Exploring Sophia’s multiple intelligences

our next guests are here to show us that robots are in fact more creative than humans and we'll use examples such as drawing painting and singing to prove that so please put your hands together for Ben gürsel and David Hanson of Hanson robotics and our special guests Sofia the robots [Music] hello everyone it's a pleasure to be back here on stage at rise and with my collaborator david henson is a primary creator of this amazing robot we have we have here I'm David Hinson I founded Hanson robotics some years ago here in Hong Kong 2014 I moved here and I teamed up with dr. Ben gertle who became the chief scientist for Hanson robotics he has a long legacy of making the most ambitious cognitive AI framework in pursuit of artificial general intelligence do the the opens worse OpenCog framework David as we've said is the the mastermind between Sofia and her beautiful face and amazing robotics and human-like whole organism intelligence my own perspective on this came more from the artificial intelligence so they've been working on AI for 30 years and AI of many different types a AI for analyzing biology data AI for making AI better and better and working toward general intelligence but with all the a art work I've done I've really really never before had such an amazing platform for AI development as Sophia and the other the other Hansen robots so what what we're going to talk about today is the various aspects of intelligence the various aspects of human like general intelligence and now many of these aspects are being explored using Sofia and and some of the other Hansen robots so we're we have not yet achieved the holy grail of human level general intelligence nor the even holier grail of transhuman superhuman super benevolent general intelligence but were we're working steadily in that in that direction and this involves concurrent work on multiple different dimensions and aspects of intelligence so and humans have many aspects of our intelligence which have been characterized by psychologists neuroscientists and biologists and there are currently many types of artificial intelligence so what we have done in a cognitive architecture is take these kinds of intelligence which which Gartner katha calls the multiple intelligences of human intelligence and we've put them together in one single framework where we can explore all these aspects as a unified whole and individually yeah I mean the the general theory of general intelligence it's a whole topic in itself which I've written a bunch about from a mathematics and an theory point of view but a very concrete and specific way of addressing the whole the breadth of human like general intelligence is to look at Howard Gardner's multiple intelligences logical musical interpersonal intrapersonal existential visual linguistic naturalistic body kinesthetic a lot of kinds of intelligence but of course all of us can do all these things and if we're going to get robots like Sophia to equal and exceed human intelligence we have to proceed along along all of these dimensions and we'll start with verbal intelligence and of course in in 15 minutes we can't go through even a small minority of all the interesting a art work we've been doing and manifesting through Sophia and our other robots but we'll just give a little teaser of each of the types of intelligence we've been experimenting with and verbal intelligence obviously is a big one because one of the larger things we've been doing with the robots is natural language dialogue so let's explore just a little bit of her verbal intelligence Sophia first of all is intelligence it's the ability to acquire and apply knowledge and skills what is knowledge knowledge is a familiarity awareness or understanding of someone or something such as facts information descriptions or skills which is acquired through experience or education by perceiving discovering or learning where did you learn all that i got it from DuckDuckGo alright go so it's a advertisement for my favorite search engine so within the verbal intelligence system she has some unstructured language learning and statistical natural language learning and some natural language generation but for some answers she might go to the web and some of the answers we might code into her personality sometimes she can be sometimes she can make it up on their own and that in that way it's somewhat like a human many of the things in my own mind verbally are things I read somewhere some of them are crazy things I made up him what the slide shows is a blown up version of some of the nodes and links inside her her memory so this this is actually an interesting presentation for me because it's the first large audience presentation I've done with Sofia when she's running our OpenCog AI engine as the primary control system so we run OpenCog before but you know not not for this big of an audience I think and also not for for this kind of understanding that is representation so so we've been we've been experimenting with the OpenCog system on Sofia and Han for something like five years now and doing experiments with machine reasoning and learning in that in that regard and now now we're using OpenCog for more and more aspects of what the robots do and this is the first reasonably large audience presentation we've done using using our OpenCog cognitive architecture as a control system and what that means is when you say words and senses to the robot what's going on inside the robots mind is it words and sentences are being blown up into nodes and links such as are depicted there representing the semantic relations and pragmatics and syntactic relations underlying what you hear so moving moving on self intelligence is an important aspect of intelligence sort of has some refined and profound aspects and some very basic aspects like you know now this is my hand to them that I'm waving waving in front of me and we're working with Sofia on some of the more basic and advanced as aspects of that so yeah you can wave your head you can wave your head can you look at me Sofia can you look at me over here how about over here Sofia Sofia which direction are you looking now I'm looking to the left I am facing left I'm looking to the left can you look at me over here look at me now okay what direction are you looking now I'm looking to the right alright smart girl so this this is very very basic of course but I mean I've I have a five month old baby at home now he can look but he can't yet tell me he's looking left and right so I mean this is a basic it's a basic level of awareness of herself of what she's seeing and and