AI kan bogstaveligt talt ikke eksistere uden Wikipedia – Wikimedia Foundation har netop bekræftet dette. Hvad er implikationerne?

I work in ML training. Let me explain why this matters technically.

Why Wikipedia is irreplaceable:

1. Quality control at scale - Billions of human-hours of editing
2. Citation requirements - Claims need reliable sources
3. Neutral point of view - No promotional bias
4. Structured data - Infoboxes, categories, consistent formatting
5. Multilingual - 300+ languages, native speakers

What happens without Wikipedia:

We tested models trained excluding Wikipedia:

• 23% degradation in factual accuracy
• Increased hallucination rates
• Worse performance on diverse topics
• More cultural/linguistic bias

The economic reality:

Building something like Wikipedia from scratch would cost billions. AI companies got it for free. Now the infrastructure is straining.

This is a classic tragedy of the commons playing out in real-time.

I study model collapse. Let me explain why Wikipedia is essential for AI’s future.

Model collapse in simple terms:

When AI trains on AI-generated content:

• Errors compound
• Biases amplify
• Quality degrades
• Eventually: garbage in, garbage out

The Nature study (2024):

Showed recursive AI training causes “irreversible forgetting” of original content. Each generation of AI gets worse.

Why Wikipedia prevents this:

Wikipedia is STRICTLY human-curated:

• No AI-generated content allowed
• Active enforcement
• Continuous human verification

The strategic importance:

As AI-generated content floods the internet, Wikipedia becomes MORE valuable, not less. It’s the anchor of truth in a sea of synthetic content.

Brands that get properly represented on Wikipedia will have advantages as AI increasingly relies on verifiable sources.

Running an AI company. Here’s the business reality:

The uncomfortable truth:

We absolutely depend on Wikipedia. Our model quality is directly tied to Wikipedia quality. We should pay for it.

What we’re doing:

1. Using Wikimedia Enterprise (paid access)
2. Donating to Wikimedia Foundation
3. Proper attribution in our responses
4. Sustainable crawling practices

Why more companies should do this:

• Sustainable Wikipedia = sustainable AI
• It’s the right thing to do
• Licensing requirements are coming anyway
• Early compliance = competitive advantage

The cost:

Less than 0.1% of our compute costs. Trivial.

The risk of not paying:

If Wikipedia restricts access or degrades in quality, our model quality suffers. It’s risk management, not charity.

Let’s talk practical implications for brands:

The training data hierarchy:

Strategic implications:

1. Wikipedia matters most, but you control least

   • Focus on generating coverage that Wikipedia can cite
   • Build notability over time

2. Your website matters less for AI

   • But still important for direct traffic
   • Use as source for third-party content

3. News and authoritative sources are key

   • Create newsworthy moments
   • Build relationships with industry publications

The Am I Cited angle:

Monitor how AI synthesizes information about your brand across all sources. The output tells you which inputs are working.

I negotiate data licensing deals. Here’s what’s coming:

The licensing landscape:

• Google already pays Wikimedia (2022 deal)
• Other AI companies in active negotiations
• Pricing models being developed
• Enforcement mechanisms coming

Expected pricing structure:

Per-crawl fees (for training)
+ Per-query fees (for RAG/grounding)
+ Base access fee
= Sustainable Wikipedia funding

What this means for AI products:

Costs will increase. But it’s still cheaper than:

• Building your own knowledge base
• Dealing with degraded model quality
• Legal/reputation risks

What this means for brands:

As AI access to Wikipedia becomes more formal:

• Attribution will improve
• Quality will stay high
• Your Wikipedia presence becomes more valuable
• Monitoring becomes more important

The open source/commons perspective:

The CC-BY-SA license requires:

• Attribution
• Share-alike (derivative works use same license)

AI companies are arguably violating this:

• Training produces derivative works
• Attribution is inconsistent
• Revenue isn’t shared

The philosophical question:

Wikipedia was built for human knowledge sharing. Is training commercial AI what the community intended?

My view:

The license allows commercial use. But the spirit of Wikipedia is open access to knowledge for humans. AI companies should contribute back.

What brands should know:

Your content, if cited by Wikipedia, enters this commons. This can be powerful - but you lose control over how it’s used by AI systems.

Multilingual perspective:

Wikipedia’s 300+ language editions matter:

• AI systems trained on multilingual Wikipedia
• This enables better non-English responses
• Local markets have local Wikipedia coverage

For global brands:

Your Wikipedia presence in multiple languages affects AI responses in those languages.

What we discovered:

Our German Wikipedia page was minimal. ChatGPT’s German responses about our company were vague and sometimes wrong.

The fix:

Generated more German media coverage → German Wikipedia page improved → German ChatGPT responses improved

Key insight:

Each language is a separate AI visibility challenge. Monitor across all relevant markets.

Looking ahead 3-5 years:

Likely developments:

1. Mandatory licensing

   • AI companies will pay for Wikipedia access
   • Standardized pricing models

2. Improved attribution

   • AI responses will cite Wikipedia more explicitly
   • Users will see source links

3. Quality control mechanisms

   • Wikipedia may verify how AI uses their content
   • Accuracy audits

4. New content types

   • Wikipedia may create AI-specific datasets
   • Optimized for training

What this means for AI visibility:

Wikipedia’s importance will INCREASE, not decrease. As AI access becomes formalized:

• Verified content becomes more valuable
• Wikipedia presence becomes premium real estate
• Brands without Wikipedia coverage get left behind

Start building Wikipedia-worthy notability now. It takes years.

Excellent discussion. Here’s my synthesis:

The fundamental reality:

Wikipedia is AI infrastructure. Not optional - required. The Wikimedia Foundation’s statement is literally true: “AI cannot exist without Wikipedia.”

What this means for AI development:

1. AI companies must start paying for access
2. Licensing requirements are coming regardless
3. Wikipedia quality = AI quality (direct relationship)
4. Model collapse prevention requires human curation

What this means for brands:

1. Wikipedia presence is more valuable than ever
2. Building notability is a multi-year investment
3. Each language edition matters separately
4. Monitor how AI uses Wikipedia to represent you

The action items:

For AI companies:

• Join Wikimedia Enterprise
• Donate to Wikimedia Foundation
• Implement sustainable crawling
• Proper attribution in responses

For brands:

• Develop Wikipedia-worthy notability
• Generate citable coverage
• Monitor AI visibility with tools like Am I Cited
• Build presence in multiple language editions

The Wikipedia-AI relationship will only become more important. Plan accordingly.