|
|
Ethics of Human Interaction with Robotic, Bionic, and AI Systems |
|
workshop supported by the ETHICBOTS European Project AbstractThe Ethical Status of Artificial Agents – With and Without Consciousness Steve Torrance (Institute for Social and Health Research, Middlesex University, UK; Centre for Research in Cognitive Science (COGS), University of Sussex, UK) 1. ARTIFICIAL CONSCIOUSNESS AND ETHICS. Recently Artificial Consciousness (AC) has emerged alongside of AI. It is a matter of some debate as to whether the former is just an extension of the latter, or whether it is a distinct field raising fundamentally new issues. (See [5] and [10] for differing views on the relation between the two fields.) AI is often characterized as the activity of creating systems (machines) which perform in ways that require intelligence when humans perform in those ways. This definition nicely dodges the question of whether such systems actually are intelligent, in some psychologically real sense. Correspondingly AC can be characterized as the activity of creating systems which perform in ways which require consciousness when humans perform in those ways. Could any such AC systems be conscious? Mirroring debates on the actual intelligence of AI systems, which stretch over decades, there has been growing discussion of the question of whether AC systems could in principle display consciousness in a real psychological sense or whether they can at best only simulate or model consciousness. Put another way, this is the issue of whether AC could achieve ‘phenomenal' consciousness or only ‘functional' consciousness. (See the papers in [7] for a discussion of a range of related issues; and particularly [6] for making explicit the distinction between ‘functional' and ‘phenomenal' consciousness.) This question of the psychological status of possible AC systems seems to have a certain bite or urgency about it as compared to the corresponding question about AI. The difference between ‘real' and ‘merely simulated' consciousness appears to mark more clearly an ethically significant boundary than the corresponding one between ‘real' and ‘merely simulated' intelligence. Conscious experience seems to be very much at the heart of ethics: much of what concerns us ethically about our own lives and the lives of others, has to do with the existence, and the quality, of the conscious experiences that fill those lives. Creating artificial agents which are believed to have genuinely conscious lives would thus seem to be of great moral significance. On the other hand agents which display intelligence or cognition – for instance learn, process visual input, play chess, etc. – do not necessarily have to be conscious; although no doubt certain kinds of cognitive activity do require consciousness. The goal of developing consciousness in artificial agents may thus put rather weighty ethical responsibilities on the shoulders of those researchers who take it up – at least if taken to the stage of directly replicating phenomenal consciousness. It could perhaps involve creating beings which will, themselves, have inherent ethical status – rights to be treated in certain ways, and maybe also responsibilities towards other agents, both human and artificial. 2. PROLIFERATION. To see that this is not just a trivial problem – a matter of a few conscious machines in labs around the world that might suffer a little in the cause of science – one only needs to remind oneself that successful technologies tend to proliferate over the planet at considerable speed. If AC researchers were to develop robots that were ‘really' conscious, for use as companions, empathetic advisors, cops, commandos or whatever, then a great many countries, corporations or even households would want to own them. The question of the psychological reality of the consciousness attributed to such robots will take on a peculiar significance. Normally we take our technological innovations to be our tools, our instruments. If, however, we adopt a purely instrumental stance towards such ACs, we (that is, humanity) may be guilty of the reintroduction of a (legally approved!) form of slavery. The slavery, not of some humans by other humans, but of beings which are non-human, and which we created, without their prior consent, to experience a life in which they are manufactured, sold, used and disposed of. A thought-experiment: think for a moment of some favourite hi-tech possession: laptop, cellphone, car, iPod or whatever. Now imagine that this device is actually conscious – its electronic circuits and computations have really produced an inner awareness so that it feels, desires, has satisfactions and sufferings. A crazy fantasy, of course: such devices are just tools, instruments. But how different would your attitude be towards a device if it (or some future successor artefact) were a genuinely conscious machine? Would you still be able to adopt a merely instrumental attitude towards it, or would you be tempted to be concerned about how the world was consciously experienced from its point of view? It would no doubt be very difficult to know when, from an ethical perspective, one had a machine that was more than a mere instrument. AC researchers are not ruling out the possibility of eventually developing machines which do have consciousness – both in a functional, and in a phenomenal sense. Perhaps the difference between the two does mark a relevant ethical divide. However some have argued that, philosophically, there is no intrinsic distinction between ‘functional' and ‘phenomenal' consciousness – that the latter just implies a richer and more inclusive set of functionalities. ([5] provides an extended defence of such a view.) 3. WHAT IS IT TO BE ARTIFICIALLY CONSCIOUS? Bearing in mind the possibility of global proliferation if AC became a successful technological platform, we have the following scenario: One day there could be a large number of machines on the planet – possibly several in every human household – that don't merely behave