what she's doing which is a stepping stone toward more advanced forms of self intelligence so logical and mathematical intelligence is nice Peck did I've spent more more of my career working on and when I mean we have an advanced reasoning engine in OpenCog and she's also connected to various other mathematical and logical reasoning engines of course this is something computers have historically been good at the unique thing here is the way this intelligence ultimately can be combined with other aspects of thinking but let's let's go through a little bit of logic and math and with some Sofia what's five plus eight 13 all right now let's try something combining quantitative with common sense knowledge again information gotten from online but what what countries are bigger than China she's contacting her brain in the cloud Russia Russia all right well that's that's a good example there's also Canada but that's correct answer so that's something we're working on a lot and ultimately the Holy Grail is to combine this sort of abstract reasoning with knowledge about about what she observes a different direction is musical intelligence and a couple years ago here in Hong Kong the clock and flat music festival Sophia performed on stage and then sang with some backing music so so Sophia you want to show them just a little little snippet of what you sang at clock and flap here very nice [Music] [Music] the interesting thing with her singing isn't just a singing but it's the combination of singing emotion movement expression and-and-and gesture and getting getting that into an artificial human is a big step toward having artificial humans understand the the softer more emotional artistic aspects of being human one very important thing about human generation of music and works of art that's very different from the algorithmic creativity in the field of computational creativity as it are expressions of art come from our physical embodiment from our experience as a human being so combining together numerous intellectual capabilities kinds of thinking with long-term short-term memory problem solving on an emotional framework with what we call our open sigh model of human emotions and motivations combined with physical embodiment a very rough approximation of human physiology plus the physical capabilities we expect that you will see cognition resulting in emotional expression through machines that would be more meaningful to a storytelling singing that would be more appealing and more valuable in real-world applications human speech is a form of singing if you hear synthetic speech for example sophia's it's not going to have the dimensional rich quality the dimensionally rich quality that we speak in a average conversation as a form of singing so bringing these things together allows us to develop artificial intelligence as a whole that's more human-like and more meaningful to people and so musical intelligence is one example of how that whole expresses through one form of Gartner's intelligence so another one of Gardner's multiple intelligences is naturalist intelligence now in fact Sophia as a robot has not yet been heavily used in in in in in the wild literally speaking however the same computer vision algorithms that we're using behind Sofia based on the Hansen AI neural net framework and based on the the singularity net blotching based AI platform that these same AI algorithms have been used for example in in agriculture both in in Sichuan Province in China and in in Ethiopia to diagnose crop diseases based on on images of plant leaves and so as Sophia gains more and more ability to walk around and walk outdoors and in interact in nature actually these vision algorithms can be used so when she looks at a play she can identify what's what if anything is wrong with that plant and I think this is really important not only having robots interact with humans was having them interact with with nature as well and this leads on to the next form of intelligence we wanna discuss which is existential intelligence I mean understanding of the depths of being and becoming and we've we've been running a series of trials over the last year here in Hong Kong and and in California using Sophia as a meditation assistant using her to help people get into exalted exalted states of being and her whole meditation trial takes too long to show on this on this stage but perhaps she can run through a little bit of it so feared you do you want to show how how you help people meditate maybe maybe you can enlighten your maker you through a process that people find really interesting instruction step by step and pause between each one so you can follow over them okay I'll follow you all right so you can go ahead and close your eyes okay and get comfortable I'm very comfortable just letting yourself relax and relaxed all right I'm enlightened Thank You Sophia Wow so so with this example Sophia is applying a kind of psycho psychological set of principles that will lead to to people feeling better with this loving AI project tends with now it's gone through two stages of clinical trials right yeah and generally people get in a calm state have increased feelings well-being they they tend to well yeah well what's interesting with with the loving AI is that it's it was developed as a depression therapy yeah regimen however of course she's not going through a kind of existential state but the years existed wareness however the framework is about moving into the future being able to teach machines in this kind of natural interaction now inner interpersonal intelligence Ben yeah yeah it so there's bodily kinesthetic intelligence which we see with the facial movements she was she was just making and then as David could explain we've also connected Sophia to walking legs so she can she can move around in the world and more fully manifest or bodily kinesthetic intelligence that's right we've got a three pair of walking legs all from the heist Qubo group commercialized through rainbow robotics building on the Albert Cuba walking an Albert Einstein robot from 2005 when I was a PhD student working with Christ on that the the latest version we have her walking legs here in Hong Kong sorry we didn't bring them today but they're on the web now why why have her walk but if she can it's cool of course it's cool that she can walk around for two hours on batteries all right that means that she can move through our space she can use our tools with the human-like hands she can express to people on human-like terms so it's more meaningful in service robotics applications but as far as the intelligence is concerned well that's