Because I will be starting an AI course next week that deals with modern ethical theory, I think it worthwhile to get my unadulterated thoughts down first, before they become corrupted by the university.
Problem: AI use is not free, and the costs are difficult to quantify. These costs include, without accounting for specificity or rank (or thoroughness): thermal, energy, air, water, land use, societal, political, military, privacy, and so on. Each of these stands on its own, and I make no attempt to delve herein, but simply to mention the complexity. The shortened version is that AI use is not free, so it appears we should be mindful of its use.
Framing: I have a conceptual issue with the notion that AI use is not equal, in either quality or intent. Whenever people discuss AI, it is within a giant bucket of all things AI, instead of being appropriately compartmentalized based on use cases. Making a meme is not the same as proofreading an environmental permit, and thus should not be held to the same standards for use. Therefore, it is appropriate to discuss AI, not in its whole-bucket entirety, but at the use-case level.
Implications: So far as we can reduce this complexity down to something digestible, I see two main items to address.
1. Given the difficulty in quantifying the AI use problem, we should be mindful of its use, generally, and limiting its use to beneficial use cases.
2. We take it for granted that certain use cases are beneficial.
There are several, immediate issues and outcomes from these two statements. The first is that we presume to know what the word “beneficial” means, and that we are the arbiters of that definition for all of humankind. The second is that the AI use problem changes over time, sometimes by the day. Perhaps it is not ethically permissible to use the latest, computationally taxing model to create a meme for your friends, but it may be using a different model. What is the price of a smile, anyway? Whereas I want absolutely the best model available for environmental protection, and I am assuming that the benefits outweigh the runtime costs.
The very, very short version is that not all use cases are equal. It would be nice for a query to initially return an estimated cost before run, “Your meme will cost 0.2 kW, and 4 mL of freshwater. Do you wish to proceed?”
Anyway, I’ll revisit this subject after my indoctrination.
11380' 