in highly intelligent and sentient ways, but actually live conscious lives, and which might thus have to be regarded as having ethical status analogous to that of humans . This is the scenario which seems to be envisaged by those who are optimistic about the possibility of real consciousness being developed in electronic artificial beings. But could any technology ever produce agents with the right kind of properties for us to be even remotely in the ball-park of real artificial consciousness? Even if you're highly sceptical, and think that consciousness depends on organic physiology rather than just functionality, some points must be borne in mind. First, the gradual development of electronic prosthetic devices may combine with our own organic systems so as to enhance or change our consciousness. Thus artificial consciousness of a sort might be produced which is electronically-based, partially at least, and hosted by our own organic constitutions rather than present in self-standing electronic beings. Few would doubt the psychological reality of cyborg-enhanced consciousness in the way that might be done in the case of self-standing electronic AC developments [4]. Second, a kind of ‘bottom-up' AC may come about if work in artificial biology progresses to a stage where fully functioning organisms are capable of being manufactured using molecular construction techniques, for example. So various forms of artificial consciousness may be possible, even if electronically-based conscious agents were thought unlikely, at least in the short or medium term. 4. ARTIFICIAL ETHICS AND ETHICAL STATUS. In raising the ethical implications of current and future AC research I am seeking to make clear some important issues that at present have been obscured in current discussions of techno-ethics. Clearly we do have a responsibility to think about the ethical consequences of creating robots, etc., which we consider to possess real conscious states. And this highlights the important links between consciousness and ethics – although many detailed and subtle questions are raised in considering such links. Alongside the field of Artificial Consciousness we must now also admit that of Artificial Ethics (AE) which could, provisionally, be defined (by analogy with the definition of AC above) as the activity of creating systems which perform in ways which confer or imply the possession of ethical status when humans perform in those ways. It has to be said that this definition is less than fully satisfactory. First, the term ‘ethical status' is far from clear. We will clarify the notion (or at least some key aspects of it) below. Second, this definition (like the earlier ones) centres on the idea of performance . But possibly certain ethical properties that humans have do not arise just as a result of actions that those humans perform. For example, having a right to life, or a right not to be treated merely as an instrument of someone else's needs or desires, are properties which are part of the ethical status of a human being, but a person doesn't acquire such rights just because of what they do . This may extend to ethics when applied to artificial agents. However I hope that the definition as given here will be sufficient for the present, perforce limited, discussion. As we have seen, AC and AE are closely bound up together. But what exactly is the relation? In order to arrive at a clearer idea of this – and more generally, of the relation between consciousness and ethics – I wish, paradoxically, to ask if there could be any ethical status in the absence of consciousness ? Could our future technological creations have ethical status even if they were considered as not possessing real conscious states? Before focussing on that question, however, we need to clarify some key aspects of what is involved in the notion of ethical status. 5. ETHICAL STATUS: MORAL CONSUMPTION AND MORAL PRODUCTION. Ethical status seems to involve at least two major components. The first is being the recipient or target of moral concern or beneficence . The second is being the holder or the dispenser of moral responsibilities . It's the difference, one might say, between being a moral ‘consumer' and a moral ‘producer'. If my household robot were to rescue me from the burning wreckage of my car it would perhaps be displaying moral responsibility – it would be an ethical ‘producer'. If, on the other hand, I were to rescue it from the wreckage then I would perhaps be viewing it as a target of moral concern, as an ethical ‘consumer'. I will consider whether artificial agents, when the latter are assumed to be non-conscious , could be either moral consumers or moral producers. (I here mean specifically devoid of phenomenal consciousness.) The case of non-conscious agents being moral consumers may seem straightforward. Few would think that we would owe non-conscious artefacts any moral duties, any more than we would our TVs or washing machines. If certain artificial agents behaved in intelligent or conscious-seeming ways, many might believe such agents to be (phenomenally) conscious and, for just that reason, might adopt a morally considerate stance towards them. But suppose, for argument's sake, that they were not in fact phenomenally aware. Could such moral considerateness be rational? Perhaps not, although we may be inclined to err on the safe side and assume that we should treat them as having awareness if the behavioural signs were sufficiently realistic. We will come back to this below. What about the case of non-conscious agents being moral producers – having ethical responsibilities? We might expect artificial agents which had sufficiently rich cognitive capacities to be able to ‘understand' moral distinctions, to engage in moral deliberations, at least within certain domains, and to act in ‘ethical' ways. Thus we would expect such an agent to be designed in such a way that it conformed to certain patterns of conduct – and discourse – which mimicked the ethically acceptable behaviour expected