind of contextual contextual stuff that's that's going on simultaneously and sorting out all of those different parts. But that's much harder to convince computer to do, or to allow a computer to right. Yeah, yeah, just programming computer to have that same kind of capability is you know,

a pretty hefty task. So in the sixties, the approach was really to create a semantic based understanding to resolve lane which ambiguity. This was kind of the brute force approach teach a computer everything you can about language so it understands all the rules and can follow along and be able to interpret commands. But that's pretty tough to do.

The example I read was was great, talking about using using a language uh interpreter where you could type in a phrase translated into a different language, and then translated back to see how well it did. Um. My dad loves to do this every Christmas. He'll take Christmas carols and run it through the Google Translator and running through about five or six languages until it gets back to English, and then you have to try and figure out what the carol is. I used to do this in high

school with Metallica lyrics. It was great. Um. But yeah, the the example they gave and the presentation I was reading about this was about translating English to Russian back to English and the phrase they used was time flies like an arrow, followed of course by the phrase fruit flies like a banana. You know, those are two different phrases that mean two different things. It's obviously a joke, but a computer would have really a lot of difficulty

with both of those. So they fed time flies like an arrow English to Russian Russian English, and it got back. Time flies enjoy arrows. So time flies, which is like the flies of artists, like, yeah, they travel throughout the eons, Ye bring us wisdom from They also enjoy arrows. They don't enjoy arrow which would mean that they like a show on the c W, but they enjoy arrows. I think that there should be a semi colon in there.

It is it is a imperative. It's saying time flies, so therefore enjoy arrows, right, and what little time we have left exactly, don't eat and drink for tomorrow week. I guess it's like, make sure, make sure you stop to smell the roses and enjoy arrows. Uh yeah, so I didn't he rate again. This shows that this this is you know, it's hard. Those words are ambiguous, like

is ambiguos was right? Um? So, as computers became more powerful in the nineteen eighties, you got to a point where your desktop computer had the same amount of power as those the most sophisticated computers of the fifties or sixties. So now we've got desktop PCs that can do this. The idea originally back in the fifties and sixties was with enough power, we can totally do this natural language processing just by removing all ambiguity as much as possible.

But at this point, once that power was really getting within our grasp, computer engineers said, it's actually the problem is bigger than that. Part of it is that these

this kind of brute force approach isn't scalable. In other words, if you want to have a working vocabulary of more than two words, if you want to have opened that up and have as as many different variations as you can imagine, you cannot take this approach because it would just it would it would be Yeah, it's not just the computer power or the storage, it's the actual man hours of programming a computer to be able to do this. So they started looking at different approaches, including using a

shift to probabilistic approaches. So we're talking about statistical approaches where no longer are we uh, speaking in definitive terms, We're looking at statistical probabilities. So it starts to break down sentences into a different you know, like kind of like tears of possible meanings and then goes with the highest ranking out of all of those. Okay, so is this sort of how Watson works at least in the last stage of what. I don't know if this is

how it interprets UH language to begin with. But when you see Watson competing on Jeopardy, the way it compares answers is it doesn't have a definitive sense of meaning. It gets probabilistic answer saying this was the closest match, it's right, and if it fell below a certain threshold, it wouldn't It wouldn't hazard a guess in the first place.

Right there was that was the rule was that if it didn't have over I can't remember what the number was, UH certainty, it wouldn't buzz in because it wasn't certain

enough that that could be the right answer. And yes, Watson would come up with multiple answers to the questions or or multiple questions to the answer, if you want to be technical with the way Jeopardy works, UH, and then would assign that that percentage of probability that that each one was the right answer, and go with the highest. But you're saying here, that's sort of the same principle applied to assigning meaning to a unit of lane exactly.

So they found that this approach was scalable they didn't have. It wasn't as monumental as trying to to just teach a computer all the different ways to put these words together. And according to some of the research I was looking at, they said that the most successful approaches were what they called supervised learning, which means that the data has been labeled or tagged by humans to give it context so that the machines can start to interpret what words mean.

