Raúl Arrabales is assistant lecturer in the computing department of Carlos III University in Madrid. He specialises in artificial consciousness and has set up a web site exclusively to consider this topic: www.Conscious-Robots.com
Although Arrabales is a robotic consciousness specialist, this interview is mainly directed to understanding human consciousness. He refers to Edelman and Tononi who address the neurological origin of consciousness and supports the collaboration between psychologists and neurologists to explore the issue. He characterises consciousness as a non-enigmatic process – it is not necessary to invoke new laws of quantum physics to understand it – and he believes that the day will come when mankind will design conscious machines.
Tercera Cultura: What is consciousness? What is the best theory or model that in your opinion explains this phenomenon? Will we be able to materialise subjective experience? What is the theory that in your opinion has come closest to explaining this phenomenon?
Raúl Arrabales: In the first place, in Spanish we distinguish between two different meanings of the word ‘consciencia’. On one hand, this word refers to the immediate knowledge that one being has of him or herself, but on the other hand, it also refers to the capacity that human beings have for judging between good and bad [as it can be translated into English as consciousness or conscience]. In fact, this second meaning refers to the metaphorical voice that tells us that we have done wrong.
My field of investigation focuses on the first concept, that of self-awareness. In fact, consciousness covers a multitude of related concepts. From this point of view, we can consider consciousness as the effective integration of various mental abilities, such as attention, feelings, the meaning of I, taking decisions, imagination, empathy, etc.
As regards the scientific study of consciousness and the models attempting to explain it, I would like to highlight two points that I think are relevant. In the first place, I believe that we still do not have a satisfactory theory that gives a complete explanation about what consciousness is and how it is produced. In the second place, although there are various hypotheses trying to explain consciousness, I do not believe that their approaches are exclusive or that they are necessarily in competition. In reality, each hypothesis approaches specific aspects of the consciousness phenomenon. What is clear, however, is that some theories have been more successful among the scientific community studying consciousness. Here we can distinguish between two types of theory: the psychological and the neurological. An example of a psychological theory is Baars’ Global Workspace theory. This theory uses the simile of a theatre play to explain consciousness. This explanation is only a metaphor that helps understanding about what consciousness is, but does not provide any indication about how it is produced in humans. By contrast, the neurological theories are focussed on searching for existing mechanisms in the central nervous system that give rise to consciousness. A relevant example is Edelman and Tononi’s Dynamic Core hypothesis that tries to explain how the neuronal connections of the thalamocortical loop give rise to the perception of consciousness. In general, there is consensus among neurologists that the thalamocortical loop is key to the generation of consciousness.
I believe that the current trend is to unite high level explanations proposed by psychological theories and hypotheses regarding low level neuronal substrata that support them. In fact, studies on consciousness always imply a multidisciplinary effort and I don’t think it would be right to cite one particular theory as giving the best explanation. It is more likely that collaboration between technical, theoretical and empirical disciplines will turn out to give a satisfactory answer in the long run. We should also not forget that the sciences of computation and artificial intelligence can be especially complementary in clarifying some aspects of consciousness.
Regarding the materialisation of subjective experience, I assume you are referring to the construction of machines with subjective experiences. Here, the difficulty lies in that the main characteristic of subjective experience is something that is not material. From my point of view, it is a process. Therefore, the proper question to ask would be: is it possible to artificially reproduce the process giving rise to subjective experience? Or in other words, Can we create a non-material mind in a machine? This question is related to the so-called “hard problem” of consciousness, which asks for an explanation in scientific terms of something that is not physical and that cannot be demonstrated by the classical scientific method. In order to study any phenomenon in a scientific way, it is essential that it can be observed by a third party. However, subjective experience falls outside this restriction because it is not, in itself, composed of physical parameters. Only I can access my own subjective experience; a scientist cannot observe it (observing an output of a brain scanner is not observing subjective experience). There is no other alternative than to believe what I say about my subjective experience. By chance, we ourselves can act as if we were scientists of consciousness and observe our own subjective experience by the technique of introspection. But to give a final answer to the question of whether it is possible to reproduce non-material subjective experiences in machines, I think that it is possible. In fact, computer programmers are accustomed to creating non-material entities starting from conventional hardware. A programme, when it executes in a computer, becomes a process that is non-material but executes on a substrate constructed on a core of silicon chips.
TC: It seems that the model for designing robots is the human being. The aim is for machines to substitute for humans in many areas and to make human work more efficient. Will there be conscious robots one day?
Raúl Arrabales: Obviously in the field of robotics the model to reproduce has always been the human being. However, owing to the fact that this is a huge challenge, ambitions have been temporarily curtailed to take other simpler animals as a model. Currently, due to the latest technical advances that have been achieved, it seems that we are returning to the original challenge of humanoid robots, but their cognitive capacities are still at a fairly rudimentary stage. I personally think that we will be able to construct conscious robots but it may be that the consciousness experienced by these robots will not be like our own. In fact, I think that the artificial consciousness of a robot should be adapted to the type of work it must undertake. Subjective experience and feelings are not required for soldering things on an assembly line but these capacities may be crucial for a social robot helping older people.
