Generative AI has seen some phenomenal advancements in recent years with platforms such as ChatGPT setting records for user base growth. But for several potential users, there’s one big problem with generative models.
Most generative AI does not have a built-in accountability mechanism, and without understanding the sources and decision-making processes behind their results, it’s hard to trust them.
The answer is Explainable AI such as Trellis Secure Chat. The AI tool from local tech company Trellis Data delivers super-fast outputs similar to Chat GPT but, equally importantly, also provides sources and user data.
Trellis Data co-founder and CEO Michael Gately said responses from this platform could be tested against evidence and came with justifications as to how the AI model itself reached them.
“For example, someone in Defence might say ‘Tell me how to improve a specific manoeuvre in an operational environment’. But if you can’t go back to the source document and know if that was a good action to do or that it led to a good outcome, how do you know whether it’s useable?” Michael said.
“With explainable AI, the user might ask ‘Your report cites XYZ, can you provide the evidence that supports that?’ and the AI can refer back to the documents that helped create the answer and say ‘This submission is compliant in accordance with Commonwealth procurement regulations part 456 Section B’.
“That kind of accountability is critical for any user that needs to be able to trust the information in their AI-generated outputs.”
Trellis Secure Chat clients can upload libraries of millions of documents for the AI model to trawl through to produce its answers. Soon to launch will be the ability to search a million documents in seconds and pull references out of vast libraries in real time.
“That changes the game,” Mr Gately said. “It underscores the difference between technology like ChatGPT, which is about generative chat, and ours, which is about organisation knowledge.”
For many of Trellis Data’s clients, this kind of AI would be unnerving if it weren’t for the fact it operated in an “end-to-end integrated environment”.
Unlike most generative text technology, Trellis SecureChat does not connect to the internet but is designed to be used in-house. This means it can operate in highly classified environments while providing the same benefits as a technology connected to the internet.
This security is further bolstered by the company’s sovereign credentials.
“From the beginning, we’ve released our own research papers, built our own model from the ground up, used our own data sets,” Mr Gately said.
ACT Chief Minister Andrew Barr recently visited Trellis Data headquarters.
Mr Gately said it was a great opportunity to share the tech with government, along with their love for Canberra and hopes for the future of the company in this city.
“Trellis Data actively seeks out local partners to work with, and we look forward to an ongoing and productive relationship with the ACT Government,” he said.
Notably, Trellis Data is rooted in a foundation of “deep learning” – an AI method that teaches computers to process data in a way that’s inspired by the human brain.
ML Research Labs, Trellis Data’s sister company, has a strong focus on research and development and is considered a leader in deep learning. Top graduates of the ANU Masters course for AI go there to deliver real-world AI benefits.
That’s why, though Trellis’ Secure Chat capabilities attract a high number of Defence, intelligence, government, law enforcement, logistics, health, financial and other clients with similar security profiles, its uses are as broad as generative AI.
“This is a pervasive technology with enormous potential for all types of businesses,” Mr Gately said.
“Imagine I’m a head of department, and on my way to an event I speak into my phone and say ‘Get me the top five points this minister made on this topic and how we can help them, and keep it to 100 words or less’ – and there it is in a moment.
“Or imagine you’re working the phones at Centrelink and someone wants to know how a change in legislation will apply to their payments, and rather than trawling through all the documentation, you can simply ask the question and trust the answer comes with justification. The possibilities are endless.”
For more information visit Trellis Data.