 Well, good evening and welcome. It's fantastic to see so many people here on a slightly stormy night out there. Welcome to the second of the lectures in this year's Gibbons series where we're looking at the steps towards the singularity, artificial intelligence and its impact. I'm Robert Amor. I'm Head of Computer Science Department here at the University of Auckland. And it's my really great pleasure to introduce Professor Hans Gusskin, who's a friend and colleague to many of us over the years. And I don't think Hans got a drink or any food because he was saying hello to so many friends in the audience as he came in. So, Hans is a Professor of Computer Science at Massey University and his qualifications come from universities in Germany and Bonn and Kaiserslautern and Hamburg. And he came to New Zealand in 1992, so he's almost lost his accent. And he came to the University of Auckland when he joined us there. And in 2007 he shifted to Massey University where he's a Professor in Computer Science. Doesn't seem like a decade that Hans left us. That just seems a few years ago that that happened, but ten years ago. So his main interests are in the era of smart environments, ambient computing and intelligence, knowledge representation and inference, constraint satisfaction and spatial temporal reasoning. And tonight he's going to talk about artificial intelligence as it can impact on society and addressing one very specific problem where with our aging population and a population whose mental and physical capabilities are declining. How can AI help in that area? And so there are some solutions. Hans is going to tell us about those solutions and the solutions that should give us a home sweet home. Thank you very much Robert. Thank you for this very warm welcome. It reminded me of my arrival in New Zealand 25 years ago where I received an equally warm welcome both in general from New Zealanders and in particular from the staff members in this department. And I will never forget that. Now when Robert said I had to catch up with a lot of friends and therefore couldn't have a drink, it's half the truth. The other half is that my wife says if you drink alcohol your accent becomes even worse and nobody will understand you. So that has to wait until after the talk or otherwise you wouldn't understand the word. Now I'm really honoured to be here. You were quite frank. When Bob approached me he said, well, look Hans, do you want to give a lecture in the Givens Lecture series? And I thought, oh, that's great. But to be frank, you are our second choice actually. But nevertheless I'm really, really happy that you're louder. Thank you. I think I should move this purpose a little bit. Is it better? No, it's not. OK, I will try to speak up a little bit. So I'm really delighted to be here. And just to give you a little bit of an idea. So it's almost my silver anniversary in New Zealand. And 10 years since I left the University of Auckland moved to Massey. I've done artificial intelligence almost my whole academic life, so to speak. There was an odd professor in Germany who said, look Hans, there is a topic. Nobody has heard of artificial intelligence. It's really hot. You do that. And I said, yeah, I do it. And since then I got stuck. And over the years I developed various flavours. And the most recent one is smart homes. Now when we talk about smart homes, there are a lot of things probably going through your head. And the term is actually quite overloaded. You can install your own smart home nowadays, which automates most of the things in your home. Or you can use it in various other ways. And the way I want to use it is to support elderlies. So this is my motivation. So here's some statistics. According to statistics, about a quarter of our population will be aged 65 or beyond. So that's one thing. So there's an expectation nowadays that we live longer. And of course, who doesn't wish to live longer a healthy life and an independent life. But at the same time that poses a problem because it's not always possible to live an independent life because as some of you might have already experienced and I start to experience it slightly, there are certain things that happen over age so I run much slower now than I run as a child, I noticed. I also find it difficult to do particular tasks. I didn't have a problem at all, like touching the floor of my nose when I'm sitting down or things like that. I mean, that's silly things, but you see where I'm coming from. So the question then holds, how do we actually support people because obviously it's not always possible to live independently or in other terms, who cares? So we need to find a solution and the question is what is that solution? We need to find an alternative solution. There are various solutions and the one you see on the slide is probably not one somebody prefers. But if you're a fan of Dan Brown, you probably noticed that in his recent book Inferno, he suggested a more subtle solution which is to make half the population sterile. Well, even that is kind of unsatisfactory. But even if we somehow manage to get a solution, there's another problem. When I think back to my parents, my mum married the guy from the next village. My grandparents never moved. Nowadays, people are very flexible in terms of moving. I'm now probably on the opposite side of the world, almost, from where my parents live. So with this flexibility actually comes a problem that we are not as easily prepared to take care of our parents or grandparents. So this is more a goal I'm looking at. Independent happy life and preferably not in a rest home very early, in the early stage. So how do we achieve this? Or what is necessary to achieve this? I'll probably do the following and have a short break and put my watch here because I otherwise go on forever. So there are various aspects in smart homes or what we