Artificial Intelligence and home automation: Putting the human (back) at the centre of technological development
How do you make a group of intelligent agents communicate in a way that artificially copies human reasoning? How do you create artificial intelligence (AI) to substitute a virtual entity for a real entity? Informatics is a universe in perpetual evolution on a number of levels, including aspects such as material, software and design. Today, informatics is opening up to a new era: an era of ambient intelligence – the human assistance – that, through home automation is slowly changing our habits and improving our daily lives, but also disrupting our societies. This certainly is a progress, but behind this lie questions on the relationship between humans and machines.
Professor Amal El Fallah Seghrouchni presents the state of art relating to artificial intelligence and communication within multi agent systems (MAS) during a ‘Science at Heart’ conference at the University Pierre and Marie Curie (UPMC). Professor at UMPC and president of SMA college, Amal El Fallah Seghrouchni is in charge of the Multi-Agent Systems team in the Information Laboratory at Paris 6 University (LIP6).
Artificial intelligence can be defined as the science and engineering which aims to develop intelligent machines.
A machine is intelligent if it manages to pass itself off as a human when faced with a human. The Turing test (1950)"
Artificial intelligence seeks to develop processes similar to rational human thinking: a chain of events starting from detection, analysis to reflection, and finally, action. Some architectures take into account the intentional attitudes such as desires, beliefs and so on and integrate the practical rationality as a ‘means-ends’ reasoning.
Multi-agent systems: a teamwork
Multi-agent systems are groups of intelligent, interacting agents (otherwise referred to as a community or society). This discipline is a sub-branch of artificial intelligence also called distributed artificial intelligence. Intelligent agents can be virtual agents – for example software – or they can be real. In the case of the latter, the objective of MAS is to simulate a group of partly autonomous individuals that interact with the environment. For example, multi-agent systems provide an interesting means to modelsocieties, and have a large range of applications extending to human sciences among others. For example, the Terra Dynamica project aims to represent a virtual town combining interacting entities (vehicles, pedestrians etc.) Even though multi-agent simulation of urban movements is complex, the applications are multiple and would thus allow improvement in sectors such as security, urbanisation, public transport and so forth.
The simulation of aerial missions, a collaboration between the LIP6 and Dassault-Aviation, is a good example of collaborative simulations. In this project, an airborne system composed of several planes takes of towards an objective (an enemy interception for example). The planes in flight communicate with each other as well as with the control tower. As the tower detects a second enemy, it informs the formation of planes whose mission must then be modified. Decisions must be taken in order to adapt the mission and reorganise the tasks between the autonomous agents (the planes). Multiple communication requirements arise from this need for coordination in a collective context, between the observer (the control tower), the decision makers (cognitive agents) and the executing agents (reacting agents).
From artificial intelligence towards ‘distributed’ artificial intelligence
Collective missions present a hybrid type of architecture that mixes agents of different levels. Two different MAS streams can be recognised: the reactive approach and the deliberative approach. A reactive agent perceives the world more than it conceives of it. It reacts to a stimulus with an action. It forms part of a simple architecture that allows the self-organisation of a large number of agents. An ant colony is one example of this. A deliberative agent can be recognised by the ‘awareness’ of its own existence within the system. It develops a symbolic representation of itself, of the environment and the other agents. It allows intentional organisation (the allocation and dependant nature of tasks, resource sharing, and negotiation/coordination protocols) as part of complex architectures (BDI) the number of agents of which depends on the degree of cognition.
These collective and collaborative situations exist in a broader context in which agents are virtual entities, one example of which is the collective performance of software dedicated to communicating within an informatics system. Artificial intelligence is thus referred to as ‘distributed’ and adapted tools must be developed in order to simulate complex situations and to reproduce sophisticated behaviour. Questions to be solved include the development of intelligent agents that operate independently at least in part, languages of communication between agents, and systems that simulate collective coordination.
The agent is characterised by four criteria: 1) its reactivity characterises its capacity to perceive its environment and react to changes in real time; 2) its proactive nature characterises its capacity to take initiatives; 3) its social nature expresses its capacity to integrate with other agents, and, 4) its autonomy – defined on several levels (power source, human intervention and so on).
Programming languages can be defined according to three main areas: the level of abstraction, the program structure, and the choice and form of the action (in order to define its temporality). Linguistics distinguishes three speech acts: locutionary, illocutionary and perlocutionary following the action induced by the oral enunciation towards the recipient (incitation, engagement and so on).
The contribution of learning arises from the powerful tool that it puts at the disposition of the individual: language. Lev Vygotsky"
Towards mixed systems resembling those of humans
In a more societal approach, communication tools between humans and intelligent systems are being developed and aims to put humans back into the centre of technological development. There are a number of areas of application within this framework: companionship gadgets, entertaining games, accompaniment to ease solitude or provide help for a handicap, life or company simulators as well as interventions in dangerous zones (de-mining, nuclear accidents).
Embodied Conversational Agents (ECA) can be introduced to isolated people to simulate a conversation, an exchange or even a presence. They can also be used to mediate a human-human communication that allows role-playing as in a job interview for example. Another use is as mediator (in the mediated behaviour sense, adopted by people with autism) to encourage those with autism to interact with people without autism. In a general way they can be used to replace humans to help in cases of lack of staff, or isolation.
The Philips iCat for example plays chess, provides commentaries on strategic moves and permits interaction in which the robot and the human talk together. The robot-cat is a game companion that can analyse human emotions according to aspects of varying facial expressions: frowning, smiling, change of focus, cheekbones...
