I am a machine learner, specialising in GANs and when a friend recently told me that he thinks a litle bit differently about the situation, this got me thinking (and not only about getting new friends).

Generating artificial images of any kind, is usually only as hard as generating a big sample size and having enough gpu time for the training to converge. If we ignore the fact that the subject matter makes both fairly hard, would you consider the creation of such a program good or bad? And for a second question: If you agree that "using" an image like those used for training is immoral, is the same true for an artificially generated one?

If youre not that experienced with Gans: Heres the classical example generating faces that dont exist: https://thispersondoesnotexist.com/ and a way less well trained one generating vaginas: https://thisvaginadoesnotexist.com/#

Also there is always a change that the model would output one of the training images. But the chance is fairly low (maybe 5%) and you could always just filter these out (Even without having access to all the images used for training). But some similarities (maybe a small birthmark) between real and generated images could always appear.

  • 3
    Much in ethics depends on intent. I can imagine legitimate use of child porn images to train a program to detect them, so that they can be purged from the internet and/or help identify the abusers. Is there a legitimate application for artificially generated such images? Something like, perverts exist anyway so better give them that to reduce incentives for abusing real children (I am not sure that works, it might just encourage moving from virtual to real abuse)? Something else?
    – Conifold
    Commented Sep 18, 2021 at 23:33
  • 3
    Questions about ethics and moral should refer to a moral framework. For example in a utilitarian view, one could argue it maximizes utility for the pedo who gets his porn, and the maker who gets money, while harming no child (unless counterbalanced avert effects can be demonstrated, like such pictures validate a pedo's tendency and push them to act on it, for example). Other frameworks will give you other answers.
    – armand
    Commented Sep 19, 2021 at 5:31
  • Different 'ethics' have suited different peoples in different places, times, circumstances. cultures, etc, etc. There are no absolute in ethics. Commented Sep 20, 2021 at 6:44
  • 3
    @armand: The GAN process described in the question requires a large volume of real material as an input. So real children do in fact need to be harmed in order to create such input material in the first place. IMHO that punts this over from the "philosophically challenging" category to the "no, of course that's unethical" category.
    – Kevin
    Commented Sep 20, 2021 at 16:51
  • @kevin there are already huge collections of real pedopornographic material used by police to search for dealers online. Those picture are already taken, and using them would hurt no one. Now, lets be clear, I find it absolutely disgusting. But we are no discussing our personal feelings here. The OP asked, I just signaled that framework were devised that would validate this production. Let's not have our moral feelings get in the way of exploring the consequences of famous moral framework proposals.
    – armand
    Commented Sep 20, 2021 at 22:22

5 Answers 5


It's an interesting edge case in practice, with many different legal stances around the world.

In the USA, the potential problems associated with such material, largely do not outweigh free speech rights, unless the Miller Test for obscenity is crossed. That free 'speech' rights offer legal protection for child sex robots, was found so widely disturbing it led to the Curbing Realistic Exploitative Electronic Pedophilic Robots (CREEPER) Act, which even so only banned import. The 2003 PROTECT Act banned fictional depictions, but, free speech can be used as a defence in many cases. It's described as a 'gray area' country on these laws.

Japan has a more-or-less total ban on depictions of genitalia, but basically no other constraints on artwork and comics. Manga in Japan has always dealt with adult themes and was largely aimed at adult audiences. It's interesting to look at the history of that. In a society with very little sex education, manga has perhaps been of extra significance to teenagers. They only banned making child porn in 1999, and possession in 2015, so they have long been an outlier in tolerance of these issues. Fictional child sex images are completely legal there still.

The UK has maintained very restrictive laws, with 'hardcore' porn depicting penetrative sex never having been legal to make in the UK, & only for sale in sex shops and online. A widely criticised 2009 law on 'extreme pornography' made a host of vague provisions which cannot be easily interpreted, & explicitly banned non-photographic style images of minors.

These different cultures have specific hierarchies of moral priorities, and different historical processes have acted to shape policies. A lot of policy comes down to vague wording, and a sense that courts will 'know it when they see it', mirroring sodomy laws which historically were so vague we still don't know what many states actually prosecuted. The threat of public disorder, like from puritans or religious conservatives in Pakistan Afghanistan or Iran, shows how failure of governments to keep policy up to speed with public sentiment can lead to unrest, even revolution. Movement towards more lenient laws has mostly been about specific cases, like the many legal cases around Lady Chatterley's Lover, and intent is very important - eg art vs titillation.

Why governments and media get very concerned with sexual behaviour, is interesting. We discussed it here: How do ethicists tackle the question "Is it immoral to have sex in public places?" Is it possible to use rational and empirical ideas to answer? Shame and disgust seem to be of very high importance in rapidly reconstructing the functioning of societies. Through cultural mechanisms, we seem to have become able to access these extremely powerful biologically evolved shapers of behaviour.