could use a smart home for. Is my voice still all right? Can you hear me or am I fading off already? Okay, thank you. Give me a signed wave. So if you wave, I know I'm too quiet. If you wave, I know I have not switched on my microphone. I've seen no waves there, no waves there. Everything's good. So first thing, assurance. I think there's not something which is particularly related to older people. I don't know if you are in a situation, sometimes you leave the house and you think, oh, have I switched off the stove or have I locked the door and you can really get paranoid about that. I do from time to time. And then you go back and check and then you leave again and say, did I actually check that? Okay, so assurance is actually something that is quite welcome. Just on an individual basis, feedback to certain things to a person living in a smart home. The other one is support. Support can mean a lot of things. If you depend on medication, it could mean that you get reminders at the right time to take the right medication. You probably all have heard about these intelligent boxes that open and the pills come out automatically at the right time. I'm always asking myself, are the people then taking the pills or falling them away? So there are always supporting elements like that. But it could also be just intelligent reminders that something has cooked long enough that you have forgotten, for example, to drink enough during the day and so on and so forth. And the third aspect is assessment. So sometimes with becoming older, it goes along a diminishing physical or mental capability. And there's a fundamental difference in my view about the tool. So the physical impairments is certainly something that you probably notice very easily. You find things harder. The danger with cognitive impairments or, for example, diminishing mental capabilities is that you might not notice it. You might forget things and so what? It doesn't affect you immediately but it affects you in the long run. So for that, we need, of course, input from the outside. So these are the three supporting elements and the key word here how some of the research in this area tries to tackle these is what is called ambient intelligence. What is on the bottom of this slide is actually a phrase not I came up with but people usually refer to ambient intelligence in literature. A digital environment that proactively but sensibly supports people in their daily lives. So what is this beast, ambient intelligence? What are the features we are talking about? Obviously, you might have guessed. I haven't said so. It could be all kind of things but as you probably know I come from computer science. It's a computer science lecture so it has to do something with computer system. But what kind of computer system? It's a system where there's ambient intelligence in there as a player with words so there's the artificial intelligence in there. So it's an intelligent system computer system that's intelligent but it's not disconnected from the world. What does that mean? It is sensitive and you probably might interpret the word in a way I don't mean it in this case or researchers don't mean it in this case just means it senses the environment. It's responsive, it acts on the environment. It's adaptive. It has some transparency so you know what's going on and this is ubiquitous. I finally managed to actually say ubiquitous. It's a difficult word for me. I was actually considering not to do research in this area because of this word. Right, so that's what it is. And people came up with more sophisticated definitions. I just put up this slide so those of you who really want to know more about it there's heaps of literature around and some of the definitions tick certain boxes so in this column the S-R-A-A-T you might have guessed is what was on the previous slide the sensitivity, responsiveness, adaptiveness, transparency, ubiquitous intelligence and some cover more and some cover less aspects on there. So how do we go about this? Well, we try to create a computer system that can act, sense reasonable at the word but don't just want to do it in any way. There are various aspects that also play a significant role and it's sometimes overlooked. One is the what is usually called human-computer interaction and you probably have a feeling of what that means but what does it actually mean in a system that's ubiquitous? I've actually haven't said what ubiquitous is. Ubiquitous is you don't see it, right? It's all over the place. And that makes sense. You don't want to have your R2D to a standing in a corner or something like that, your mainframe. You don't want to see it. It doesn't want to intrude. You don't want to intrude have the system intruding on you. So it needs some but you want to interact. So traditional computer interfaces really don't do it, right? You can't expect an older person to go to a terminal all the time and use the mouse and select something. That is disruptive. So we need to find new ways of interaction with a a million intelligence. And the other thing is security. People are usually quite concerned about their privacy, right? You know? People are usually concerned about their privacy. Although they don't act on it just to give you some idea security cameras or internet security cameras are quite a hype nowadays. Who in the room has a internet camera? Yeah. Makes perfect sense. I installed one in my house. My daughter is living with us. Sometimes the parents are not at home and she gets frightened living in Parmi. It's not as populated so you can feel a little bit. It says somebody out there. I installed a camera and I looked around and I got one and I just browsed to familiarise myself with a few issues around it and so on. And it turned out it's amazing of how many security cameras you can access through the internet. There's a site in the US and it lists