AI How It Works

OmniIndex AI Technology Q&A with CEO Simon Bain

OmniIndex enables the same analytics available on plaintext data to be performed on encrypted data. Data is stored in the user’s own blockchain and is protected from unwanted third-party eyes and Ransomware attacks. This means customers can add their most sensitive and confidential data to collaboration, productivity, and analytics tools with no risk of exposure.

Hidden within the depths of this Data Platform, is the OmniIndex AI engine.

Q1: Thanks for doing this. Let’s begin with what inspired you to develop OmniIndex’s AI technology?

I believe that the next and best UI will be voice activated and chat based. As such, I needed a method to allow The OmniIndex Platform to recognize and interact with the user’s input. This offers a fast and very direct way for customers to get answers from their data. 

Q2: Interesting! Can you give a brief explanation of how OmniIndex’s AI works and what makes it special?

OmniIndex’s AI engine is very specific in its tasks (Narrow). Its job is to determine what a user is requesting, then based on a range of dictionaries, lexicons and thesauruses (referred to as ‘ontologies’), work out an optimal response. These responses are then put through a separate probability matrix before being sent back to the user. In doing this narrow task, the AI has to process and action various aspects of the input. For example:

  1. Determine what a user is requesting. 
  • Is it a question
  • A statement
  • A command

2. Create an actionable request from that

  • Get answers
  • Respond with like (Conversation)
  • Run the command

3. If ‘get answers’ or ‘respond with like’

  • Process responses with a probability engine
  • Send responses back

4. If ‘run the command’

  • Check the command is the correct one, and send response back

Our AI engine is special because it is able to work with data using an ontology. (Dictionaries, Thesauruses & Lexicons.) These can be added to match the specific industry requirements of a customer, for example Health or Finance, without learning requirements for these new sources with it ready to answer questions about the user’s data immediately. (Regardless of whether the data is fully encrypted or not)

Q3: Great! Can you describe a real-world use case for this technology and how it has been applied in that scenario?

Sure. One of our customers, Health innovation Hub (HIH), regularly uses our AI. 

HIH holds sequenced data about their clients and our AI is able to answer specific questions with regards to that data. Despite the data being encrypted at all times using our patented technology to ensure there is no risk of exposure. The following is an example query that might be asked by HIH, with the AI’s process in answering the query broken down into steps.

“How many clients prone to gastric anomalies have signed up within the last 6 months?” 

The AI’s process for answering this is:

  1. Natural Language Understanding (NLU) analyzes the content to determine if there is a question, statement, or command. It does this by looking for question words and phrases. In this case:
  • ‘How’
  • ‘How many’

2. Next the NLU analyzes the content to look for any verbs and relevant words that give the context to those question words and phrases. In this case:

  • ‘Many’
  • ‘Have’
  • ‘Within’

3. Once done, the NLU looks for what item(s) the identified action needs to work on. In this case:

  • ‘Gastric’
  • ‘6 months’

4. As items have been identified, the NLU has the basics to create a knowledge base and it can therefore look at the words that surround the item for more context and clarity. To do this, it first identifies any such words, and then defines them to see if they are needed to clarify the command. In this case:

  • ‘Anomalies’
  • ‘= irregular, not normal’

5. The NLU can now put together a statement based on the context of the processed question and the identified data the question is being asked of. In this case:

  • ‘Retrieve number of DNA_SEQUENCE_TYPE where signups have been created within 6 months.’

6. The AI now checks its data store to see if any programmed commands align with the statement. If they do, then they will perform the given query.

7. The final step is to return the results to the user. This is only done once the AI has checked that the required authorization and authentication is still in place. 

Q4: Thank you. And so is it this sort of scenario that sets OmniIndex’s AI apart from the competition?

Yes and no. Remember, our AI is a Narrow AI Engine and unlike ChatGPT or Bard, it is focussed on a specific set of tasks. Its ability not to have to digest millions of lines of text certainly give it an edge because it means organizations can have an AI expert on their team providing insights on their encrypted data within just a few moments with a specific understanding of that industry and field.

Q5: Thank you. To finish, how do you ensure that the OmniIndex AI is accurate, reliable, and trustworthy?

Great question. We try to do this by having a probability engine within the AI. This engine takes the results and then passes them back through the inputted text, checking and scoring against what was inputted. These scores are then rated and the information collated and returned. 

As for the data that’s inputted into the AI, we work with our customers to ensure the dictionaries and other sources the AI digests are the right and appropriate ones for the client. There is no machine learning or additional source material added beyond this ontology, and so the chance of the AI being given biased or inaccurate information to work from is reduced. 

However, it is not impossible and this is something we always remain conscious and vigilant about.