In many ways, the raison d'etre of computers is to perform tasks that humans either cannot do or do not want to do, from the mundane (such maintaining database records and sorting them into order) to the tedious (such as performing the millions of calculations required to produce even a short-term weather forecast). From this viewpoint, the assistance of particular sectors of society who are disadvantaged in some physical or mental capacity, thus performing some task that they cannot do easily, is a natural use of computers, from the development of speech synthesisers for those who cannot talk and automatic readers for the blind through to powered legs to assist those who have trouble walking.
As computers become more integrated into our everyday lives, and as we start to expect to interact with them in a more ‘natural’ way (that is, not just through a keyboard and mouse), so we start to expect real intelligence from them in the form of relevant responses and an understanding of what we are doing. It is this challenge of ‘ambient’ computing that can be seen as one of the hardest: as humans, we generally know instinctively what tasks other humans are engaged in, and can usually guess why, and so we respond appropriately in conversations and other social interactions. To date, our interactions with computers have been on their terms, and the assistance that they have rendered has been carefully specified to be particular to some specific task. We would not expect a database program to assist in writing a letter, nor a speech synthesiser to help us plan travel routes, and so we select the technology that we use for each task carefully.
While there is research in a great many aspects of ambient or ubiquitous computing, one area that is receiving particular interest is that of assisted living. This is an area of profound importance for human society at the current time: improvements in medicine mean that a ‘baby boomer’ now entering retirement in the Western world can expect to live for another 25 years on average, and as they enter the group of ‘oldest-old’ (those over 85 or 90) they will need significant assistance from carers and other medical professionals. As birth rates in the Western world drop, so the increasing proportion of the elderly in the population (sometimes called the ‘grey tsunami’) means that providing full-time human care for the elderly – even if it were economically viable – is not plausible. However, in order to be actually useful, an assisted living solution will need to have the ability to interact with, and interpret the actions of, the humans that it is assisting.
Thus, one of the particular challenges that ambient intelligence engenders is the requirement for the computer to understand the situation of the human, and therefore to choose and tailor the response appropriately. In addition to an interpretation of the requirements of the human—exactly what activity the human is attempting to perform, and what help they might require—a lot of this problem is concerned with context, i.e. the specific environment in which the humans are interacting with the computer and that modifies their activities and responses. While there are many aspects to context awareness, two of the most important are those of time and place: when and where activities happen.
This observation that spatial and temporal aspects of context awareness are important leads, as so often in human intellectual effort, to the possibility of knowledge transfer: there is a rich literature of spatio-temporal reasoning in symbolic Artificial Intelligence, and so this is a natural place to begin work on reasoning about context. From a sound basis there, it is to be hoped that the extension to other areas of context, whether environmental (e.g., temperature and humidity) to personal (e.g., emotion, boredom level) to the relationship with other activities (e.g., I just had went for a walk, so I don't want to go for another one just at the moment) will be simpler.
In the previous discussion an important and rather loaded word was used: reasoning. Looking at the problem of assisted living from the viewpoint of artificial intelligence – and it is hard to think of any reason why this is not the natural stance – it is clear that both logical and statistical methods will be needed: systems will need to be able to recognise what activities people are engaged in, deduce why that could be the case, identify relevant contextual variables and generalise them appropriately to other activities. They will also need to be able to identify the potential role that the system should play as assistant and reason about how their interactions will modify those of the human. In the papers in this book the approaches to context awareness, arising as they do from the spatiotemporal reasoning community, are primarily symbolic: witness the worlds ‘planning,’ ‘knowledge representation,’ and ‘reasoning’ in their titles. This leads to approaches that in some sense echo the concepts of spatial and temporal cognition, with the benefit of outputs that can be explained by the system and that are therefore more likely to be trusted, an important concept in all aspects of ambient intelligence, but most especially assisted living: unless the system can be queried to understand why it decided on the particular actions that it produced, it can be rather hard to identify defects in training and rectify them appropriately.
The more we expect a computer system to interact with us on our terms, the more understanding of human drives and behaviours the system needs. That is the implicit target behind commonsense situational awareness, one of the main aims of the papers in this book; we expect an artificially intelligent system to act (and react) in a way that we, with our human knowledge, would consider to be commonsensical. This is an ambitious goal, not least because we are tuned so well to recognise human behaviours and respond to them, and so anything that nearly, but not quite, meets our expectations produces a strong adverse reaction, even when it is other humans responding in ways that we do not expect. It can be hard to articulate the feeling we experience, let alone explain it: something is just ‘off’. Even defining what commonsense is turns out to be a challenge; like art, we might not know about it, but we know what we like.
The papers in this book tell us about the diversity of the field of ambient intelligence in the wide range of approaches and the variety of applications that they suggest. There is consideration of how space, time, and other context can be represented, modelled, and reasoned about, for use in mapping, multi-agent interactions, assisted living, and even emergency responses, using many techniques from the gamut of artificial intelligence. It is this variety that represents one of the strengths of the area, since the weaknesses of one approach can be offset by the capabilities of others.
To close, when thinking about the current state of commensensical in ambience intelligence I'm put in mind of the famous quotation from Samuel Johnson on a small part of the female bid for equality: “Sir, a woman's preaching is like a dog's walking on his hind legs. It is not done well; but you are surprised to find it done at all.” With developments such as those in this book, the equivalent of bipedal movement for assistive technologies comes a step closer.
Massey University, New Zealand