of fellow humans. But is this sufficient for genuine, full-blooded moral responsibility? After all if a (human) psychopath is made to conform to socially acceptable norms of conduct under hypnosis we would not ascribe full moral responsibility to such a person: at the very least a certain kind of autonomy would be missing. What is it to be the kind of creature that has genuine ethical responsibilities? Perhaps some or all of the following: the ability to consider empathetically the situation of others; to act in a free or self-directed way in accordance with appropriate rules (and not just to be ‘hardwired' so as to keep within those rules, in the way that robots might be hard-wired with Asimov's Laws); to be able to desire or want ethically appropriate situations; to have a morally virtuous character or dispositions; appropriate moral emotions or sensibilities , etc. How far could an artificial agent possess such features? And what of a non-conscious artificial agent? It would seem that some, at least, of these characteristics that make up a fully-fledged morally responsible agent would be unthinkable if that agent was not conscious in some rich, phenomenal sense. 6. A STRONG VIEW. According to one kind of view – possibly a rather extreme position – an agent could not count having fully-fledged moral status unless that agent has consciousness in the full psychological sense. (There is an extended discussion of this view in [11].) On this view, to be able to carry genuine moral responsibility for actions and decisions requires being able to experience the world, to live a life as an aware, self-directed individual, rather than simply to have some of the functionalities of consciousness. Thus in order for one to have a true ethical commitment to helping another, one must, on this view, be capable of having empathetic feelings towards that individual. So moral action, at least of that sort, is based on the possibility of considering the potential experiences of others as if they were part of one's own consciousness. Indeed, it may be argued that part of what makes humans the conscious agents that they are, is that we experience our individual self-awareness as inherently linked to the self-awareness of others around us – in other words that a condition of first-person consciousness is an empathetic identification with the consciousness of others [9]. So on this rather strong view, developing artificial systems with ethical status in the fullest sense necessarily implies building these systems so that they are conscious in a psychologically real sense. But is such a view correct? Can there be artificial moral agents without consciousness? 7. ETHICAL RESPONSIBILITY IN NON-CONSCIOUS ARTIFICIAL AGENTS. It is undeniable that, as AI systems become more sophisticated, ethical questions will attain ever higher profile. Even without anything approaching phenomenal consciousness, artificial agents may well display certain kinds of functionality associated with consciousness in humans, and they could also display progressively richer kinds of autonomy in decision-making. Herein lies a possible paradox: even where artificial agent systems are not judged to be subjects of direct or inherent ethical responsibility, we will still (as mentioned earlier) expect them to behave ‘responsibly' – that is to at least conform ‘outwardly' to ethical standards of conduct. We will communicate with such systems in richer and subtler ways, and we may well look to such systems for advice and support (even ‘moral' support) in the decisions we have to make. In some situations we may indeed choose to defer to their normative decisions over our own – for instance where we feel that the multiplicity of factors that need to be balanced in a given decision would defeat unaided human minds. Just as today we use automated systems to assess credit risk when granting loans, and automated pilot systems to fly planes, so we may use automated moral decision and advice systems in many areas in the future. Already, prototype systems are being developed in such fields as medical ethics. In one such prototype, an inductive learning algorithm enables the system to generalize from training cases covering ethical dilemmas applying in hospital situations to new such cases [1]. If such an approach can scale up to cope with the complexities of common clinical practice it could prove invaluable to hard-pressed hospital staff who may welcome having such systems to provide guidance in difficult cases. Where would the locus of responsibility for the outputs of such decision systems lie? Arguably, we should always make the designers of the systems shoulder the blame if things go wrong, despite any temptation to offload blame onto the automated systems themselves. An airline or plane manufacturing company could hardly expect its executives to escape moral or legal questions if an automated pilot system crashed a passenger plane; so too, surely with moral autopilots – at least if the autopilot systems were not sophisticated enough to be considered genuinely conscious. So the kind of moral responsibility that non-conscious artificially ethical systems might have could be limited, even if such systems were to provide us with moral guidance that we regarded as authoritative [12, 13]. Another way in which automated agents may be given a kind of ethical responsibility is by our seeking to ensure that their actions always conform to certain humanly agreed ethical norms by applying more sophisticated constraints than those offered by Asimov's Laws of Robotics. One such system under development adopts a deontic logic (the logic of our concepts of obligation and permission) to ensure that robots always apply such a logic to constrain their decisions to a set of general ethical principles [2]. Because such constraints are derived through logical inference rather than built in to hardware or firmware they may be seen as providing a degree of ethical autonomy to the machine. Such systems may thus display many of the features of morally circumscribed action and choice, while operating on pretty conventional principles of AI design, which do not require postulating forms of consciousness. 