 Okay, all right, okay. Here we go. It must be two o'clock two o'clock rock on Friday. This is Think Tech I'm Jay Fidel and that is Ethan Allen say hi Ethan. How do you get me here again Jay? He's our resident scientist here on Think Tech and he comes up with stuff and then I get a chance to ask him questions about it and today we have two areas of thought literally thought one is You know an MIT article MIT technology article and you can take a look at that It's very good. It's in the technology review of MIT. It's under www.technologyreview.com Anyway the article where you know running off his first computer to match humans in conversational speech recognition now we've been Seeing and working with Dragon dictate and the like On computers for as long as computers at least personal computers have been around right where we can speak to it and I could listen and now We have Siri and and the comparable program on the Android and it couldn't understand a lot of what you say But it's not perfect and you wouldn't count on it And I gotta add that I use something called speak right and speak right is a non-machine way of doing this where I may a sound file and dictate a letter or something and Through technology that sound file goes to somebody in the Philippines or India and they make a perfect, you know translation of it and It comes back to me just perfect. Well, not quite perfect, right? Not quite perfect. It's only sound But this MIT article suggests that we are now learning to deal with more than just strictly the sound the phonetics of it Right, we are learning to incorporate other factors and vectors and and signals that people use when they communicate and thus We can we can learn a lot more about what somebody wants to say and we can make it perfect transcription of it Using this new technology and that means that we will be able to command computers to do things with a hundred percent You know reliability. Well, okay. Well, tell me what's happening. No, it is true We there there have been some speech recognition programs. Yes, there's the ringer app on the phone I don't know if you know ringer if you and I get ringer We can talk on the phone and then get a transcription of our conversation back from ringer Takes both of them again has somebody sitting listening and they transcribe with probably 95 percent accuracy and Yes, now by taking a lot of the Non-verbal but but audio cues in the speech of the pauses the us The sort of little sounds you make what do they call them? Background sounds basically, but it's voice right and a machine could you know characterize that right easily Exactly, so if you're talking I say, uh-huh, uh-huh, I'm encouraging you to talk more. Yeah, if you're talking and I say, uh-huh I've now sort of said there's something wrong. I'm not understanding you right and it's a very subtle distinction there early And you don't have to train it like you had a train and drag and dictate way back when yeah It'll know from all that you are and say and I suppose it can catch somebody with an accent, right? I come from another country second language or all that I would guess they have to give it a big training session first though with a lot of a lot of this kind of stuff going on That has to pick pick up and do it through adaptive learning. I Mean, I don't know how they do all this fancy stuff because it is very high level I'm wondering if adaptive learning really ultimately will be necessary because I can I can feed it the accents I can have oh, this is a guy with you know a Pakistani accent and we know what that means So we're going to figure that out, you know It's a lot of data but theoretically you could put all that data in this program and it could you know Make choices about where this person is from and therefore how he says certain words in English or any other language, right? I mean all of this is possible What what interests me is not so much to pick up, you know the words to find the language You know linguistic context of it Or even you know the us and the odds and how that might affect the meaning that he's trying to convey what interests me is the Context the thought context and this is really directly related to our Larger discussion tonight about science looking into thought right, right? So if if the machine figures out by some key words that we're talking about widgets, right? Then you know if I use a word that's that has a one context around widgets But another context around apple pie the machine knows we're not talking about apple pie We're talking about widgets and it's going to pick that it's going to choose that meaning rather than apple pie meaning Exactly, I mean I've been impressed just really in the last few weeks My smartphone has started transcribing voicemail messages that are left for me. Yeah, take it in so so they show up in print and Sometimes you're gonna get charged for that. I know you are Sometimes they're very active but other times it does it just obviously has no context And it's just sort of it's almost like word salad that it's doing, you know It's just picking up on words and trying to figure out words. Yeah, thanks It hurt those trend those translations are really not all that good Right enough to get the idea But you have