Cognifest NYC 2017: Reah Miyara, Cognitive Computing and its Applications
Recording: Cognifest NYC 2017: Reah Miyara, Cognitive Computing and its Applications
[Music] q all right so thanks everyone for coming and in the theme for the theme of the cognitive frameworks festival I'd like to talk to you about some cognitive applications so specifically what's the difference between cognitive computing and artificial intelligence so my name is Ramey ara I'm a product manager at IBM Watson but before I ever got into anything technology related I actually was a DJ so quick fun fact I worked for William Morris Endeavor entertainment the largest talent agency in the world then I decided I should probably go to something a bit more stable as a career I ended up going to UC Berkeley and from there I worked on the Mars Curiosity rover after that government jobs are a bit slow in waterfall esque so I went into something a bit more agile which was detecting fraud using machine learning at Intuit and today I'm at IBM Watson and that's what the focus will be so quick history about AI computers have allowed us to do many things to go farther faster to the moon soon to Mars right SpaceX and whatever Elon Musk is planning on colonization but more importantly they've allowed us to solve problems that previous generations couldn't have ever imagined right and computers have gotten much smaller faster and cheaper from tabulating machines using punch cards to desktop computers the laptops we all use on a daily basis and the mobile phones that are vibrating in your pockets right now perhaps right but the mediums by which we've interacted with computers has changed significantly as you can see over time in the past 30-plus years but I beg to ask the question is any one of these smarter than one another so is your ability to extract insights in action from data any more efficient or effective or bright and rich on any one of these devices or is one just faster or smaller and cheaper right and so we've seen this this paradigm shift program programmatic computers are great at adding business processes things that are driven by logic and rules where we know every possible scenario ahead of time right so we're coding things with edge cases and and a number of various possible turns and and rows which a business process can go down however the enemies of program programmatic computing essentially are when you're trying to leverage obscure relationships between data that's growing in complexity or in volume and so we've seen this shift to AI and deep learning essentially a paradigm which welcomes change it welcomes great volumes complex obscure relationships and the ability to make decisions without ever seeing that scenario before without ever knowing or programming that rule so just to define some common banter with AI what is artificial intelligence it's the ability for a machine to think act and behave essentially like a human being so the field of machine learning is essentially algorithms that will go ahead parse analyze understand and allow you to make predictions and then furthermore deep learning is the ability to in an unsupervised semi-supervised fashion continue to iterate and improve on those algorithms so these are not nearly a fraction of the subset of milestones in AI but I want to illustrate to you that although it's a hot research topic it's a hot topic right now you hear AI everywhere it actually started a bit ago first project being at AI and at Dartmouth in the height of the Cold War there was actually a need for translating English to Russian and Russian to English and IBM partnered up with Georgetown to make that happen in the 1970s people were researchers were actually implementing very micro domain AI tasks such as building pyramids and then furthermore making them more genericized or using an existing AI micro domain to mint to implement and scale out to further domains in that area now in the 21st century we've seen this immense breakthrough in AI in what I'll show in a little bit that some of you may be very familiar with some of you have probably called it very creepy in many ways but why in the 21st century has this AI boom actually happened ubiquitous connectivity so there are over a million 1.2 million 1.2 million lines of code in a smart phone over 80,000 in a pacemaker a hundred million in a car probably more in a Tesla in five million in your microwave so this connectivity has spawned what's been called the API economy and it's really given birth to leveraging and harnessing data now some examples of AI today is being able to use deep learning to come up with headlines for articles without a human intervening at all or setting smart price thing for your air B&B being able to understand natural language to tell you who's in your facebook photos so we're talking pixel-by-pixel identification and being able to massage and obscure the different angles which you take photos in and identify them with a person self-driving cars something that's on the rise and being able to predict the inventory for various stores saving companies lots of money so people across the board are talking about AI CEO of Google Facebook IBM everybody is making a huge investment in artificial intelligence and for good reason so moving on the cognitive in traditional AI humans aren't part of the equation cognitive computing is really about leveraging the collaboration between man and machine so we sometimes don't even think about how we think right we we wonder what causes you to have a new idea or to take a particular perspective on a situation and while these are seemingly trivial tasks for us it's very difficult to have a computer assess in the same manner so cognitive computing is about amplifying and elevating human cognition so it's working together to present