TC: The philosopher Víctor Gómez Pin claims that artificial intelligence is an illusion, that there is no such thing, while neurologists, such as Adolf Tobeña consider that intelligent machines are already with us. What is your point of view? Which opinion seems to be mistaken and what would you say to expose the error?
Raúl Arrabales: These days, the term Artificial Intelligence is used to designate a well established field of scientific investigation but one which is only 50 years old. I think that the concept of intelligence should be considered as the capacity of a being to survive and adapt to the environment in which it finds itself. I have no doubt that the property of intelligence can exist in a machine. The level of intelligence that can be developed by different entities in accordance with their design is something else. When we speak of intelligence and of consciousness, we often talk in absolute terms. But I prefer to talk of qualitative levels or quantitative measures (although establishing these measures may be complicated). So I agree that today we do not have machines with the same intelligence as a human being, but we do have machines with some intelligence. I think that we have machines with the intelligence of a mosquito, for example. One of the objectives of artificial intelligence is to equip machines with intelligence similar to that of humans; for this the capacities of common sense and imagination are required. The problem is that today we do not have machines that are capable of successfully confronting new situations in complex environments, but we do have automatic learning implementations that are capable of learning and generalising in some controlled environments. For example, we have artificial intelligence systems capable of automatically detecting harmful intrusions into computer networks.
TC: What does the term computability mean? How is this notion used to reject the possibility that a machine can become conscious? Do you think it will be necessary to resort to new laws of nature to understand consciousness and therefore to design conscious computers?
Raúl Arrabales: Some investigators equate the human brain to a computer because, just like a computer, it receives inputs, processes them (that is, computes) and finally produces an output. But the brain produces something more than just a physical output (behaviour). It also produces what we know as a subjective experience. The problem is that the concept of a philosophical zombie is a priori possible from the purely scientific point of view. A zombie is a thing that processes the inputs of the senses and produces an adapted behaviour but has no subjective experience at all. How do we know that a person feels something when they cry out with pain? It could simply be a programmed response in the brain, just like a robot could be programmed to say, “Ow!” when a pressure sensor is activated. What normally happens when we see a person cry out in pain is that we infer that they feel pain in the same way that we would feel pain in the same situation. That is, “we put ourselves in their place” and feel empathy. However, if we see a robot cry out in pain, how do we know that it is really feeling something? In this case, we cannot establish an analogy with ourselves because it is a completely different machine. Does this mean that it is not possible for a robot to have subjective experience? The key lies in knowing if the subjective experience is produced by the specific biological substrate that human beings have. Some researchers, such as Penrose and Hameroff, argue that subjective experience is produced by quantum phenomena that take place in the microtubules present in the neurones. If this was the case, then we could only build conscious machines using quantum computers that reproduced these types of processes.
In fact, I think that it is not necessary to resort to quantum mechanics to understand consciousness and that we can create conscious processes in machines by considering only the known laws of nature. I see consciousness as a specific process in execution. For me, consciousness is not a property of matter; it is simply a process.
TC: There have been some trends, such as strong Artificial Intelligence, that have extreme points of view about intelligence. Do you think there is any basis for such theories?
Raúl Arrabales: In reality, strong AI aligns with the original objective of the subject of AI itself: to create machines that are as intelligent and as conscious as humans. What happened is that the objectives had to be relaxed due to the false expectations that they had created. We have realised the tremendous complexity of this task, but that does not mean that this branch of AI has been abandoned. In fact, in the last few years we have been witnessing a resurgence of this hard branch of AI materialised in the two lines of Artificial Consciousness and General Artificial Intelligence. Personally I think that they are fields that deserve exploring further to determine the effective limits. What I am sure of is that there is no basis for claiming that strong AI objectives are not reachable.
TC: Do you think that the educational potential of computers is being fully realised? What could be introduced into schools to make learning more effective? In the past, rich families could hire tutors for their children, who received personalised and efficient education. Can computers fulfil this role in a more democratic way?
Raúl Arrabales: Of course, these days it would be a pity if the potential of computers were not used to a large extent in education. Information technologies in general are an excellent tool for improving education. Neither teachers nor students should reject such powerful tools, especially as they are now available at low cost, at least in the developed world. As with all tools, the work of the teacher is to show students how to make the best use of them. In the future we can expect teaching robots, which will present a much more natural interface for education.
TC: According to Varela, a nervous system is not an information processing system because, by definition, information processing systems require clear inputs. Can you explain what he means?
Raúl Arrabales: I would rather say that a nervous system is an analogue information processing system, where the information received from the senses can take continuous values. By contrast, a conventional computer is based on digital circuits where the inputs take discrete values. No matter which way we think about the term ‘processing’, in both cases information is being processed and outputs are being generated.