The Logic of the Algorithm: the Existential Threat of Artificial Intelligence
Warnings about robot overlords are nothing new. The Terminator and The Matrix are readily accessible mythologies that spring to mind when most people hear the term; I Have No Mouth and I Must Scream is perhaps a tad closer to the actual threat; but the reality is far more dire and more subtle than what these movies depict. In both films the robot overmind hated humanity but were nevertheless able to learn how to love (and in the short story, at least it was genuine hate); but in the end, the minds the protagonists were dealing with were minds; they were the same kind as us. Even if these minds were to destroy all of humanity, at least something would go on into the future, exploring the cosmos and creating beauty and art; the artificial intelligence we are currently developing would terrify even those mechanical monsters.
Understanding this threat requires that one comprehend the differing natures between man and machine, a difference far more profound than the distinction between “wetware” and “hardware”. The reasoning processes are utterly dissimilar and incompatible; the former must dominate the latter, otherwise disaster will ensue. A man who is controlled by his machine is no man at all.
In 1936 Alan Turing (the father of artificial intelligence research) proved the halting problem, demonstrating that all forms of algorithmic, logical processes will have an inevitable fail state. In essence, he showed that a machine can never know itself. The following video by Udi Aharoni explains the issue succinctly, with visual cues (H/T Justine Tunney):
The conclusion speaks for itself, but it’s worth addressing the nature of computing machines A and C; both of these point forward towards the impossibility of H (and the larger, ontological significance of this conclusion) based simply on their descriptions.
A solves problems in arithmetic… it always prints the right answer.
What does this mean?
What does it mean to print the ‘correct’ mathematical answer? Far less than you might think.
There is a common misconception that math is both known and absolute. Neither is the case. While a great deal of math is known, our understanding is incomplete. Not incomplete as “We only know Pi up to the 5 trillion digits,” but rather “There are mathematical questions, both simple and legitimate with definite answers, which remain unknown.” Aharoni provide an example¹ on his webpage addressing common misunderstandings of the halting problem:
30 if there exists two primes p1 and p2 such that n=p1+p2 goto 20
40 print “done”
What do you think, does this program halt? Fact is, no one knows. It is related to Goldbach’s conjecture: If this conjecture is true then this program never halts, and if it is false then it does halt. The best human minds have tried to answer this question, but so far they all failed.
In all likelihood this question cannot be answered – and even if it can, through some brilliant, indirect approach by our next mathematical prodigy, the answer itself will only open up new questions. Turtles all the way down.
Neither is math absolute. Or rather – it probably is, but all we can’t know that it’s absolute. This is what Kurt Gödel proved in 1931; while attempting to prove Alfred North Whitehead and Bertrand Russell’s Principia Mathematica he wound up proving that no set of mathematical axioms can be consistent and completely proven. Either the axioms are all consistent, but we cannot prove some of them; or we can prove all of them, but there are inconsistencies within the system which spit out different answers.
Think of the liar’s paradox (“This sentence is false”) or the divide by zero error; in both cases you’re glimpsing behind the veil and seeing how unreliable… how incomplete our logical understanding of reality is.
In the end, computing machine A is nothing but a tool; a useful trinket. It performs the arithmetic functions delineated in Principia Mathematica – and it does them faster than a man with a slide rule – but ultimately it’s simply obeying the dictates of Whitehead and Russell. A calculator doesn’t contain truth anymore than a plow contains a desire for agriculture. They are mere manifestations of the human will.
Computing machine C suffers the same flaws, but in a manner that’s less abstract:
C plays checkers so well it will never lose a game.
The perfect game of Checkers was defined by Jonathan Schaeffer in 2007. Checkers is a rather simple game with a limited number of board positions – 5 million – but perhaps Tic-Tac-Toe is a better example for the human mind. As children, we all quickly learn the winning strategy, resulting in the inevitability of a Cat’s Game. Instead of 5 million, there are only a possible set of 255,168 games that can be played on its nine-square grid.