security cameras so you can look into living rooms into garages of people. How does this happen? Well, you buy this security camera, right? Where do you buy it? Don't want to name a particular provider in New Zealand nor overseas. Maybe you get it somewhere brought in from overseas by email, per mail. Sorry. You install the thing. It works. You're delighted. Perfect. Well, these things come with standard settings, right? And one of the standard settings is the standard password. And they are not so many manufacturers and they are not very creative about their passwords or logins. Let me guess. If you go into your router, right? Who locks in with admin? Yeah. Because you're so proud to be an admin? No, because if people call it admin and so if I have to guess that's the login, right? Now, up to probably a year or two ago when you bought a router from one of the major providers in New Zealand it had a standard password. Recently they changed it and now the standard admin password are the last so-and-so many digits of the serial number of your model. So at least it's personalized. So if you don't bother to change that I know your password. So I can lock into your router, right? I open a port. Perfect. Your camera has the same password anyway and one of the favorite passwords for the camera seems to be one, two, three, four, five, six, seven, eight. Okay, there might be another password so you're off. And the same thing of course in principle can happen in an admin intelligent. So let's look at something which is a little bit more straightforward which is the sensor side of ambient intelligence. Catch my breath a little bit keep my voice a little bit difficult so I can perhaps say something with a lower voice Robert mentioned the weather he actually ordered that especially for me, me from Parmesan North always being windy in a windy area so we ordered some wind so I feel at home. Very good. Okay. So what type of sensors do we actually use? You are all familiar with sensors to some extent. If you come home, I guess most of you have security light around your home, right? There's a sensor in there, a passive infrared sensor, switches on. Perfect. We have other sensors instead of motion sensors. So what we could sense as well is light sensors in your security light do that already because they don't switch on usually during the day. Some form of radiation, temperature all kind of things. We can of course have a sound sensor if there's noise in the house. Another thing is solids, liquid and gases. So think about the tap running. How do you notice that? Well, you put something in the pipe and if flow goes through it senses this and the system recognizes the tap is open. Now we had an interesting project I get a little bit sidetracked but I'm hoping that this is making it easier for you to relate to this type of technology. We had an experiment in Parmas north of an elderly gentleman in his 90s blind living independently in his home and we equipped the home with sensors. Now the problem is although there is a kivi ingenuity there are a few things you can't do in New Zealand. So you can't just go in as a computer artist and cut the pipes. First of all it's not good anyway but even if you think you can fix them later and put something in there it's not legal. So one of our colleagues invented something that listens to the pipe a little device we just glued on and looked for noises and he did some other things to filter out for example voice which is a different form of noise. So you can do quite a few of things to get around the problems. Okay, the last one I haven't mentioned position, direction, distance and motion. So one thing we are currently doing is trying to track people in their movements because these movements might give us an indication of whether something is wrong with a person. Now of course this is a non-trivial task, a passive infrared sensor doesn't really do the trick what people usually do nowadays or what has been suggested in the literature experimental setups around the world is putting pressure sets into the floor so these are pressure mats and they recognize if somebody steps on that. This is actually a nice thing here there are no pressure mats in this floor I was expecting one because this is recorded this lecture and when we do that at Massey University we have to stand on the pressure mat and the camera zooms in as soon as you go off camera zooms off and if you are a very moveable person like I am because a student recently complained not to me, to a colleague of mine with a short message you are making me sick. Alright so these are the sensors then the question is well they are all quite limited they just send certain aspects how about cameras talking about cameras people usually are not having to have cameras everywhere they are at least in some places perceived as invasive thank you although the prices have dropped quite significantly they are more expensive than these cheap sensors so if we talk about infrared sensors or just sensors doors they are less expensive and because you get an image doesn't necessarily mean that the job is done and I have the pleasure to have some in the audience who are quite familiar with computer vision and they build whole careers on this and not definitely not because it's a trivial task so just having cameras doesn't do all the work for you and it turned out depending on what you want to achieve it's not really essential okay so the other aspects you need to consider is wired versus wireless sensors and there are various pros and cons wired sensors they are a little bit cheaper but you pay for the wiring and if we think about getting smart home technology into houses in New Zealand we can't just wait until all the houses have been replaced with new houses that have the equipment wired in and retrofitting houses although I can tell you retrofitting