As organisations incorporate artificial intelligence (AI) into their operations, records managers need to understand its impact on information management. Generative AI, in particular, is transforming the field by enhancing capabilities such as classification, search, and summarisation. This article presents five key insights into how AI in information management is reshaping records management, offering essential information for professionals aiming to effectively leverage these advancements while addressing potential challenges.
1. Generative AI: The Current Focus in AI in Information Management
Artificial intelligence (AI) is a large field with a deep history, but all the current attention is on generative AI technologies. These are machine learning models trained on extremely large corpora of text, images, audio, and video. For example, both DALL-E 2 and ChatGPT were based on OpenAI’s GPT-3 model. Generative AI tools may focus on specific tasks, such as the generation of text or images, or be multimodal and capable of consuming and generating both textual and audio-visual content.
Generative AI tools aren’t just about creating realistic images or convincing conversations, they have a diverse set of applications including translation, summarisation, captioning, classification, information extraction, image recognition, transcription, analysis, and ideation.
Key things to consider when assessing generative AI tools are:
- their base models: underlying these tools are models that have been pre-trained on large corpora of content such as internet harvests. This content can be the source of risks like harmful biases, inaccuracies (facts known by the models can become stale), and intellectual property concerns. The base models have varying sizes (measured in millions or billions of parameters). Bigger sized models generally have better output but are more costly to run.
- their capacity for fine-tuning: after pre-training the base model, there is often further training with smaller but higher quality corpora. This fine-tuning can customise a tool for a particular task (e.g. a chatbot) and fix issues (e.g. adjust for biases). Because the training sets are much smaller, this process can be repeated frequently. Some services allow customers to use their own data for fine-tuning.
- the size of their prompts: the “context window” is the amount of input that a system can operate on measured in “tokens”. Context windows are generally measured in thousands of tokens, but some newer services can take millions, which equates to books of text or hours of video. Larger context windows allow for customised responses by loading prompts with relevant source materials alongside the query.
- their integrations: some tools offer integrations that give them new capabilities. For instance, to operate as agents (e.g. by sending emails or setting calendar reminders) or perform actions like web searches to improve the quality of their output. “Research augmented generation” allows end users to populate databases with their own content for querying by the models.
2. Advancements in Generative AI and its Impact on Information Management
Generative AI had its big coming out moment in 2022 with the launch of two revolutionary tools by OpenAI: DALL-E 2, a text to image service, and ChatGPT, a large language model powered chatbot. A flurry of new tools and services have been released in the wake, including Copilot (Microsoft), Gemini (Google), Llama (Meta), Mistral and Claude (Anthropic).
What’s notable is that the pace of advance shows no signs of slowing down. OpenAI’s scientists wrote a paper in 2020 that observed that the “intelligence” of large language models scales predictably with the size of models, the size of datasets used to train the models, and the amount of compute used. This rule holds true today and it means that we’re not waiting on new technical breakthroughs to see further advances, simply enlarging the models and the amount of data used will be enough.
3. Applications of AI in Information and Records Management
The capabilities of generative AI to retrieve, analyse, summarise, and classify information makes it a great fit for many records and information management processes.
Applications of generative AI for records and information management include:
- classifying and labelling records and information
- enhancing search and discovery (by providing answers to queries as well as lists of results)
- summarising records and information (e.g. give me the history of this project)
- prompting end users for file naming, filing, labelling
- identifying vital records and sensitive information
- analysing patterns of records creation and use across an organisation.
A challenge of deploying generative AI in information and records management is that it will require high quality information as inputs. This might be in the form of controlled data for fine tuning, “prompt stuffing”, or retrieval augmented generation. Without this type of customisation, generic pre-trained models are likely to give results that are confidently incorrect. Organisations that have good records and information management controls in place will be best placed to take advantage of the technology.
4. Broad Applications of AI Beyond Records Management
Of course, the potential of generative AI goes well beyond records management. It is radically changing how information is created and used in workplaces and wider society. Generative AI tools, such as Microsoft Copilot, are already being used in workplaces to draft content (e.g. suggested email replies or images for presentations), summarise and transcribe (e.g. create meeting minutes), translate, automate workflows, perform analysis, interact with clients (online chatbots), write code, and even monitor security (Microsoft Copilot for Security).
5. Ethical and Recordkeeping Risks: The Role of Records Managers
While AI offers many benefits, it also raises significant ethical, information security, privacy, and record keeping concerns. For example, AI systems can perpetuate biases present in the training data, leading to unfair or discriminatory outcomes. The content generated by AI systems may contain inaccuracies or hallucinations. AI systems present a new set of security risks (e.g. prompt injections and data poisoning) and can risk leaking sensitive data. Robust governance and human oversight of AI processes is necessary to manage these risks and records managers are critical stakeholders.
Google, in its Secure AI Framework Approach, identifies data lineage, metadata, and control over the creation and destruction of data as being important parts of what it terms “AI data governance” and these are all areas where records managers have expertise. Records managers should also be involved in negotiating how corporate information is shared with AI systems, in governing how AI generated content is identified and controlled, and in defining how AI is incorporated into corporate record keeping processes.
Harness AI in Your Business: Embrace Opportunities, Control Risks
Recordkeeping Innovation offers a comprehensive AI Governance service designed to help organisations navigate the complexities of integrating AI into their information management practices. Our service ensures that AI technologies are implemented responsibly, ethically, and in compliance with regulatory standards. By providing expert guidance on data governance, privacy, and security, we enable organisations to leverage AI’s full potential while maintaining the integrity and trustworthiness of their records and information systems. Click here for more information.
About the Author
Author Bio:
Richard Lehane has worked as a records, information, and archives specialist for over seventeen years. Specialising in digital archives and electronic recordkeeping, Richard has extensive experience in implementing innovative digital systems, process improvement, records training delivery, policy and procedure development, strategic planning, and establishing digitisation programs. He is dedicated to advancing the standards of information management through the integration of cutting-edge technologies, including artificial intelligence. Connect with Richard on LinkedIn.
