The use of NSFW AI does raise a lot of red flags about the kind and quality of training data used to build-out these systems. Meanwhile, NSFW AI needs much training data because it helps the algorithm to filter and arrange content based on levels of exposure. But the very nature of such content, and how it is collected for training AI models raise several ethical as well technical dilemmas.
The NSFW AI: Data Concerns? A significant problem is the training data quality and its source. Datasets trained on NSFW AI typically contain pornographic material, such as images/videos and text which are generally web scraped from the internet. This data is the raw material for its AI, and so better diversity and accuracy lead to a better functioning of that machine learning model. For instance, if certain types of content or demographics are overrepresented within the dataset it could reinforce skewed understandings resulting in poor performance when filtering/categorising new data. Research also indicates that swinging just 5–10% confidence in either direction can even lead to a 40% increase false positives of negatives respectively, which kind defeats the purpose of your AI doesn't it.
It is also a significant problem, though not the main one that non-consensual material might be part of the training datasets. Because NSWF material contains sensitive, it is necessary that such images which inspires AI should be ethically sourced with consent to its use. But in many cases, the data is collected without frank consent from parties involved --an arrow to heart of ethics. This first became really apparent in the 2019 scandal about AI for deepfake video detection, where copyright/otherwise unauthorized content frequently showed up as part of training data.
Further concerns are being raised around the legality of using such datasets Being that this is illegal in many places there for, legal action can be taken against the platform running the AI but also against developers. This risk only further emphasizes the need for strict data governance. As organizations like the Electronic Frontier Foundation (EFF) have argued, AI training data should probably be regulated in some form, with an emphasis on transparency and accountability about how this data is captured and employed.
Then you have the issue of data protection and ensuring no personal or sensitive information is exploited by fraudsters. AI models for NSFW often require a large amount of personal data as training set which might be a potential privacy risk. From a legal and reputational perspective, these breaches or improper handling of sensitive content can be extremely serious. One vulnerability of particular concern is that related to handling sensitive data such as was seen in the 2020 Twitter data breach where millions of users had their phone numbers and other personal information compromised — bringing into light issues surrounding AI applications.
AI ethicist Kate Crawford and others have long noted that “the seam of machine learning is based on data, where if the base of it suffers from a lack of diversity or does not represent reality faithfully enough, then the entire AI output will also reflect this.” This is especially important for not-safe-for-work (NSFW) AI, as the ethical and legal issues relating to training data play an integral part in shaping how trustworthy or fair a model can be.
In the end, NSFW AI provides important nuance to how we approach explicit content moderation while highlighting a greater issue around training data. Because at the end of day, responsible and effective development of NSFW AI systems will require strong guarantees on ethical data sourcing, bias correction as well as privacy protection.
This is Just the Peaks BakeBakeInfo — NSFW AI advertisingCHECK out for these topics and further reading on nsfw ai at this resource.