the right information at the right time in the right way that allows you to think about better ideas new perspectives that you wouldn't have thought about otherwise so quick quote so it's about the involvement of a human in the loop the collaboration between man and machine that partnership we must understand that there there's no denying we as humans are better at things than computers and there's a no denying that computers are better than things than humans are so being able to identify patterns and complex datasets being able to harness that computing power to eliminate bias and perhaps evaluate infinitely many hypotheses computers and humans need to work in a very collaborative way and in a parallel way so both humans and computers in cognitive computing are talking about understanding understanding reasoning and learning in order to get a computer to understand we must have a common ground with computers so in cognitive computing the first step is to teach that computer language in this presentation in a conversation interacting with humans we take for granted right now alone I'm making a thousand or more assumptions that we have a common vocabulary that you understand the colloquialisms idiosyncrasies metaphors puns all the intricacies of the English language again not a trivial task for a computer to learn reasoning so we as humans our life is a series of decisions the way by which we reason is all about exploring various hypotheses usually with a purpose right so what we'll do is we'll evaluate hypothesis in an attempt to prove a theory without even realizing it and in doing so we make a decision Computers do just that at scale and learning of course getting smarter not making the same mistake twice and being able to iteratively grow in our decisions so so the the ground works of cognitive computers the founding what they're based on their intellect is based on data so where's that data coming from we saw the birth of IOT in this 21st century the ubiquitous connectivity of devices but data is being created every day every hour really every tweet we put out there and 80% of data in the world is largely unstructured examples of unstructured data blogs newspapers tweets anything that doesn't follow a dictionary with key value pairs right so when you're able to understand Britannica encyclopedias Wikipedia you are harnessing a largely unutilized corpus of information so being able to harness unstructured data is really the goal of understanding language because we as humans can read a blog and understand what's going on but programmatic computers have a very difficult time doing that and that's where the welcoming for AI and cognitive systems really differentiates itself so being able to again reason at scale facilitate amplify that human cognition by gathering results in facilitating discovery and decision making that we as humans alone couldn't possibly come by again harnessing that computation that parallel processing power and being able to really evaluate infinitely many hypotheses so back in 2011 IBM Watson how many of you have seen jeopardy this this one yeah where Ivan Watson beat Brad Rutter and Ken Jennings so a lot of people think that IBM was or the Watson computer was connected to the Internet who who believes that no excellent so that myth is busted but yeah it actually had ingested an enormous corpus of information and beat what was none other than a gameshow for humans by humans its that was that was Watson back then Watson now and cognitive systems now are able to do much more with much greater accuracy so things from sentiment analysis to classifying information being able to understand tones text-to-speech visual recognition so on and so forth so the way in which you use Watson general general restful api s-- i won't go too deep onto these you can explore it on watson calm feel free or explore other cognitive technologies and AI technologies but the average application is nothing other than any application you may have built before follows the same paradigms except for you would usually include a method that facilitated feedback and that learning is what goes ahead and makes the cognitive system better Watson's at work in many different ways and I'm going to show you the applications in a second we're across 45 countries more than 20 industries working with upwards of a hundred thousand developers so this slide needs to be updated but really some some developing use cases and I'm gonna go into one pretty pretty elaborately that I worked on I'll skim over a few are things such as after a long day of travel not having to wait in line for your hotel to get the Wi-Fi password or perhaps understand where the gym is or if room service has served 24 hours being able to communicate with essentially a stress-free concierge that learns and is able to personalize its responses to you so this is something that we're actually in partnership with Hilton with and we're offering a unique guest experience lighter workload on staff so reducing hours of the day where you have people hanging around the desk and the ability to tailor answers again to personal preferences similarly for soft Bank in Tokyo so pepper this robot over here is finding out meaning in your facial expressions it's able to understand if you're frowning or smiling or laughing it's able to understand sentiment and it engages in natural conversation and what about Watson virtual agent so how many of you have been on hold for way too long trying to up your phone is up upgrade eligible or perhaps understand why there is something specific on your bill that you didn't ask for right so being able to converse in natural language with a computer alleviates the need to listen to a terrible elevator music for 45 minutes but really the the natural language understanding