AITLETIA

AITLETIA is an intuitive web application that maps current scientific knowledge, allowing innovators to navigate what is known while showing the maturity of each topic and its relevance to society.

Comment

Inspiration

Most digital tools help us find answers. AITLETIA explores an earlier and harder problem:

Can artificial intelligence help us identify valuable questions that have not yet been clearly formulated?

The project is inspired by the Theory of Nescience, a mathematical framework I have developed for studying the limits of knowledge using ideas from information theory and algorithmic complexity.

AITLETIA translates those ideas into a practical tool for exploring scientific and technological domains. Its goal is not to claim that AI can discover objective unknowns, but to help researchers, engineers, students, and innovators identify promising areas that may deserve further investigation.

What it does

AITLETIA creates an interactive map of a research knowledge.

It organizes topics according to metrics such as maturity, relevance, structural position, and relationships with neighboring topics. Users can explore known subjects, inspect underdeveloped regions, and review candidate research questions associated with possible knowledge gaps.

Conceptually, AITLETIA looks for opportunities using a combination of uncertainty, relevance, and connectivity.

The current prototype maps thousands of topics and includes tens of thousands of candidate research ideas, from multiple areas of knowledge.

How we built it

AITLETIA combines structured knowledge, quantitative analysis, graph visualization, and generative AI.

A Python pipeline collects and organizes information from public knowledge sources. A FastAPI backend exposes categories, topics, and candidate unknowns through an API.

The frontend uses HTML, CSS, JavaScript, Web Components, and D3.js to display an interactive radial map.

Nescience-inspired metrics estimate topic maturity, while additional signals provide context about relevance and relationships. Generative AI then converts detected gaps into understandable candidate research questions.

The workflow is:

  1. Structure the research domain.
  2. Identify potentially interesting gaps.
  3. Analyze their context.
  4. Generate candidate questions.
  5. Leave novelty and value assessment to the user.

During OpenAI Build Week, we are using GPT-5.6 and Codex to search for the uknowns, strengthen the backend, refine the interface, and accelerate experimentation.

Challenges we ran into

The main challenge is that an unknown unknown cannot be directly observed. AITLETIA must rely on imperfect indicators such as low maturity, weak connectivity, or missing relationships.

Data quality is another limitation. Public knowledge sources are incomplete and unevenly structured, so an apparent gap may reflect poor documentation rather than a genuine research opportunity.

Novelty is also difficult to verify. A generated question may already have an existing literature under different terminology. For this reason, AITLETIA presents questions as candidates for investigation, not as confirmed discoveries.

Visualization was another challenge. Research domains are dense and multidimensional, while the interface must remain clear and usable.

We also found that unconstrained idea generation often produces generic or implausible results. Better questions require structured context and supporting evidence.

Accomplishments that we're proud of

We are proud to have turned an abstract theoretical idea into a working application.

AITLETIA can already:

  • map thousands of scientific topics;
  • represent known topics and candidate unknowns;
  • rank possible gaps using several signals;
  • generate tens of thousands of candidate research ideas;
  • and let users explore them through an interactive interface.

We are especially proud that the project keeps human judgment at the center. AITLETIA does not claim that every generated question is novel or valuable. Instead, it helps users explore possibilities they might otherwise overlook.

What we learned

We learned that detecting a gap and explaining it are different problems. A score may identify an unusual topic, but users also need to understand why it was selected.

We also learned that unknownness is not binary. It is better represented through several complementary signals.

Generative AI performs much better when it works inside a constrained, evidence-based process rather than being asked to invent ideas without context.

Most importantly, we learned that AI can support discovery without replacing expertise. Domain specialists are still needed to evaluate novelty, feasibility, ethics, and scientific value.

What's next for AITLETIA

The next step is to evolve AITLETIA from an exploratory prototype into a more rigorous AI-assisted research companion.

Our priorities are to identify thousands of uknowns and generate millions of questions, strengthen literature and novelty validation, expand into additional domains, and provide clearer explanations of why each candidate was proposed.

We also want users to be able to refine a question, review evidence, identify assumptions, compare alternatives, and create a preliminary investigation plan.

The long-term goal is not to automate science, but to give people a better instrument for navigating uncertainty and discovering which questions may be worth asking next.

Built With

Share this project:

Updates