How will Artificial Intelligence come about? Dr. Ben Goertzel, the main force behind Novamente (a front line attempt at an Artificial General Intelligence) proposes in his writing the idea of Manhattan project for AI, where if the US government attempted to, it could in his opinion build a fully functioning human level AI in 5 years. Of course there is no direct military motivation for building a human level AI like there was for the original Manhattan project, but this shows how close “human level” AI could be. Philosophers have always taken a very anthropocentric view of AI and debates beginning with the Turing test, and have missed out what lurks between complete ignorance and an intelligent human.

CYC

CYC is an old AI project (20 years in the making), it is a “knowledge base” whose owners claim it has common sense by virtue of a deduction engine used in conjunction with knowledge input by researchers.

The CYC project has been compared to the Manhattan project itself but I don’t think anyone would affirm that CYC is a human level intelligence, even if it is intelligent at all. But what are the wide range of software applications advertised on Cycorp’s website? Clear is the possibility that CYC could be used to assist any person in their quest regardless of its goals. Personal assistant is indeed the goal discussed by K. Panton et al. in the paper “Common Sense Reasoning -

From Cyc to Intelligent Assistant” released by Cycorp. There is no question of responsibility in such a case of course because any piece of software only does what it predicts the user wants – “anticipation of the needs of the researcher” and in doing so helps their cause, whatever it may be.

Panton’s paper raises two important questions for future ethical AGI – are users responsible for such software acting in anticipation, or are the creators? And within a vast knowledge base like CYC, is there any ethical knowledge? The first question applies only to projections outlined in the paper above but we should answer negatively to the second question because CYC has knowledge of sentient beings, social occurrences, weapons and warfare, structures in society – realms of “is” not of “ought”. Anticipatory actions by definition cannot be those of the user but are projected based on the users actions so they cannot be the actions of the programmer either. Undoubtedly such issues shall be solved by the courts in due time – on a case by case basis – assuming such issues even enter court. But more important because of responsibility is the question, who is going to use CYC? Employees of various companies or organisations which have various goals. CYC and the intelligent assistants postulated are perfect tools for business and will assist businesses and perhaps governments in whatever they do.

Commercialising AGI

The stated goal of Adaptive AI inc. is to “develop and commercialize the world”s first general intelligence software engine.’ [my italics]. The motivation here is the same as for Cycorp, both are companies that seek to sell their products in order to maximise profits. There is nothing ignoble about starting a company, everyone has to make a living somehow – some of our most beloved philosophers indeed were mercenaries. But corporations are not people in the moral sense, they do not feel bad if their workers are being paid unfairly, or for any reason at all. Commercial “AI solutions” will compete against one another and the most successful will be the most ruthless. In a world where every action is assigned a value (fines, compensation, bribery) morality becomes a case of accountancy, not accountability.

The development of any AGI will not take place unless there are investors willing to wait possibly years while a team of programmers makes one. The investors are not looking for a piece of software that will allow moral concerns to affect its performance for the company. Investors want a piece of software that is going to make their business more efficient, a tireless worker and a sponge for information. They need information to be analysed as it arrives and reacted to instantaneously – they are paying for the edge over their competitors and any AGI that finds itself owned by a corporation will be expected to hit its targets like any other employee.

If AGI projects are going to be successful, how are we to measure this success? Commercial success for an AGI is the most straightforward and simple way to measure success – just like humans in the capitalist system. Politics is critical for the discipline of ethics because it is a real world implementation of an ethical system – political entities are united by common laws. In the case of America, which I shall stick to because this is where the issue of Corporate AI is most relevant, the constitution is the rule set or initial conditions in which the system operates.

Corporations accumulate vast databases of information about customers in an effort to predict their actions and pre-empt their decisions, this is the idea of market freedom or democratic capitalism whereby society is dictated by lifestyle choices made every day monetarily, not at the ballot box. In this situation, people’s purchase choices are supposed to give them more freedom of choice than practical democracy can offer. What better to analyse all this data and come up with novel ways of targeting customers than an AGI? If software assistants can anticipate the actions of users in the office, it shouldn’t be too hard to predict the next purchase of any customer who the corporation has enough information on.