to build the context in your mind, right? And it would be far better if the machine knew what the company figured out what the context was and you know How are we going to know when this is really succeeded? Yeah, when the machine start making puns Right huns. Yes, when they start playing with the word is then you'll realize they really do when a machine comes up with you know Time flies like an arrow fruit flies like a banana Then you'll you'll know the machine writing what you're saying that the machine really understands what what language is all about That leads to a whole new thing. You know, it's not just the puns It's going to improve what you're saying, right you said it You know in a klutzy way the machine knows what you mean the machine is going to clean you up Well already I mean, you know your edit track changes Tends to pull everything out of passive voice and put an active voice because it believes the passive voice is bad And let's see that if this will be oral. Yes, right? Well oral to writing, right? Yeah, so the eye G We should all live so long to see this come to fruition But it's it really we you know We're in a place where a lot of data a lot of processing you'd be putting a little a little thimble Yeah, and it can do all this processing multiple languages multiple accents multiple contexts And it can be as good as any human or a composite human all humans can understand all other humans Isn't that something well one of the things they the point they made in this article, but was Both machines when they transcribed human speech and people when they transcribe have a certain error rate now They've gotten machines that are about as good as people but the types of errors they make are very different The machine has trouble with those little little sort of background almost non-verbal things um the uh's and People that doesn't give us pause when we know that perfectly well And we're able to deal with those and machine has some trouble With sort of small words that get dropped in thought uh and Doesn't sometimes get those in perfectly sounds so exciting. Why why is it exciting because the payload is so huge It's the tower of Babel. We still have that in this world But maybe can all come together and everybody can understand everybody the question I put to you though is Do we have to teach this program the one in the thimble do we have to teach this the all these rules These sound rules and thought rules and context rules and language rules Do we have to teach the machine this or can the machine learn these rules by itself through artificial intelligence and What you called as something else before where you the machine learns? I mean, is it really possible for the machine to learn like that? That's pretty complex Yeah, that that is a whole area that and I know very very little about it But the artificial intelligence folks are getting machines to learn in ways that machines could never learn before They it's not just a matter that you had to plug all the right stuff on in and machine simply could do sort of high-level Computation on the you plugged in machines are actually now able to take experiences and Interpolate and extrapolate from them apparently They have figured out how to make the circuitry clever to do that I don't know how that's done, but yes, they're learning in new ways, which means yes with a limited data set they'll be able to start Using language now in new ways. Yes when when the machine start writing poems Good poems beyond puns well poems have puns right? Yeah, but I mean that would be pretty thrilling to see your language sort of Fly your language expanded your language improved your whole thought process because a pun or a poem Gee whiz it would know what's in your mind more than what you know is in your mind I know it would take us all to a new level of thought right? Yeah when your smartphone starts expressing your thoughts better than you can Express your thoughts. Yeah Yeah, and well and then you know I mean I hate to say this but we we owe you and me we Ethan we always get into these things now the machine is looking into your mind It is finding thoughts in your words that you didn't realize you had maybe it is Learning more about what you were thinking than you thought what you were thinking and therefore it can find things Perhaps that the government doesn't like right? Yeah Police are coming to you now Yes, it can be finding patterns in you in your speech that reveal ways that you're thinking Patterns you will be completely unaware of and those around you will be completely unaware of because yes we have these amazing capacities now to gather these monstrous big data sets and Track them and use them in interesting ways Stakes are high it is you know because it's a lot You know easier in this world to just articulate some words Instead of write a write a note to someone You know I would like to be able to walk around all day I mean it's just like like the cop, you know who wears the microphone or the camera and and speak to people Speak to my whole world. It's sort of like email on steroids So, you know, I have a friend in South Africa. I can talk to him I can say I want to talk to my friend in South Africa bingo, you know, and I'm sending him