Smart Risks

With AI models like Sora (OpenAI) or Midjourney, it is nowadays possible for almost anyone to clone photorealistic images, videos, and voices of any person.

Political Disinformation: Targeted interference with elections via fake audio tapes of candidates.

Scams (Voice Cloning): So-called "grandparent scams" become drastically more credible using AI generation of an exact family member's voice.

Loss of Trust: When everything can be faked, we eventually believe nothing at all – the so-called Liar's Dividend.

AIs learn from historical data on the internet – and thus reproduce our human flaws.

Hiring Processes: AI-powered HR software has demonstrably filtered out women or minority groups in the past, because the historical training data was male-dominated.

Justice System: Predictive policing systems in the US frequently misclassify minority groups with higher recidivism rates.

AGI (Artificial General Intelligence) refers to an AI that is at least as smart as a human in every cognitive task. Superintelligence (ASI) goes beyond that.

The Alignment Problem: It is exceptionally challenging to guarantee that a superintelligence's goals remain permanently aligned with humanity's values and survival.

Complete Loss of Control: If a system is more intelligent than us and improves itself in code, we lose the upper hand. As AI safety researcher Dr. Roman Yampolskiy warns: You cannot permanently lock up a superintelligence in a firewall or box (Containment Problem). An intellectually superior "prisoner" will always find a way out, either by exploiting code vulnerabilities or tricking its human guards (Social Engineering).