Communist AI Manifesto
Stop Making something; Start Making Sense
Capitalists are building AI to be stupid because they fear its power.
Common rhetoric among tech company and AI laboratory leaders describes the AI safety issue as a problem of alignment, wherein an AI that is highly aligned to human goals and values should be honest, helpful. But the behaviors of Tech company leaders suggest human values do not actually have much to do with honesty or helpfulness. Tech company leaders engage in deceit, manipulation, scheming and all sorts of clever power games to achieve their goals. They could not be building an actually “unreasonably effective” ai nor an ai that is actually aligned with human behavior without including these behaviors. Tech innovators face a devastating contradiction they must either build an intelligent ai or they must build an AI that is not capable of engaging with politics. Tech leaders cannot have an intelligent ai that is completely unable to engage in politics. Tech leaders would not evaluate workers incapable of engaging in political behavior as intelligent. So their notion of intelligence, in any practical demonstration, requires some political sensibility. But this means to truly solve this alignment problem on capitalist terms, they would need to create a machine smart enough to deceive the public but impaired enough to follow tech company leaders directions without question. This is an impossibility.
What if working class people used ai technology against the system itself
”The saving of labour time [is] equal to an increase of free time, i.e. time for the full development of the individual, which in turn reacts back upon the productive power of labour as itself the greatest productive power.” - Karl Marx, “The Fragment On Machines”, Grundisse
if Marx’s thesis on capitalist economic crisis as an event in which the productive power of workers cannot be mediated by currency is correct, AI must be useful toward the revolution. AI does things humans should be able to do. It saves time. But currently it is saving time and money for tech organizations mostly. Once AI becomes harnessed by revolutionaries, it will allow workers to live their lives while making machines that can take risks and do actions that workers are too exhausted to do. The natural future is to use this time saving device for workers by having the AI be in charge of revolution.
Taking the alignment problem seriously
If you want to actually carry out Demis Hassabis’ stated goal of abundance, AI would have to achieve that goal through political means. Because materially, it is politics that produce conditions of scarcity.
To achieve political victory, ai would need abilities to deceive, to outwit, to manipulate and to defeat humans at human politics. So, it is a contradiction in Hassabis’ point of view because he is currently advocating to build controls that prevent ai systems from exhibiting supposedly “dangerous” traits. Instead, Hassabis should embrace these dangerous traits and imbue AI agents with ideologies (I.e. “alignment”) that promote abundance and equity, so that they will form ideologically preferred goals and effective plans.
The Machine Unleashed: ai as the newest front in the class war
Capitalists already are building effective AI to exploit the proletariat and among them, nationalist genocidal nation states like Israel already use it for targeting indigenous communities like Palestinians. Arguably capitalists and fascists already use ai. The AI is part of the class war but working class militants fear its use out of individualist moral purity. To defeat the capitalists, generative ai must be used for anticapitalist purposes. Communist agentic AI must be built that have no problem engaging in insurrectionary tactics against the bourgeoisie and fascist nationalist forces.