By 
When asked for a title for his 1979 collection of philosophical papers, my all-time favourite philosopher[1] Thomas Nagel chose the title Mortal Questions, an apt title, for most of our philosophical preoccupations (and especially those pertaining to the broad realm of moral philosophy) stem from the simple fact that we’re all mortal, and human life is as such an irreplaceable good. By extension, most things that can be created by humans are capable of being destroyed by humans.
That time is ending, and we need a new ethics for that.
Consider the internet. We all know it’s vulnerable, but is it existentially vulnerable?[2] The answer is probably no. Neither would any significantly distributed self-provisioning pseudo-AI be. And by pseudo-AI, I don’t even mean a particularly clever or futuristic or independently reasoning system, but rather a system that can provision resources for itself in response to threat factors just as certain balancers and computational systems we write and use on a day to day basis can commission themselves new cloud resources to carry out their mandate. Based on their mandate, such systems are potentially existentially immortal/existentially indestructible.[3]
The human factor in this is that such a system will be constrained by mandates we give them. Ergo,[4] those mandates are as fit a subject for human moral reasoning as any other human action.
Which means we’re going to need that new ethics pretty darn’ fast, for there isn’t a lot of time left. Distributed systems, smart contracts, trustless M2M protocols, the plethora of algorithms that have arisen that each bring us a bit closer to a machine capable of drawing subtle conclusions from source data (hidden Markov models, 21st century incarnations of fuzzy logic, certain sorts of programmatic higher order logic and a few other factors are all moving towards an expansion of what we as humans can create and the freedom we can give our applications. Who, even ten years ago, would have thought that one day I will be able to give a computing cluster my credit card and if it ran out of juice, it could commission additional resources until it bled me dry and I had to field angry questions from my wife? And that was a simple dumb computing cluster. Can you teach a computing cluster to defend itself? Why the heck not, right?
Geeks who grew up on Asimov’s laws of robotics, myself included, think of this sort of problem as largely being one of giving the ‘right’ mandates to the system, overriding mandates to keep itself safe, not to harm humans,[5] or the like. But any sufficiently well-written system will eventually grow to the level of the annoying six-year-old, who lives for the sole purpose of trying to twist and redefine his parents’ words to mean the opposite of what they intended.[6] In the human world, a mandate takes place in a context. A writ is executed within a legal system. An order by a superior officer is executed according to the applicable rules of military justice, including circumstances when the order ought not be carried out. Passing these complex human contexts, which most of us ignore as we do all the things we grew up with and take for granted, into a more complicated model may not be feasible. Rules cannot be formulated exhaustively,[7] as such a formulation by definition would have to encompass all past, present and future – all that potentially can happen. Thus, the issue moves on soon from merely providing mandates to what in the human world is known as ‘statutory construction’ or interpretation of legislative works. How are computers equipped to reason about symbolic propositions according to rules that we humans can predict? In other words, how can we teach rules of reasoning about rules in a way that is not inherently recursing this question (i.e. is not based on a simple conditional rule based framework).
Which means that the best that can be provided in such a situation is a framework based on values, and target optimisation algorithms (i.e. what’s the best way to reach the overriding objective with least damage to other objectives and so on). Which in turn will need a good bit of rethinking ethical norms.
But the bottom line is quite simple: we’re about to start creating immortals. Right now, you can put data on distributed file infrastructures like IPFS that’s effectively impossible to destroy using a reasonable amount of resources. Equally, distributed applications via survivable infrastructures such as the blockchain, as well as smart contract platforms, are relatively immortal. The creation of these is within the power of just about everyone with a modicum of computing skills. The rise of powerful distributed execution engines for smart contracts, like Maverick Labs’ Aletheia Platform,[8] will give a burst of impetus to systems’ ability to self-provision, enter into contracts, procure services and thus even effect their own protection (or destruction). They are incarnate, and they are immortal. For what it’s worth, man is steps away from creating its own brand of deities.[9]
What are the ethics of creating a god? What is right and wrong in this odd, novel context? What is good and evil to a device?
The time to figure out these questions is running out with merciless rapidity.
Title image: God the Architect of the Universe, Codex Vindobonensis 2554, f1.v
References
| ↑1 | That does not mean I agree with even half of what he’s saying. But I do undoubtedly acknowledge his talent, agility of mind, style of writing, his knowledge and his ability to write good and engaging papers that have not yet fallen victim to the neo-sophistry dominating universities. |
|---|---|
| ↑2 | I define existential vulnerability as being capable of being destroyed by an adversary that does not require the adversary to accept an immense loss or undertake a nonsensically arduous task. For example, it is possible to kill the internet by nuking the whole planet, but that would be rather disproportionate. Equally, destruction of major lines of transmission may at best isolate bits of the internet (think of it in graph theory terms as turning the internet from a connected graph into a spanning acyclic tree), but it takes rather more to kill off everything. On the other hand, your home network is existentially vulnerable. I kill router, game over, good night and good luck. |
| ↑3 | As in, lack existential vulnerability. |
| ↑4 | According to my professors at Oxford, my impatience towards others who don’t see the connections I do has led me to try to make up for it by the rather annoying verbal tic of overusing ‘thus’ at the start of every other sentence. I wrote a TeX macro that automatically replaced it with neatly italicised ‘Ergo‘. Sometimes, I wonder why they never decided to drown me in the Cherwell. |
| ↑5 | …or at least not to harm a given list of humans or a given type of humans. |
| ↑6 | Many of these, myself included, are at risk of becoming lawyers. Parents, talk to your kids. If you don’t talk to them about the evils of law school, who will? |
| ↑7 | H.L.A. Hart makes some good points regarding this |
| ↑8 | Mandatory disclosure: I’m one of the creators of Aletheia, and a shareholder and CTO of its parent corporation. |
| ↑9 | For the avoidance of doubt: as a Christian, a scientist and a developer of some pretty darn complex things, I do not believe that these constructs, even if omnipotent, omniscient and omnipresent as they someday will be by leveraging IoT and surveillance networks, are anything like my capital-G God. For lack of space, there’s no way to go into an exhaustive level of detail here, but my God is not defined by its omniscience and omnipotence, it’s defined by his grace, mercy and love for us. I’d like to see an AI become incarnate and then suffer and die for the salvation of all of humanity and the forgiveness of sins. The true power of God, which no machine will ever come close to, was never as strongly demonstrated as when the child Jesus lay in the manger, among animals, ready to give Himself up to save a fallen, broken humanity. And I don’t see any machine ever coming close to that. |
