On AI

Recently, opinions on AI use have become increasingly polarized. Be it AI slop, AI in schools, or AI’s insane energy use, it is not unreasonable to view AI destructive force. This is all despite increased usage among the general population. There’s a common saying I have seen that goes “I thought AI would take all the hard jobs and we could do art all day, instead AI’s doing all the art and we’re doing the hard jobs”. It sucks to watch artists, a group who already struggle in terms of finances, find it harder still to find work. Nevertheless, if we want to examine AI, both the benefits and consequences, first, we must take a step back from the individuals effects AI is having.

Due to the functioning of our global market economy, one company or country deciding to progress slower with AI only hurts them, as their competitors will simply not slow. That is point one. Unless we are to have a Butlerian jihad (destroy all the AI unanimously which will never happen as we can’t even unite around climate change), our functioning as separate states with independent companies within said states means that AI progress will not stop.

The second point is that there is a good chance that this progress will likely go beyond anything thought previously possible. Every month or so, a new AI model comes out that beats the benchmarks of the last, and when we compare the models from 2022 to 2025, it’s a completely different ballgame. To think in 2022 that AI would be able to code better than many developers and make entirely realistic looking videos by 2025 would be unbelievable. How will models look in 2030?

If AI continues to improve, we should thus consider the possibility of widespread structural unemployment. Structural unemployment is when someone loses their jobs due to advances in technology, such as the milkman. If AI reaches the point where work such as accounting, lawyering, and general white collar work can be laid off and replaced by AI (which is probably much cheaper), it is only rational that companies will do this.

The cons to this are obvious. Large amounts of people will lose stable employment. Firms and investors will control AI driven productivity, further increasing wealth inequality. Income volatility increases, worker protections and bargaining power decrease, and wealth/power are consolidated more.

A short term response would be increasing taxes on firms that benefit more from automation and using that money to create public employment in areas underallocated by the market (infastructure, ecological restoration, care work, etc). Historically think of FDR’s Works Progress Administration. In the long term, AI could reduce working hours and create better living standards, but only when paired with deliberate government intervention. I have heard anecdotally that in China, people are supportive and optimistic regarding AI in work because they see the government as using AI to do the aforementioned things. Compare this to the US where, besides those at the top, people are mostly scared. These reflect the different aproaches taken by the government.

AI can multiply currently existing economic and political incentives. As a tool in of itself, AI is neutral or mildly negative (AI slop). Structurally, AI in profitmaximising systems means increases in power for those already at the top. This is done by eliminating jobs, hollowing out expertise, and increasing capital and state power through data. Restrictions are needed on the systems around AI mainly, as again AI progess is inevitable but must be used in ways that reduce general inequality instead of increasing it.

AI’s energy consumption is a legitimate concern, but two points are worth considering. Firstly, efficiency does improve over. Much of why Deepseek gained attention is that is was able to achieve similar results to models for a fraction of the cost. Secondly, AI can be broken down into different uses. High impact problems like medical diagnostics or scientific research (two linked studies: 2024, 2025) is undoubtably more critical than helping make a grocery list. Considering selectivity, many instances can justify the cost.

Regarding AI in politics/government, this may be out there for many of you, but it could be implemented in helpful ways. For example, if AI could analyze bus routes and identify a more productive route, that would be a useful tool. The question is not wether AI can be used in government, but if it can be used to enchance democratic decision making. How far this should go remains an open question, one that should be grounded in governmental design rather than cult-like enthusiasm.