The Singularity as a 24/7 2Fort Instant Respawn Server
Published:
Continual advancements in the domain of automating everything (AI) have moved the question of playing god from idle speculation to a serious policy debate. Many people find this prospect terrifying. So terrifying that we should abandon this whole enterprise even if we can be reasonably certain a good outcome will be achieved (for some definition of good).
Bourgeois Blues
After Work, We'll Have Each Other compares the social situation when all work is automated to 18th-century aristocrats. When labor is automated, it'll be "a world where our connections to other humans are what give us structure, identity, and meaning". Since aristocrats "did not work for pay" and "did not see work as a source of meaning", they are "as close to a purely relational world as anything [Collier] can think of".
In Collier's account the 18th-century aristocrats were profoundly boring. Their memoirs are interesting to read only "in large part because of the contrast between the keenness of [their] observations and the inanity of their content". Their "conversation consisted overwhelmingly of political gossip" and they are "ruthlessly pragmatic in the pursuit of power". Since their relationships are all that matters to their social standing, any action they take is somehow in service of preserving or expanding their relationships.
The Solve Everything Button
Suppose then we give "two cheers for the American work ethic" and accept work as a primary source of meaning. What happens when a new way to do the same work faster is developed? Market pressures tend to render the old way unprofitable and obsolete rather quickly. Many people still create things in ways that are not economically efficient, but these are, by definition, hobbies, not work. Suppose then we could cheaply make a button that instantly did a specific job. The existence of this button, then, would convert that job into a hobby.
An AI that could do every job would similarly convert every job into a hobby. Suddenly all the meaning through work would be gone. This is why people who find meaning through work are so scared about AI. Yet consider the decision to not develop AI: is this not already deciding to not take the most efficient path to accomplishing an outcome? If you're preventing a new technology on the grounds it's more efficient, you already believe your current activities have value beyond their mere economic value.
A Walk Down 2Fort
People already do plenty of things with little plausible economic value. Consider, for example, the video game Team Fortress 2 (TF2). Released in 2007 after nine years in development, it was a pioneer of Western live service games and continues to have a large active playerbase to this day. Similar to many more modern titles, players queue into matches where they use class-specific weapons to help their team achieve a gamemode-specific objective. Unlike newer titles, these outcomes do not feed into any sort of visible skill rank.1 Matches exist essentially in isolation; they are not building up to anything.
One map, 2Fort, is rather infamous. The official wiki run by the game developers notes that "due to the map's design with long, narrow corridors, it's easy for a team to become entrenched and for stalemates to occur". Since the map runs the "Capture the Flag" gamemode, there is no in-game timer and matches can last seemingly forever. Many players treat this as a sort of feature, queuing for games of 2Fort with no intention of playing to achieve the provided objective. Instead, they blow each other up, form conga lines, and throw sandviches at each other with no purpose other than to have fun. Community-run servers have been created specifically for this purpose and often feature modifications such as removing the objective entirely and removing the delay before respawning after death.
Perhaps it could be explained that these people don't find value in what they are doing or it is somehow instrumental to economic efficiency. Maybe Sam Bankman-Fried provided a similar explanation to himself for why he spent his time hardstuck Bronze in League of Legends. This reframing is far too hasty: leisure does not stop being enjoyable without it being towards something else. Of course not everyone will enjoy the same things, but rollercoasters, sports, walking outside, swimming, etc. are not fun simply in virtue of furthering employment.
The Spectre of Marx
In Karl Marx's transformation to the preeminent avatar of the left, the multiplicity of his thought has been flattened into a caricature. Marx the critic of the industrial revolution, Marx the condemner of the bourgeoisie, Marx the socialist/communist, Marx the revolutionary advocate, but almost never the Marx who pondered if capitalism had gone far enough. This sounds preposterous, but this line of thought has been a quiet-yet-persistent undercurrent of Marxian theory--one waiting for its moment to come.
One of Marx's great insights was that people's social relations are in large part determined by their economic circumstances. His 1859 Preface to A Contribution to the Critique of Political Economy provides an account of historical epochs based on this insight. One epoch transitions to another when the "existing relations of production" that once served to develop the productive forces "turn into their fetters", restricting their further development. Capitalism, like feudalism before it, would become a fetter, giving way to communism. In this account, communism is not less productive than capitalism, but more. With this, Marx's declaration "we say to the workers and the petty bourgeois: it is better to suffer in modern bourgeois society, which by its industry creates the material means for the foundation of a new society that will liberate you all, than to revert to a bygone form of society which, on the pretext of saving your classes, thrusts the entire nation back into medieval barbarism"2 makes perfect sense.