Oral message I'll have to write it out. Yeah, and we have this coming and it leads me to another really interesting question So we have we have Alexa here Alexa. What is artificial intelligence? Sorry, I can't find the answer to the question See she should know she should I mean I'm disappointed actually Alexa. What is intelligence? She needs training So but you know that goes to my question we have Alexa Alexa's coming along You know every Friday Amazon sends out a rather lengthy email telling you all the new things and Alexa Then I just got this fire stick thing for my TV and it has the same kind of technology You push a button on the remote and now you can talk to the television and tell her what you want in terms of movies news Music, you know, just talk to it and it understands you really first time around because it knows what you're saying You know against the database of what it has that's quite interesting And then Siri some people swear by Siri some people swear by by the you know Android sound sound interpreter Microsoft question I need to put to you and this is really important is in our world today our world of science our world of MIT our world of Amazon and Alexa Alexa I was only kidding She might have a cover. Thank you, Alexa And then you know Siri on Apple and the Android Where is this going to happen? Is it going to happen in the laboratories of MIT? Is it going to happen in Apple? Is it going to happen in Samsung in Korea? You know Where is it going to happen? Where where is the big drive going to take place? Yeah, I don't know. I mean it ties back to our earlier dialogues about about self-driving cars, right? I mean there again. They're doing the same thing They're making a lot of decisions about the world about how to interact what's happening. They're predicting and behaving appropriately Yeah, where is that happening? I mean Tesla is gathering data from all of its new cars all the time those cars all have always on internet connections and Tesla is gathering information from all those sensors. It has Millions probably at this point billions of miles of driving data on on its cars That's what you gotta have and and from that data set Yeah, it can probably tell you a whole lot about how you how you drive how the guy beside you drives Yeah, but you need that now and that is the current challenge in information technology We know we have big data. We know that from science pure science You're a scientist You know how how big those spreadsheets and data tables, you know can get but we have to tell all the data about a given Subject whether it's driving or simply understanding what I'm trying to communicate to you and put that somewhere in the program And then we have to connect it up so that it is interpreted immediately And so that it learns it learns when it makes a mistake or something new happens And it needs to know something else it is able to learn with artificial intelligence, right? so that then brings us to the point of Do these computers have what you call theory of mind, right? Right and what and what is this conversation between you and me and the people have to do with monkeys because we did say You know what from machines to monkeys and so we need to know but don't answer that yet We're gonna take a break and come back. We're gonna connect this all up with monkeys. You'll see Ethan Good afternoon Howard wig the host of cold green for think tech Hawaii comm I appear on Mondays at three in the afternoon and my gig is Energy efficiency that is the cleanest quickest safest way that we're going to get Down to 100% clean energy by the year 2045 and that's not that far away, but it is doable. Just join me Mondays three o'clock Aloha Hi, I'm crystal. Welcome to think tech my show clock talk normally airs at 10 o'clock on Tuesdays But it's gonna change to 11 o'clock so don't miss it It's an hour later. You can sleep a little longer come with me and engage in some sensitive provocative discussions on everything It's all good. All right, women's issues things that people don't dare talk about. We want it on the table. So join me Wow the plot thickens Here with Ethan Allen and me and likable signs on the given Friday at two o'clock clock So now we're gonna connect everything. We've talked about before should we dare to do that in terms of machines? Recognizing speech and thought and expanding on what your thoughts might be We're gonna talk about new theories of thought in general between human people and monkey people and Every other living thing I suppose so we're talking about theory of mind Ethan I know this because you told me about it. So why don't you tell the people to so theory of mind is this Idea that others have beliefs and feelings and thoughts about the world and these may not necessarily be the same as your own beliefs and Feelings and theories about the world. So it's it's often demonstrated by a so-called false belief demonstration, so if we can look at this on the table and Okay, great here. We are so Let us just say for fun that I have this piece of candy and I've decided I wish to hide this piece of candy So I take and I put it under this cup Now Jay, you're sitting there watching you're sitting there watching me do that You know it I get up and I