The Implications of Large Language Machine Learning Models in Tutoring

The Implications of Large Language Machine Learning Models in Tutoring

Since ChatGPT was released to the public via OpenAI the education community has been in constant chatter about its implications for learning. A large language model is a machine learning system that uses vast amounts of data to generate human-like language and perform natural language processing tasks such as text generation, translation, summarization, and sentiment analysis.

If you have not experienced ChatGPT you should. It’s definitely an exciting next generation in how we think about searching on the internet. For almost a quarter of a century Google has been the cornerstone of where we find information online. Google presents relevant links based on search. With ChatGPT information is presented in natural language and feels much more like a chat with someone very knowledgeable about the question you ask. 

Although the conversation feels fairly natural and the answers are cogent, they are not always correct. That being said, if you have a decent understanding of a subject you will probably find that the chat feels authentic and most of the time pretty accurate.  

When the AI was released everyone at Pearl was interested in jumping quickly and trying to understand the implications for tutoring. We spend a lot of time thinking about equity and imagine a world where anyone that needs a tutor could have a tutor. None of us is convinced that AI could replace a human, but I do think that AI has the capacity to make someone better at tutoring or give more opportunities for people to teach.

Here are some interesting implications for the tutoring industry:

MATCHING

Without compromising any person’s identity a tutor admin can give a list of characteristics of a learner and then ask to see corresponding characteristics of a tutor. Below is an example of a question and ChatGPT’s answer:

Question:

“Based on the research, what are 5 strong characteristics to look for in a tutor that would match well with a 9 year old Asian dyslexic child coming from a poor socioeconomic background learning in English whose first language was Chinese?”

ChatGPT:

“Great question! Here are five strong characteristics to look for in a tutor that could match with an Asian 9-year-old dyslexic child coming from a poor socioeconomic background learning in English whose first language was Chinese:

  1. Patience: A tutor who is patient can help the child feel comfortable and confident as they learn. Dyslexia can be frustrating and challenging for a child, especially when they are also trying to learn a new language. A patient tutor can help the child to feel supported and encouraged.
  1. Multisensory approach: Dyslexic children can benefit from a tutor who uses a multisensory approach to learning. This means that the tutor can use different senses, such as sight, sound, and touch, to help the child learn. For example, the tutor could use visual aids, like pictures or diagrams, or use physical activities to help the child understand the material.
  1. Experience working with dyslexic children: It’s important to find a tutor who has experience working with dyslexic children. This experience can help the tutor understand the unique challenges that dyslexic children face and how to tailor their teaching methods to meet the child’s needs.
  1. Cultural sensitivity: A tutor who is culturally sensitive can help the child feel understood and respected. It’s important to find a tutor who is familiar with the child’s cultural background and can be sensitive to their unique needs and experiences.
  1. Flexibility: Finally, it’s important to find a tutor who is flexible and can adapt to the child’s needs. Dyslexic children can learn at different paces and in different ways, so the tutor should be willing to adjust their teaching methods to meet the child’s individual needs. Additionally, the tutor should be flexible in terms of scheduling and availability to ensure that the child can receive consistent support.”

SCHEDULING

Once ChatGPT or some version of this type of AI is available, an admin could ask AI to construct a calendar based on the data for two tutors.  Here is an interesting example of ChatGPT’s response to a basic scheduling prompt.

Person:

I am a tutor that needs to meet with my student three times a week for 30 minutes each time.  I would like to know the dates I can meet with that student in school based on a typical school calendar.

ChatGPT:

The specific school calendar will depend on the school district and location. However, in general, most schools in the United States operate on a traditional academic calendar, with two semesters per year (fall and spring) and breaks for holidays and summer vacation.

To determine the possible meeting dates, you can start by looking up the school district’s academic calendar online or contacting the school directly to get a copy of the calendar.

Assuming a typical school calendar, here is a general example of possible meeting dates for a tutor meeting three times a week for 30 minutes each time:

  • Fall Semester: September to December (approximately 16 weeks)
    • Possible meeting dates:
      • Monday, Wednesday, Friday: September 6th to December 16th (excluding holidays)
      • Tuesday, Thursday, Friday: September 7th to December 16th (excluding holidays)
  • Spring Semester: January to May (approximately 16 weeks)
    • Possible meeting dates:
      • Monday, Wednesday, Friday: January 10th to May 20th (excluding holidays)
      • Tuesday, Thursday, Friday: January 11th to May 20th (excluding holidays)

Keep in mind that these dates are approximate and may vary depending on the specific school calendar. It’s important to confirm the exact school calendar with the school or district to ensure that the meeting dates are accurate.”

