KPMG has obtained a patent from the United States Patent and Trademark Office for a blockchain-based method which improves the selection, curation and management of data used to train machine learning and artificial intelligence (AI) models.
“Using blockchain allows both KPMG and our clients to confidently stand behind the responsible use of data entrusted to us,” said Marisa Boston, KPMG director and patent co-inventor with KPMG Director Rupa Shah.
“With this approach, we are able to help companies manage data, identify and mitigate issues such as bias or unnecessary access that could compromise data integrity, and drive increased confidence in insights driven by AI.”
The challenge
Organisations – across both public and private sectors – recognise the potential for data-driven business models, because of their potential to enhance competitive advantage. However, current compliance regulations require careful data monitoring for ‘rightful use’. Additionally, in order for machine learning models to become more accurate and effective when reflecting human intelligence and reasoning, those models need improved data quality and data security.
KPMG’s new patent – U.S. Patent No. 10,528,890 – demonstrates how to use blockchain to store, track and trace data used throughout the AI lifecycle. This includes:
- training, deployment and monitoring of AI models and their algorithms
- the establishing of an AI chain of custody that verifies how and where data and models have been generated and used.
The patent is one component of KPMG’s intellectual property portfolio – which focuses on developing and deploying technology-based solutions that address complex business challenges via:
- strategic investments in emerging technologies
- partnerships with leading technology companies.
The KPMG approach
KPMG’s method improves on existing shortcomings in current data management approaches. It can help organisations operationalise a trust paradigm within the AI lifecycle – by using its blockchain technology to:
- manage data provenance and lineage
- support integrity in models and algorithms.
This makes it possible to:
- identify the most relevant available data
- record how multiple parties interact with that data
- manage data permissions and approvals for use in AI algorithms
- identify when potential bias occurs or has been introduced through ‘training data’
- monitor training data history and quality
- track the deployment and utilisation of models.
By automating key components of the data management process, using smart contracts, there is an enhancement of the existing human procedures required to maintain data records and controls across large organisations. Outputs and insights gleaned by AI become explicable. In turn this increases confidence in results among stakeholders.
“Technology is both the impetus for and solution to many of today’s most complex business problems. However, it’s critical that new innovations be developed and deployed responsibly,” said Mike Nolan, KPMG Vice Chair, Innovation & Enterprise Solutions.
“This blockchain method demonstrates how KPMG is focused both on uncovering the full potential of converging emerging technologies such as AI and blockchain, and on their responsible implementation.”
Enterprise Times: what does this mean
The new KPMG method employs blockchain to demonstrate how to protect data so that it is trusted and used responsibly in an artificial intelligence lifecycle. With trust and transparency enabled by end-to-end data provenance, AI can become a reliable technology.
The appeal of AI to enterprises and their managements is substantial. But business leaders need confidence in the quality of the results before they can leverage AI as part of broader digital transformation efforts. The KPMG patent could materially assist this.
Beyond helping clients with data management, KPMG says it plans to use its new technology in its own data strategy and digital transformation efforts. That will also be both challenging and interesting: Enterprise Times looks forward to learning more.