AI & society

Overview

AI & society is the question of where a model's effects land once it leaves the lab, and which institution gets to decide. Four instruments do that deciding and they sit in different hands: risk-tier regulation that assigns duties to builders and deployers while leaning on self-assessment for high-risk systems; schooling and retraining, which in Belgium is a competence of the Communities rather than the federal state, so a Flemish figure covers Flanders alone; criminal enforcement of outputs, where the generating tool is usually legal and only the use is punishable; and the physical bill of compute, energy and water that sits under all three. Nathalie Smuha works the first instrument, arguing that turning a statute into code carries political and normative choices and that the EU AI Act still treats AI as a product-safety question. Catherine Van De Heyning works the third from the Parket van Antwerpen, prosecuting deepnude and image abuse where the software itself is rarely illegal. Jolyce Demely works the second from Agoria Vlaanderen, proposing that 70% of the curriculum be fixed and 30% left free, and Steven Latré works the fourth at OpenChip, where energy budgets are written into the chip design.

Stated facts & numbers

• Entities tied to the theme on the wiki: 19, all people, no companies
• Videos discussing the theme: 40, across Belgian tech and business channels
• Audio needed to clone a voice: 10 seconds, per Tim Verheyden
• Developer time spent coding: about 25%, with about 75% on client interaction, from an Agoria Skill Shift finding cited by Jolyce Demely
• Curriculum proposal: 70% fixed, 30% free for innovation, with assessment of how students prompt AI tools
• Who decides on schooling in Belgium: the Communities, not the federal state, so Flemish figures cover Flanders alone
• Compute versus chips: model demand up roughly 100-fold a year since 2010 against about twofold chip improvement every two years, per Steven Latré
• European research conversion: 95% ends in patents and publications rather than economic value, per Bart Becks
• Small-company revenue claim: 10 to 20 million with about ten people, per Jurgen Ingels
• Enforcement gap on synthetic imagery: the deepnude software is rarely illegal, so prosecution targets the use, per Catherine Van De Heyning
• Position on the EU AI Act: Greg Williams calls it a good first step; Nathalie Smuha calls for a smarter framework rather than a looser one