anything in New Zealand house is easier than a house in Germany and do things like that you can usually access the space over your living room and things like that but it's still an effort so wireless is a little more expensive but it has downsides like it uses batteries and so on and possibly it's not as robust again a trade off people need to consider now reasoning this is actually where my work lies I am not an electrical engineer so the sensors we use are usually out of the box sensors except for the little developments my colleagues do I am interested in reasoning so making sense out of what comes in from this huge amount of sensors so on this picture which is incredibly small I realised that which is really amazing that I put something like that on my slide I had a talk yesterday to my students how to avoid presentation mistakes and I am just thinking ok this is number 12 at least that counts so far anyway the first one is actually not being anxious but when I stared at this room at the beginning I was kind of terrified but I am also at the same time very delighted so these are sensors what do these sensors produce these sensors produce some form of data they transmit some form of data this data goes into a magical box and this is our immune intelligence this is essentially a computer system and it's amazing what you can do nowadays on small computers so we are not talking about these rooms full of machines even with simple very small computers you can achieve quite a bit some of you are familiar with Raspberry Pi I guess quite a few interesting projects you can do and we let our students do which already goes in that direction and provides some aspects of immune intelligence so out comes some recognition of some activity so let's say the immune intelligence says oh yeah the older person in the house just got up and is preparing her breakfast and with this information we can do various things we can just ignore it if it's a normal activity or if it's an unusual activity let's say it happened at 2 o'clock in the morning we might give some support for example calling a carer so that's the general idea of the flow of what we want to achieve now we call this activity recognition some people call it behaviour recognition I don't know which is the most more appropriate term so we want to first find out what is going on in the house now what we get is something similar as shown on this slide this could be a typical data stream so really row by row so in comes a signal at approximately 5 past 6 that in the living room the television was switched off and the curtains were closed about 3 minutes later things like that come in of course if you look at that that doesn't well it might tell you something but if your computer system looks at it without any reasoning process it doesn't tell you anything it's just data it's no information at this stage no knowledge that comes out of it data that comes from the sensors these observations are not activities so we need to think about how we turn observations into activities how we can map a sequence like this to the occupant is making breakfast or having a shower and so on now this might sound easy but it's actually incredibly difficult and there are various things that get in a way and one thing that gets in a way is variation luckily it doesn't get in a way completely because people have their habits and it's amazing what kind of habits people develop but in principle if you think of a simple task like making a cup of tea with milk so do you put the tea or the coffee into your cup first and then the milk or the milk first and then the coffee it might not be the best example but you can see there are variations in there if you have your bathroom routine do you brush your teeth first and then shave first who knows so the observations as a result of that might be quite different second thing that gets in a way interwoven activities what does interwoven activities mean it means that sometimes we interrupt an activity and do something else for example if you put your clothes in the washing machine I guess you are not switching it on and then you stand in front of the washing machine and wait until it's finished most people don't do that I think they do something else and even if it's just having a cup of coffee but that gets in a way because the cup of coffee has nothing to do with doing the laundry unless you think doing laundry is really a ritual which involves coffee drinking but it could be something else you do in between and most people actually do so if you look at the observation what really fits together is just a part the observations that belong to these orangey errors on the slide and are labelled as laundry whereas there's something in between making coffee so we need to take care of that at one thing we did this and now it becomes technical hidden mark of model who are familiar with hidden mark of models wow that is great I wasn't well, I wasn't a long time ago but okay so it is what is called a probabilistic model probabilities you have probably heard about probabilities probably so it's something that's not quite certain to toss a coin so it's a 2050 chance that you see head when it comes down we know all these kind of things so somehow this approach incorporates probabilities sorry about you guys who already know about hidden mark of models and are now bored to death okay for you I have this slide the rest ignores it so what does a hidden mark of model actually do well imagine yourself getting blindfolded okay and I put you in my car which I haven't here but let's say I had a car here and I drive around and we get out somewhere in Auckland and you get certain signals okay so let's say you heard some church bell ringing okay then we walk along and you hear a ship horn right and you walk along and you get some other input maybe you smell something funny so what