Dr. Deepak Padmanabhan:

When we talk about AI, we generally talk about it as a monolithic database but there are different flavours of AI. One of the earlier flavours of AI, which is now called good, old-fashioned AI and is no longer in the picture, is something called symbolic AI, which essentially is about authoring rules. These systems operate by way of rules so we can trace the provenance of the decision back to a particular aspect. It is a little bit like industrial automation. If a printing press is not working, we can drill down to determine which particular part of the machine is causing it to break. In that sense, we can work backwards and find out which rule caused it to fail. There are those kinds of AI systems that have an inherent transparency to them, in contrast to the systems that we have now like large language models. Maybe LLMs could imbibe some sort of flavour of those and there are explorations in that direction. Dr. McQuillan talked earlier about small LLMs and there are ways of infusing human knowledge into LLMs in a very transparent way. There are explorations but it is still early days.