I think the most important distinction between AI and a human teacher is that students don’t ask complete and intelligible questions all the time. In your comments above, you posed questions that were formatted to what AI might expect. You gave the full equation in asking your question. However, it’s the questions of students who are unclear and don’t understand the conventions of mathematics, or possibly the conventions of English, whose questions are likely to pose a problem for AI. If a student tries to ask, “Why do you plus the 3 and not minus the 3?” then the bot doesn’t have the context to answer the question. Human teachers, on the other hand, can use their eyes and experience to not only correct the student’s language gently, but also to know the context without having to ask. We can also see students’ writing and get an idea of where a student is making their mistakes to help us ask the right questions. Not that AI couldn’t evolve to read text or respond to visual stimuli, but I have to figure it will take more than a few years for the programming to be developed that would help a bot to read the writing of some students.
So much of what makes AI amazing right now is knowing how to craft great prompts, or ask good questions. Kids are inherently bad at this because they don’t know what they don’t know!
I think that in reality AI and human teacher will form a team. AI acts as an assistant to humans rather than replacing complete jobs. To do this you only need to succeed at the first two, the other helps but may not be necessary. Take a look at Amy.app if you are interested in this, I think they have the closest product to this on the market.
Hey Dan, what I’m really curious about is exactly what you stated in your last paragraph. How can they partner in effective ways is exactly what we should be thinking about.
Just about everyone in education thinks “If Education’s sole purpose is knowledge transfer, I want no part in it!” But I think most educators are in it for how can we help these kids become their best possible adult.
I don’t see AI achieving that goal, but I do see AI achieving the knowledge transfer goal.
So what can we do to help teachers and AI be the most effective, together?
I suspect the problems with ChatGPT's current form are a bit less transparent than most analyses, including this one, suggest. The trouble isn't that ChatGPT doesn't know how to solve problems correctly--it's that it knows both how to do it correctly, AND incorrectly. In your case, your prompts are all quite vague about what kind of answer you expect--do you want the correct answer or not? I entered your prompts and got more or less your answers. However, when I augment with a clearer description of my desired discourse (e.g., adding things like "this is a math equation", and "Please answer correctly"), I got correct answers.
Of course, this is if anything *more* challenging! How would learners know what prompts to give an AI so that it doesn't (deliberately?) mislead them by being wrong (albeit wrong in a compellingly humanlike way).
I agree. I don't think AI can ever replace teaching unless /you/ want it to. I think teaching that comes from the heart looks and feels very different than that which comes from a script and it's impact is probably more distinguishable in the long-term. I also agree that teaching is a conversation. AI can answer questions, but can it /ask/ the right ones to carry a conversation and produce learning? Finally, for me, having a teacher as a role model is huge and carries so much benefit no matter what stage you're in. AI can never replace that unless we redefine what teaching is.
I don't believe artificial intelligence can ever replace the human experience of teaching and learning. In the first example, there is no opportunity for a student to ask Why? It's strictly a memorization of steps without meaning. Could AI be developed to cover that pedagogical content knowledge? Maybe. But the caring aspect will never be realized by AI. When I think about the impact smartphones have had on decreasing human conversation and relationships, I think artificial intelligence would only lead to more of this lack of human connection.
I love the way you've posed this as a wondering. This is the first I've seen anyone talk about the social emotional learning parts of teaching. Personally, I believe that AI is far more likely to become tool that teachers (heavily?) use rather than replace the teacher entirely.
I take the point to be why teach what a computer can do faster and better. This includes algebra and has for years. Who needs a computer to tell me how I should do what I don't need to do. Let a computer be the automaton.
I think the most important distinction between AI and a human teacher is that students don’t ask complete and intelligible questions all the time. In your comments above, you posed questions that were formatted to what AI might expect. You gave the full equation in asking your question. However, it’s the questions of students who are unclear and don’t understand the conventions of mathematics, or possibly the conventions of English, whose questions are likely to pose a problem for AI. If a student tries to ask, “Why do you plus the 3 and not minus the 3?” then the bot doesn’t have the context to answer the question. Human teachers, on the other hand, can use their eyes and experience to not only correct the student’s language gently, but also to know the context without having to ask. We can also see students’ writing and get an idea of where a student is making their mistakes to help us ask the right questions. Not that AI couldn’t evolve to read text or respond to visual stimuli, but I have to figure it will take more than a few years for the programming to be developed that would help a bot to read the writing of some students.
This is incredibly insightful, especially, "students don’t ask complete and intelligible questions all the time."
So much of what makes AI amazing right now is knowing how to craft great prompts, or ask good questions. Kids are inherently bad at this because they don’t know what they don’t know!
I think that in reality AI and human teacher will form a team. AI acts as an assistant to humans rather than replacing complete jobs. To do this you only need to succeed at the first two, the other helps but may not be necessary. Take a look at Amy.app if you are interested in this, I think they have the closest product to this on the market.
ChatGPT is a baby. In a year it will be unrecognizable.
Hey Dan, what I’m really curious about is exactly what you stated in your last paragraph. How can they partner in effective ways is exactly what we should be thinking about.
Just about everyone in education thinks “If Education’s sole purpose is knowledge transfer, I want no part in it!” But I think most educators are in it for how can we help these kids become their best possible adult.
I don’t see AI achieving that goal, but I do see AI achieving the knowledge transfer goal.
So what can we do to help teachers and AI be the most effective, together?
I suspect the problems with ChatGPT's current form are a bit less transparent than most analyses, including this one, suggest. The trouble isn't that ChatGPT doesn't know how to solve problems correctly--it's that it knows both how to do it correctly, AND incorrectly. In your case, your prompts are all quite vague about what kind of answer you expect--do you want the correct answer or not? I entered your prompts and got more or less your answers. However, when I augment with a clearer description of my desired discourse (e.g., adding things like "this is a math equation", and "Please answer correctly"), I got correct answers.
Of course, this is if anything *more* challenging! How would learners know what prompts to give an AI so that it doesn't (deliberately?) mislead them by being wrong (albeit wrong in a compellingly humanlike way).
I agree. I don't think AI can ever replace teaching unless /you/ want it to. I think teaching that comes from the heart looks and feels very different than that which comes from a script and it's impact is probably more distinguishable in the long-term. I also agree that teaching is a conversation. AI can answer questions, but can it /ask/ the right ones to carry a conversation and produce learning? Finally, for me, having a teacher as a role model is huge and carries so much benefit no matter what stage you're in. AI can never replace that unless we redefine what teaching is.
I don't believe artificial intelligence can ever replace the human experience of teaching and learning. In the first example, there is no opportunity for a student to ask Why? It's strictly a memorization of steps without meaning. Could AI be developed to cover that pedagogical content knowledge? Maybe. But the caring aspect will never be realized by AI. When I think about the impact smartphones have had on decreasing human conversation and relationships, I think artificial intelligence would only lead to more of this lack of human connection.
Hooking up ChatGPT to Wolfram Alpha can help with proposal 1. I see this as the "easiest" of your three hurdles, but still enormously difficult.
I love the way you've posed this as a wondering. This is the first I've seen anyone talk about the social emotional learning parts of teaching. Personally, I believe that AI is far more likely to become tool that teachers (heavily?) use rather than replace the teacher entirely.
I take the point to be why teach what a computer can do faster and better. This includes algebra and has for years. Who needs a computer to tell me how I should do what I don't need to do. Let a computer be the automaton.