do you do the first thing you probably say where the heck am I right don't see anything where the heck am I okay could be any type of places now Auckland is a better example because there is a huge number of places but let's say it was somewhere in your house or in a restricted area okay I could be either in my garden or I could be in my living room or I could be in my kitchen okay I don't know probabilities but then you get observations in suddenly you hear something that ship horn suddenly you think ah so I must be now at the harbour and you develop a model over time we call states where you are and we want to find out or in your process while being blindfolded you want to find out what I have actually done with you in Auckland where I let you and you might or might not be successful at the end you might answer yeah I know you dropped me off Upper Queen Street then we walked down Queen Street we ended up at the ferry building then we went to I don't know the bridge and so on or you might say well we could have been Upper Queen Street or it could have been some other street in Auckland and it could have been 50-50 chance but then something else happened so I think in the end we ended up to that particular sequence that's what the hidden mark of models do so what does this have to do with activity recognition so I skipped this oops there's just a few features why we use these let's go to this slide it looks very technical but it's not really so on this what is this colour it's a pink pinkish in the middle you see these O's so here come your sensor data you see the cover door opening the fridge door opening you see some activity in the living room and so on so this is a signal to our system so our system does the same thing you did in Auckland when being blindfolded but it makes some assumption and it tries with these hidden mark of models to find a good match for a part in the observation so this is what the principle is about there are a few things to consider because this comes in as a flow at some stage the activity change so we need to find the right window so there was a little research around this so when is it actually relevant to listen to certain hours to certain sensor information and when do we stop but you can do a few tricks and that's quite a successful approach so after this then we have a good idea of hopefully a good idea for this going on in the house of course these things have to be trained trained means in AI lingo you just observe people in the house and you tell the system like you train or teach a human being what certain observations actually correspond to that's called the ground truth so we say this type of observation is actually breakfast but at some stage we don't stop this and we expect the system to actually do the job on its own okay right so let's give you some idea of how you can do it in principle now what does it actually tell you when it tells you some things for example if you don't observe that a person has used a shower for a few days you could imagine that this might be something that is not acceptable or maybe it is depending on your lifestyle I shouldn't really jump to conclusions here okay but you see where I'm coming from so certain things that are missing you could interpret as abnormal and potentially dangerous the same thing the other way around that happen too often so let's say a person opens the fridge every 24 hours maybe there is some eating disorder going on or maybe the person prefers white wine over red wine which of course you keep in the fridge but again then every 24 hours might not be something you want to accept but there is a more subtle way of things unacceptable abnormal behaviours and that's behaviours that happen in the wrong context okay so having for example dinner at 6, 7 o'clock at night quite perfect but if suddenly you realise the dinner is taking place at 11 o'clock and the next night at 2 o'clock in the morning it's still the dinner activity which is perfectly fine but the context is wrong or let's say the person puts on the heater and is in the middle of summer now this is an incredibly bad example given the late summer at least perhaps north one okay so I will think of better examples sorry about that but you see out of context context information spatial information everything with where are things happening do they happen in the right place so taking a shower in the bathroom is quite normal but outside it's not so normal temporal information we had that example in emotional states and so on all this information can feed in and then either can boost the recognizer process which I haven't referred to but what I referred to is to recognise abnormalities so that's the idea then the next question is where do we do this so we are now moving on from activity to context so the image that we have in mind is to use context maps these are maps where we record when certain things happen so let's say you have certain activities lunch, watching TV, ironing and so on and they happen at a certain time so whenever we observe that somebody is having lunch we record which day it is which month it is which season it is in which room it happens and so on now the thing that's important here when we talk about temporal data it's not just a time stamp right because some kind of abnormalities happen because they happen at the wrong time of the day some abnormalities happen because they happen on the wrong day that you suddenly do things on the weekend you did during the week or during the wrong month or the wrong season right so we have to distinguish between these and the same holds for spatial data now in the following I just mentioned a few things for those who are familiar with techniques of AI so there are various again probabilistic models around based on neural network approaches and one keyboard here is Boltzmann machines for instance a particular