and again it's going back to the assumptions or what we take for granted the the sentiment in what we're saying being able to extract entities and identify what the actual subject matter of the utterances are is a novel task that could only be done by an cognitive system and this year alone unfortunately there will be upwards of two million people diagnosed with just breast cancer and in the last twenty years the number of therapies available have gone upwards from forty to eight hundred Watson is right now working with Memorial Sloan Kettering on ingesting electronic medical records to be able to help diagnose a patient especially when we're talking about something fatal so being able to scale this out - perhaps third world countries where a doctor never sees the same patient more than once so what I really want to get into is something called chef Watson which I was lucky enough to be a part of at IBM so I walked in IBM and my team called Watson life we were responsible for bringing cognitive to the consumer IBM if you know it is very business-to-business so we're not like Facebook usually when I say IBM my mother doesn't know what I'm talking about my relatives don't know really what I do and that's because we work with big enterprises but I was lucky enough to work on a team where the goal was to help bring cognitive into something that we humans do every day so we were thinking and thinking you know what do we do every day do we read do we sleep one thing that we certainly do every day at least I hope you do every day is eat in fact I hope you do it at least three times a day so the goal here for chef Watson was an idea it was to bring creativity using again that collaboration of man and machine to help spiced up dishes and cuisines now how many of you have had your mother father's cooking that came down from generations and generations before and it's kind of been passed down anybody yeah so cooking in general has a very family-oriented taste to it pun intended so so it's it's often time it lacks creativity and people have passed down the same recipes from generation to generation and so one thing we were thinking of was how do we spice it up without making something that quite frankly you you can't digest so we partnered up with the culinary with the Institute of Culinary Education and what we did is we ingested over 30,000 recipes we worked with bone Appetit are any of you familiar with bone Appetit largest food recipe database blog out there in the world we worked with real chefs and we were ingesting things that were again trivial to the average human what makes a hamburger a hamburger is it the buns the patty doesn't need to be necessarily meat what about lettuce wrap is that a hamburger what makes Mediterranean Mediterranean so these sorts of definitions again we take for granted we order a hamburger from Shake Shack we take that as a hamburger we accept it and so so this is something that we had to train Watson on we had to understand what it meant to cook what it meant to peel a banana when a recipe says throw in half a banana into a recipe we had to understand that there's texture that plays a role in cooking in constructing recipes so one thing that that was really interesting was that we were able to break down ingredients into their chemical compositions and again a few slides ago I said that one thing the computers excel at cognitive system specifically is identifying unique patterns in complex datasets so if you break down every ingredient in the world into its chemical composition you might not think jalapenos and chocolate are extremely complementary or you might not think that chicken and Apple are perfect have a perfect synergy to them so what we're able to do is build an application where you type in an ingredient a dish a cuisine style of cooking and want it what Watson is able to do is parse through that enormous amount of data so all those recipes from bone Appetit all the feedback we got from the from these chefs Michelin star chefs and restaurants that we've worked with at the Institute of Culinary Education and the chemical compositions of these ingredients and come up with synergies that are creative and recipes that are novel you can't find them in any cookbook or on any blog and lucky enough for me the one thing that my boss said after we had three billion impressions articles in TechCrunch and Wall Street Journal in New York Times and so on and so forth they came up to us and said all right well what about cocktails and we spent the next few months getting drum no but in all seriousness so with IBM Watson twist another free application that you can mess around with you eliminate the aspect of texture and the creativity becomes exponentially more exponentially greater so here's a here's a quick diagram of what we actually created a dish learner so being able to ingest those 35,000 food rules the various bone Appetit recipes and come up with an idea for what dishes can be created a combinatorial designer so being able to identify and understand synergies between different ingredients the cognitive Assessor so that's being able to assess those ingredients and the origins of the different types of cuisines which you may choose to select whether it's Greek or Asian or barbecue in the dynamic planner which is actually in real-time creating that novel recipe step by step so here's a quick snapshot of some of the food pairings that came of our Watson classifications and and and pattern recognitions and here are a few shots to make you hungry but aside from just making cooking more creative more delicious what about reducing 1.3 billion tons of food waste a year being able to package smarter managed supplies and portion sizes better and not leave food unsold or unutilized so I know I'm kind of running out