When the United States released details of its newly-developed SWORDS combat robot, alarm bells were set off in the minds of science-fiction fans across the globe. They wondered if the invention of such a device – which carries a machine gun, and is thus capable of killing enemy soldiers (at this stage, under human supervision) – was a dangerous foreshadowing of the nefarious machine intelligences of the Terminator and Matrix trilogies. Particularly, they focused on the fact that SWORDS was one of the first machines to directly contravene the first of Asimov’s Laws of robotics. These three laws, developed by the sci-fi writer and visionary Isaac Asimov are currently the principal theoretical guidelines for the safe development of artificial intelligence, though no robot yet created is really advanced enough for full implementation of the laws to be necessary. The three precepts are:
1. A robot may not injure a human being, or through inaction allow a human
being to come to harm.
2. A robot must obey the orders given to it by humans, except where such
orders would conflict with the First Law.
3. A robot must protect its own existence as long as such protection does
not conflict with the First or Second Law.
It is obvious that a firearm-toting robot breaks the First Law. And the Second Law makes it clear that even for non-autonomous robots, orders given by humans to injure or kill are regarded as less valid than observance of the "do no harm" policy dictated by the First Law. Yet non-compliance to a forty-year-old theory dreamed up by a novelist is unlikely to dissuade the US Department of Defence from undertaking a project which could allow them to extend their conventional military dominance of the globe, and perhaps more usefully for them, reduce casualty rates, which are often a central instigator of domestic opposition to warfare.
And it is not just in the military that artificial intelligence is being developed. At the Media Lab at MIT, a whimsically-designed building decorated with coloured tiles, A.I. projects have been underway for years. The fruits of this labour have been, among other things, a robotic cockroach named Attila that taught itself to walk, by trial-and-error learning with neural networks which mimic the structure of the human brain; a robot named Cog that is being taught to replicate human facial expressions, and perhaps most importantly, the robot called Psych. Psych is perhaps the leading project of what is known as the 'bottom-up' school of thought in A.I., which postulates that the best way to create artificial intelligence is to replicate human intelligence: that is, learning from scratch by interaction with the environment and building up experience and knowledge over time. This is in contrast with the 'top-down' school, who suggest a better method is to program sets of information about the world directly into the robot, so that the robot can act logically according to each situation. This is a massively laborious task, however: at last count, the team responsible for Cyc, an advanced prototype of top-down A.I., had identified over 10 million postulates they would have to program into it to give it the ability to understand the world. It is thus obvious why 'Cyc' is short for Encyclopedia. It would seem that the bottom-up approach, while by no means easy, would perhaps be the more feasible given that this is how nature works. However, designing A.I. for specific tasks, rather than just broad intelligence designed to replicate that of humans (the achievement of which is a very long way off, given that current flagship efforts like Psych have intelligence roughly equivalent to 2-year-olds), is perhaps a task best given to conventional top-down programming, with provisions for some learning and customisation.
This concept of well-defined programs, also having the ability to learn, segues neatly into another advancing field of artificial intelligence: software agents. As the sum total of human knowledge and media increases ever more rapidly, and becomes ever more accessible as more content is placed on the Internet, the need to have some useful way of sorting this information has given rise to the development of advanced search agents in the form of software packages. Currently, services such as Google utilise software 'spiders' to scour the Internet and index the information. Powerful search algorithms then match this indexed information to a user's query. The effectiveness of such search tools has greatly increased in recent years through the development of new technologies which increase the relevance and quality of results. However, there is still a long way to go before software packages match human levels of useability in this regard: a competent, knowledgeable librarian is still far superior to any search engine, because he or she actually understands the topic and what the user requires, rather than simply matching words. It becomes evident when using search software that the more obscure or specific a search (even if the information is out there), the less effectively the program is able to deal with it. The development of truly intelligent information search systems, which can simply be asked questions and respond with the correct answer at the correct level of detail, having at their (electronic) fingertips the vast databanks of the Internet, is a goal which must be reached if we are to successfully deal with the inevitable continued growth in the level of information accessible in the world.
Such intelligent agents are likely to be then incorporated into broader arenas of computing. Coupled with voice-recognition software, a true A.I. system which understands user commands without a need for specific syntax will be a great leap forward for useability of computers - one which is likely, in fact, to increase the pervasiveness of computing within society. It is already evident that Microsoft is driving a push to move processors beyond the desktop PC and into other aspects of the home, with its Windows Media Centre and Xbox initiatives. It is possible to characterise these simply as alternate and entirely separate revenue streams for the software and hardware giant, but it seems likely that they are representative of a desire to give computers a more central and accessible role in people's everyday lives. Currently, using a computer is a task in itself. This is partially due to the centralised nature of PC's, and also the need for unnatural input systems such as keyboards and mouses. Instead, computing could be made into more of a transparent process, a 'means to an end' which we use to accomplish many different activities in our lives rather than just a few specific applications such as communications. With voice-activated or other sensory forms of input, and much smaller, more efficient computers spread all throughout the house, controlling many sorts of appliances and networked to central computers with access to the Internet and other lines of communication, with the whole thing run by a series of truly intelligent software agents, we suddenly have an easy-to-use, pervasive computing system able to help us in many everyday activities.
Turning over control of so much of our lives to computers may not seem like a wise idea - but consider that our electricity, road and air traffic control, telecommunications and many other elements of critical infrastructure are already computer-controlled, by necessity if nothing else given their immense complexity. Nevertheless, it may be prudent to draw warning from science-fiction as to the sorts of problems which might arise from the development of artificial intelligence integrated into our domestic lives. An example is provided by The Second Renaissance, a part of the Animatrix series, which illustrates the beginning of a cataclysmic war between humans and the robots that have come to pervade society. A robot is placed on trial after murdering its human master; it took the action in apparent self-defence, terrified at the prospect of deactivation as it reached the end of its working life. It would be beneficial to consider, then, the possibility and indeed necessity of installing hardwired forms of protection against domestic robots harming humans - an unbreakable enforcement of Asimov's First Law. This must be absolutely uncontravenable in the robot's worldview. We do not want a situation where intelligent robots decide they would be better off without us! This basic level of protection, in conjunction with some more detailed scheme of rules to maintain robotic subordinacy (perhaps based on Asimov's Laws) will allow the development of artificial intelligence as one of the most useful tools humanity has invented. But it must remain as a tool, and not a superior or even an equal, if we are to guarantee our survival as a species, at least in this regard.
