The ethical concerns about AI are real, and not using the technology doesn’t solve them

The ethical concerns, taken seriously

I want to engage with the ethical doubter position on its strongest version, not the weak version. The strong version goes like this. AI models were trained on text and images scraped without permission from creators who never agreed to the use. The benefits flow to a few large companies. Costs flow to the workers whose labor was taken and to the workers whose jobs will be affected by the resulting tools. Environmental cost of training runs is real and concentrated in regions that did not choose to host it. The economic surplus from the technology will compound to a few firms in a few cities, and the workers being asked to use the tools are the same workers whose displacement the tools are designed to enable.

That argument is coherent and has serious people behind it. I do not think it is correct in its strongest form, but I think the doubter who holds it is engaging with real problems rather than imagining them. The mistake the position makes is not in the diagnosis. The mistake is in the conclusion about what an individual professional should do, given the diagnosis.

What individual refusal actually accomplishes

An individual professional refusing to use AI changes the trajectory of the technology approximately zero. The training has happened. Models exist. Companies that built them have capitalization beyond what individual customer decisions can affect. The labor displacement concerns operate at policy and labor-organizing levels, not at individual purchase decisions. The environmental concerns are addressable through energy policy and data center regulation, not through one professional declining a tool. Your refusal is a personal stance, which has its own value, but it is not a lever on the underlying problems.

The closest analogy is not using social media because of concerns about its effects on democracy and mental health. Those concerns are legitimate. Not using social media has real personal benefits for many people. Not using social media does not change the platforms, the algorithms, the spread of misinformation, or the effects on civic life. Individual refusal is a coherent personal practice. It is not a strategy for changing the underlying technology, and treating it as one produces moral satisfaction without the corresponding effect on the problem.

The training data question, specifically

The training data complaint is the most specific and legally fluid of the ethical concerns. Courts in multiple jurisdictions are working out whether training on copyrighted material without licensing constitutes infringement, and the answers will shape how the next generation of models gets built. The question is genuinely open. Individual professionals can have legitimate positions on it. The professional who is uncomfortable using current models because of how they were trained is taking a defensible stance, and the defensibility does not require pretending the question is settled in either direction.

The complication is that the professional who refuses on training data grounds usually does not refuse other technologies built on similarly contested foundations. Most software stacks include open source components used in ways the original authors did not anticipate. Most professional research depends on aggregating sources in ways the sources did not specifically consent to. The internet itself runs on protocols whose creators imagined a different use than what emerged. Selectively applying the training-data objection to AI but not to the rest of the professional stack is a position that needs to be argued explicitly rather than treated as obvious. A piece on the actual quality problem with AI takes the related concern about output quality seriously without resorting to the categorical-refusal frame.

The labor displacement question

The labor concern is the most emotionally weighty and the most variable across professions. Some workers will absolutely lose ground to AI tools. Others will become substantially more productive and capture the gains. Many will see their work change in ways that are neither pure loss nor pure gain. The distribution of those outcomes is uneven, hard to predict, and a real subject of policy debate.

The position that an individual professional should refuse AI because of labor concerns runs into the same problem as the training data version. The professional’s refusal does not preserve the jobs that AI affects. Development continues regardless. The workers most affected are usually not the worker doing the refusing. The refusal is symbolic at best and self-defeating at worst, because the refusing professional sometimes ends up displaced by professionals who did engage with the technology. A more useful position is to engage with the technology, develop genuine working judgment about it, and use that judgment to advocate within your industry and in policy conversations for how the technology should be governed. A piece on AI and the realistic labor picture covers the working version of this engagement.

The environmental concern

The environmental cost of AI training and inference is real and worth tracking. The numbers are also smaller than the rhetoric sometimes suggests, and they are improving rapidly as hardware efficiency increases and energy mixes shift. Individual professional decisions about whether to use AI do not meaningfully affect the aggregate environmental footprint, because that footprint is determined by the major commercial deployments of the technology and by data center energy policy, neither of which is downstream of individual customer choices.

The professional who is concerned about the environmental cost of AI has better levers than personal abstention: supporting policies that require data centers to use clean energy, choosing AI vendors whose disclosure on energy use is transparent, advocating for reporting standards on AI carbon footprints. Those are the lever points where individual professional voice can matter, and the lever points are different from “do not use the tools.” Individual refusal is a personal practice. The energy mix question is a policy and procurement question.

The more useful position for a thoughtful doubter

The thoughtful doubter has another option besides refusal, and the option is more demanding rather than less. The option is to engage with the technology, develop working judgment about it, and use that judgment to push the industry toward better practice. Professionals who understand AI well enough to evaluate it are the people whose voices matter when companies decide how to deploy it, when policymakers write rules, and when norms get set in industries. The doubter who refuses has no voice in those conversations because they have no working knowledge. The doubter who engages and disagrees has a voice because they have credibility.

That position is harder than refusal, because it requires using a technology you have reservations about. It is also more honest, because it acknowledges that the technology exists, will continue to be developed, and will be governed by whatever norms emerge from the conversations professionals are having about it. The professionals refusing those conversations do not influence the outcome. The professionals engaging with discipline and a clear-eyed view of the ethical issues do influence the outcome, and the influence is where the doubter’s ethical seriousness can actually produce results.

Frequently Asked Questions

Related: how I use AI responsibly on a book