- AI Natives
- Posts
- AI Natives #23 - The EU Acts
AI Natives #23 - The EU Acts
The EU AI Act comes to life, while AI companies are trying to simulate worlds
Kornel Grunwald
December 15, 2023
Hey there, #TheAINatives! 🤖
Happy to have you here joining us with the twenty-third issue of The AI Natives.
EU AI Act changes a lot!
So, we have EU AI Act. Here is a nice summary of its’ implications:
Certain applications are strictly prohibited:
Systems that categorize based on sensitive traits (such as political, religious beliefs, sexual orientation, race, etc.)
Indiscriminate collection of facial images from the internet or CCTV for facial recognition databases
Use of emotion recognition in workplaces and educational settings
Social scoring based on personal traits or social behavior
Systems designed to override human autonomy
Technologies targeting individuals' vulnerabilities (age, disability, socioeconomic status, etc.)
Applications posing substantial potential risks to health, safety, rights, the environment, democracy, or legal governance are considered high-risk:
Mandatory registration in an EU database
Compulsory impact assessment on fundamental rights (applicable in sectors like Insurance and Banking)
Requirements to mitigate biases, uphold non-discrimination, and respect rights
Prohibition on influencing election outcomes and voter behavior
Right for citizens to file complaints and seek explanations for decisions made by high-risk AI systems affecting their rights
Guidelines for General AI Systems. This category includes foundational models:
Must comply with transparency standards
Required to maintain technical documentation
Obligation to adhere to EU copyright laws
Detailed descriptions of training content
Certain systems must undergo model evaluations, address systemic risks, perform adversarial testing, report serious incidents to the Commission, ensure cybersecurity, and disclose their energy efficiency standards
___
A note on 🚩, 💚or 🟠 being next to the news title in the sections below. This is expression of my take, whether I see the news as positive or dangerous. We will delve deeper into dangers of AI going onward, as there is not enough being discussed in that space.
Here we go with the rest of the big news⏬
CONVO OVER COFFEE ☕
Big topics to discuss with friends and colleagues
The AI Act is on🟠
Following extensive negotiations and final discussions focusing on facial recognition and fundamental AI models, the European Union has provisionally agreed on the AI Act. This legislation, potentially serving as a model for worldwide AI regulation, sorts AI systems by their level of risk and introduces a dual-tier framework for general AI systems, as well as measures for systems deemed to pose an "unacceptable" risk. This Act now confronts the task of keeping pace with the fast-changing landscape of AI technology.
My take: as much as I am happy to see AI Act to ban a lot of very dangerous practices regarding social and behavioral fabric, I still believe it is a bit premature. I do not agree with some of the votes from the public claiming that now Europe have more regulation than AI companies. AI Act limits all companies operating in EU, so shields users from tools available on the continent regardless of their origin. Yet, we are to see what specifically this regulation will produce as an implementation standard. I hope we won’t see anymore ‘cookies’ pop-ups-like effects.
Runway wants to build Worlds 🟠
Runway recently announced a significant shift in AI research towards General World Models (GWM). These models aim to create an AI that comprehends and simulates the real world and its dynamics. Unlike existing models, which are limited to specific settings like video games or driving, GWMs will encompass a broader range of real-life scenarios. This ambitious project involves overcoming challenges like constructing comprehensive environment maps and modeling complex human behaviors. Runway ML is spearheading this initiative, inviting interested parties to join their team.
My Take: The evolution from Linear Language Models to General World Models will mark a pivotal moment in AI development. This trend is gaining traction across major tech firms, including OpenAI and Meta, due to its potential in creating more versatile and realistic simulations. SpatialTech is the keyword for all of you, and Multimodal AI, which integrates diverse data types, is the next frontier for businesses, particularly in complex areas like video and 3D simulations.
Other news:
Reply