The robot-cat speaks and acts according to the person facing it, inspires emotions, creates and maintains a dialogue. Conversational agents represent a step forward for artificial intelligence because of the action and decision-making taken without direct human input.
Ambient intelligence: putting the human at the centre of technological development
Human assistance, or ‘ambient intelligence’ designates MAS to be used to assist humans in their daily lives. It is being developed specifically for people with reduced mobility or those who need specific help. More systematically, human assistance is entering our houses as home automation that integrates different systems affecting the habited environment (including lighting control, security, energy use) in an effort to coordinate them. It aims to intelligently coordinate all white goods that, communicating together, ensure greater comfort and optimise the security of individuals while reducing energy consumption.
EDF, in collaboration with LIP6 has developed the SMACH project (Multi-Agent Simulation of Human Behaviour). For EDF, the goal of the project is to develop a tool that allows the study of a user’s home behaviour and the impact of so-called ‘New generation’ intelligent products. Ergonomists have worked in partnership with informatics researchers to transcribe everyday behaviour and simulate the behaviour of a house’s inhabitants in order to improve comfort, and above all, to reduce energy consumption. Three areas of research have been developed at LIP6 as part of this project. First of all, are studies based on the simulation of multi-agents that (generically) reproduce human behaviour while stressing the ease with which they can be used by people unacquainted with informatics. Secondly, an interesting tool has been developed in conjunction with the Institute for research and development (IRD). The tool is trying to allow collaborative programming in interaction with the user who gives advice and prompts modifications throughout the programme’s development phase. Thirdly, research teams envisage the introduction of automatic learning functions, so that objects from daily life have the capacity to adapt to the introduction of a new device or to the behaviour of individuals.
According to Nicolas Sabouret, professor-researcher at the LIP6 in the MAS team, the SMACH project team can announce that, during a test simulation, human participants were unable to differentiate actors played by the multi-agent systems from those controlled by humans - the start of a Turing test.
Although SMACH is a long-term project, it covers a lot of ground and allows the development of a number of broad areas of research that meet societal expectations. Surprising and encouraging results have already been obtained. It has for example been shown that it is possible to reduce the energy consumption of a habitation by 60% with only a slight modification of the habits of inhabitants. It has also been found that a tool simulating energy consumption on a large scale would enable the optimal distribution of nuclear energy as well as the anticipation of peak and off-peak periods, according to regions, current happenings (sports events for example) and a large number of important social parameters.
These systems function thanks to cutting-edge technology objects fitted with sensors, microprocessors, built-in software, means of communication and advanced informatics architectures. Ambient intelligence is thus at the heart of current technological advances and also allows social and societal studies - which invasive character should be regularly re-evaluated. The technical evolution is no longer only the subject of attention from engineers and scientists.
Assistance system for older people and those losing independence is a social issue residing at the heart of a society in which the ratio between older and younger people is on the increase. In France, 2005, there were 2.2 people in the workforce for one person non-working. Due to the growth in the number of elderly people, it is predicted that in 2050 there will be no more than 1.4 people in the workforce for every one person outside of the workforce (source: Insee). So it is that Amal El Fallah Segrouchni would dearly like to see the introduction of MASs to cater for people in such diverse domains as individual transport, home automation and health. The aim is to conceive of systems that understand their environment and, without being invasive, are capable of self-adaptation.
Since the advent of the computer, the evolution and ubiquity of informatics material have given rise to new demands in terms of adapted networks and widely available architecture. Challenges that confront ambient intelligence include managing the heterogeneity of agents, enabling an openness to change (offering to all the possibility of enriching the system), the capacity to dynamically manage tasks, and the usability of the man-machine interface.
Ambient intelligence is thus a project for the future that must take software, the man-machine interface, human trust in machines and security/confidentiality of privacy into account. Scientists in other domains such as sociologists, software and informatics security engineers, and philosophers must work at all levels of development closely linked to the development of informatics.
 Towards an Empathic Chess Companion, I. Leite, A. Pereira, C. Martinho, A. Paiva, G. Castellano, http://www.lirec.org/biblio/1706 Find out more: 1) Scenarios for ambient intelligence in 2010, ISTAG UE (2001), ftp://ftp.cordis.europa.eu/pub/ist/docs/istagscenarios2010.pdf 2) Fundamentals of Multiagent Systems with NetLogo Examples, text book by José M. Vidal, (2010), http://multiagent.com/ 3) Massive is the premier simulation and visualization solution system for generating and visualizing realistic crowd behaviors and autonomous agent driven animation for a variety of industries, http://www.massivesoftware.com/ 4) Contribution des Sciences Sociales dans le domaine de l’Intelligence Artificielle Distribuée, thèse d’Isabelle Jars (2005), Université Lyon 1, http://tel.archives-ouvertes.fr/docs/00/04/98/85/PDF/These_IJ_2005.pdf 5) iCat, the Chess Tutor : An Affective Game Buddy Based on Anticipatory Mechanisms, these de master 2 de I. M. dos Santos Carvalho Leite (2007), https://dspace.ist.utl.pt/bitstream/2295/153068/1/MScThesis_IolandaLeite.pdf 6) Système domotique Multi-Agents pour la gestion de l’énergie dans l’habitat, thèse de S. Abra (2009), http://stephane.ploix.com/IMG/pdf/these-shadi-soumission.pdf