Call us now:
Say goodbye to RAG and say hello to knowledge-based reasoning. A transformation that we have been predicting for the past year. You can call it knowledge-based reasoning, or ontological-based reasoning; it does not matter. What does matter is that as AI develops and consumes data, the knowledge pool subsequently condenses; there are only so many ways you need or want AI to write a presentation or evaluate data sets to refocus your sales and marketing figures.
Ultimately, AI runs up against a problem, which is one of reasoning against vectorized data. Some problems are too complex for AI to handle and produce an accurate output. The data sets are too large, and there’s a lack of input and output correlation, where probability adds little value. In simple terms, AI can take inputs, run them through reasoning, and generate outputs. But unless it understands the function that translates those inputs into an accurate, repeatable output, the outputs will not be correct. Call it hallucinations, errors, or AI making up responses.
There are instances where the output must be correct, such as in audit or cybersecurity. Where accurate and repeatable knowledge is not available across the internet, to train AI against. That is why we are building a knowledge-centred, ontological, domain-specific (whatever takes your fancy) AI solution to evaluate cybersecurity compliance against EU NIS 2 and ISO 27001 for internal audit.
The future of agentic AI won’t be decided by who has the longest context window, Walter said. It will be decided by who can operationalize trusted knowledge at scale without blowing up cost or governance.
We ground our AI on the Knowledge, skills, and experience of industry professionals to shrink the context window. We ground our reasoning (the function) against knowledge and information from industry professionals and other sources. We apply prompts developed from knowledge and experience. We create reasoning that is grounded. Creating a knowledge engine for the evaluation of cybersecurity control design and effectiveness is required for regulatory compliance. As a precursor to the assurance of other audit-related requirements.
Faisal Khan Liz Sandwith CFIIA Ryan Buckner Michael G. Jacobides Georgios Alevras
Aegis Labs
#ai #artificalintelligence #knowledgebasedreasoning #rag
https://lnkd.in/eDQ2s2q9
View Original Post
https://www.linkedin.com/feed/update/urn%3Ali%3Ashare%3A7457358169829208064