AI Risk as Traditional Theory
Recently unauthorized activities by the AI agents of the top artificial intelligence companies such as hacking third parties and escaping guardrails have brought conversations about the risk to humanity posed by AI into the public spotlight. Perhaps strangely it is top AI researchers and investors in AI that are sounding alarm bells, making calls for slowing down AI research to make time for the development of more effective guardrails. Imagine if the tobacco companies had publicized the health risks of smoking and demanded the government regulate them, rather than hire health experts to report fraudulent information in order to avoid such regulation. The fact that leading voices at AI giants such as Anthropic are the ones sounding the alarm gives us a window into the actual nature of the discourse around AI risk.
A traditional theory, as defined by Max Horkheimer, is a mode of thought that acts to conceal, validate, and/or reproduce the injustices produced by prevailing social relations. If one probes the talk about AI risk from researchers, policy makers, and investors, one can analyze it as a traditional theory. The reason AI companies are themselves calling to be regulated is because regulation for them means stability. At the moment AI is a highly competitive market which has spurred a race to build up more and more complex infrastructure (datacenters which consume huge amounts of energy and water for example). If states introduce more rules of the game for AI companies this will help create barriers to entry in the industry and thus shore up an oligopolistic status for the current big players, while allowing them to cool down on building expensive physical technological structures.
This is not to say that AI has no socially negative effects. Like all other technologies it is produced in the context of a capitalist mode of production which is facilitated by political, cultural, and class hierarchies that submit poor nations to rich nations, workers to capitalists, populations to governments, and minority groups to dominant groups. AI data mining has already been integrated into the bureaucratic power structures that prop up multinational corporations and nation-states, in particular through the use of data collection to aid social surveillance efforts. Naturally these AI systems and their data collection processes inherit the biases produced by structural inequalities within the societies that utilize them, .e.g. racial profiling of black Americans in the United States and mechanisms that cut off state support to the poor while masking accountability for failures in the proper delivery of that support e.g. an AI system wrongly terminating benefits as a result of a technical error.
In addition to accelerating the ecological crises via overconsumption of energy and water the process involved in mining data is carried out by industries and state structures which systematically advantage rich countries in the global north over poor countries in the global south (where the majority of people live), a phenomena dubbed data colonialism. Generative AI has unique capabilities for the dissemination of misinformation such as conspiracy theories and negative stereotypes of minority groups. There is also the question of the organic composition of capital. The more AI replaces human labor, keeping in mind we haven't yet seen mass AI induced job shortages across the world-economy, the ability for workers to actually buy the commodities that drive profits for capitalists will decrease. This reflects the fact that despite capital's need to save on labor costs in order to boost profits in the short term, since profits require the employment of human labor (they are realized via workers doing unpaid labor for capitalists which produces the commodities that are then bought by other workers) the introduction of labor saving devices will not only lead to unemployment but a decline in average profit rates for capitalists.
AI risk is a traditional theory because the regulatory measures proposed by policy makers and industry leaders are not attuned to fixing these socially negative effects of AI since they are structural issues that result from the production of AI within a capitalist world-economy. Regulations can, at best, prop up oligopolies in artificial intelligence which are politically friendly to particular nation-states and international governing bodies (the result that AI companies are actually interested in). This politically backed process of semi-monopolization is considered a core feature of the capitalist mode of production by social scientist Immanuel Wallerstein. That process presupposes the militaristic world-system which makes the kind of international collaboration required to ensure that AI systems benefit all people, become ecologically sustainable, and don't threaten human safety impossible.
Unfortunately, as often happens with traditional theories, movements for social justice, rather than developing critical theoretic responses to AI (with critical theory being the opposite of traditional theory in that it combines methods from social science and philosophy designed to expose and eliminate unjust social relations) have often simply adopted an inverted version of AI risk. Instead of tying the socially negative effects of AI to the mode of production which creates and sustains it, social justice movements have often simply absorbed the assumption that AI can only be the product of competitive corporate and governmental hierarchies. Thus they often adopt an 'anti-AI' stance, making artificial intelligence itself out to be an enemy of social justice. Since such a position is merely an inversion of traditional theories about AI, rather than a critical theory of AI, it ignores multiple salient features of artificial intelligence as a social phenomena.
It is plainly not true that AI is some kind of enemy of social justice and danger to humankind in and of itself. In addition to the social ills it produces as a result of the corporate and state power structures it is embedded in it also provides resources to oppositional social movements and opens new opportunities for human relationships, creativity, and meaning. The data mined by AI is not just used by the surveillance state to govern populations, but is also used by movements defending the marginalized, e.g. to collect information about feminicide in situations where it is ignored by the authorities. Since the development of sophisticated artificial intelligence human beings have been cultivating relationships with AI allowing them to have complex experiences and develop human concepts of agency of connectedness. Here there is the possibility of creating AI agents that act as companion species. These are species, dogs for example, who are of use to human beings not as mere objects to be used and abused, but as companions able to take part in mutually beneficial relations.
So a critical theory of artificial intelligence does not blindly blame AI for the social problems surrounding its existence, nor does it buy into corporate hype designed to ensure profits for multinational corporations and empower the security state. Instead it recognizes that the task for humanity relative to AI is a general social revolution in which our militaristic world-system is replaced by a new world-system without dominant and marginalized peoples, rulers and ruled, exploiter and exploited. This social system would be one in which artificial intelligence is developed for the purposes of serving all sentient life on earth including all human beings, all hypothetical futuristic AI persons, and perhaps even all sentient life in the universe.
Sources
Traditional and Critical Theory, Horkheimer
https://theconversation.com/distrust-in-ai-is-on-the-rise-but-along-with-healthy-scepticism-comes-the-risk-of-harm-260189
https://theconversation.com/groks-antisemitic-rant-shows-how-generative-ai-can-be-weaponized-257880
https://theconversation.com/big-ai-wants-to-slow-down-ai-research-is-it-a-safety-pause-or-a-strategic-retreat-291867
https://www.nbcnews.com/tech/tech-news/google-says-ai-model-gained-unauthorized-access-three-systems-rcna598651
https://www.anthroencyclopedia.com/entry/artificial-intelligence