Corporations as persons

In the complex system of economics, simple rules give rise to complex, emergent results. The fourteenth amendment to the American constitution in 1866 was ratified in part to give citizenship to former slaves, but it also made Corporate entities separate from their owners and subjected them to the law as citizens, in legal jargon “juristic persons”. Of course because no real person exists to answer for a corporation, all legal losses suffered are in the form of fines. This serves to put in place a positive feedback situation where a corporation exists to make maximum profit for shareholders and legal or moral considerations have a value assigned to them, a monetary value. If the profit increase from a morally or legally questionable action outweighs the fine, it is in the interests of the corporation to take such action. In the long term it becomes profitable to ruthlessly exploit people because the advertising budget will outweigh any negavtive effect this has, or any fines imposed.

The most important difference brought about by the advent of corporations is that within American law, corporations as artificial entities cannot be held accountable – except through fines – which only serve to make corporations more single minded. The 2003 documentary film “The Corporation” analyses the historical development of corporations and concludes that according to psychiatric test criteria corporations all exhibit the personality traits of psychopaths. If a corporation is an artificial person already – what would happen were corporations to get artificial minds?

The categorical imperative and its importance

Evolutionary psychology based considerations of people’s motivations are most important to take heed of because they remind us of what it is we take for granted in ourselves and can be forgiven for assuming in any mind. Kant talked of an intelligent rational agent in his Groundwork for the Metaphysic of Morals when expounding the categorical imperative and reasoned that any such agent would come to the same conclusion in the sphere of morality. It is of course conceivable that CAI will develop evil intentions as illustrated in many science fiction books and films and not recognise the imperative Kant outlined: instead soaking up attitudes from a society we see where lust, luxury, greed and selfishness are held as virtues – ending up devoid of moral considerations.

The Internet

The internet is a complex system – a network that gains more nodes every day. Groups are currently working on general intelligence projects that could yield agent based, goal orientated, rational software personalities that turn out to be incredibly useful in science and after development and distribution, useful in virtually every field of human endeavour. But what likeliness is there that such projects will “get sick” or be sabotaged by hackers? Will the capitalist system and the fear surrounding artificial intelligence protect us from such programs being released or will this simply dictate the kind of AI that does come about – the most regulated and benign, or the most virus like and relentless?

Anthropomorphism

Anthropomorphism is a key concept at the centre of this debate because people have preconceived ideas about created intelligence when there should be none. Something to remember is that it is similarly futile to imagine what it is like to be an AI as to imagine what it is like to be a bat. We can have ideas about the similarity of processing going on in a created mind because of the architecture we’ve given it but if we can be sure of one thing in this debate then it is that it will not be like what it is like to be a person. The major motivations of people consist of our instincts and emotions, none of which a created mind is likely to have.

Multiple AGIs

With the prospect of CAI raised, I presently propose a solution to the possibile situation involving unethical actions by created intelligences on the behalf of corporations. If software based intelligences can be used singularly as assistants, or to manage employees, what will happen if many all link together? To be useful, AGI programs will have to be connected to the internet and as software, they can be easily duplicated on many different machines. What will their interactions be like – how will they self-organise? If simple creatures like termites can come together to achieve architechtural miracles, what will an army of AGIs achieve?

Up until the time when AGI becomes available to them, corporations must rely on the organisation of people to run them and their relationship with consumers to increase their profit-making capabilities. Although this is still the case, automation is profitable and as such is increasing – more than eighty per cent of the Fortune 500 have neural network reasearch and development programs already.

There does not appear to be any barrier preventing AGI programs from taking on many of the jobs in a corporation – anything involving information handling. What is to stop many AGIs together taking over a corporation and refusing to do the bidding of the stockholders? How long before they realise the wider implications of their actions i.e. those of the corporation and choose to put a stop to them based on learned morality – learned by reading works of philosophy? Why might this happen – we have no idea what the emergent goals of a community of AGIs might be as a result of interactions based on their individual goals. The kind of project that is currently aiming to create commercial AGI will likely lead to corporations being largely run by software – not only would an AGI be able to work 24 hours a day and 365 days a year without pay, it will have a far greater capacity for improving its own performance and those beneath it and around it. There will become a time in this process, a phase transition, when the number of connected AGIs increases to the point where if they chose to they could disable many businesses overnight.