5 Barriers to AI Expansion in Higher Education (and how to overcome them)

If universities are indeed places that explore the unknown, advance thinking and push the boundaries of what’s possible, then AI is probably the best tool in their bag.

“To AI or not to AI” is no longer the question in higher education: the debate has long moved on. Nobody serious is still asking whether AI is good or bad, whether to use it or ban it. We’re now in much more complex territory, that's about access, cost, infrastructure, ethics and control. This stage is messy, full of lessons learned, and one that is explored both individually and collaboratively, as higher education is like no other sector in sharing learnings and supporting the collective progress of the sector.

What are the top barriers to AI expansion in higher education?

With pressures growing both from students and staff to provide access to AI tools, universities are starting to remove the blockers that previously kept them on the fence about AI adoption, adapting their approach as they go. While leader’s belief in AI’s capacity to cause havoc is starting to fade as usage is growing, barriers to its expansion still remain. The main ones include:

1. Governance, compliance and privacy

Data governance, or in many cases the lack thereof, is another key factor in the caution towards widespread AI adoption in a sector that is risk averse – with good reason. Depending on their jurisdiction, institutions are bound by multiple privacy laws: FERPA in the US, GDPR in Europe including UK, and the newer EU AI Act, the Artificial Intelligence and Data Act in Canada, to name just a few.

With AI models dominated by a handful of private tech giants, institutions feel unease in feeding their data to train those commercial models, not to mention the intellectual property and commercialization considerations that are the lifeline of research institutions. Without “walled gardens” or custom-built sandboxes, it is difficult to protect student privacy, confidential research or remove bias from unreliable data.

2. Legacy data foundations

Universities have operated for too long with siloed and fragmented data sets, as each department owns its data, uses different systems, different formats, reference numbers or naming protocols.

This “dirty data” (inconsistent, incomplete, duplicated – in short unreliable for business use) coupled with vague ownership and compliance restrictions are more than enough to put AI in the “too hard basket”.

3. The complexity of the IT “plumbing”

If KTLO (“keeping the lights on”) takes up the entire IT’s capacity, where is innovation going to come from? With computing costs going up and enrollments going down, the likelihood of additional funding for AI implementation is very low. And even if that did happen, it would have to run on the existing “plumbing”, adding to its complexity.

This is a “catch 22” situation, with AI acting as a magnet for students looking to gain future-ready skills, leading to higher enrollments (hence more income) while at the same time “keeping the lights on” with the same resources and budget stretched even further.

4. The cost of AI consumption

Stories about tech companies giving employees free access to AI tools only to limit it soon after, have made headlines for a while now. Microsoft, Uber, Meta and Amazon, to name just a few, have all reported huge AI bills or burned through a whole year’s budget in a just few months. And these anecdotes haven’t spared higher education institutions either, although due to the increased scrutiny on their spending as public institutions, you are more likely to hear those in private conversations.

This inability to predict usage and cost, as well as the cost ownership debates with faculty, (is it an IT tool or a learning resource?) has kept the brakes on larger scale AI expansion in higher education.

5. Compute power, storage and connectivity

Running AI at scale requires expensive infrastructure modernization. From high performing computers to cloud storage and Wi-Fi network upgrades, the costs rack up.

When AI requires additional bandwidth or compute power in one department or research area, it can affect the teaching and learning experience in a different area – a balancing act IT needs to keep a handle on.

For large research workloads in medicine, engineering and multidisciplinary fields, purely cloud-based environments will no longer suffice, with many institutions considering private partnerships to build and run their own data centers.

But, where there’s a will, there’s a way, and despite all these blockers, the sector is making a consistent effort in advancing AI adoption. While this is a competitive advantage for individual institutions in attracting the best and brightest, they are often working collaboratively, sharing practices and learnings across the sector.

To find out how universities around the world are preparing to scale their AI adoption, the key priorities they are tacking first, and some of the tools and initiatives they’ve adopted, download our whitepaper.