elizaonsteroids logo
Deutsch

Artificial Intelligence and Consumer Deception

May 4, 2025 · Alexander Renz | Translations:

The term “AI” creates an image for consumers of thinking, understanding, even consciousness.
LLMs like GPT meet none of these criteria – yet they are still marketed as “intelligent.”

🔍 Core Problems:

  • Semantic deception: The term “intelligence” suggests human cognition, while LLMs merely analyze large amounts of text statistically. They simulate language without understanding meanings or pursuing goals. The model has no real-world knowledge but instead makes predictions based on past training data.

  • Lack of transparency: Users typically receive no systematic disclosure of the limitations of generative systems, such as:

    • No true understanding of text (no semantic grasp)
    • Frequent hallucinations (fabricated facts with no basis in reality)
    • Lack of traceability (black-box behavior)
    • No consciousness or intentionality – the AI does not “know” what it is saying

  • Marketing vs. reality: Many companies use anthropomorphic terms (“assisting,” “thinking,” “learns from you”) and visual or linguistic tools that exaggerate system capabilities. This creates a false expectation of autonomy and reliability in consumers.

⚖️ Legal Assessment:

  • Section 5 UWG – Misleading about essential characteristics of a product:
    Using the term “intelligence” or suggesting true autonomy can constitute misleading commercial practice if essential information about function, limitations, or risks is withheld.

  • Violation of Section 3 UWG (ban on unfair business practices):
    Particularly problematic is concealing system-related malfunctions such as hallucinations – especially in sensitive domains like education, healthcare, justice, or consulting.

  • EU AI Act (Regulation on Artificial Intelligence – final version adopted in 2024, effective mid/late 2025):

    • Obligation for transparency in generative models (Art. 52 AI Act):
      • Disclosure that content was generated by AI
      • Documentation of technical limitations and possible risks
      • Ban on manipulative interface design (dark patterns)
    • AI applications interacting with humans must be clearly identifiable as such (Art. 52 para. 1)

  • Legal comparison USA / EU:

    • The US lacks unified AI legislation, but since 2023 the FTC has warned explicitly against “AI Washing” – marketing products as AI-based when this is not, or misleadingly, the case.
    • The EU, in contrast, is introducing a precedent-setting regulatory framework with the AI Act.

✅ Proposed Solutions

  • Mandatory notices for generated text:
    e.g. “This text was generated automatically. The system does not understand content.”

  • Ban on anthropomorphic branding:
    No visualization as “smart assistants” with eyes, voice, or emotional speech if no cognitive capabilities are present.

  • Strengthen consumer education:
    Awareness campaigns about the differences between:

    • Statistics and meaning
    • Machine learning and human thinking
    • Output illusion and actual competence

  • Standardized risk disclaimers:
    Especially in sensitive areas like medicine, legal advice, or child welfare, a risk disclaimer should be mandatory.

  • Clarify liability issues:
    Who is liable for damage caused by AI output? Providers must be held accountable if systems are misrepresented or risks downplayed intentionally.