Meta, the company behind Facebook and Instagram, tests its AI chatbots, which are computer programs that speak like humans, with data. This data is considered potentially not allowed. Meta wants to improve the safety of its programs with this. However, this approach leads to much criticism. It is about where the boundaries for research lie.
Meta's approach is important. It shows that large tech companies are gaining more influence over how sensitive data is handled. When large companies use data that may not be allowed, even for good purposes, the rules become blurred. This affects everyone who uses AI services. Who guarantees that traces of this questionable data will not end up in finished products? It's about trust and the security of our digital world.
The tech company Meta uses content that is legally and ethically questionable for internal safety checks of its AI programs. According to Golem.de, Meta itself calls this an 'AI safety benchmark.' Critics see this as a clear crossing of boundaries. The data used could potentially contain unauthorized material. Meta wants the AI programs to become more resilient against harmful content. They should also not generate such content themselves.
For private individuals, this means: If large tech companies like Meta process such data, a risk increases. Sensitive or harmful content could enter AI systems. You could unknowingly speak with an AI that was trained with problematic sources. This can affect the quality of the answers. In the worst case, it leads to unwanted or false information. Your trust in AI helpers could suffer.
Companies face a problem. Anyone using AI services from providers like Meta must ask: How openly and data-protection-compliantly were the programs trained? The use of potentially illegal material in training is a major risk for compliance with rules. This can lead to liability issues. This happens if the AI later creates unwanted content. Companies must carefully check which data flows into which programs. This protects their reputation and legal security. It's about who owns the data and who controls its origin.
The opportunity lies in KIs learning through such tests. They can better recognize and filter out extremely harmful content. If the programs become more resilient against attacks and manipulations, future AI systems could be safer. This could, for example, more effectively combat deepfakes, which are fake videos, or the spread of hate speech. The AI learns to recognize and block these patterns.
