Ambient Intelligence (Artificial Intelligence)


In recent years much research and development effort has been directed towards the broad field of ambient intelligence (AmI), and this trend is set to continue for the foreseeable future. AmI aims at seamlessly integrating services within smart infrastructures to be used at home, at work, in the car, on the move, and generally in most environments inhabited by people. It is a relatively new paradigm rooted in ubiquitous computing, which calls for the integration and convergence of multiple disciplines, such as sensor networks, portable devices, intelligent systems, human-computer and social interactions, as well as many techniques within artificial intelligence, such as planning, contextual reasoning, speech recognition, language translation, learning, adaptability, and temporal and hypothetical reasoning.

The term AmI was coined by the European Commission, when in 2001 one of its Programme Advisory Groups launched the AmI challenge (Ducatel et al., 2001), later updated in 2003 (Ducatel et al., 2003). But although the termAmI originated from Europe, the goals of the work have been adopted worldwide, see for example (The Aware Home, 2007), (The Oxygen Project, 2007), and (The Sony Interaction Lab, 2007).

The foundations of AmI infrastructures are based on the impressive progress we are witnessing in wireless technologies, sensor networks, display capabilities, processing speeds and mobile services. These developments help provide much useful (row) information for AmI applications. Further progress is needed in taking full advantage of such information in order to provide the degree of intelligence, flexibility and naturalness envisaged. This is where artificial intelligence and multi-agent techniques have important roles to play.

In this paper we will review the progress that has been made in intelligent systems, discuss the role of artificial intelligence and agent technologies and focus on the application of AmI for independent living.


Ambient intelligence is a vision of the information society where normal working and living environments are surrounded by embedded intelligent devices that can merge unobtrusively into the background and work through intuitive interfaces. Such devices, each specialised in one or more capabilities, are intended to work together within an infrastructure of intelligent systems, to provide a multitude of services aimed at generally improving safety and security and improving quality of life in ordinary living, travelling and working environments.

The European Commission identified four AmI scenarios (Ducatel et al. 2001, 2003) in order to stimulate imagination and initiate and structure research in this area. We summarise two of these to provide the flavour of AmI visions.

AmI Scenarios:

1. Dimitrios is taking a coffee break and prefers not to be disturbed. He is wearing on his clothes or body a voice activated digital avatar of himself, known as Digital Me (D-Me). D-Me is both a learning device, learning about Dimitrios and his environment, and an acting device offering communication, processing and decision-making functionalities. During the coffee break D-Me answers the incoming calls and emails of Dimitrios. It does so smoothly in the necessary languages, with a re-production of Dimitrios’ voice and accent. Then D-Me receives a call from Dimitrios’ wife, recognises its urgency and passes it on to Demetrios. At the same time it catches a message from an older person’s D-Me, located nearby. This person has left home without his medication and would like to find out where to access similar drugs. He has asked his D-Me, in natural language, to investigate this. Dimitrios happens to suffer from a similar health problem and uses the same drugs. His D-Me processes the incoming request for information, and decides neither to reveal Dimitrios’ identity nor offer direct help, but to provide the elderly person’s D-Me with a list of the closest medicine shops and potential contact with a self-help group.

2. Carmen plans herjourney to work. It asks AmI, by voice command, to find her someone with whom she can share a lift to work in half an hour. She then plans the dinner party she is to give that evening. She wishes to bake a cake, and her e-fridge flashes a recipe on the e-fridge screen and highlights the ingredients that are missing. Carmen completes her shopping list on the screen and asks for it to be delivered to the nearest distribution point in her neighbourhood. All goods are smart tagged, so she can check the progress of her virtual shopping from any enabled device anywhere, and make alterations. Carmen makes her journey to work, in a car with dynamic traffic guidance facilities and traffic systems that dynamically adjust speed limits depending on congestion and pollution levels. When she returns home the AmI welcomes her and suggests that on the next day she should telework, as a big demonstration is planned in downtown.

The demands that driveAmI and provide opportunities are for improvement of safety and quality of life, enhancements of productivity and quality of products and services, including public services such as hospitals, schools, military and police, and industrial innovation. AmI is intended to facilitate human contact and community and cultural enhancement, and ultimately it should inspire trust and confidence.

Some of the technologies required for AmI are summarised in Figure 1.

AmI work builds on ubiquitous computing and sensor network and mobile technologies. To provide the intelligence and naturalness required, it is our view that significant contributions can come from advances in artificial intelligence and agent technologies. Artificial intelligence has a long history of research on planning, scheduling, temporal reasoning, fault diagnosis, hypothetical reasoning, and reasoning with incomplete and uncertain information. All of these are techniques that can contribute to AmI where actions and decisions have to be taken in real time, often with dynamic and uncertain knowledge about the environment and the user. Agent technology research has concentrated on agent architectures that combine several, often cognitive, capabilities, including reactivity and adaptability, as well as the formation of agent societies through communication, norms and protocols.