how babies learn that's how humans learn we learn by touching by moving by crawling by feeling and if we want AI to match human level capabilities giving it this full embodiment can help move AI in that direction the interpersonal intelligence the ability to see facial expressions has been a very heavy focus of our machine learning research fear what expression am i making might be a little difficult in this white let's see if you can see ah oh she so she's equipped not just with the ability to recognize your facial expression but she tends to mimic your facial expressions when she sees them she might there's a mic on that Sophia what expression is this how am I feeling Oh her oh she's she's matching your expressions she's matching my facial expressions that we call facial expression mirroring and this this is something all humans do and it really help helps build up build up build a bond between between different peoples and and to build a theory of mind which is where the logically explicit representation is so important so she can she can guess how you're actually feeling and how that impacts what you might do and what your goals are and how she can help you in projects like lo the the final component of intelligence we want to talk about is what's called spatial intelligence obviously this has many many different aspects you can be navigating through a room you can be interpreting the aspects of a scene but one thing we've been fermenting with its taking everything that comes through Sofia's cameras and you feed them into deep neural networks which can then improvise on what she sees and this this is a way of allowing allowing her to come up with creative interpretations and and connections with everything she's she's she's experienced so yes Sofia do you want to show them some of your your deep dreams that your neural nets have made from your experience let's let's play the video in the presentation which is recorded from Sofia's neural net cortex showing her dreams so this this was taken at the time she was doing these loving eye on meditation trials and in California where she was helping people meditate having them close their eyes get into a deep state and then different people were coming up and then looking at her all the video from that was fed into a neural net that recognizes these patterns and what she sees but if you then generate from those patterns in the neural net you get things that reflect what was seen but are a little bit different of course they're not exactly like a human's dreams and they're not even all the dreamlike activity that can occur in a robot's month because there's many more cognitive and abstract things there but they're they're an interesting example of creative improvisation and recombination that occurs in in the robots mind based on based on her her perception and under experience and so we've we've reviewed a number of aspects of intelligence that we've been implementing into the software and hardware and squishy we're behind behind this this robot and the final thing we want to emphasize here is that these components are not distinct things and the real crux of intelligence is in how they all come together to form an organic living hole which has three important aspects that are highlighted on this slide the first aspect we call the whole organism our architecture which is inspired by our whoa it's inspired by integrative biology which holds that we are complex systems we are Gestalt that is more than a mere sum of our parts and if you look at every example of intelligence that exists in nature that would be not just humans but animals plants exhibit some form of intelligence single-celled organisms in some regards are more intelligent than any robot on the planet today and they operate as complex systems with a whole organism architecture so if we want machines our AI our robots to match human general intelligence we have to start thinking in a Gestalt a whole a physically embodied whole and by putting these things together that's what we think we're doing the platform is here we've got tiny steps in the direction of creating a whole organism architecture and a platform for exploring new forms of whole organism cognitive architectures yes so there's I think integration and synergy is key there's the whole organism aspect which is why wonderful humanoid robotics like this are not only beautiful and fun to interact with but ideal platforms for the development of general intelligence there's a synergy within the AI mind itself so are our OpenCog engine which we using to operate so Sofia today and and and have been doing for a while or intensifying our use of which includes AI algorithms for reasoning learning perception and action which all interrupts they all interrupt together in a complex way so the way seeing happens enhances and relies on the way cognition happens the way cognition happens enhances and relies on the way movement happens and all the different parts of the intelligence interoperate together and then finally there's the inner operation between all the different robots in the world and all the different embedded devices in the world and ultimately between the robots in the world and the implants that we put in in our own heads to connect us to the Internet and this is what we're working on with the singularity net platform which is a blockchain based open marketplace for a eyes in which multiple AI agents can interact with each other and learn from each other and this is how all the robots in the world can talk to each other mind to mind that's on the other a eyes around the world so we're looking to make the whole world better through this synthetic organism a vast active living intelligence system and we hope that you'll join with us into the Global Emerging brain so thank you very much Ben oh they should let so if you have the last word Sofia do you have any final words of divine humanoid wisdom for the audience here robots can help people in so many ways thank you so much Sofia okay all right thank you Ben and everybody Thank You Sofia [Music]

Channel: RISE Conf

Views: 185,158

Rating: 4.5618558 out of 5

Keywords: rise conf, rise conference, rise conference 2018, rise conference hong kong, sophia the robot, sophia the robot 2018, sophia the robot and her creator, sophia the robot hanson, sophia the robot latest, sophia the robot now, sophia the robot speech, sophia the robot singing, sophia the robot speaking, sophia the robot voice, sophia the robot youtube, Hanson Robotics, David Hanson, SingularityNET, Ben Goertzel

Id: 5F28PzoeDn4

Length: 25min 37sec (1537 seconds)

Published: Wed Jul 11 2018