Autonomous AI Agents in Enrollment Management and Student Success - Part 1: Riding the Waves of AI

Autonomous AI Agents in Enrollment Management and Student Success - Part 1: Riding the Waves of AI

The dizzying speed of innovation in Artificial Intelligence (AI) resembles a high-speed car chase.? The road flashes in front of us. New options and obstacles appear as fast as we can possibly navigate them.? It leaves little time to wonder if we’ve taken the best path or what unseen obstacles lie ahead, given the blinding pace of new information coming straight at us.? While slowing down to make better choices and consider the options is prudent, there is a counterbalancing pressure to keep up.? We hope that our institution will go just fast enough to remain in the chase, avoiding the high costs of doing nothing - becoming a “have not” in higher education and allowing others to race in a more advanced league.

This chase appeals to our competitive nature and to our strong desire to find innovative ways to serve our students, faculty, and staff.? It is also one that likely raises our blood pressure a bit.? There are unknowns on the horizon, and we want to make the best possible decisions for our institutions, students, and stakeholders.? In order to successfully navigate the changing course of technology, we need to have a good understanding of what the map of innovation looks like.

The Four Waves of AI

Salesforce has identified four waves of AI [1]: predictive, generative, autonomous and general intelligence.? What are these and what wave are we riding now?? Let’s pump the brakes for a moment and consider the road behind, around, and in front of us (at least as far as we can see today).

As discussed in an earlier article [2], predictive analytics is not new in higher education.? We have long aggregated and analyzed data to understand the likelihood of something happening or not happening.? The use of data to inform decision-making is at the root of Strategic Enrollment Management (SEM), and some have mused that this separates enrollment management (EM) from SEM.? Whether scoring the likelihood of someone to enroll, or understanding the relative weight of variables in persistence, higher education has come to expect that predictive analytics are standard tools.

What changed?? Increases in computing power and cloud computing allowed predictive analytics to move from research-based studies and limited use, to widely-available tools that can be placed in the hands of end users.? There was also an increase in expectations for the amount, timeliness and rapid analysis of data.? Predictive AI is capable of generating analyses more quickly than humans, even assessing which models provide the most relevant results. It is a tool that, once generally available, became an incredibly useful to higher education practitioners.? For example, Salesforce’s Einstein was embedded into the platform where end users could see its assumptions and results, rather than just a predictive score.? It could recommended next best actions, to focus their work with prospective or current students.? This is especially helpful for new staff who may be less familiar with the routines of enrollment.? Neural networks approaches surfaced insights that may have eluded even veteran staff.

Generative AI exploded after OpenAI launched ChatGPT.? This built upon the machine learning and deep learning advances in predictive AI.? However, there were some important innovations that made this appear to be quite new and different (note: these are very high-level descriptions and there is no attempt to provide exhaustive information on any of these innovations in this article):

  1. Natural Language Processing.? Rapid improvements in the way that machines could understand human languages allowed the application of Generative AI to computer interfaces.? It predicts the meaning of languages by understanding when those words have been used in the past, in similar or exact combinations.
  2. Natural Language Understanding.? A subfield of natural language processing, it uses algorithms to understand the context, intent, and “meaning” behind human communications with the machine.
  3. Multi Directional Vectors.? This was a significant advancement in the way that machines understand languages.? It allows the nuance of languages to be understood, since one word can have several casual or formal meanings, depending on the context of its use.

As this new level of Generative AI was released, it was fun to provide prompts and receive incredible mimicry of human interaction at jaw-dropping speed.? These innovations also explain advancements that occurred in chatbots and interactive voice systems.? Old, clunky chatbots and phone systems were frustrating.? They had few options or choices, and these were often too general for most users. Trying to say or type questions often resulted in frustration.? Comedians made light of these frustrations, noting that what you said or typed into the chatbot was often horribly misunderstood.??

These advances have provided significant improvements in human-machine interfaces? You may have noticed it when speaking to a digital assistant, such as Siri or Alexa, or you may have called an interactive voice system that was able to listen and respond to you accurately and fluently.? Perhaps you’ve recently engaged with a chatbot that simply started with, “How can I help you today?”? The difference between old and new chatbots is startling.? The newer versions, enabled with AI, can more quickly and clearly assess your issue or question, and offer more intelligent responses. ? It is more likely to point you toward a knowledge article that accurately addresses your question or issue.? If you are signed into an online account, it can provide very personalized responses, including the status of your application, account, etc.