This is also very similar to what we talked about when we when we talk about semantic web and tagging the heck out of everything that we can think of, so that computers have multiple ways of of classifying words. It's ontologies right there. There this ability to organize language in a way that a machine can have a better grasp at meaning, and to create enough cross references within any body of work or multiple bodies of work so that machines can start to figure out how they relate

to one another. Yeah, yeah, and I like that. Uh. One of Google's approaches for this translation um application, which you know, if you've never gone on Google Translate and used it it can be. It's first of all, it's really useful. I mean, there are there if you're using something like Chrome, you often will have the option to translate a web page written in a different language, whatever

your native language you've set for Chrome. If it's a different language that you visit, it will give you the option to translate the page into English or whatever one of your native language happens to be. Um, that's really useful. It's it's pretty good. It's not, you know, perfect, it's it's never going to be as good as a human

translator who is skilled in translation. In fact, Google has said like, this is a good start for people to help you get the general meaning of something, but if you want to be really precise, you really need, at least for now, a human translator. Sure, some people would definitely argue with your use of the word never there though, UM and uh. One one of them is a relatively recent addition to the Google team, and he's one of the people who's who's work in natural language in general

has basically changed the entire field. And I'm talking about Rake Hurtzwil, also known for his UH. I'd say I was going to say specifically the singularity, but yeah, he's he's one of the the guys who often talks about this idea of the singularity where we we reached the point of no return, and his version of the singularity is kind of super happy, everyone live forever version of singularity a much less terminator than some other people talk about it being yeah, funny you always hear the super

happy everybody live forever or the dystopian hellscape work. And I want to I just want to point out that to the robots, that second one is everybody is happy and less forever. I'm just saying it's all a matter of perspective. I, for one, welcome our robot over lords. But you were actually Lauren talking about Kurtswild himself. Well, I fair enough that that that is absolutely a part of the discussion to consider whenever you're talking about kurtswhile

and and I'm I'm very fond of him. I say this in utmost fondest, But the dude is a little bit wacky, do he He? He can come across as a very enthusiastic and sometimes in ways that you wonder if they reflect reality. Well, however, a lot of his predictions have come completely true, um, which which is not how a whole lot of futurists work. So so i'd say that that's pretty impressive and he's done a lot of work in this specific field. Oh yeah, and well,

I mean he's he's a terrific inventor. He he invented the first flatbed scanner, the first software that could optically recognize any typeface rather than a typeface that had been specifically designed for the software. UM, also the first text to speech synthesizer UM and then combined them together to create the Kurtswhile Reading Machine, which was the first device that could read text allowed for the visually impaired. And that was like in nine seventy six. Yeah, he was

like what by then twenty seven years old? Smart dude. Yeah. UM. He also one of the companies that he worked for, I mean, he was leading the team, but one of the companies that he created also invented the first commercial speech recognition system and that was in seven UM. As of today, he's working for Google, is their director of Engineering, though in a in a practical sense, he considers his

job to be bringing natural language understanding to Google. And speaking of that, Google actually has the at their research at Google website, they have the natural Language Processing page and uh that includes lots of research done at Google and by people associated with Google on the subjects of natural language. So if you want to get really technical, I mean I'm talking about we're talking we're talking about technical both on the level of computer science and just

the way linguistics work, the way language works. Go check out some of these articles and uh, you know, not all of them are meant for the layman, right if you're not into this. For example, a discriminative latent variable model for online clustering might be a little beyond someone

who's just curious about natural language processing in general. Uh. And or a data set of syntactic ingrams over time from a very large corpus of English books like these are These are the types of things that you'd have access to if you want to go and look at them on this research page, and it will tell you a little bit more about the different challenges that engineers

are facing when coming at this problem. And also we should make it clear not everyone is approaching this exactly the same way, and there are lots of people working on it, so we may see different implementations at different levels of sophistication in various fields before it becomes a universal.

Oh absolutely, which is one of the cool things about the field, really, and I mean especially because this this technology has application beyond being able to yell at machines or or I mean you can yell at machines now, but being able to yell at machines and have them respond appropriately. I want my toaster to feel ashamed when

it has burnt my text. Yeah, that's that's true. You can yell at a machine now, like you can tell a machine, probably in some way or other, that it did something wrong in a very rudimentary way, with some pre programming. What's really important is to be able to subtly shame that and and through with through a very nuanced way, belittle it and make it feel insecure, and then be able to take a picture of it and upload it and make a meme out of it. Important. Okay, well,

well you guys. You guys are joking. I hope, um otherwise I'm kind of upset. But but but this does tie heavily into artificial intelligence. I mean, if we're gonna have convincing robot buddies, they have to be able to understand us without us having to to feed them punch cards. Yeah. In fact, artificial intelligence that you know, you see, you would see a lot of improvements as we improve computer's ability to handle natural language processing. You'd see improvements in

artificial intelligence skyrocket as well. We've also seen this in other areas that are related to but are not directly connected to natural language processing. For example, you mentioned that Kurtzwild had created that scanner that could that could recognize different type sets, different fonts. Uh. You know there's that security uh measure that was around for a really long time but has officially ended capture. You know what I'm

talking about. Capture has officially ended because we've gotten to the point now where machines are able to recognize characters even after being distorted about as well as humans are. I'd imagine this is thwarting way more humans. Yeah, I was about to say, I'm sure that they can do it better than I can, because I am terrible at

