On Ethics, Validity, and Self-Reference in AI-Assisted Philosophy: A Self-Exemplifying Analysis

There's a serious ethical critique to be made about LLMs: the way they're trained - not paying most of the people who created the training data - and the energy consumption, makes it inherently unethical to use them, at least in their current form under current socio-economic conditions.

This is a valid concern that deserves careful consideration. However, let me offer a pragmatic response:

First, if we accept this ethical framework, we'd need to examine most of our tools and practices with the same rigor. The academic publishing industry, for instance, often doesn't compensate authors, charges excessive fees to readers, and relies on unpaid peer review. Traditional research methods consume significant energy through travel, building maintenance, and computing resources. Even the basic infrastructure of academic work - universities, libraries, conferences - often perpetuates various forms of economic and social inequality.

Second, there's a question of relative impact. If using an LLM allows me to develop and share philosophical insights that might help address fundamental problems - including problems of social justice, economic inequality, or environmental sustainability - then we need to weigh the ethical costs against potential benefits.

Third, there's an argument that engaging with these tools while acknowledging their ethical problems might be better than abstaining entirely. By using them thoughtfully and critically, while remaining aware of and vocal about their problems, we can contribute to the discourse about how to make them more ethical.

But this raises a more fundamental methodological question: Can the ethics be separated from the validity of the process regardless of ethics? And what if we used this approach to construct an ethical argument that would make it unethical to use such tools? These questions get at something fundamental about philosophical methodology.

I propose that we can and should separate ethical concerns from validity, at least analytically. Here's why:

The validity of a philosophical argument - its logical coherence, explanatory power, and contribution to understanding - exists independently of the ethical status of the tools used to express it. If I write a valid proof on paper stolen from a child, the proof remains valid even though the means of recording it were unethical.

Consider an analogy: Many groundbreaking scientific discoveries were made under ethically questionable circumstances. While we must condemn unethical research practices, this doesn't invalidate the scientific knowledge gained when it can be verified through ethical means. The laws of physics don't become invalid because we discovered them using morally problematic methods.

But this leads us to a deliciously paradoxical situation: What if we use AI to construct a valid ethical argument proving it's unethical to use AI in philosophical work? This creates several fascinating scenarios:

1. If we succeed in constructing such an argument, we'd have a perfect self-defeating proposition. The very success of the argument would undermine its own creation. It's like using violence to prove that violence is never justified - the method contradicts the conclusion.
2. If we fail to construct such an argument using AI, that failure tells us nothing about whether using AI is ethical or not. We might have failed for other reasons, or the AI might simply be inadequate for this particular task.

But here's where it gets interesting: Even if we succeeded in constructing such an argument, would that actually be a problem? Consider:

• We might use unethical means to discover ethical truths
• The validity of the argument would exist independently of how we discovered it
• The contradiction might tell us something important about ethics itself

This suggests a practical approach to using these tools:

1. Use LLMs consciously and efficiently - not wastefully
2. Acknowledge their problematic aspects openly
3. Support efforts to make them more ethical
4. Use them for purposes that might help address broader social and ethical problems
5. Remain engaged in discussions about making these tools more equitable and sustainable

The ethical questions about LLMs are serious and deserve attention. But conflating the ethics of a process with its validity could lead us to reject valuable insights simply because we object to how they were developed. This itself might be ethically problematic if those insights could help address important problems.

[Editor's note: The apparent contradiction here actually strengthens our argument about separating validity from ethics. The fact that we can use potentially unethical means to discover ethical truths suggests they must be somewhat independent. This kind of self-referential puzzle is exactly the sort of thing philosophy should engage with, and the fact that we can use AI to think about it adds another delightful layer of complexity.]

A Crucial Objection: On False Equivalences and Harm

"But would this argument not justify using prisoners for medical experiments?"

