The future of student support: integrating AI and human expertise
By now, you’ve likely heard a lot about ChatGPT. It’s an artificial intelligence making waves in higher education — and everywhere else — for its amazing ability to respond to its users prompts and generate creative stories and answers. We’ve written before about the role of AI like this in higher education, and today I’ll go a step deeper and explore what ChatGPT and similar language models mean for Mainstay and our partners.
Comparing Mainstay’s Engagement Platform with ChatGPT
Complementary, not competitive
Mainstay’s engagement platform and OpenAI’s language models were built for two different purposes.
- Mainstay’s bots are built to proactively support students and drive positive outcomes
- GPT-3 was developed to generate human-like text in response to a wide range of user’ prompts
- ChatGPT, a variant of GPT-3, was designed as a chatbot to have conversations with people
Because ChatGPT was trained on a smaller, conversation-focused dataset than GPT-3, it excels at the specific task of chatting conversationally with people. It learns how people respond to prompts, and mirrors it. It does so remarkably well, making it a perfect fit for complementing Mainstay’s strengths.
ChatGPT | Mainstay |
|---|---|
Solely reactive. Like many other chatbots, ChatGPT generates text in response to individual prompts. What if you could use ChatGPT’s language generation capabilities and proactively reach out to students to suggest that they take action? | True two-way engagement. Mainstay does much more than automatically answer student questions. Our partners also use our platform to proactively engage target audiences with the right information at the moment it’s needed (like before a key deadline.) Our Best Practice Collections offer partners a full calendar of proven, research-backed communications to spark student progress and drive positive outcomes. |
Lacks human intervention or checks and balances. ChatGPT advises its users, “The system may occasionally generate incorrect or misleading information and produce offensive or biased content.” Because it is built to generate answers based on prediction, it will sometimes guess or create answers that sound reasonable, but are inaccurate or potentially inappropriate. On its own, ChatGPT does not allow for human intervention to ensure that its responses are accurate and reflect well on your institution. | Accurate answers that keep a human in the loop. It’s true that Mainstay’s bots cannot answer as broad a range of questions as ChatGPT. However, we are 100% confident that the answers our chatbots deliver contain accurate information because they have been finely tuned for each partner’s needs by our expert [human] team. Our AI is transparent and admits when it cannot answer questions, which ensures that a person can provide an accurate answer and expand the bot’s relevant knowledge with usage over time. When our platform detects that a person needs to get involved because the AI cannot answer a question accurately or because a student is asking about a sensitive topic, we flag the question or conversation so that a person can seamlessly take over the conversation. |
Potential vulnerability to bias. AI models are only as good as the data they are trained on. Any deficiencies — such as bias, intentional or not — in the dataset may be mirrored in the text it generates. A model like ChatGPT that is trained on a dataset of typical conversations rather than the most supportive responses can only generate language reflective of typical dialogue, not best practices. But what if an AI with ChatGPT’s language generation capabilities was trained specifically to support improved equity in higher education? | A focus on historically underserved students. From our inception, we have designed our platform to improve access and equity for first-generation, minority, low-income, and non-traditional students. We provide not only the technology to have automated conversations with students, but also the messaging shown to have the greatest impact on outcomes for all students. |
How Mainstay leverages GPT-3
We see our platform as an extension of your team, and our outlook is that AI like ChatGPT and similar language models like GPT-3 are complementary to what we do and represent an opportunity for Mainstay. We’re actively pursuing opportunities made possible by using Davinci (the most capable GPT-3 model OpenAI has made available), plus our own technology to support students more effectively, with even fewer demands on higher ed staff member’s time.
The keystone of this work is our research partnership with Dr. David Yaeger at University of Texas at Austin. With Dr. Yeager’s expertise, we have explored how chatbots can respond to students’ messages about their emotional state and motivation in a way that is supportive, empathetic, and most likely to lead to positive outcomes.
Imagine having an AI trained by the best cognitive scientists in the world:
- A student texts your Mainstay bot and asks a question it isn’t confident it can answer accurately
- The bot instantly responds to the student, telling them that it doesn’t know how to respond and will check with the team
- The bot notifies a staff member and suggests an answer to the student’s question based on how the best cognitive scientists in the world would respond
- The staff member can then edit and approve the bot’s response before it’s sent, keeping a human in the loop and ensuring that answers are accurate and appropriate
We are actively exploring possibilities like this, and others, and are excited for what the future holds for our partners.
In conclusion
The rise of artificial intelligence revolutionizes the way we approach education and engagement with students. AI, such as ChatGPT, is a promising tool in education, but it’s essential to ensure that it’s used ethically and complemented by human expertise to deliver the best outcomes for students.
Let's talk AI
Mainstay works with research partners like the Yale Center for Emotional Intelligence, the National Institute for Student Success, and The Partnership for Ed Advancement to develop messages proven to drive better student outcomes.
If we’ve piqued your interest, let’s explore how we can work together to spark a brighter future.
Not ready to connect, but want to know more? Check out our white paper, All about Behavioral Intelligence.