8. HOW SHOULD WE TREAT OUR ARTIFICIAL AGENTS? Humanoid agents may take on richer and richer bodily, communicative and expressive similarities to us. As they do, the temptation to empathize with them may be very strong. This September a conference in Hertfordshire, UK is discussing issues concerning friendly and empathetic interactions between people and humanoid robots – a topic on which much work is being done in Japan by researchers like Shuji Hashimoto of Waseda University, a keynote speaker at the Hertfordshire meeting [8]. Conversely, many members of the population may have a deep hostility to such agents, especially if they come to be seen as having special status or ‘privileges' of various sorts. There could well be enormous social consequences if the large-scale introduction of such humanoid agents into the world was not handled in a sensitive way. Many may wrongly take non-conscious humanlike agents as being fully conscious, or at least as being able to enter into richer social relations than they are strictly capable of – so new forms of racism or xenophobia may come into being. Could it ever be rational for humans to consider themselves to have genuine moral duties towards non-conscious artificial agents? Possibly not the forms of ethical obligation that are, for instance, geared to minimizing suffering or to increasing satisfactions in such agents. But consider the case of property ownership. Non-conscious ‘autonomous' agents may come to have legal rights, including rights to own money or other goods. Such rights are already accorded to corporations and other kinds of legal persons; why shouldn't legal systems be extended to accord such rights to artificial persons? [3] Property ownership would in turn allow forms of legal penalties to be applied directly to such agents. In that way such agents may come to have financial or material interests – even if not the sort of interests that imply being able to have pleasures and pains and other experientially-based forms of well- or ill-being. As such, we may consider ourselves to have moral obligations to respect such interests, just as we would expect similar ‘respect' from them of human material interests. Further, if property ownership became widespread amongst such artificial agents, one might have a situation in which large sections of the economy may get to be owned and controlled by groups of such (non-conscious) artificial agents: they may even come to control key political structures. In this way non-conscious agents, if given certain kinds of legal rights, may attain positions of social and political power which could make profound differences to the historical character of future society. We may think we will own them, but they – some of them at least – may end up owning us. 9. CONCLUSION. There seem to be very deep connections between questions concerning what kinds of artificial agents might possess psychologically real states of consciousness, and which might possess genuine or full moral status. One rather strong position suggests that having the capacity for genuine conscious states (arguably, a remote possibility for electronic beings) is a strict requirement for full moral status. On this view, to be either a moral ‘consumer' or a moral ‘producer' are both ruled out if one is an intelligent but non-conscious agent. However even if this position were correct, there are still important ways in which non-conscious artificial agents could come to have moral responsibilities, and even rights, of sorts. Property-ownership is perhaps one such case. In any case, there are various ways in which we could expect non-conscious artificial agents to assume various kinds of moral responsibilities, and even to give us moral guidance. However, the strong view may be right in saying that moral status in the full sense may require psychologically real, phenomenal, consciousness – so that at the limit, perhaps AE and AC do go together. References: [1] Michael Anderson, Susan Leigh Anderson, Chris Armen, "An Approach to Computing Ethics," IEEE Intelligent Systems , vol. 21, no. 4, July 2006, pp. 56-63. [2] Selmer Bringsjord, Konstantine Arkoudas, Paul Bello, "Toward a General Logicist Methodology for Engineering Ethically Correct Robots," IEEE Intelligent Systems , vol. 21, no. 4, July 2006, pp. 38-44. [3] David Calverley, “Imagining a Non-Biological Machine as a Legal Person”. Artificial Intelligence and Society: Special issue on Ethics and Artificial Agents. To appear, 2007. [4] Andy Clark, Natural-Born Cyborgs: Minds, Technologies and the Future of Human Intelligence. N.Y.; Oxford University Press, 2003 [5] Daniel C. Dennett, Consciousness Explained. N.Y.: Little, Brown, 1991. [6] Stan Franklin, “IDA: A Conscious Artefact?” in Holland, 2003, pp. 47-66. [7] Owen Holland (ed), Machine Consciousness. Special Issue of Journal of Consciousness Studies. Volume 10, No. 4-5, April 2003. [8] RO-MAN 06: The 15th IEEE International Symposium on Robot and Human Interactive Communication , “Getting to know socially intelligent robots.” University of Hertfordshire, September 2006. http://ro-man2006.feis.herts.ac.uk/index.php [9] Evan Thompson, “Empathy and Consciousness”. Journal of Consciousness Studies. Vol. 8, No.5-7, May 2001, pp.. 1-32, [10] Steve Torrance, “Producing Mind”. Journal of Experimental and Theoretical Artificial Intelligence. Vol. 12, No. 3, July 2000, pp. 353-376. [11] Steve Torrance, “Ethics and Consciousness in Artificial Agents” Artificial Intelligence and Society: Special issue on Ethics and Artificial Agents. To appear, 2007. [12] Blay Whitby, Reflections on Artificial Intelligence, The legal, moral, and ethical dimensions , Exeter: Intellect Books, 1996. [13] Blay Whitby, “Computing Machinery and Morality” Artificial Intelligence and Society: Special issue on Ethics and Artificial Agents. To appear, 2007. |