The AI train has already left the station. Governance is trying to catch up.
Universities and other tertiary education institutions collect vast troves of data, from private student details, research material, financial records, course outlines and schedules, payroll receipts, and facilities systems information. This creates a complex data environment that can prove wieldy and chaotic if proper management systems are in place.
Like most industries, the higher education sector has moved toward AI integration to help streamline complex operations and unluck operational efficiencies. When combined with other tools such as advancements in modern data platforms and cloud infrastructure, AI should, theoretically prove an invaluable piece of the data management puzzle for educational organisations.
However, in practice, institutional data contains some unique characteristics that cannot be readily conquered by standard, off-the-shelf AI platforms. Institutional data often behaves like dialects of a shared language. Different departments can use the same word but with significantly different meaning attached. Departments can use varying technology platforms, and operations can rely on human intervention and shared understanding. The result can be a patchwork of siloed data definitions, legacy applications, and workflow processes reliant on shared trust and institutional memory.
The perils of feeding data that relies on human interpretation into AI models has proved a huge challenge for many education institutions. They have been forced to produce unambiguous definitions for words and phrases that human users instinctively understand. AI implementation has removed that safety net.
Feeding old data that was only intended for human use into AI platforms is unlikely to be a smooth exercise that generates positive outcomes. Definitions, assumptions that every human user automatically makes, and natural filters are not applied by machine learning, often leading to chaos that cannot be easily unravelled. This process forces institutions to address the often-archaic way data has been processed and produce universal definitions that apply beyond human interpretation.
Another pressing issue for education institutions has been the scope of access that should be afforded AI models. Should a single AI tool be granted unfettered access to all institutional data, or should AI be treated as a security threat, whereby tiered access is granted on a need-to-know basis. Just as junior employees are not permitted to log into any database at will, and access information reserved for board level personnel, AI needs clear parameters and restricted access to complete specific, designated tasks.
In addition to setting parameters, education institutions have also faced the question of whether AI models should be granted read-only access or possess the ability to make changes to data. Again, treating AI like a foreign entity or security threat and limiting its power has been the most prudent use of this technology. Customer service employees cannot log into accounting software and change a company's financial projections, and AI should have similar limitations.
Tertiary students are inquisitive creatures. To cope with the volume of student inquiries, many educational institutions have deployed chatbots to quickly address FAQs. But how does a chatbot respond when asked which lecturer is most likely to pass substandard assignments, or which fraternity throws the wildest parties?
Successfully deploying chatbots has proven a valuable lesson for institutions on providing specific parameters for AI and limits to its tasks. Chatbots are only given access to generic information like semester start dates, enrolment deadlines, the time and place lectures will take place, library opening hours, contact information for staff, and other curated data that can be freely shared without risk. Building such guardrails is essential to a successful AI deployment.
As institutions moved from basic information retrieval toward empowering AI to act, new ethical and procedural questions were posed. Could AI be trusted to eventually suggest course adjustments, adjustments to academic planning, or guide users through complex workflows? The technology certainly has the scope, but the best results have been achieved by incorporating human supervision into the equation.
This is ideally addressed by building pauses into a system wherein human confirmation can ensure adequate oversight. An advisor can clearly see a summary of what the AI intends to do, takes time to review, before deciding whether to proceed. This crucial step transforms automation into assurance. The simple goal is to let AI handle the repetitive and technical tasks, thus freeing people from such tasks, while simultaneously ensuring accountability.
Data governance frameworks are traditionally based on incremental change. The same rules do not apply to AI, which changes daily. Governance policies written for static systems cannot remain up to date with generative models that learn new behaviour with every release. This speed of obsolescence demands an adaptive governance model with each AI use case outlining what data is being accessed, the definitions that apply, and the audit trail that exists.
Traditional data governance, with its focus on access, accuracy, and security is not designed for additional dimensions such as how AI models apply reason, how they explain their work, and how people interpret the responses. Managing these dimensions demands an AI governance group that can combine technical, security, academic, and administrative expertise.
In traditional setting such as education institutions, the fear of technological change is an ever-present cultural factor. These fears can range from being replaced by the machines, to trust in its ability to work accurately. For AI adoption to be successful, it must be accompanied by shifts in organisational culture.
This shift requires AI to be embedded into people’s daily work activities. The tools must meet staff members where they work, inside existing systems and workflows where adoption feels natural. When AI can reduce points of friction, enable speed and efficiency improvements, and facilitate better decisions, employees feel enabled and empowered by its adoption.
Like all industries, the education sector must embrace emerging technology to remain relevant and competitive. The road to AI adoption is, however, a complex undertaking. It raises a host of data governance issues that demand significant expertise. Off-the-shelf products can struggle to operate in such environments unless expertly configured. Questions such as tiered access, human oversight, and the ever-evolving nature of AI technology can prove beyond institutional IT departments and their governance experience.
Infotrust specialises in helping the education sector manage complex data environments, unlocking upside potential while ensuring robust and compliant security. We help build organisational resilience, drive efficiency, while providing the peace of mind that can only be enjoyed when AI is strategically deployed within exact guardrails, and supported by elite data protection and cyber security experts.
For more information, contact Infotrust today.