type of approach that can be used for that however I would like to go into something which I think is a little more intuitive and a little perhaps easier to get your head around with because these Boltzmann machines again like the activity recognizer you train and you hope that something is coming out that is right in most cases that is the case and sometimes not so let's forget about all this stuff and say how would we actually in a naive way tackle the problem and let's get back to these probabilities so I'm now assuming that you're all familiar with probabilities I don't assume anything so let's have a certain this so what is A and C and P so it's a probability of something happening A activity and C context what is the probability that an activity happens in a particular context that both happens so there are all these formulas around and you don't have to really be bothered about that but what we are actually interested in is what is on the bottom line which is so called conditional probability and there is some way to calculate this which I have put on there conditional probabilities what is a conditional probability well a conditional probability how can I best explain this it's difficult enough to find a coin tossing a coin ok let me try to find an example that brings it closer one day a man was caught when entering the aircraft because he was carrying a bomb and of course nowadays the assumption is t ends with ist right but this man had a perfectly valid explanation right he said I read in a newspaper that there is a certain chance that you enter an aircraft and there is a bomb on the aircraft right but it is very unlikely that there are two bombs on the aircraft so I thought if I carry a bomb on the aircraft I am safe conditional probabilities right so this person didn't know anything about conditional probabilities because the conditional probability is actually the same as probability of one bomb on the aircraft so conditional probability of two bombs given that I am carrying a bomb is the same as right that's what conditional probabilities are about sorry not bombs but that kind of stuff ok so we would need them because we want to condition the activities on our context so given that it's a Monday what is the probability that the person does the laundry the problem is you need a lot of these and you need to calculate a lot of these or first of all you need to observe a lot of these and that's computer scientists are notoriously lazy so they don't want that so we need to find something else so they came up with something that is called a Bayesian network what is this beast again probabilistic graphical model but I give you an example this example is in the middle of the slide so what does it mean I know something I know A and A influences B ok let's say A is me carrying a bomb B is a bomb on the aircraft ok now think smart homes A is observing that the light goes on in the living room B is me watching TV ok so obviously the light is going on or the TV goes on has an impact on how I feel about whether I would assume that the TV is on or not other things don't so if the security light goes on outside doesn't have an impact on whether the TV is on you would say they are independent so that's what people call conditional independence so in this case E is conditionally independent of all the stuff that is not directly related to E so let's see the in a nutshell what Bayesian network it's about there are not too many people in this room because otherwise I have trouble when leaving this room because they will tell me this was too not right but the reason I'm trying to give you the idea of it because I now want to build one so how do I do that in the smart home context weekday weather season we have our behaviours, watch TV, ironing lunch at home and we have sensor readings the TV is on, the iron is on so let's say you have a little sensor in your iron on your TV so we create something then mimics what is going on in the smart home so here is an example we made up Mary goes out for lunch in summer if the weather is fine, rarely in spring, autumn and nearly never in winter so if it's nice to go out if it's not nice don't go out we all know TV program isn't all the same every day so maybe she prefers to have in her view it's good on Monday and Fridays sometimes on Saturday and Sunday if she thinks otherwise it's not for her she would iron while watching TV rarely during lunch nearly never otherwise so if you have that kind of information and it's a little bit of a made up simplified one we construct something like this, this is our easy network and I was going fully through this but just to give you the idea there are these bubbles nodes in there and we have for example lunch at home and that means that signals to us behind this is a probability that will get a value it's likely it's not likely and that is influenced by the weather and the season so when we run this network we run it it does magical things behind seeing some calculations and we suddenly get something like these tables there and if you go back to the lunch at home example suddenly you say t0.78 and f0.22 and what that means is that is roughly 25% chance that is not the case and 75% chance is the case so these are the things we get out of this now just to I skipped the next time the next thing the details of the next slides but one thing you need to consider also if you deal with probabilities is that there is a requirement that probabilities add up to 1 so if I toss a coin there is a 50% chance that I have tail and 50% chance that I have head so together I have 100% chance that the coin comes down somehow right now the problem is if you don't have information complete information this might cause trouble for example let's assume I don't know anything about which is really fair to say because I actually don't know anything about your