We are seeing the first uses of Generative AI to create communications.? Machines can be trained to look at specific information, such as an institution’s website, and draft communications from them.? One example is a knowledge article, used to provide an online visitor with helpful infomration on a common topic. You could ask the AI assistant to draft a short summary of how to apply for scholarships, housing options, or other common topics.? Another example is a conversation summary.? When Generative AI is turned on during calls or text exchanges, it can quickly (and usually very accurately) create a draft summary that can be added to the comments record of the student. These are incredible time savers for staff, allowing them to spend less time writing comments, and more time interacting with students and others.

There is some very interesting work going on in automated communication creation, using Generative AI and security layers to personalize communications.? Personally identifiable information (PEI) gets masked during the generative process, and the result is an email (or other communication) that reflects the specific data of the recipient.? An example would be communications about specific academic program interests, combined with contact and application history, to provide a nuanced draft email about next steps in the enrollment process.? Another would be a specific communication to a student showing academic difficulties, with targeted support steps.? Salesforce uses a sophisticated “trust layer” to protect sensitive data while leveraging the power of external large language models.

Today, we are experiencing the third AI wave, Autonomous AI, or Agents.? This is still emerging but promises to truly deliver on the early expectations held for Generative AI in service and operations.? The Agent combines predictive and generative AI with automation.? In service scenarios, it could offer to guide a student through the enrollment process steps, checking to see what steps have been completed, what have not yet been completed, and offering assistance along the way.? During an interaction with a current student, it may identify a process that needs to be completed, such as a petition to change status or request a leave of absence, then start the automation to get the petition through its many steps.

In the back office, Agents could help complete routine tasks, such as checking new data for outliers or omissions, then creating reports for human review.? Answering requests from faculty and staff in human resources or IT departments to respond to and process routine requests and needs could free up staff to spend more time on complex issues.? The outcome here is not to replace people but to free them from routine work and allow staff more time to perform truly human work, interacting with students and others to foster a sense of community, belonging, and care.

Some early adopters of Agents are companies that have a high level of customer service or interactions.? These recent deployments can now assist online customers with routine requests, such as returns or exchanges, taking the task from initial “how can I help you” through a completed process.? There are likely several similar interactions in education that could be handled this way, enabling more complete support for students 24/7/365.

As the third wave is just starting, we look ahead to the fourth wave, General Artificial Intelligence.? We’ve seen this in the realm of science fiction through characters such as Data from Star Trek: The Next Generation, or even earlier with Rosie the Robot from The Jetsons.? J.A.R.V.I.S. from Iron Man may fall somewhere between the realm of Agents and General Artificial Intelligence. To achieve General Artificial Intelligence, AI must be able to perform multiple forms and types of algorithms in harmony.? Today, AI can perform admirably in limited ways.? The algorithms that can generate knowledge articles from your website can’t drive your car or operate your smartphone.? There is a great deal more innovation and advancement required to get us from the current, very impressive, AI waves to this fourth one.? It also provides an insight into just how impressive the human mind is, given all that technology can do and the great gap yet to be closed to achieve General Artificial Intelligence.

Three “Must Haves” for Success with AI

As enticing as it may seem to pop something out of the box and start using it, there are three critical considerations required for using Agents in any situation, especially enrollment management and student success.? Each of these requires thought and strategy to ensure that the institution spends wisely, improves (rather than decimates) its brand and market position, and achieves the greatest return on investment possible.

Integrated Data

All artificial intelligence is only as “smart” as the data it can use to associate relationships.? If you only supply it with academic history data, it can only predict the likelihood to persist based on how a student does in courses.? It can’t use financial or social factors that can heavily influence outcomes if they are stored in other data systems or not harmonized with the way academic data is stored.? Similarly, if the AI sees only the information in a CRM that includes contacts (email, phone, SMS, events) with the admissions team and not social media interactions, visits to website pages, contacts with other departments that may use other CRM (or no CRM) systems, or if marketing and event data are in other systems, it can only predict the likelihood to apply or enroll from the limited data within its sight.? Additionally, it loses the ability to personalize generative communications that are based on the context of those other interests, contacts and experiences.

See a previous article for more insights on how AI and analytics require higher education to revise its data strategies [3].