Recent work has attempted to exploit these techniques for AmI. In (Augusto and Nugent 2004) the use of temporal reasoning combined with active data bases are explored in the context of smart homes. In (Sadri 2007) the use of temporal reasoning together with agents is explored to deal with similar scenarios, where information observed in a home environment is evaluated, deviations from normal behaviour and risky situations are recognised and compensating actions are recommended.

Figure 1. Components of Ambient Intelligence

Components of Ambient Intelligence

The relationship of AmI to cognitive agents is motivated by (Stathis and Toni 2004) who argue that computational logic elevates the level of the system to that of a user. They advocate the KGP agent model (Kakas, et al 2004) to investigate how to assist a traveller to act independently and safely in an unknown environment using a personal communicator. (Augusto et al 2006) address the process of taking decisions in the presence of conflicting options. (Li and Ji 2005) offer a new probabilistic framework based on Bayesian Networks for dealing with ambiguous and uncertain sensory observations and users’ changing states, in order to provide correct assistance.

(Amigoni et al 2005) address the goal-oriented aspect of AmI applications, and in particular the planning problem withinAmI. They conclude that a combination of centralised and distributed planning capabilities are required, due to the distributed nature of AmI and the participation of heterogeneous agents, with different capabilities. They offer an approach based on the Hierarchical Task Networks taking the perspective of a multi-agent paradigm for AmI.

The paradigm of embedded agents for AmI environments with a focus on developing learning and adaptation techniques for the agents is discussed in (Hagras et al 2004, and Hagras and Callaghan 2005). Each agent is equipped with sensors and effectors and uses a learning system based on fuzzy logic. A realAmI environment in the form of an “intelligent dormitory” is used for experimentation.

Privacy and security in the context of AmI applications at home, at work, and in the health, shopping and mobility domains are discussed in (Friedewald et al 2007). For such applications they consider security threats such as surveillance of users, identity theft and malicious attacks, as well as the potential of the digital divide amongst communities and social pressures.


One major use of AmI is to support services for independent living, to prolong the time people can live decently in their own homes by increasing their autonomy and self-confidence. This may involve the elimination of monotonous everyday activities, monitoring and caring for the elderly, provision of security, or saving resources. The aim of such AmI applications is to help:

• maintain safety of a person by monitoring his environment and recognizing and anticipating risks, and taking appropriate actions,

• provide assistance in daily activities and requirements, for example, by reminding and advising about medication and nutrition, and

• improve quality of life, for example by providing personalized information about entertainment and social activities.

This area has attracted a great deal of attention in recent years, because of increased longevity and the aging population in many parts of the world. For such an AmI system to be useful and accepted it needs to be versatile, adaptable, capable of dealing with changing environments and situations, transparent and easy, and even pleasant, to interact with.

We believe that it would be promising to explore an approach based on providing an agent architecture consisting of a society of heterogeneous, intelligent, embedded agents, each specialised in one or more functionalities. The agents should be capable of sharing information through communication, and their dialogues and behaviour should be governed by context-dependent and dynamic norms.

The basic capabilities for intelligent agents include:

• Sensing: to allow the agent observe the environment

• Reactivity: to provide context-dependent dynamic behaviour and the ability to adapt to changes in the environment

• Planning: to provide goal-directed behaviour

• Goal Decision: to allow dynamic decisions about which goals have higher priorities

• Action execution: to allow the agent to affect the environment.

All of these functionalities also require reasoning about spatio-temporal constraints reflecting the environment in which an AmI system operates.

Most of these functionalities have been integrated in the KGP model (Kakas et al, 2004), whose architecture is shown in Figure 2 and implemented in the PROSOCS system (Bracciali et al, 2006). The use of reactivity for communication and dialogue policies has also been discussed in, for example, (Sadri et al, 2003). The inclusion of normative behaviour has been discussed in (Sadri et al, 2006) where we also consider how to choose amongst different types of goals, depending on the governing norms. For a general discussion on the importance of norms in artificial societies see (Pitt, 2005).

KGP agents are situated in the environment via their physical capabilities. Information received from the environment (including other agents) updates the agents state and provides input to its dynamic cycle theory, which, in turn, determines the next steps in terms of its transitions, using its reasoning capabilities.


As most other information and communication technologies, AmI is not likely to be good or bad on its own, but its value will be judged from the different ways the technology will be used to improve people’s lives. In this section we discuss new opportunities and challenges for the integration of AmI with what people do in ordinary settings. We abstract away from hardware trends and we focus on areas that are software related and are likely to play an important role in the adoption of AmI technologies.

