Deep and dynamic access to data
Multiple regulatory and compliance agencies use ASIC data as the key reference on company structures, office holdings and shareholdings. Access to this data can be costly, is based on a static view provided by dashboards or external partners and often requires multiple requests. This is particularly the case where it is necessary to understand relationships between companies and people that are indirect (for example looking through multiple shareholding layers).
To address this, the Data Analytics Centre created ‘Chat To ASIC’, a chatbot powered by OpenAI GPT4 and ASIC data hosted in a graph database. The chatbot offers deeper, more dynamic access to ASIC data down to 10 levels of shareholding. It integrates with cutting edge enhancements in large language models that improve the quality of the responses and limit the models to using only specific data to generate responses. Additional business data can be added, based on business needs, to enhance and provide more tailored insights.
The NSW AI Assessment Framework was used to evaluate the responsible and safe use of AI for this project. The chatbot has been designed to ensure that that risks such as hallucination are minimised. It has also been designed with safeguards such as raised explainability, multi-metric manual and automated evaluation, a detailed log of questions and answers for model maintenance, guardrails and jailbreaking safety measures, and private and secure access to OpenAI.
Achievements
As a result of Chat To ASIC, users can ask a wide range of queries with various levels of complexity and can customise requests. This has reduced costs (from $10-150 AUD per one-off query to external providers to ~10c USD per query plus ongoing platform costs) and resulted in time savings. It has the potential to enable users to extract valuable, high-level insights by linking ASIC and other regulatory data, which will also support a rich understanding of the networks formed in regulatory jurisdictions by the relationships between relevant stakeholders.
Engagement
The initial trial phases focused on understanding the impact of generative AI on This project has involved partnership and engagement with a small number of NSW regulatory agencies, with the intent of expanding engagement to other regulators as the product matures.
Department leading the project
Department of Customer Service – Data Analytics Centre