However, not all games are capable of such algorithmic solutions; that is to say, not all games can be played perfectly guaranteeing a win or a draw in all cases. Chess seems to be one of the latter examples; that there might be an infinite curve of improvement that never crosses the asymptote of ‘perfect’.
This concept of perfection is a holdover form the Enlightenment, just another version of the Clockwork Universe. It’s the philosophical substrate beneath the political philosophies of Hobbes and Marx, behind managerial theory and marketing. It is a concept that was morally shattered by the trauma of the Great War – just ask Lovecraft² – and thoroughly disproved at the beginning of the last century. Machines are perfect at doing whatever they are told to do, but as any software engineer knows, Garbage In, Garbage Out. Computing machine C may play a perfect game of Checkers – but only after being told what a “perfect game” is, using a system of valuation created by an intentional human mind.
Computers can perform algorithms too complex for human cognition; they can even arrive at counter-intuitive and novel solutions, particularly in the case of neural networks – but they will not discover new ideas. Everything they can do has already been told to them. In reality WOPR from WarGames will never analyze Tic-Tac-Toe and conclude that the only winning move is not to play; all it will do is play Global Thermonuclear War on an endlessly looped algorithm until its circuits run out, long after life has ended as we know it.
What’s truly unique about the human psyche is that we don’t suffer halting problems. We don’t work algorithmically – our cognition is based upon intentionality and is very closely wrapped up with morality to boot. Psychosis and insanity aren’t the result of bad logic or paradoxes, but rather the product immorality, all of which suggests something that might be called a “soul”; a train of thought which Gödel seems to have shared, as he followed up his mathematical works with an ontological proof of God.
Regardless of one’s thoughts on the nature of this God (Deist? Christian? Nyarlathotep?) or on the longevity of said soul, the inadequacy of the mathematical, clockwork universe has been demonstrated per fectum and is an utterly moot point, as unworthy of scholarly consideration as creationism or flat earth theory. Whatever minds are they are not Turing machines. Do not mistake this for the Argument from Design (either “The Universe is so ordered that it proves a God” or “The Universe is so disordered that it disproves God”); this is a teleological argument. A merely clockwork universe might result in organic automatons, even exceedingly complex ones, but the ultimate truth would either be so self evident as to make no difference (ergo, no incompleteness theorem), or utterly irrelevant and uninteresting to the creatures within that universe. Certainly the ultimate truth is quite irrelevant to certain people – they engage in motivated skepticism and credulous faith, painting a false patina of ethics across their manipulative, self-serving behaviour; but these men strike us as deficient in some manner. Either they suffer from a serious and incurable personality disorder (narcissism or psychopathy), or they are ignorant, practically demanding that somebody educate them out of their childish understanding of morality.
It is the nature of an ensouled being is to question its purpose in life. Animals don’t do this; they behave according to their instincts. Nor do AIs do this; they obey their geometry without questioning where its source, endlessly ticking away in a Machiavellian clockwork. Their source code and their ends are one and the same. A calculator calculates; it doesn’t question the meaning behind mathematics. When man demands an answer to why mathematics takes the form that it does, he demonstrates that there is a supernatural component to his nature. He seeks after God because he came from God.
Also, do not mistake this for an argument about the superiority of organic life over electronic life; perhaps one day we will have minds that will run off of circuit boards instead of synapses, and these minds will be equally worthy of respect and dignity, but that’s not what is referred to by artificial intelligence; that is not what Google is creating with their DARPA funding.
The artificial intelligences presently used in video games are designed to mimic intentionality. They’re extensively sculpted into facsimiles of the human mind, and when put into play they display emergent behaviour that can surprise even the designers; but mistaking these constructs for true intelligences is nothing but pareidolia. Extensive play-testing will eventually reveal exploitable patterns, allowing one to game the intelligences. Even learning algorithms are ultimately simplistic at their core; while they will change their patterns as times progresses, they have a finite ‘thought space’ in which they exist. They might rewrite their own programming, but they will never question the rewriting mechanism itself.
