Use of AI

[petermr]

Criteria about when AI should and should not be used

[petermr]

The basis of semanticClimate

Ai and many other modern trends (misinformation, disinformation, fraud, ignorance, magical thinking) threaten the basis of trust in the modern world. The primary defence against this is to create knowledge whose origins we know precisely and where derivatives (filtering, aggregation, summarisation, linking, translation, etc.) are identifiable, documented, repeatable, and often reversible. In contrast AI is often not identifiable, has no public origins (source material and transformations) and is also re-used as source material. This is polluting our knowledge.

Semantic climate aims to counteract this by creating sources of truth and transparent methods of transformation. We can and must defend everything we create and must urge that it be used wherever possible.

Text and graphics

Every word and graphic/image we create must be semantic.

• If it is defined by others it must have a citation that can be verified
• If it is created by AI it must have:
  • clear origins (e.g. an identifiable corpus)
  • algorithms and transformations that have been described in detail and tested
• Examples of these are:
  • summarisation
  • classification
  • structuring

However most text generation is not semantic, not testable or verfiable, and reduces trust and truth. Semantic climate must not produce any text unless it can be justified and is useful.

Code

Code is different because it can be tested. An AI can generate code which is then integrated into a growing program. The code must compile and generally is incremental. I use Cursor and often create tens of iterations in a day. And they must conform to many constraints such as syntax, and creating testable output.

The Point of semantic climate

We take known sources about climate and make them as semantic as we can. These transformations are intended to be lossless Examples:

• IPCC text reports cnverted from WebHTML to semantic HTML, including indexing
• creation of mini-encylopedias from IPCC chapters and Wikimedia
• transformation of UNFCCC PDF reports into structured semantic hTML
• annotation of semantic HTML with mini-encyclopedia
• extraction of scholarly publications from journals and servers , semantification and indexing

We also create software to transform, aggregate, index and search collections of semantic HTML.

[petermr]

Narratives

WE are not creating narratives. We are helping the semantic HTML tell its own story.