those things. Dear a B. I don't know. The interesting thing to me is that the relationship between making captures harder and harder to recognize, because if you remember, you know, maybe six or seven years ago, these captures were a lot easier to read back then, and they've just gotten increasingly difficult. The reason they've gotten increasingly difficult. Is people started building programs that were better and better at figuring

out what those words were. And once you get to a point where it's very hard for the average person to figure out what the word is, much less a machine, it's no longer a useful system in the first place. Just as you were pointing out, if we can't figure

it out, then it's not good for us. But the but it was an example of how artificial intelligence was improving in the field, and even the people behind capture we're saying, I'm not upset because it means that we're getting better computers it Now, granted, it means that we have to design better and better security systems, but that's a small price to pay for the fact that we are advancing artificial intelligence through this response. And then you know,

it's it's this constant response. It's response from the people who are trying to break that security system to the people who want to make it more secure. Now, granted, for a long time, the easiest way to break the capture system was to pay a bunch of people a tiny amount of money to just do it for you, like to just physically look at the screen and type it in for you, which, sadly that's the way most

of the capture systems were broken. So another application is and this is a really important one, big data, which we've talked about before. Yeah. That that's the kind of convenient catch phrase for the fact that every two days we're creating as much data as as it took humanity to generate from the don of history up until two thousand three, Right, So we are generating and and normal. It's impossible to exaggerate how bunch of data we are

generating on a daily basis. It's crazy. I mean, you just just take a look at these statistics or the little fact from YouTube about how more than a hundred hours of footage gets upload YouTube every minute. So a lot of this data probably wouldn't be useful to most people for any reason at all, not a single parts, not on the individual right, but an aggregate. It can be really really fascinated. And what's crazy is you can see in a chaotic system a rise of patterns that

at least computers could see this. We couldn't. It's just too much information for us to be able to process well. But the thing is that computers can't either because they are unable to contextualize all of this stuff that's written in natural language, right, so they would need to have

an ability to understand natural language. Not even if you're talking about a system that just collects data from various like app that have been optimized so that computers can at least sort through stuff so that a human can later come in and take a look at the data and make meaning out of it. Even then you can

start to see patterns. But just imagine what's possible once the computers themselves understand what is they're looking at, They're not just classifying it based upon tags that we've put in, but can actually have this natural language processing that allow it to make those draw those conclusions itself. Oh right.

And and that kind of combination of the understanding of big data with artificial intelligence is is what people like Kurt's while talk about being the next step towards this singularity um, which which is, you know, computer is becoming as colloquially intelligent as people and humanity therefore progressing to a new technological state of evolution. Mm hmm, yeah, yeah, I mean it's it's exciting to me because I like this idea that we're going to learn more and more

about ourselves in ways that we had not anticipated. Maybe we eventually learned that, uh, ultimately, when you really dig down, we're no more complex than your typical ant colony. Maybe that's what we learn eventually when you when you really look at the big, big picture. Or maybe we learned that there's something really complex going on and that we can actually use the information to help people in a measurable way. That would be fantastic. Yeah, and you know that.

I think the idea is that if we if we can teach you know, high processing capacity computers to really understand us, then we can we can create better versions of ourselves. Speak for yourself, Lauren Jonathan three point oh, which is the current version I'm running, is pretty darn awesome. Now that now you had that whole vision thing corrected, it's pretty you know, I'm happy Joe and I are

programmed to not argue. Yeah, alright, Well, now that we've well know, we established that when the bottom line is, it's still very much a challenge for compute its to understand natural language. Is a challenge for us to be able to teach them how to do that. But it's it's a process that's seen a lot of progress over the last few decades, and that progress is increasing. It kind of a I don't know if you'd say an exponential rate, but it's really amazing the developments that are

coming out of all these different companies. And again, the implementations we see today may seem pretty simple on the surface, but it took a lot of work to get there, and it just it's just kind of a stepping stone. It's just a stepping stone for for more incredible applications in the future. So that kind of wraps up this discussion, but we wanted to remind our listeners that you can get in touch with us. Let's know, what sort of

futuristic topics you really want to hear about. You can email us our addresses f W Thinking at Discovery dot com, or drop us a line on the numerous social networks we frequent, which include Twitter, Facebook, and Google Plus. With the handle f W Thinking and don't forget go to FW thinking dot com to see all the video, to read the blog posts, to listen together podcasts, check out all the other information we have up there. It's fantastic and we look forward to talking to you again really sick.

For more on this topic and the future of technology, visit forward Thinking dot com m brought to you by Toyota. Let's Go Places