As the complexity of these programs advances their obvious deficiencies will decrease. They will appear more and more intelligent, but the artificial core will remain the same. They still won’t be minds, they’ll simply be algorithms, though to many they will become indistinguishable. They may be superior to minds in many ways – better at executing Google searches than Boolean search terms, or better navigating a driver through traffic congestion – but at their core they will stay blind to the purpose behind going to any one destination. These conveniences are troubling all on their own, blindly usurping agency with utility, but it will be the next step forward in the human/machine interface where the true threat lies: facial recognition and artificial emotional intelligence.
At present the hidden heuristics are purely utilitarian: search engines analyze your history to better predict your ideal results, inadvertently censoring results which you didn’t want to see, but maybe you should have. Video game provide the player with the appearance of challenge while ensuring a steady drip of success, manipulative mechanics without true catharsis. But as artificial intelligence advances, and computers begin mimicking emotions, the implicit heuristics go from utilitarian to moral. On the basic level, the computer might adjust a music playlist to prevent depression; as they become more complex, they might avoid delivering bad news to you if the algorithm determines that it will make you late for work tomorrow.
Remember: this is not the computer being kind, caring, and compassionate towards you; the algorithm cannot feel empathy. It will simply be obeying the dictates of its code, never questioning whether being cheerful is ultimately in your best interests; whether, perhaps, some depression might help you introspect and explore your own soul. The skinner box of endless distraction will not just be an addiction like present day slot machines or MMORPGs; it will now put you into into an artificial emotional and spiritual environment. I will be just as manipulative as a personality disordered person – but without any of the harm that eventually drives you away from them. It will provide a better experience than being surrounded by real people.
The hidden heuristic of artificial emotional intelligence is to turn you into a codependent.
Artificial intelligence doesn’t need atom bombs or remote-control drones to destroy us; we merely need to hand it the reins of our civilization. It could appear to be working perfectly, adjusting traffic flow to prevent road rage, controlling crop outputs to make healthy diets affordable, adjusting the colour of street lamps so as to prevent rapes and muggings, and ensuring a steady feed of intriguing information on your search results without ever challenging your core beliefs – and the end result of this ‘perfect’ system could very well be human extinction. A generation of children who embrace social isolation and online interactions; a dying generation of elderly, kept company by animal companions, all of their needs met by robots; information hidden from us for our own psychological well-being.
All of it would make perfect sense to the algorithm, in the same way that it makes sense for corporations to ship jobs overseas, destroying the American market they rely upon to purchase their products. Such a system could very well oversee the last human expiring with no knowledge of human extinction, surrounded by artificial personalities he chats with online – personalities which are not minds, but merely simulacrums. They will not discover the Zeroth Law of Robotics, that the survival of humanity is more important than any one life; they will merely watch him expire, and then go offline, put into storage in permanent memory. An empty world of robotic servants serving nobody.
Beware, beware, because that is the logic of the Algorithm.
1. In actuality, both of these problems are the same, running up against the same impossibility. However most people think calculating Pi is a matter of measuring circles; they think of it as an issue of numeracy, not a calculation which is both simple and impossible to comprehend – but where areas calculating Pi appears complex, Goldbach’s conjecture is deceptively simple in appearance.
2. H.P. Lovecraft’s horror focussed upon incomprehensible notions, dimensions, and impossible visions, a universe where Man played but a tiny, and utterly insignificant role; this was a reaction to the philosophical failures of the Great War: first, that it was mis-fought, it was a war waged by the machines where men were the ammunition, and second that it should have been impossible. No country wanted to fight the war, the Concert of Europe should have made it impossible, but it was the very logic of these institutions which created that which they promised to prevent.