Deterministic AI in a Gen AI world

Image left to right: Korina Konopka, Kevin Stephens and Richard Napier

Richard (who had flown in to Sydney from London specifically for these training sessions) took the Kyte team through advanced use cases for OIA and I was lucky to have some extended conversations with him outside of the training as well. Here are my key take aways from those conversations.

Is there a future for OIA amongst the generative AI explosion?

Short answer, absolutely - in fact there may be more of a need for a deterministic solution as more and more business focus on generative AI.

While generative AI is great at making up content, there is no guarantee that it won’t make up responses that are false. This is known as hallucination and while this may be perfectly acceptable for some use cases, there are times when potentially allowing incorrect answers to go out to your clients could leave you open to reputational damage or even serious litigation (such as health care or public service). While newer AI models can show a reasoning chain, each step in the chain is still susceptible to hallucination and entering the same inputs does not always yield the same result.

This is where deterministic AI is required, and this is exactly what Oracle Intelligent Advisor is built for. Running inputs through the policy model will always produce an expected result. Every permutation of response can be catered for and also be built out into pre-defined test cases saving time and effort during development and test cycles.

However, the most important factor here is that the results show a complete decision chain about why the model reached the determination it did. Inputs can be traced through each step of the model and the full decision chain handed to your customer at the end of the process. If you don’t get the result you expected, it is easily traceable to the rule(s) that influenced the decision.

What are the best use cases for OIA?

One of my favorite things about OIA is its versatility. It can do simple things like be used as a data collection front end such as a form on a website or an audit tool however it really shines when the rules are built to handle complex domain specific information. Picture a legislation or regulation heavy industry where SME’s hold the body of knowledge. OIA can be built to effectively become a knowledge expert, containing rules that accurately represent that expert domain knowledge, defined in one place that can be used across the whole of your business by multiple systems. OIA is system agnostic and can just as easily be hooked up to your Salesforce CRM as it can to your Oracle Digital Assistant.

An often-missed use case for OIA involves deployment into multi language environments (such as Government websites catering for non-native language speakers). Translations are applied to the model, meaning that rules are still written in your preferred language (such as English, French or Spanish) and then translated at run time by the UI component of the solution. This means you have a single centralised rule base that supports multiple languages, instead of multiple single language models that all need to be maintained separately.

Lastly, OIA is perfect for any determinations that need to use temporal reasoning. Any time you need to understand a value or situation at any given point of time (such as the balance of an account on date x or if the balance drops dropped below a value for a period of x months) OIA can use simple rules formula to work it out. No complex programming required, just rules written in plain language that visibility aligns to the policy or regulation being determined.

These are just some of the scenarios we covered over the week but the ones that really stood out. Again, I’d like to thank Richard for his time and enthusiasm discussing all things OIA!

For anyone interested in a more detailed conversation about getting the most out of their OIA implementation or what OIA could do for their business, please don’t hesitate to reach out to the Kyte Consulting team. We have broad experience with OIA across Federal and State Government, Education and Technology sectors.

About Richard Napier

After 8 years in case management and ERP software roles, Richard Napier joined Siebel Systems in 1999 and took up the role of managing the nascent Siebel University in Southern Europe. He subsequently was Director of Business Development and Education for InFact Group (now part of Business & Decisions) for 8 years. He now runs Intelligent Advisor IT Consulting OÜ. Owner of intelligent-advisor.com, he also is Co-Founder of the Siebel Hub.

Source: Oracle Intelligent Advisor - Intelligent Advisor IT Consulting OÜ / Richard Napier

Author: Kevin Stephens