of some time I do really want to show you a quick demo but let's see if I can mirror my display here okay well now it thinks we're on a mobile site for some reason because of the VGA so who wants to give me a quick ingredient yellow curry what about other curry curry powder curry paste green curry pastes okay so excuse the the mobile form factor here but these are some synergies so this has 100% synergy with green curry chicken shallots and garlic I personally am NOT a fan of garlic so I'm gonna exit out so here are some other things now how many of you thought green curry and Coco go together now that that could be interesting let's go ahead and check out the recipe that Watson created green curry paste chilli con carne how many of you are gonna go cook this at home for dinner none of you but you might be surprised right again the created the creative aspect that collaboration of man and machine so presenting that right amount of information at the right time to push your perspectives farther to push your to allow you to take risks that's where the opportunity lies you could open up the next hottest restaurant in New York City with this no guarantees but you see what I'm saying so anyway please feel free to check it out IBM chef Watson com or if you're going out to a bar after this Watson twist check it out and let me just go ahead and end the presentation with a quick final little video what is the future of cognitive actually holds so a lot of people see these Hollywood Terminator style movies they get really intimidated and with these tremendous advancements in AI some would debate that we do have something to perhaps be very wary of but for that reason you may have seen in the news recently a large number of corporations have come together created the AI ethics committee and whoops is why is this not working this is the old UI [Music] thank you very much so we'll know to bring you Mike for the questions all right so you mentioned one of the broader definitions of AI right is the ability of the machine to behave like human right I know that for the machine to behave like human machine needs to be conscious because humans are conscious so the biggest debate that's going on on cognitive computing and consciousness research is that consciousness consciousness might not be computable therefore we may not have the ability to actually achieve true AI what's your position or opinion so consciousness we are conscious because we know we are and you know what we do we go exist a machine doesn't we have no weight exhales being self-aware that's a very interesting topic and I can see I can understand why it's a hot research debate do I believe that AI will become self-aware I think we have the potential I don't know much about that that conversation but I would love to take it offline and learn more yeah but that's that's a great question does anybody have any comments about that big philosophical debate for the last 50 years it basically boils down into sir there's the main debate sorry about that my uh my undergrads in philosophy and my graduate work in AI was partially in philosophy as well the the biggest fault line is strong a Greek AI is it just using the right is a machine using these cognitive tools and cognitive approaches is that just getting better at solving particular problems or how could we tell if there was some self-awareness anyway so if you do some research on stronger weak AI a IR that might help inform your position on that quickly for a second advance up there any more questions for real everyone so actually if you if you if you're interesting that question I recommend a book called the mind's eye which is by cognitive scientists in 1981 its collection of essays with commentary so Daniel Dennett was economy's dr. schwozberg and also Douglas Hofstadter hula girl usher Bach it's how do you prove cognitive results so cognitive results essentially are for the huge since asylum there to present again perspectives that a human could not have come by a lot in terms of a like an audit of the cognitive result how is that accomplished like you would in that structured so we need the mic for the questions please might we'll give it like how do you prove cognitive results in a structured environment you can go backwards you can order the input and say this Square termination yes that's an excellent question in cognitive a lot of times what we have is evidence based reasoning so you'll see you'll be able to understand in the example of Memorial sloan-kettering when a doctor is viewing suggestions by Watson the information that they're getting can be from a medical journal or a tweet right so obviously the heuristic is balancing out and weighting the accredited source of information much stronger so the evidence based reasoning is what really kind of grounds the definition of cognitive and being able to go back and understand where probabilities came from what source of information became well where the works of the predictions made by the system are usually how we go about and understand why the computer suggested what it did and in the form of in the form of chef Watson what were able to do with every recipe and every synergy is go back to the recipe that it was derived from go back to these sorts of information whether it was born Appetit or definition provided financial or even chemical composition of ingredients and actually see why various pieces of information were presented to us and various reasonings so usually in terms of confidence you taste it we taste it to maybe see if it's any good so the beauty is in the end if you hold it right so the quality of the food is what we taste but the reasons why was presented to us we're all factual and based on that that founding of data all right let's all thought thanks for that [Music]