hygiene routines so so you ask me what is my belief anybody of you can ask me what is your belief Hans that I shower in the morning and I would say don't know so I say I don't believe it 0 so you think I'm not showering in the morning right no that's not what I said just that I don't have any knowledge yeah but if you say it the opposite must be true right no not really so then you say yeah Hans you did it all wrong you should have said 50% right because if you don't know like the coin 50% chance okay you ask me again I say 50% okay that's right so show in the morning show in the evening show on midday now the 50% don't work right because 50% in the morning 50% midday 50% in the evening wow 150% that you shower at all super clean okay so message here is there are formalisms that do that you can use and I skip these that do all that stuff okay and the last thing is you could even talk about imprecise concepts as opposed to probability oh now it's getting taken imprecise concepts are as opposed to probabilities what is he talking about okay example what is probability versus fuzzy logic okay give you an example again I want to talk about a fuzzy model and I want to talk about a probabilistic model and they both belong to their set with a probability or a degree of 90% so what is the difference if I say yet for example a bottle contains portable liquid with a probability of 90% or I say I have a fuzzy bottle that contains a subset that belongs to the set of portable liquids with a degree of 90% is it the same no it's not the same so here's a scenario a fuzzy bottle means that you filled it slightly from the slightly polluted well so it's not the best it's not your New Zealand spring water which we soon won't have anymore but maybe you fill it from the tap and that's reasonably good it's not your cleanest water so that's your fuzzy bottle the probabilistic bottle is I give you a 10 pack of avian bottles and in one of them is a poison that drops you dead so here's the difference both are 90% portable liquid but some is a degree some is a probability so again you can talk about this and rephrasing things in in terms of fuzzy theory so you could ask to which degree is a shopping event a Sunday event and you can play around with this and you get information okay the last thing I would like to talk about is dependencies between activities and then the formal stuff is over so what we did is what computer scientists often do they download from the internet what is called data sets so data sets is a kind gesture of other researchers that they make available observations they have made in their setup it's very convenient so you don't have to set up your smart home you get this data set and one famous sets of data sets is the so-called CASAS data set from the University of Washington and we played around with this they recorded a number of activities and then we analyzed them and not surprisingly some activities follow other activities more likely than others so if you have a particular activity let's say breakfast it's more likely that the next activity is leaving the house because you go to work and it's less likely that the next activity is going to bed well I run out of sensible examples because whenever I say something I immediately think maybe there are people who have breakfast and go back to bed which is totally sensible but anyway so they are given a particular individual and this depends so that's the form part so what is actually going beyond all this reasoning stuff where do we go from here detect if an activity is normal abnormal in context and initiate a reaction but how about privacy and security coming back to these topics that I had on one of the first slides so what type of information will be gathered who will this information interact with and how will information be encrypted and sent so security where does this reside and why is the information needed in the first place and these again probably not from a computer scientist point of view purely but these are sensitive issues these are issues that have to be resolved and there is actually a project I am involved in but I am really a little bit of an outsider here because this is actually a project that is run from the business school and sponsored by HRC which is a health research council and they want to look at how you can sensibly use that and now we are using sensibly not in the sense of sensing but how we would use sensibly use and how can you integrate it with social media technology so the picture that serves as a metaphor for this project is in the centre is your elderly person in the smart home and there is your circle of friends around close friends, relatives and then there are the wider circle like healthcare providers and so on and they are all connected and other things of communication and the idea behind this project or the question behind this project is to find out which kind of technology is acceptable where and to which degree and I guess one reason why this technology hasn't really spread widely is that there are still people are hesitant to just accept this technology so to conclude there seems of research and I am confident all of you are now so hyped up about it you will immediately go to the library which I think is over there and get the 10 news publication on smart homes and read all whole night about it tomorrow you go to any shop buy yourself some equipment and build your prototype like many people all around the world because there is no integrated package you can buy we are getting there there are still hurdles things like standardisation and so on and again there is another new research area but you can get already some smart devices for example some of you might have heard about NEST just to give you one example it's a controller for your air conditioning unit or your heat pump and it learns how you control the