Suddenly it seems, artificial intelligence (AI) is everywhere in our lives. We have Alexa in our homes, smart apps on our phones, and chatbots popping up prominently on many of our favorite websites. Recommendation systems used by Netflix and Amazon have normalized our expectation of smart commerce, and we're starting to get used to commercials that tout broad business applications for AI. Nevertheless, anxiety and uncertainty persist about what AI is and what it does.
Within education in particular, many are left wondering what the most immediate and impactful (or disruptive) applications of AI will be for our institutions and students. In theory, AI techniques could improve decision-making and help us achieve better results at every level of the educational endeavor—but exactly how remains uncertain.
In a recent conversation with Jim Mathews, our chief technology officer, we discussed today's reality and the future promise of AI in education, through the lens of the work his team has been pursuing at EAB. AI is one of our technology team’s five strategic pillars, in addition to Quality, Privacy & Security, Service Oriented Platform, and Cloud, some of which we will explore in more depth in future posts.
First of all, it seems like there are lots of different types of AI—Can you explain the continuum?
Yes, and even before that, it’s helpful to always have two questions in the back of your mind when thinking about AI. The first is—what are the abilities and limitations of AI and the second is how well do those abilities and limitations align with a specific business problem. AI is good for lots of things and not so good for plenty of others. And even among the list of worthy problems I think AI will eventually help us solve, there are a number of instances where we don’t have enough data yet for AI to be relevant. AI requires data at scale.
There is no monolithic, single definition of AI; it's really a collection of capabilities, all sharing the common core principle of automating the extraction of information from data towards a useful end. One way to understand the different types of AI is to think of them as being on a continuum, based on levels of autonomy.
At one end of the spectrum you have fully autonomous systems where machines make decisions and take actions—think of a self-driving car. At the other end, you would put simple analytics, where data is summarized and presented to a person for interpretation and decision, which is only a small step up from that person having the raw, unmanipulated data. But I think the best way to think about the continuum is to ask yourself, "how much autonomy is desired and valuable and can you attain a suitable performance level such that you feel comfortable with that level of autonomy?"
One other thing to consider is that AI isn’t that great at defining "success", so it's very constrained to the specific definition of success it was trained against. This makes it very important to use it as an aid, but not an autonomous system when success could reasonably mean different things to different people. AI informed-decisions also inherently have an uncertainty band—they're far from perfect. And since people have a different tolerance for uncertainty, people will always want to retain final control of how the information is used in many applications.
Why does AI matter for education?
What AI does really well is pattern recognition and playing the odds. It performs best when you have many repeated observations of a situation and can compare outcomes against a starting state. Given this strength, some of the best and most interesting problems for AI are situations where individual humans face a decision infrequently, but humans overall as a group face that decision frequently.
Within education, decisions like, "which students should we focus extra attention on to prevent them from dropping out," "which high schools are most likely to yield the most students if we send an admissions counselor to do a college visit," or "where can we make the most impact by adding an additional course section" fit that scenario. These are the kind of use cases where AI can add tremendous value because individuals might not have an opportunity to make that decision frequently enough with clear feedback to learn from their choices—and mistakes could be costly.
AI can also power the next generation of education technology, continuing the historical trend of technology automating high-frequency, routine actions previously performed by people, increasingly even those requiring relatively high skill. That is a trend that has been progressing for 50+ years. For example, when handheld calculators displaced the slide rule and made it much easier for anyone to calculate something efficiently without special training. AI can extend that into more complex interactions like Google’s well-known automated scheduler for service appointments.
Where is EAB using AI in its work today?
At EAB, we're focused mainly on what I would call "Augmented Intelligence," which lies somewhere in the middle of the AI autonomy continuum. We share AI-derived insights with our users at the moments when they are making important decisions, to make them more effective and efficient in their daily work.
Machine learning, which underlies AI capabilities, is embedded in the DNA of our technology products, and has been since we launched the Student Success Collaborative in 2012 with a predictive analytics platform. The reason AI feels like such a natural extension of EAB’s work is that AI could be described as a mathematical translation of the "best practice" concept, one of our core concepts in EAB research. Since AI improves with scale and gets better as it is applied to different scenarios, we believe our AI work offers a different kind of best practice, complementary to our qualitative research.
Today, AI appears in all of our technologies. In student success, it powers how we stratify graduation risk for users of our student success management system (SSMS). Our algorithms and models predict a student’s likelihood to succeed across a variety of time horizons and outcomes. In our enrollment Pipeline Analytics offering, which is one of our latest initiatives, AI helps schools predict which students are most likely to enroll. We also use AI to identify likely candidates for our Adult Learner Recruitment offering, where targeting potential students can be challenging through traditional methods.
What’s next for EAB and AI?
The next frontier for us to realize the promise of AI, especially within the domain of student success, is to figure out how to (carefully) bring augmented intelligence directly to students at some of their most pivotal moments like course selection, planning, and major selection.
When students choose a major, or even a course, their decisions have long-term and costly consequences—does their choice align with prior work and maintain them on a path to timely graduation? What’s more, students making these choices have to balance multiple variables, such as interests, aptitudes, and long-term personal goals, not to mention cost. If you apply our criteria for a good application of AI—scenarios that are low-frequency for an individual, but high-frequency in aggregate—these types of academic choices fit perfectly. These are decisions where augmented intelligence can be very powerful. We’ve discovered in prototype work that we can bring targeted insights to help students more successfully navigate these choices.
We know we must be very thoughtful about how we choose to present and share AI-derived insights directly with students, but it's exciting to consider the potential benefit to students to help them make more informed choices.
Looking a little further into the future, and further down the autonomy continuum, we are thinking about how AI can help provide "always-on" support for students. Imagine chatbots and intelligent assistant integration like Amazon’s Alexa, supplementing the critical human work our members do to support student success.
Any final thoughts?
Yes: Things that are new (and maybe even a little disconcerting) when we first encounter them quickly transition to a normal part of our daily experience and even quicker into our expectation set. So even if some of these AI concepts and capabilities feel new, I encourage you to join us on this journey, and collaborate with us on figuring out the best applications of AI in higher education and student success. Five years from now, I’m certain the students we all strive to serve will have benefited.