It seems as if ChatGPT came out of nowhere and is now everywhere. Even for those who were familiar with AI, ChatGPT generated a wow moment in terms of how quickly it spread. In this blog, we discuss where ChatGPT came from, and what you as an educator should know about it to help your students.
While I refer to ChatGPT in this blog, nothing here is really about that particular product from OpenAI, I'm referring to any tool of its kind that is going to emerge. And we know for sure that others will emerge. We've already seen that a competitor has come out from Google, for example, called Bard.
Where did ChatGPT come from - an intro to Natural Language Processing
These tools are the most visible and advanced embodiment of a type of AI called Natural Language Processing (NLP), which is basically computer programs that can understand language. Natural Language Processing is used in digital assistants, assessing movie reviews, etc. Advanced NLP works by taking complete sentences, removing part of the sentence, and trying to see if it can predict what the missing part is. For example, if I were to say - I am very hungry, because I have not, and I left the rest of it blank, then the models might predict “eaten today” as a completion of the sentence. The full sentence would be - I'm very hungry because I haven't eaten today. To learn how to complete the sentence, the AI has to understand how words work and interact with one another.
What makes these AIs very cool is that they can learn from any sentences they find, and the internet is full of text content. So these models learn from massive piles of data on the internet. Because of that, they create sort of like a digested processed version of all the knowledge they have access to. This makes them very powerful, very cool, and very dangerous. The internet is filled with fascinating, correct, accidentally incorrect, and purposefully incorrect information.
How does ChatGPT work?
ChatGPT, is a proprietary product and we can't say exactly how it works, but we know generally how these tools work. Your first question might be: "But I don't give ChatGPT incomplete sentences?" It converts your question in the background to an incomplete sentence and then all the AI does is try to complete the sentence. Next, humans provide feedback on the completion and the AI learns further from that. You might read on the news that there are actually armies of humans around the world providing feedback to help these tools get better. Is this a good thing? Probably. Is it a great thing? Maybe not because what those humans believe may not match what you believe. And they may even have unconscious biases that the AI then learns.
How do ChatGPT detectors work?
There are tools that can detect ChatGPT - like this one. These tools learn by studying the patterns of text written by ChatGPT and text written by others (such as human sources). By learning the patterns of expression, the tools try to differentiate the two. This also can work on a per-human basis. For example - it is possible to train AI tools to tell which human wrote the text (see an example here of an AI built by an 8th grader to differentiate Shakepeare and other playwrights). However, when it comes to ChatGPT detection using methods that use these techniques, you should remember that this is a race between two AIs, ChatGPT will get better and the detectors will get better and they will always race. So - these tools are interesting and maybe useful but they are not guaranteed to work.
What is Prompt Engineering?
You may also have heard of a term called Prompt Engineering. Tools like ChatGPT are not exactly programmable, but since they work by completing sentences, the key to getting it to do what you want is to generate the best incomplete sentence. This is called Prompt Engineering - the incomplete sentence is the Prompt. Prompts can be anything from partial sentences to whole paragraphs or many examples before the last incomplete sentence. As these tools become more popular, we can even expect entire classes just on how to design good prompts.
What can we expect in the future?
First, you should expect a lot more of these tools. They will be very easy to use. They will be really fun and informative. Many times they're correct. But just like you expect of anything else on the internet, they're not always correct. And they can often be biased and they can often have opinions that may not match yours. They're also susceptible to being manipulated. The way you ask your question can have a lot to do with the answer. It is possible to trick these AIs by asking questions in tricky ways to get the AI to respond in all sorts of inappropriate ways.
What can we as educators do?
This will be a long journey, but here are a few things you can do immediately
The first is - these tools are going to be used by students, no matter what. They are too easy, too fun, and too informative. And, frequently, kids can learn a lot from them. However, they can be mistaken in all sorts of ways. I've noticed from my personal experiences that even when they are wrong, they can be very convincing and it takes an expert to figure out the difference between the response and the truth. If you can help the students understand how these tools learned what they know, it can help the students understand how to be appropriately careful.
On the cautionary side:
Help students understand what the rules are around ChatGPT in both your school and in standardized tests. For example - many schools have banned ChatGPT use for essay writing, and it would be unwise for students to use ChatGPT for submissions to competitions or other standard tests unless the test administration explicitly permits the use of these tools.
You should always double-check the answer. Many schools already help students understand what reputed sources are, and double-checking with reputed sources is still a very wise move.
Any time a student is engaging with these tools, they are essentially communicating with an unknown internet entity - they should know this and protect their privacy
On the positive side:
ChatGPT can be very helpful to engage students. For example, I was able to generate 100 questions on a course I am teaching, in the space of about five minutes. And I found the questions super useful. I was able to get it to generate three wrong answers and one correct answer for every question, again, in a few minutes. I double-checked those to make sure that the correct answer was in fact correct. And each wrong answer was, in fact, wrong!
I have also seen students become more engaged in their studies (from STEM to literature) by interacting with ChatGPT. For example - this is what it gave me when I asked for a poem on chemistry for 8th graders.
Chemistry is science so great,
It’s about the stuff that makes everything we create.
From the atoms that make up the air we breathe,
To the molecules that give water its ease.
There are elements that we know so well,
Like oxygen, carbon, and nitrogen as well.
They combine in ways that can be wild,
Making reactions that are sometimes mild.
When elements bond and stick together,
A compound forms that can last forever.
And if you’re brave enough to make a reaction,
Be careful, because it could cause a big chain reaction!
Chemistry is science so fine,
It’s all about the building blocks that make up our time.
So the next time you breathe or drink or eat,
Remember the chemistry that makes it all complete.
Taking action
In general - the more you know about ChatGPT and the more you can explain to your students how it will impact their life - the more likely they are to be successful at navigating it safely. I encourage my students to exercise their judgment, combine information from many sources and form their own opinion. It's a useful tool. It's but like any tool, it's not a replacement for their brain.
If you would like more information about ChatGPT and managing this topic with your students, please check out our writings on Forbes on this topic.
You can also find lots of resources on this website if you would like to explain AI to your students in general. You can start here.
If you would like to chat 1:1 and get more advice on your particular situation? Sign up for a free 15 min call here.
Comments