TUTOR ASSISTANT

We’re starting to get into the interesting part now. Obviously, it is imperative that the teacher’s knowledge confirms, in some way, whether the ChatGPT response is helpful. As an example, ChatGPT can help tutors contextualize or better consider a student’s learning style or interests in order to support their tutoring. Let’s see if ChatGPT builds some math problems based on two factors (age and affinity) based on my son’s interest in trucks and construction equipment.

Person:

My son is 12 years old and has a strong affinity for trucks and construction equipment.  Based on his age and interests please produce 3 great ways to teach him long division. 

ChatGPT:

If your son has a strong interest in trucks and construction equipment, incorporating these interests into his long division lessons may help him stay engaged and motivated. Here are three great ways to teach him long division:

  1. Use visual aids: Use pictures or diagrams of trucks and construction equipment to explain the steps of long division. For example, you could use a picture of a dump truck to represent the dividend and use smaller pictures of construction equipment, such as bulldozers or excavators, to represent the divisor and quotient. This can help your son visualize the process and make it more concrete and meaningful.
  1. Turn division into a game: Create a game that uses trucks and construction equipment to teach long division. For example, you could create a game where your son has to use a dump truck to carry a certain number of building materials and use excavators to divide the materials into equal groups. Each time he successfully divides the materials, he can move to the next level of the game. This can make the learning process more fun and interactive.
  1. Use real-life examples: Find examples of how trucks and construction equipment are used in real-life scenarios that involve division. For example, you could use examples of construction projects where trucks have to haul and divide large quantities of materials. This can help your son see the practical applications of division and make it more relevant to his interests.

Remember, every child learns differently, so it’s important to find an approach that works best for your son. Incorporating his interests and making learning fun can go a long way in helping him stay motivated and engaged in the learning process.

ASSESSMENT

Assessment is the final interesting implication for tutoring.  At Pearl, we believe the key to tutoring is quick formative assessments that verify student learning and identify areas where knowledge or understanding might be lacking. Once again, this is a great way to leverage several variables in a question. The following are two examples that you may find interesting:

Person:

Please give me three formative assessment questions that will help me validate that my 9 year old student understands long division.

ChatGPT:

  1. Can you explain the steps involved in long division and how they relate to each other?
  1. Can you solve a long division problem independently and explain how you arrived at your answer?
  1. Can you apply long division to solve real-world problems and explain how you arrived at your answer?

Person:

Please provide me with 3 problems to solve that would assess my 9 year old’s understanding of the basics of long division.

ChatGPT:

  1. 84 ÷ 6 = ?
  1. 145 ÷ 5 = ?
  1. 232 ÷ 8 = ?

Note: These problems are designed to test the student’s understanding of the basic steps of long division. When assessing your student’s responses, make sure to check that they have correctly used the divisor, dividend, quotient, and remainder, and that their final answer is correct. It’s important to provide feedback on their work and address any areas where they may need additional support or practice.

SUMMARY 

When tutoring programs consider using ChatGPT or any of the other emerging large language model AI tools, they must validate the tools carefully.  It is possible to speed up certain processes, but there is a cost associated with it.  There is a possibility that AI is extrapolating biases that are not aligned with your ethics.  Additionally, tutors should cultivate value that reflects their approach to tutoring and not rely too heavily on computers.  Relationships are at the heart of tutoring, so it’s important to foster connections and care throughout the process. Nevertheless, keeping up with the changing world is essential.  There is no doubt that our students will benefit from this new tech; therefore, as educators to the next generation, we need to keep our eyes on the ball. 

Now, for kicks, let’s let ChatGPT have the final word ?

Person:

What are the implications of Large Language Machine Learning Models in Tutoring

ChatGPT:

Large language machine learning models have several potential implications in tutoring, including:

  1. Personalized Learning: These models can help provide more personalized learning experiences for students by tailoring the content and pace of the instruction to their individual needs, interests, and learning styles.
  1. Immediate Feedback: The models can provide immediate feedback to students on their work and progress, helping them identify areas where they need additional practice or support. This can help students stay engaged and motivated and can also help teachers more effectively monitor student progress.
  1. Automated Scoring and Grading: These models can help automate the grading and scoring of student work, freeing up time for teachers to focus on providing individualized instruction and support to students.
  1. Improved Accessibility: These models can help make tutoring more accessible to a wider range of students, including those with disabilities or who speak different languages.
  1. Improved Instructional Design: These models can help identify patterns in student learning and suggest improvements to instructional design and content, helping teachers to more effectively support student learning and improve outcomes.

However, there are also some potential challenges and limitations to the use of large language machine learning models in tutoring, including issues of privacy and data security, the potential for bias in the algorithms used to analyze and evaluate student work, and the need for teachers to be trained in how to effectively use and interpret the results of these models.

The End…