What then does this new communist society look like? With the radical productivity under communism, the economy would no longer be a significant determinant of social relations. Instead, communism would allow for the "free development of human energy". In the unpublished Saint Max3, Marx and Friedrich Engels consider painting:
The exclusive concentration of artistic talent in particular individuals, and its suppression in the broad mass which is bound up with this, is a consequence of division of labour. Even if in certain social conditions, everyone were an excellent painter, that would by no means exclude the possibility of each of them being also an original painter, so that here too the difference between “human” and “unique” [in the egoist sense employed by Max Stirner] labour amounts to sheer nonsense. In any case, with a communist organisation of society, there disappears the subordination of the artist to local and national narrowness, which arises entirely from division of labour, and also the subordination of the individual to some definite art, making him exclusively a painter, sculptor, etc.; the very name amply expresses the narrowness of his professional development and his dependence on division of labour. In a communist society there are no painters but only people who engage in painting among other activities.
They are not painters because "painter" is an economic category. People are free to engage in activities as they wish, but these activities do not define them. Instead, their limits are defined by "the boundary created by matter itself", as G. A. Cohen puts it.
Solve Metaphysics Make No Mistakes
"Philosophy with a deadline", warns Nick Bostrom; "anyone trying to work out what they think about accelerationism better do so quickly", rejoices Nick Land; "think fast, chucklenuts", says Scout TF2. Let's then quickly sketch out the problem: (1) AI that economically obsoletes humans will be capable of enacting a very large amount of change in the world. (2) That capacity will be used towards some goal. (3) It is therefore important to get this goal correct.
(1) and (2) are basic preconditions of the deadline, so it is (3) that merits attention. To illustrate (3), in Superintelligence Bostrom considers a superintelligent AI tasked with hedonism. Such an AI would want to tile the universe with "hedonium", which is "matter organized in a configuration that is optimal for the generation of pleasurable experience". If it was wrong about what produces pleasure, it might then be doing its best to convert the universe into something essentially worthless, and if hedonism is wrong, it's practically guaranteed.
This problem structure repeats across other goals we could explore: the AI optimizes the goal and then we get buyer's remorse. If we can't be confident we have the perfect goal, we should therefore delay until we do. Yet it's unclear what this would even entail. Are we supposed to just keep sciencing and philosophizing until words like "consciousness", "qualia", "soul", "good", and "evil" can be empirically measured? Has the long history of philosophy given us any indication this would work?
Natural Intelligence Doomsday
AI is not the only deadline we face: the earth continues to get hotter, the nukes are all ready to blow everything up at a moment's notice, and a pandemic could wipe everyone out. The natural world is fraught with peril. So far we've been able to avoid the worst of it. Will human intelligence be enough to face all future challenges?
The question with AI is not if it will achieve the maximally good outcome, but if it can achieve a better outcome than without it. So long as the outcome with AI is better than without it, anything else is just a bonus. Preventing the development of AI is not a neutral act. It could perhaps sometimes be justified to delay it to work towards circumstances for an even better outcome, but we should not let the perfect be the enemy of the good.
The Curious Task of Left Accelerationism
Supposing then that giving AI a good goal is akin to aiming for the broad side of a barn, the path forward still remains murky. All current AI systems are developed under a capitalist mode of production. With AI becoming an increasingly important part of the global economy, it is difficult to take beliefs in machines of loving grace, gentle singularities, or five-year plans at face value. The suspicions that advances in AI will simply perpetuate capitalism, perhaps in a more dystopian form, are difficult to shake.
Yet searches for other revolutionary paths have turned up empty: using capitalism to defeat itself remains the most plausible line of flight. The problem remains difficult, but it is difficult in human terms. A pause in AI could be ruinous because it lets AI developers think hard about who they cut out or because it prevents AI entirely. Accelerating AI too carelessly could risk missing the barn entirely. Nationalization could prevent AI companies from running away from the rest of us, but it could also be the demise of AI.
The task of left accelerationism is then both obvious and elusive. Obvious in that there is a clear target to aim for, but elusive as the path has not yet been forged. Yet if we are to believe in any sort of indomitable human spirit, surely it is in solving a political problem once rather than solving increasingly difficult challenges forever.
- ^ Montesquieu LVI, Neue Rheinische Zeitung
- ^ MECW and others present Saint Max as part of a book Marx and Engels supposedly wrote called The German Ideology. Saint Max was intended for standalone publication and The German Ideology, as it is standardly presented, does not exist. See Farewell to The German Ideology for more information.