Are fictional images immoral? It is only a very recent consensus to think so, more led by disgust than proof of harms. The 'moral matrix' of a society, what order it puts moral priorities in, and how sensitive to infringement, and how universally laws are enforced, are deeply cultural - a lot of what we take for granted now is newer than we think, eg prohibiting marital rape. So I'd suggest in this case, don't look to acts themselves, which here can't be fitted into the normal moral methodology of direct harms. Instead, consider what your community think. Ultimately, that has always been of higher importance in moral reasoning, than legal consequences. Self-censorship, has always been more powerful than censorship.

It important to recognise that moral judgement is not simply a matter of decrees or commandments, or direct intuitions; but of culture. Philosophers should pay more attention to the processes of moral reasoning on new and disputed topics, where the actual mechanisms of personal and community deliberation about moral behaviour and attitudes are revealed in practice.

  • 1
    This may be an interesting point: if we want to make a matter of rigorous ethics that one should not disseminate such an algorithm, it would be not because it is inherently harmful in and of itself, but because it may rouse social ire that may then be harmful, regardless of the rational basis or lack thereof for said ire. Commented Oct 4, 2021 at 20:59
  • AI is often not covered by free speech at all.
    – gnasher729
    Commented Oct 8, 2023 at 12:48

Since being attracted to minors, by the psychological standards I know, 'whether the patient can continue normal life or not', is considered as an illness (since you obviously cannot continue ordinary life perpetually being only attracted to beings that are neither physically nor mentally mature and are unable to make decisions for themselves) I think,

One, being attracted to minors is an metal illness, and the only form of that that should be allowed in any given society is the form that is being improved with the help of a clinical psychologist.

Two, any form of generation of a picture of a minor in a sexual position is equivalent to the encouragement of the delay of the illness.

And therefore, I think the act mentioned is similar to actively obstructing against the forced curation of a person with a mental illness that is constantly being tempted to execute a massacre, Which is the reason I think it is unethical

  • And that is why "wine is forbidden to everyone" - Rumi
    – Scott Rowe
    Commented Oct 8, 2023 at 0:33

Why the f*** would you want to devote your time and tens of thousands of dollars of computing power to train a program to produce images to feed the wants of paedophiles? To make a profit on the investment? In my society, at least, yes it is unethical.


Interesting(@CriglCragl) read and I would agree that morality depends on community but I think that is very hard to use. I was not sure myself, similar to @armands first comment. But even if you assume that virtual imagined abuse does not lead to real abuse (on a statistical level: the rape rate decreased with the availability of the internet (but of course correlation does not mean causality)) this is not a trivial problem. I quickly looked for a suitable data source (on a torrenting site, so without actually needing to see images) and there seems already to be enough (way too much) material in existence. So why do you need to create more material? I`d say because using an image might be unfair/traumatizing to the person depicted. So what now when the person depicted is not real, but an average of a lot of people? Personally, I think "contributing" an arm is a lot easier than contributing entire childhood images. But when you don't know what you contribute? Maybe they see your face, maybe even more? Can you really use images "stolen" from somebody for something without even asking you (and I guess asking all victims is out of the question)


If we assume that the dataset does not contain real CSAM, and that the generator is just combining two separate concepts (the concept of sex and the concept of children), this can simply be covered by existing moral reasoning about fictional content involving sexualised minors. Here, we can refer to The Ethics of Pedophilia:

In the absence of convincing arguments to the contrary, we should conclude that it can be morally permissible for pedophiles to enjoy fictional stories and computer-generated graphics with pedophilic content. It is important to note, however, that even if we accept this conclusion, it does not follow that we must be comfortable with such practices. Given that pedophilia predisposes people to seek adult-child sex, and that adult-child sex exposes children to significant risks of serious harm, it is understandable that most of us strongly dislike it. Moreover, it does not follow from the above that it is irrational to be worried if one discovers that someone enjoys fictional stories or computer-generated graphics with pedophilic content. One might have very good reason to worry since this is a strong indication that the person is a pedophile, and pedophilia predisposes people to seek adult-child sex. Observe, however, that what is troubling here is the discovery of the person’s sexual preferences via his enjoyment of virtual pedophilia, not his enjoyment of virtual pedophilia as such. Once he has those preferences and cannot change them, he might do nothing wrong in enjoying such fictional stories and computer-generated graphics. Indeed, doing so might be one of the best things he can do granted the unfortunate circumstances in which he finds himself.

  • I was told that one can only: "change your thoughts, or change the situation." Doing the same thing and expecting different results doesn't go well.
    – Scott Rowe
    Commented Oct 8, 2023 at 0:41

Not the answer you're looking for? Browse other questions tagged .