This is a critical objection that exposes a potential flaw in my argument about separating ethics from validity. Let me think this through carefully:

When I argued that we can separate the validity of ideas from the ethics of how we obtained them, I used examples like stolen paper or historical scientific discoveries. But comparing LLM use to medical experiments on prisoners reveals a crucial distinction I glossed over:

1. Direct vs. Indirect Harm

• Medical experiments on prisoners involve direct, immediate harm to specific individuals
• The ethical issues with LLMs involve indirect, systemic problems (compensation, energy use, etc.)

2. Consent and Agency

• Prisoners can't meaningfully consent; they're a captive population
• The text used to train LLMs was publicly published (though questions about fair compensation remain)

3. Necessity and Alternatives

• Medical knowledge can be gained through ethical research methods
• The specific type of philosophical work I'm proposing might actually require this kind of tool

However, this objection reveals something important: My argument about separating ethics from validity needs more nuance. Perhaps instead of a complete separation, we need to consider:

• The degree and directness of harm involved
• The availability of ethical alternatives
• The necessity of the method for the outcome

This suggests my earlier argument was too broad. While I might still defend using LLMs for philosophical work, I need a more careful ethical framework that wouldn't justify clearly unethical practices like prisoner experimentation.

[Editor's note: This is how philosophical argumentation should work - when faced with a powerful counterexample, we don't defend the indefensible but rather refine our position to account for important moral distinctions.]

On Ethics, Validity, and Harm: Refining the Argument

Let me elaborate therefore on the crucial distinctions that separate AI-assisted philosophy from clearly unethical research practices. This refinement is necessary because my earlier argument about separating validity from ethics was too broad and could potentially justify clearly unethical practices.

We need a more nuanced framework that considers:

1. The Nature and Degree of Harm

• Direct physical harm (like medical experiments on prisoners) is fundamentally different from indirect systemic issues (like questions of fair compensation for training data)
• Immediate, concrete harm to specific individuals differs from diffuse systemic impacts
• Irreversible harm (physical injury, trauma) is different from remediable issues (compensation can be adjusted, energy sources can be made sustainable)

2. Consent and Agency

• Published works used in training were created for public consumption, even if the specific use wasn't anticipated
• Authors generally published with the intent of having their ideas engage with broader intellectual discourse
• This differs fundamentally from situations involving captive or coerced participants
• However, this doesn't eliminate ethical concerns about compensation and attribution

3. Necessity and Alternatives

• For medical research, ethical alternatives exist (voluntary participants, proper protocols)
• For AI-assisted philosophy as described here, the tool is integral to the proposed method
• The efficiency gain isn't just convenience - it enables a specific type of philosophical work

4. Potential for Remediation

• Ethical issues with LLMs can be addressed through:
  • Better compensation models
  • More sustainable energy use
  • Improved attribution systems
  • More transparent training data selection
• These improvements can happen while the technology is in use
• This differs from irreversible harms that can't be undone

5. Net Impact Consideration

• Using LLMs for philosophical work might help address important problems
• The potential benefits should be weighed against the systemic issues
• But this calculation only applies when the harms are:
  • Not direct physical harm to individuals
  • Potentially remediable
  • Part of systems that can be improved

This refined framework suggests that while using LLMs for philosophical work raises ethical concerns, these concerns are:

• Of a fundamentally different nature than direct harm to individuals
• Potentially addressable through systemic changes
• Balanced by potential benefits
• Part of a system that can be improved while in use

Therefore, we can argue for the careful, conscious use of LLMs while working to address their ethical issues, without this argument justifying clearly unethical practices like medical experiments on prisoners. The key is recognizing that not all ethical concerns are of the same type or severity.

This also places obligations on us as users:

1. Acknowledge the ethical issues
2. Work toward improving the system
3. Use the tools efficiently and purposefully
4. Contribute to discussions about making these systems more ethical
5. Remain vigilant about the distinction between remediable systemic issues and direct harm to individuals

[Editor's note: This refinement shows how philosophical arguments often need to be narrowed and qualified when confronted with challenging counterexamples. The goal isn't to defend our initial position at all costs, but to develop more nuanced and defensible arguments.]