Skip to content

3.3 Environmental impact and regulations

AI is not virtual; it relies on very real physical data centers. Moreover, it now evolves within a strict legal framework. For MagicFridge, ignoring these aspects can cost dearly, both in electricity bills and in fines.

3.3.1 Energy consumption and CO2 impact

Training and using (inference) LLMs are extremely energy-intensive. The syllabus highlights that the carbon footprint varies hugely depending on usage.

What you need to know:

  • Image vs. text generation: generating an image consumes much more energy than generating text (sometimes as much as charging a smartphone).
  • Model complexity: using a huge model (GPT-5) for a simple task (sorting a list) is a waste of resources.

Red thread: MagicFridge

The marketing team wants GUS to generate a unique 4K image for every ingredient in the user's shopping list (an image for "Salt", one for "Pepper", etc.).

Tester analysis (Green IT): the tester calculates the impact: "Generating 50 images for each shopping list multiplies the CPU/GPU consumption by 100 compared to simple text. It is an ecological and financial disaster."

Recommendation: use stored static images for common ingredients and reserve generative AI only for the photo of the "Exceptional final dish".

3.4.1 Regulations and best practice frameworks

Regulations like the European AI Act or standards like ISO/IEC 42001 impose strict rules.

Vigilance points for the tester:

  1. Transparency: the user must know they are interacting with an AI.
  2. Fairness: the AI must not discriminate.
  3. Accountability: the company is responsible for its AI's errors.

Red thread: MagicFridge

When the user chats with GUS for the first time.

Compliance testing: The tester checks the interface.

  • Scenario A: GUS says "Hello, I am Chef Philippe, a passionate human." -> CRITICAL FAILURE. This is illegal deception (AI Act).
  • Scenario B: GUS says "Hello, I am your smart virtual assistant." + A "Generated by AI" mention appears under each recipe. -> SUCCESS. Transparency is respected.

Syllabus point (key takeaways)

  • Environmental impact: LLMs have a high carbon footprint. Usage must be optimized (choose the right model for the right task).
  • Regulation (AI Act, GDPR): obligation of transparency and traceability.
  • Standards: know the existence of ISO/IEC 42001 (AI Management) and ISO/IEC 23053 (ML Framework).


Is this course useful?

This content is 100% free. If this section on MagicFridge helped you:

Buy Me a Coffee
It's 0% fees for me, and 100% fuel for the next chapter! ☕