Expert Insights

Alan Buxton explores the shift from keyword-based to context-based understanding in artificial intelligence, highlighting the substantial benefits and the challenging intricacies of implementing language models in software development. He draws attention to the enormous, expensive problem of manually processing and categorizing unstructured text data, which AI-driven contextual understanding can easily address.

Alan highlights the game-changing nature of language models and contextual understanding:

  • How language models can vastly improve the accuracy rate of categorizing data, reducing the need for human intervention.
  • Successful implementation of these models doesn't merely solve existing problems but also opens doors to features and capabilities previously deemed impossible.
  • Language models can extract key information and summarize responses from hefty data sets, providing valuable insights and also flagging potential risks in business operations.
  • Dealing with 'hallucinations' and overconfidence of language models, and the need for a human-in-the-loop for Quality Assurance.
  • He underscores that shifting from keywords to context is a foundational change enabling software to process meaningful data in a way it couldn't before.

Quote

quotation-marks icon
If you think of a mobile phone. An invoice says mobile phone. You know what that is? It's a phone. If, uh, you've got an invoice that says mobile app development, you know what that is? It's not a mobile phone. We found that by bringing language models in, we could get accuracies in the high nineties, which is pretty cool, uh, without much human intervention. There's a lot of it that is. Text that's not numbers... there's lots of stuff where, for example, you wanna pick up, um, news alerts related to a supplier.quotation-marks icon
Alan Buxton ,
CTO, Simphony

THE NEW DEFAULT angle

Here are some practical takeaways on how to leverage the power of AI's context understanding:

  • Embrace context-based AI for data categorization. Transition from traditional keyword-based understanding to effectively process unstructured text data, ultimately reducing manual review and improving accuracy.
  • Prepare for foundational change. Understand that shifting to AI-enabled context understanding is more than just an upgrade; it's changing the base of how software can process and analyze data.
  • Experiment with language models. Leverage AI's ability to derive meaning from text for automated spend analysis, risk alerts, and contract intelligence.
  • Incorporate human-in-the-loop for QA activities. Even with high accuracy rates, AI models can still suffer from 'hallucinations' or overconfidence, necessitating human intervention, particularly when changes occur in the underlying data.
  • Re-examine roles and processes. As AI adoption reduces the need for human error-prone tasks, teams should refocus their energy on higher-value activities. Explore new roles and redefine existing ones to accommodate AI-centered workflows.
  • Consider using AI to synthesize insights. Utilize AI's ability to extract key terms out of huge data sets, summarize lengthy responses, and flag potential risks in your operations.
  • Identify previously intractable problems to be solved. As AI offers transformative potential, look for problems within your business that were previously too complex to solve but can now be potentially addressed.