Embracing ambient intelligence
IST Results: What are the keywords for driving forward ambient intelligence?,John Taylor (JT): Attentional and conscious control. Computer systems - whether in wearable clothes, mobile phones, vehicles or any other machine - will increasingly be linked to embedded sensors. The danger is that all the incoming information will result in sensory overload, which may delay decision-making by the computer. The human brain has developed ways to cope with such overload, by filtering out all but the most important sensory inputs and making responses to achieve the person's main goals. We could call this state of mind 'consciousness'. IST Results: Is the aim to create consciousness in computers, so they function better?,JT: Research in the fields of brain science, psychology and philosophy suggests that attention, or focusing specifically on something, is the gateway to consciousness. Attention includes setting goals and monitoring progress towards them. There is a growing consensus that consciousness is linked to awareness and a sense of inner self. But the question is, can we implement this in machines? IST Results: You began your career as a particle physicist. Why did you later decide to study consciousness? ,JT: Being an applied mathematician, I realised you could apply maths to anything and everything. By applying it to the brain, you should be able to get consciousness. Some three decades ago I developed a neural model of consciousness from neural networks, which has now been extended by using attention, to give the model of consciousness a firmer basis in brain activity as observed by brain imaging techniques. This is related to the fact that we now realise, ever more strongly, that attention, arguably the brain's highest control faculty, is necessary for consciousness. Along the way, I and colleagues also developed a hardware chip with on-chip learning from looking closely at living brain cells in action, so as to speed up neural network processing. IST Results: How can your computational neuroscience work be used to advance software and hardware?,JT: It has extreme relevance, though we're still working hard to determine how much. Individual computer chips today contain from 10 to 100 million transistors. Within ten years, this figure could rise to more than a billion transistors. But many questions still remain about the best way to handle the information flow and processing on such a chip. I believe we will need clearly defined local as well as global interaction. So there will be local regions with one or 10 million transistors. But how will they all communicate? Even though local processing may be well defined in each cluster, how will the clusters work together? That comes back to the global communication problems on such a chip. IST Results: How can chip programmers handle such routing problems? ,JT: I believe brain architectures can give us new guidance for designing effective new chips with a billion or more transistors. We could build systems with active neurons and columns, which could be modified in learning and attentional control ways. It's still early days, but we already know that attention requires amplification and inhibition features, as well as control structures to knock out certain regions and amplify others. Attention is essentially filtering out things that you don't want to know about and processing a localised aspect. Local processing is really focusing on something that can cause danger. Future chips will need to be more flexible, so they can be adapted on the spot. IST Results: Do we really need 'creative' computers, as you suggest?,JT: We want a computer that can be spontaneous in creating new ideas - which is a very important feature of human brains. Don't forget that the human mind is infallible about itself, since it tells itself: I know that I am experiencing. That infallibility allows us to function as a unit. So the 'I' knows that 'I is I', a concept that computers do not have. It is possible to have distributed computing, but there is no person - no 'I' - in there. However, a robot which contained a 'personality' would be an improvement on existing models. It could, for example, get itself out of a tight spot by thinking creatively. A computer with an 'I' inside could also recognise other personalities around it - the humans that it is interacting with - and respond according to tried-and-tested human rules. Conscious machines would also be better at self-monitoring, using software or attentional agents. IST Results: You are involved in Oresteia, an IST project looking at 'intelligent' software agents that provide guidance to users. What are the challenges for this project?,JT: Over the next decade computing power will both increase enormously (Moore's Law) and disappear increasingly into the environment (mobiles, wearable agents, etc.). These trends will result in computing problems - such as scalability, awareness and autonomy - which could be tackled through the creation of attentionally controlled agents, ultimately with conscious attributes. Our project is interested in robots that can work with people, for example when driving vehicles. Control systems could take over when they sense the driver faces certain problems, such as fatigue or even a heart attack. Questions remain about the power of those systems and how they will relate to drivers in situations that could cause conflicts of interest. For example, people will want to get to their destination by a certain time, but they won't want to be driven too quickly around corners. An artificial-driver system must therefore sense many things. It must take account of various demands and learn how to drive properly, according to who is in the vehicle. IST Results: Which sensors is the project studying?,JT: We are building software attentional agents which take inputs from a variety of different sensors. They include biosensors - on the surface of the body of a driver (the project's main focus) or a patient in a hospital. We can measure things such as heart rate, blood pressure and pressure applied by hands to a wheel. Sensors on the car itself measure acceleration, velocity and turning trajectories, while others assess the external environment, such as the condition of the road. Lastly, there are sensors to look at other cars around the vehicle and the dangers they may represent. Each class of sensor has to be treated differently, yet combined to make an overall decision. The system could, for example, offer guidance to someone with high blood pressure approaching a traffic jam. After assessing the state of the person and the environment, it might suggest overtaking or changing route. Over time, the sensors will learn to adapt to the users of vehicles and their needs. Some people call this multimodal, many-sensor environment a 'disappearing computer' environment - currently a major research area in Europe. The computer 'disappears' because it becomes distributed over many computing elements across space. IST Results: Is there any commercial interest in Oresteia's research results?,JT: Yes. We want to place our attentional agent system in real hardware and then talk to car manufacturers and safety institutes. Some of them are developing similar systems of their own. Oresteia is focusing on the attentional control structures, to make them at least as efficient as those in the brain. IST Results: What is your long-term research goal?,JT: To drive forward implementation of a brain-like multichip system and its use in understanding human experience. That will lead to robot control systems and their implementation. Researchers could then also conduct experiments on such systems - checking their relation to the brain-science results of normal subjects and those with diseases such as depression, schizophrenia and autism. Source: Based on an interview with Professor Taylor and information from Oresteia (Modular Hybrid Artefacts with Adaptive Functionality). The IST Results service gives you online news and analysis on the emerging results from Information Society Technologies research. The service reports on prototype products and services ready for commercialisation as well as work in progress and interim results with significant potential for exploitation.
Mots‑clés
Pays
United Kingdom