# _AI Alignment_ or _AI Control_ it's probably many many things. Inspired by Alec Harris's assertion that [alignment work is 8x more promising that control work]("https://blog.redwoodresearch.org/p/the-case-for-ensuring-that-powerful), I want to explore that tradeoff between control and alignment research a bit more. And I want to define some terms that are at the core of Harris's argument, mostly for my own understanding, but also because I found his argument compelling, almost self-evident, and am surprised the argument needs to be made at all. It seems that roughly 90% of safety research dollars are being wasted, endangering us all in this arms race against the emergence of destructive autonomous agents. I won't try to define AI, and instead try to be precise in my thinking by using terms such as LLM, agentic (or self-prompting) LLMs, chatbots, and rarely _NPCs_. My education, training and profession has been the deployment of safety-critical control systems for software with the lives and welfare of many humans at stake. I have no I was not familiar with the acronym ToC, nor current state of funding for AI safety profile is not aligned with Harris's at it does not yet describe the "party-line" for AI safety researchers and their financial supporters. My goal is to speculate about possibly willful ignorance of the elephant in the room when it comes to autonomous systems control discussions. I think I'm allowed to speculate about the will of research communities and corporations, considering that those self-organized systems speculate about the intent and will of software systems such as LLMs. In his blog this week Cory Doctorow pointed out that whenever you ask yourself "how can this one thing [AI] have such a contradictory set of features and effects?" the answer is that "it's not one thing, it's **two** thing**s**." Cory Doctorow was referring to AI and its two contradictory uses: 1. within a productivity-enhancing centaur relationship, empowering us to spend more time thinking about ways to improve our workflow, rather than 2. within a reverse centaur where the AI is configured to exploit and replace humans in the modern information economy that has evolved into a zero-sum every-[privileged]-man-for-themselves economy ### References 1. 2026-07 Alec Harris [alignment work is 8x more promising that control work](https://blog.redwoodresearch.org/p/the-case-for-ensuring-that-powerful) on lesswrong.com 2. 2026 Richard Campbell: [after all the hyper, what's next for software engineering](https://www.youtube.com/watch?v=uWnUnMphmPM) at NDC Copenhagen