Firm Removes AI Training Dataset Over Child Abuse Content

Discovery Of Child Abuse Content In AI Training Dataset

A recent study discovered that there was child sexual abuse content in LAION-5B. This large artificial intelligence data set was used to train a number of well-known text-to-image generators.

However, the creator has deleted the said model for precautionary reasons. The data collection was used to train models for AI picture generators such as Imagen, Stable Diffusion, and others.

Large-scale Artificial Intelligence Open Network, or LAION, is a nonprofit organization based in Germany that creates open-source artificial intelligence models and provides the training data sets for a number of well-known text-to-image models.

Implications Of AI Model Training

Researchers at the Stanford Internet Observatory’s Cyber Policy Center found over 3,200 suspected cases of child sexual abuse material (CSAM) in the LAION-5B data set, according to a report released earlier this week. David Thiel, Big Data Architect and Chief Technologist at the Stanford Cyber Policy Center, confirmed that third parties had verified much of the content as CSAM.

Although the existence of CSAM doesn’t guarantee that it will “drastically” affect the results of models trained on the dataset, Thiel pointed out that the volume of CSAM could still have some impact. However, he added that it might not dramatically alter the model’s output except for its ability to teach concepts related to sexual activity in children.

According to LAION, 5.85 billion image-text pairs are included in the LAION-5B dataset, which was made available in March 2022. LAION withdrew the LAION-5B and LAION-400M datasets as a precautionary measure for safety reasons.

AI-Born Creations Denied Patent Rights By UK’s Top Court

Dr. Stephen Thaler, a computer scientist, tried to file two patent applications under the name of an AI model he developed (DABUS). However, the UK Supreme Court rejected his applications. DABUS is a food container or a flashing light beacon.

The ruling upheld earlier ones by the UK High Court and Court of Appeal. The court concluded that a patent must have the support of a “natural person,” and they emphasized that DABUS lacks the properties of being a person in any form.

Recall that the UK Intellectual Property Office rejected two patents that Thaler filed in October 2018 and August 2019. However, the computer scientist filed an appeal to reverse the court’s decision.

The persistent issue was that Thaler never claimed to be the creator. Instead, he asserted that he was merely responsible for the invention of DABUS, and this tool creates a food container and a flashing light on its own.

However, Thaler’s claim was rejected by lower UK courts based on the Patent Act of 1977. The Court of Appeals ruled in September 2021 that only a person and not a machine can possess patent rights.

The UK Supreme Court said that the present ruling does not establish a precedent for similar cases in the future, regardless of the dismissals. The UK Supreme Court clarified that its decision does not introduce an extra condition for patent eligibility. Additionally, it does not establish new grounds for rejecting patent applications.

According to multiple reports, Thaler also made attempts to get patents for DABUS in Australia, South Africa, Europe, New Zealand, and the United States. Regulators in these countries rejected him, except South Africa, which authorized DABUS in July 2021.