Ethics and Guardrails

Artificial intelligence has no moral compass.? It sees data as objects of relative weight without regard to how those data impact human thoughts and lives.? WE are the moral compass of AI.? Understanding how AI works and its potential to create bias and toxicity is critical to its use in any organization, but especially in an educational institution.? Warren Buffet famously said, “It takes 20 years to build a reputation and about five minutes to ruin it.”? Given the speed of computing today, perhaps it is now five nanoseconds?

Imagine that Predictive AI scores only those students who have historically enrolled at your institution at high rates and you act on that, focusing on their conversion.? You miss the high-achieving scholars who are harder to convert, and those who historically have not attended your school.? While you may hit the overall enrollment number, the shape of the class and adaptation to changing demographics will suffer.? Perhaps even worse, imagine a marketing message that is crafted by Generative AI without review, containing toxic language that it found associated with a particular program or student group on the Internet.

Best practices today include assembling an AI Ethics work team or standing committee to assist enrollment management and student success efforts.? Such a team must learn enough about AI to understand how it works, and the potential for it to benefit or harm indivisuals or the institution,? It should include people of varied backgrounds and perspectives, so that they can spot potential bias and toxicity as you work to implement, then maintain, AI tools.? Salesforce has a wealth of resources on ethics in AI, including how AI design impacts its ethical use [4].

Business Process Mapping

The innovation of Agents to handle tasks requires that we understand and can map the business processes that we will allow the agent to complete.? For example, if we allow the agent to interact with a current student inside an authenticated space (i.e., logged into an app or website) and that student wants to change her major program of study, what are the steps that the agent will then take to intake the right request form? To whom does it go first, second, third? What actions are required at each step and what are the paths from those outcomes???

Once business processes are mapped, additional questions should be asked about what functions should be included/allowed in the agent’s domain and what should only be performed by a human.? Will we allow it to update the SIS without human review?? Incremental approaches and testing will be key to gaining comfort in the use of an agent for interactions with students, faculty and staff.

Conclusion and Next Steps

The emergence of Autonomous Agents signals a new phase of the artificial intelligence era.? The third of four waves, it builds upon Predictive AI, which has been used for some time in higher education, although recent improvements in computing speed and the availability of cloud computing made it more accessible and more capable of real-time analyses.? Generative AI also built from that root, but used new techniques of Natural Language Processing (NLP) to create more human-like interactions and responses.? The distance from the first wave, Predictive, to the second wave, Generative, took years.? The next leap, Autonomous Agents, was less than two years.? While the distance between it and the fourth wave, General Artificial Intelligence, is great, the time span between now and that wave may be shorter than we can anticipate.

Given the rapid innovation in this field, it is critical for individuals and their institutions to understand and harness AI.? Having a strong and active data strategy is not a luxury but a precondition of AI use.? There are numerous resources available to learn how AI works, as well as the ethical considerations of its use and application in higher education.? Now is the time to put strong policies and practices in place to ensure that the institution understands the design, potential bias and toxicity, as well as the immense potential benefits of AI.? The institution must also ensure the availability of integrated data, and that its business processes can support AI, and allow the institution to achieve the maximum ROI from its investments.

This article began with “Part 1” in the title.? This was intentional but open-ended.? The applications for Autonomous Agents in enrollment management and student success are just now emerging.? We have strong ideas about where agents can be extremely effective, and those will be honed through implementation and observation.? As use cases unfold, the next parts of the story will be written.? Get ready and stay tuned!


[1] https://www.forbes.com/sites/johnkoetsier/2023/09/12/salesforce-building-toward-artificial-general-intelligence-for-business-ceo-marc-benioff/

[2] https://www.dhirubhai.net/pulse/from-evolution-revolution-predictive-analytics-higher-tom-green-yoowc/

[3] https://www.dhirubhai.net/pulse/todays-tech-landscape-calls-higher-education-revise-its-tom-green-nprle/

[4] https://www.salesforce.com/company/responsible-ai-and-technology/

Jason Barnes

Director of Enrollment Management Technology

4 周

I enjoy the illuminating correlation between the capability to build complex AI and Agents along with the basic need for a solid foundation of mapped business processes. Also, in my current environment, I'm excited to explore using Salesforce Agents.

回复

要查看或添加评论,请登录

Tom Green的更多文章

社区洞察

其他会员也浏览了