Figure 2. The architecture ofa KGP agent

The architecture ofa KGP agent

A focal point is the observation that people discover and understand the world through visual and conversational interactions. As a result, in the coming years we expect to see the design of AmI systems to focus in ways that will allow humans to interact in natural ways, using their common skills such as speaking, gesturing, glancing. This kind of natural interaction (Leibe et al 2000) will complement existing interfaces and will require that AmI systems be capable of representing virtual objects, possibly in 3D, as well as capture people’s moves in the environment and identify which of these moves are directed to virtual objects.

We also expect to see new research directed towards processing of sensor data with different information (Massaro and Friedman 1990) and different kind of formats such as audio, video, and RFID. Efficient techniques to index, search, and structure these data and ways to transform them to the higher-level semantic information required by cognitive agents will be an important area for future work. Similarly, the reverse of this process is likely to be of equal importance, namely, how to translate high-level information to the lower-level signals required by actuators that are situated in the environment.

Given that sensors and actuators will provide the link with the physical environment, we also anticipate further research to address the general linking of AmI systems to already existing computing infrastructures such as the semantic web. This work will create hybrid environments that will need to combine useful information from existing wired technologies with information from wireless ones (Stathis et al 2007). To enable the creation of such environments we imagine the need to build new frameworks and middleware to facilitate integration of heterogeneous AmI systems and make the interoperation more flexible.

Another important issue is how the human experience in AmI will be managed in a way that will be as unobtrusive as possible. In this we foresee that developments in cognitive systems will play a very important role. Although there will be many areas of cognitive system behaviour that will need to be addressed, we anticipate that development of agent models that adapt and learn (Sutton and Barto 1998), to be of great importance. The challenge here will be how to integrate the output of these adaptive and learning capabilities to the reasoning and decision processes of the agent. The resulting cognitive behaviour must differentiate between newly learned concepts and existing ones, as well as discriminate between normal behaviour and exceptions.

We expect that AmI will emerge with the formation of user communities who live and work in a particular locality (Stathis et al 2006). The issue then becomes how to manage all the information that is provided and captured as the system evolves. We foresee research to address issues such as semantic annotations of content, and partitioning and ownership of information.

Linking in local communities with smart homes, e-healthcare, mobile commerce, and transportation systems will eventually give rise to a global AmI system. For applications in such a system to be embraced by people we will need to see specific human factors studies to decide how unobtrusive, acceptable and desirable the actions of the AmI environment seem to people who use them. Some human factors studies should focus on issues of presentation of objects and agents in a 3D setting, as well as on the important issues of privacy, trust and security.

To make possible the customization of system interactions to different classes of users, it is required to acquire and store information about these users. Thus for people to trust AmI interactions in the future we must ensure that the omnipresent intelligent environment maintains privacy in an ethical manner. Ethical or, better, normative behaviour cannot only be ensured at the cognitive level (Sadri et al 2006), but also at the lower, implementation level of the AmI platform. In this context, ensuring that communicated information is encrypted, certified, and follows transparent security policies will be required to build systems less vulnerable to malicious attacks. Finally, we also envisage changes to business models that would characterise AmI interactions (Hax and Wielde 2001).


The successful adoption of AmI is predicated on the suitable combination ofubiquitous computing, artificial intelligence and agent technologies. A useful class of applications that can test such a combination is AmI supporting independent living. For such applications we have identified the trends that are likely to play an important role in the future.


Artificial Societies: Complex systems consisting of a, possibly large, set of agents whose interaction are constrained by norms and the roles the agents are responsible to play.

Cognitive Agents: Software agents endowed with high-level mental attitudes, such as beliefs, goals and plans.

ContextAwareness: Refers to the idea that computers can both sense and react according to the state of the environment they are situated. Devices may have information about the circumstances under which they are able to operate and react accordingly.

Natural Interaction: The investigation of the relationships between humans and machines aiming to create interactive artifacts that respect and exploit the natural dynamics through which people communicate and discover the real world.

Smart Homes: Homes equipped with intelligent sensors and devices within a communications infrastructure that allows the various systems and devices to communicate with each other for monitoring and maintenance purposes.

Ubiquitous Computing: A model of human-computer interaction in which information processing is integrated into everyday objects and activities. Unlike the desktop paradigm, in which a single user chooses to interact with a single device for a specialized purpose, with ubiquitous computing a user interacts with many computational devices and systems simultaneously, in the course of ordinary activities, and may not necessarily even be aware that is doing so.

Wireless Sensor Networks: Wireless networks consisting of spatially distributed autonomous devices using sensors to cooperatively monitor physical or environmental conditions, such as temperature, sound, vibration, pressure, motion or pollutants, at different locations.

Next post:

Previous post: