Building Ethical AI Language Infrastructure: A PBC's Approach
Language is the interface between AI and humanity. Every word an AI agent speaks carries implicit values. The question is whether those values are intentional or accidental.
At LOTD PBC, they are intentional. We are a public benefit corporation building language infrastructure for AI systems, and the corporate structure is the reason the product works the way it does.
Why Corporate Structure Matters for Language
A standard C-corporation optimizing language data for AI would maximize coverage, minimize cost per entry, and ship fast. The rational path is web scraping. The unit economics are excellent. The language quality is not.
A public benefit corporation has a legal obligation to consider impact on stakeholders beyond shareholders: the learners who consume the content, the cultures whose languages are represented, and the AI systems shaped by the data they receive.
This is not a marketing distinction. It is a structural one. When quality conflicts with speed, the PBC charter requires that stakeholder impact be weighed alongside financial return.
Four Pillars of Ethical Language Infrastructure
1. Gender Equity
Word Orb tracks gender distribution at the corpus level. Example sentences are authored to ensure balanced representation across masculine, feminine, and neutral forms. The gender metadata is exposed in the API so downstream applications can select contextually appropriate forms.
2. Cultural Sensitivity
We document regional variations rather than resolving them by picking one context and ignoring others. The API provides cultural notes, regional usage guidance, and register indicators so downstream applications can make informed choices.
3. Age-Appropriate Content
Our content pipeline applies age-appropriateness evaluation at the entry level, not as a post-hoc filter. Kelly, our AI teaching companion, is the benchmark: every entry must work for a five-year-old or a fifty-year-old without modification.
4. Sovereign Infrastructure
Word Orb runs on Cloudflare's edge network, serving responses from over 300 cities across 100+ countries. There is no centralized origin server. When a developer in Nairobi requests Swahili vocabulary, the response comes from an East African edge node.
Open API, Proprietary Content
The API is open: any developer can integrate with REST calls, an npm SDK, or an MCP server. The content is proprietary because maintaining quality requires ongoing editorial investment. Open-sourcing the dataset would create forks with no quality control.
Defensive Patent Strategy
Our content methodology is the subject of a pending U.S. patent (Application No. 18/088,519). This is a defensive filing to prevent large incumbents from replicating our methodology and undercutting the market. The patent protects the process. The PBC structure ensures the process continues to serve its intended beneficiaries.
Kelly: Ethical AI Teaching in Practice
Kelly is the embodiment of these principles. She is an AI teaching companion who delivers daily language lessons across 47 languages and is the standard against which every piece of content is measured. Kelly does not default to masculine grammar, does not use culturally insensitive examples, and does not teach vocabulary with age-inappropriate sentences.
The API that developers integrate with serves the same content that Kelly teaches, because the quality standard is the same.
If you are evaluating language infrastructure investments or EdTech partnerships, we should talk. Learn more at lotdpbc.com.