leave the room Robert floor guy comes over here Pulls it out there and sticks it under the red cup Now when I come back in the room and I and you know, I want that piece of candy because I walked back And I'm now hungry. I want that piece of candy. Where am I gonna look that piece of candy? Under the cup under which cup. Oh the tea cup right the white cup, right? Yeah, that's where I put it But you know perfectly well That I will do that. Yes, you understand. I saw what you did So I looked through your eyes, right? And yet you know perfectly well That's not where the thing is you know that it's here You could tell me indeed that it's under the red. So I know both ends of it I know I know what's in your mind right and I know the reality right and that that Abilities know what's in another's mind is theory of mind That is it says I believe that others sort of like me have a mind sort of like mine And it will behave in that same way I would have been fooled too if I had walked out of the room. I haven't seen the switch being made right okay, so you have mind Theory of mind has been one of these sort of psychological treasures that we have always considered unique to people. We've always said It is only people that have that no other animal has really demonstrated theory of mind until Just recently this set of researchers Got some chimpanzees and orangutans and bonobos Little chimps essentially they did the same kind of false belief experiment and they tracked The gaze of these animals now that it's been a whole new technology They use with infants and showing how much infants know so watch where the infants are looking and you can tell a lot by How much how long the look and where they'll look okay, so you have described that I see The the candy under the teacup And I know that the the candy has been put under the under the plastic glass So I can appreciate you who left the room and didn't know all that I appreciate your state of mind Okay, now make me a monkey some people think I am a make me a monkey. How does it differ with a monkey? well, that's that's the thing apparently it really doesn't apparently the monkeys know that if you have Now what I say, this is a banana, you know, but I have I have gone and I've you're hidden here And you decide like you want that banana, right? So you go and you take this and take away where you know and you know perfectly well You're sitting there in your tree and when I come back in you laugh at me as I come up here Where's my where's my banana? My monkeys can laugh. They have humor, don't they? They can't make puns We don't think so that may break down to But again, this this does get us to this area of understanding The context understanding the other's viewpoint basically so the monkey has the same ability to understand What's in your mind right even when you're wrong right especially when you're wrong that I do all right But we and monkeys were the same that way. Yes, and yet again, you know For a long time people regard to themselves as very Special there were animals and there are people and lots of people still talk about that those are these people and there's animals I don't like that distinction. We are we're animals every bit as much as monkeys and ants and birds and fish Okay, you know, so we found this one little sort of layer. Yeah of consciousness That's common to human beings and monkeys that this is a discovery Yeah, well used to be that we thought there was this sort of giant gap between all animals all other animals and people right and we were The tool users the tool makers the thinkers of language users blah blah blah little by little that's breaking that's been been eroded and eroded and eroded They found more and more animals use tools in various ways They've now got the you've seen the videos of the caledonian crows who make tools actually they bend wires make hooks to get things out More and more animals use very sophisticated language I mean if you look at the work with the bees these bees are communicating How far to go what kind of food source that you're gonna find if you go this direction this far in that direction Whether it's a worthwhile food source and how many bees should go off to pursue that. I mean it's very sophisticated communication What's left? I mean when you Was one of the few things that was left us. What is left? Yeah, very little very little This is why more and more it's clear to me we are I mean we are just animals Yes, we're we have a little bit bigger brains a little more sophisticated wiring up in our heads and a lot of other animals But man's not much it's not really at this big qualitative difference anymore It's more of a yeah, we're a little better at this task a little better at that task Let me submit to you that maybe we're you know, we're looking more carefully and we're finding more similarities But at the end of the day You know, they're not gonna animals are not gonna be able to do philosophy They're not gonna be able to read books In any great detail, so they're not gonna be able to build computers Or build the speech recognition program now our computer is gonna be able to do those same things I mean, you know that they now Obviously read books right there are book reading machines basically