temperature over time and then does it automatically and of course I focus on smart homes but this is just one of the areas where this is used this technology can be used the other of course are smart cities controlling things in the city for example lighting probably for somebody living in Auckland not really an issue because the city is alive all night but if you go to Palmerston North the lights are on but nobody is home everybody is home that's the problem so there is a lot of waste of energy you light all the streets it would make much more sense to keep a minimal lighting for security reasons traffic picks up especially in more rural areas the lights come on a big thing for messy you probably know messy university was an agricultural college before it long ago turned into a university so part of the heritage is that agriculture is a big thing messy so smart agriculture or some people call precision agriculture as a hot topic and smart retail Amazon perhaps there has been some discussion recently New Zealand Amazon so there is a lot more and with that I should really stop and thank you for your patience and it was a pleasure talking to you thank you we do have a little bit of time for questions so if there are questions can you jump us up in the pan please I am going on out on a limp here a little bit because I am not an expert I am not a psychologist but reading a little bit in a wider area so one of the issues is actually emotional state and stress that people have through these diminished capabilities and they go hand in hand what you just mentioned the extreme case of the wish to live or the wish to die is sometimes influence by these factors and not only that sadly it also influences the person who is healthy so the person who takes care of a person with diminished mental capabilities there are studies that has a higher probability that they will die now with this technology this technology by no means make a decision about about these issues but it might help to ease a little bit on the emotional stress that people suffer through the onset of diminished mental capabilities because if you don't want to go to a retirement home and there have been studies that people actually want to live in their own home then the prospect of that might make things even worse and might actually lead to an extreme like you just mentioned that the people switches from now I'm happy to live to I had enough so it could but I don't have strong evidence about it I don't have data to back this up but I know that others actually do work in this area I think it was in Germany or some other European country where they actually built retirement homes where they equipped the rooms in a certain style let's say the 50s or the 60s or what the people are familiar with and the reason being if you suffer from dementia what goes are more recent events right so you suddenly feel not at home in a modern environment but if you recognize the room you have lived in or the style you have lived in as a kid or a young adult so they are they are going to see they are pursuing the same goal easing the stress on people but with different approaches I hope that answers some of your question I think there is a Japanese product which is just the Keto the only thing that sits in the house is Keto and that was good enough for a lot of people it's written that mum and dad had put the Keto on three times a day and they were okay not much a little bit so there are two issues one is driven by what goal do you want to achieve so it's just a particular assurance or something let's say you know that a problem would be that a person doesn't drink enough or doesn't do a certain thing enough a particular sense might give that indication it certainly doesn't help you to it's not enough to recognize complete activity what kind of activity is going on so from that point of view it's not sufficient as part of this project with the Hidden Markable there was a PhD project in the last phase the student actually tried to answer this question and the way to answer this is you have a set of sensors so you go the other way around you throw everything into the house so full throttle and then you analyze the information gain you get from computer science point of view with each of the sensors so the same in principle the same thing you do with decision trees and then you try to find a set of most informative sensors so that was the idea behind that and there was limited success we didn't get very far with it unfortunately but it's a very it's a very relevant question actually because we want to keep the cost down right so innovation and privacy computational resources and so on but I think that's absolutely correct and in a certain way these pill boxes are pretty much the same they focus on one aspect because if there's an issue, if there's medication it's really essential and that's what it's all about this does the job and I'm happy to share them we're putting up the slides of the lectures where they shade so they'll be on the given website excellent, great I'll do some first I have done a few attempts at data gathering for the message related stuff a few years ago as I hope we should have to interest it I might just need permission for the people to do it secondly I still I still have my old ACOS I'm in the mess in 1932 which was I think by the time when they attempted to publish it any possibility of possibly being published well out of blue eye it's difficult for me to say but if there's still something in there it's not for sure but it's something you need to look at it in more detail because the field progress quite dramatically in 1992 here we are with the 25 years okay can I ask you and joining with me and this isn't the end of it next week there's another one so Marcus Freen from Victoria University is going to be here talking about deep learning whatever that is