720 Australian and Brazilian Children Better Protected from AI Misuse

LAION removed children’s photos from its dataset after investigations revealed their unauthorized inclusion, which enabled the creation of harmful deepfakes. This action confirms that personal data can be successfully removed from AI training sets, acknowledging the severe risks when such information is exploited to create unanticipatable harms. However, significant concerns persist because only a minuscule portion of the dataset was reviewed, leaving many unidentified children’s images within the system. Furthermore, previously trained AI models retain the removed data, meaning the damage is not fully erased. The open nature of this dataset allowed for this scrutiny, highlighting how private AI datasets remain unexamined and opaque compared to open resources. This situation underscores the urgent need for legal frameworks to protect children’s privacy in data usage. Governments are currently considering specific legislation to safeguard minors' online rights, offering a critical opportunity to establish robust protections. These laws are essential to prevent the misuse of personal data in open data initiatives and to ensure that transparency in AI development does not come at the cost of individual safety.

Source: hrw.org
Published on 2024-09-04

“Copyright traps” could tell writers if an AI has scraped their work »

Open Data

Related news