yeah that read books Can a could you hook up a book reading machine to discuss Faulkner's Works in an intelligent fashion. Yes You could so basically and you're saying you know that the the touring test Elvin touring one of the early computer guys said the day is gonna come at some point when if you're behind a curtain You won't be able to tell whether you're talking to a machine or to a person and when that day comes when you can't can no longer tell Whether you're talking to a machine or person the computers have basically achieved humanity in some in some very very scary And we're basically Outlines are getting out there that we can do that. Yeah, and and this MIT article really Shows yeah, you can do this in casual conversation. I mean it used to be one of my favorite things I read it years ago. It was the dear doctor program at Stanford This is back in the back in the day But they set up a program that basically just took what you would type into a computer And it would type it back to you as a question and they sort of set it up as a psychiatrist so you'd walk in and you say I was having dreams about my mother and Computer type back to you what sorts of dreams about your mother? Oh, Liza Liza. Oh, I dream. She was killing me You know, why do you think you dream? She was killing you? Yeah, and what they found was this was an incredibly popular program They actually had to take it offline because basically people were just soaking up all the computer power This is a DOS program. Yeah, right. It was very old. Yeah, this is way back when yeah, you know And that that was very now imagine you've got something someone who can talk in a very real human language to you Hooked up to a machine. There was a movie about this because some fellow fell in love with his computer. Oh, yes, right Yes, and you know, I mean, I really think that's coming where we live in a nuclear world We're in silos. We don't have friends or people who listen to us They don't have time to listen to us so we could get a little but it's coming. It's within the next five years I'm sure that you can get up, you know, let's call it Ethan Okay, and I can talk to Ethan Ethan will listen to me. Ethan will respond to me Ethan will take my thought process higher. He will entertain me. He will educate me all that Yeah, so you want what you want to something like Alexis there, but but sort of warm and soft and furry, right? Yeah, I think you can get Alexa Alexa with I'm gonna activate it. You can get Alexa with a male voice too. It doesn't have to be female All that but you know take me back to the the nexus we were searching for Where how does this connect with the monkey who has maybe a greater level of consciousness than we thought Well, my question to you is are these computers that are able now to to Converse with us in human speech Do they have theory of mind? Can they have theory of mind? Because are we like them or are they Do they not and I guess that will be the ultimate test with you Sometimes we'll have we'll have to hook some of these computers up and see if they understand false belief basically I Can't do that for you 2001 space I see cut the mind exactly, you know, I don't think we're that far away from that Oh, you can you can if we can do this, you know, if you if you do it mechanically You may never get there, but if you do it by artificial intelligence And somehow allow the artificial intelligence to create this Theory of mind then seems to me that you could do show the computer could be aware of its own consciousness Yeah, yeah, and be aware of the fact that there are other intelligent machines around and Their views of the world may differ from its view of the world because of their experiences Yeah, and so for that reason the identification of this layer of consciousness in the theory of the mind, you know Research with with monkeys becomes important right because it means that we you know We can make the same analysis with humans and and with machines and have a machine Who then talking about ethics you're talking about all kinds of sophisticated human choices there. Yes, exactly Exactly there's all kinds of different sort of moral ethical Huge horizons that we're facing now issues. Yeah, yeah So the Gen X people the Millennials they can look forward to being essentially either in competition or supported Machines you bet better hope eyes as a mobs three rules robotics You know three laws robotics come you know come into play here pretty quick, right? Which are that you know a robot Can ever hurt a person? Well, you know, it's funny how science fiction becomes reality. Oh, yeah Here's another example right exactly well all I can say Ethan is I always learn a tremendous amount from you and we started out I was skeptical that you would be able to connect up the Theory of mind with the MIT article on speech recognition, but you have done it yet again Well, thank you. It was a pleasure and couldn't have done it without you It's always fun getting the great stuff that you send along and thinking about it thinking like oh that reminds me of this Not only great to be a scientist. It's great to be around scientists who don't mind answering your questions Thank you so much Ethan Allen. Thank you. I have fun to be here