Talking with computers: How chatbots can help with learning

Credit: Steve Rainwater

This article was originally published on Jeremy Riel’s Medium blog

The recent rise of chatbots and their educational uses

Followers of tech trends may have noticed the word “chatbot” appearing in an increasing number of news publications over the last few months. Chatbots, or automated software for chatting with a computer, have been around since at least the 1960s. With the recent rise of popularity in chat apps among youth, however, they have reached a newfound fad status in the last year or so. But what is it about these talkative software-based robots that has people excited and companies and organizations scrambling to set up their own chatbots? More importantly, what potential do they have for learning? In this post, I tackle what defines a chatbot, what they can help people do, and how educators can think pedagogically about this old technology that’s gotten a lot of new interest.Chatbots have found their way into discussions around silicon valley and educators circles alike. Advances in ubiquitous chat apps among teenagers have prompted a new way of thinking about how people interface with information and organizations. This has encouraged tech developers to bring about whole new generations of talking robots that you can chat with as if they are on your friends list. Businesses and organizations are using it, opening chat windows for customer service, help desks, FAQs, and for directly placing orders for things. Need a shirt of a certain color or a new pair of shoes? Just send a message to a chatbot at Nordstrom or Macy’s and they’ll help you right out! You can have CNN’s chatbot send you news whenever something interesting happens in the world (which, in the age of every-hour news, your feed could be flooded!). Want to know when you need an umbrella? There’s a chatbot for that. The White House even set up a chatbot to help field questions from the public, although the jury’s still out on how well their experiment met expectations.

You can even play games with chatbots. I recently added a chatbot called Streak Trivia to my Facebook Messenger friends list. This bot hosts a daily true/false trivia game for anyone who wants to play. It’s simple — the bot will ask you many true/false trivia questions. The person who has the longest streak of correct answers wins.

A new game of Streak Trivia is beginning — Are you in or out? You only have a couple minutes to respond!

A brief, long history of chatty software

There are a lot of new chatbots that have arrived on the scene in recent months, with new ones being created every week. But why are they so popular again, considering this is a technology that has been around since the 1960s?

Simply put, a chatbot is a robot that chats, usually in text messages. Chatbots don’t exist in the physical world unlike their more fleshed out cousins. Instead, they are software that offer pre-programmed or computer-generated responses to whatever users put it. They also tend to come with a lot of limitations, in that they can only do what they were designed to do. It’s a digital chat buddy that can only respond to a certain range of questions.

Chatbots have likely gained popularity again today because dialogue is a more natural interface for interacting with the world, and chat apps have gained a significant foothold in the daily interactions of most people with handheld digital devices. We, as humans, communicate through conversation on an almost daily basis. Instead of the clicks of a mouse, and even the touch of a tablet, a more intuitive interface may be the text and speech we use to interact with other people.

Chatbots aren’t anything new, but they may be enjoying the right conditions today in how people use technology to communicate for them to gain popularity and attention from developers. We’re seeing a resurgence in automated chatting due to the increased popularity in chat apps by the under-25 age group in recent years, toward perhaps a more intuitive interface for interaction: dialogue. It’s a buzzword that we’re starting to see pop up in education again as well, with educators and policy makers wondering how we can leverage these chatty robots to improve learning. But the history of chatbots show that it’s not an easy tech to get right. It’s tough to create a robot that can talk like a human.

Although this seems promising, it is a tough tool to make well. Since the 1960s, computer scientists have been working on ways to have computers understand and work with language in ways that make sense (a field called natural language processing). It is wonderful that chatbots are back in the news, as it represents some advancements in the field and a focus on their potential. However, they’re not so simple to put together, especially for education. As technology developers renew their interest in chatbots, it will be important for educators to critically examine the offerings that chatbots provide and to evaluate how chatbots can be used toward pedagogical goals.

One of my earliest experiences with a chat-like program was in the text-based video game Adventure, which incidentally is a great illustrative example for someone wanting to understand chatbots better. The game was written in the 70s, but I got my hands on it with my old DOS computer in the 80s. As a kid, I played the role of the adventurer who was exploring a dark cave. All actions of the game had to be input as text, and you received text in response from the game. There were no pictures — only words. Everything the system output was in response to what you put in as the player. If the system didn’t recognize your words, you had to repeat yourself in a different way or find a new set of keywords to give the system to progress in the game.

A screenshot of Adventure. All commands in Adventure have to be input with text. The computer is looking for specific words and phrases, very similar to how many chatbots today function.

You can play a working version of Adventure right in your browser — give it a try and see how far you can get!

Over the years, things started to get better for chatbots. Developers of chatbots tried to mimic human conversation better and programming got more sophisticated. The developer community is a pretty robust group, with the community competing annually for the coveted Loebner Prize for most human-like chatbot (have a conversation with “Mitsuku” to see a cool, functioning example of a recent Loebner winner). Each year, new approaches are invented and previous approaches are refined. Today, machine learning principles are increasingly used in advanced chatbots to allow the system to learn how humans interact with it, giving it better responses. Regardless of the approach, though, the premise remains the same: a user will ask the chatbot a question, or make a comment — and the chatbot will reply with a suitable response. If it didn’t understand, it will say it didn’t understand and ask the user to repeat or try saying their question in a different way.

Many modern chatbots still don’t know much about language, though. In a recent game of the Streak Trivia chatbot I mentioned above, I tried to input my commands in a way it didn’t understand (because my typing wasn’t exact). It prodded me to communicate with it in a more productive way with a touch of humor.

Streak Trivia’s identity as a microwave and not HAL 9000 is a subtle reminder that despite the advances in chatbots, we have nothing yet to worry about with chatbots taking over the world.

Thinking pedagogically about chatbots

Despite the constraints of chatbots, the education world quickly saw benefit in early chatbot technologies of the ’60s and ’70s. Educational technologists invented approaches like cognitive tutors and “intelligent” interaction systems in the early days of AI and chatbots to work with learners without needing the intensive time required for one-on-one coaching. Although these systems have a loyal following of researchers, the recent attention and buzzword status of chatbots might bring much-needed experiments and investment that the field needs to realize any educational potential. The newest generation of chatbots have some capabilities that can reach students in ways that may not have been achievable with past technologies.

  1. Personalization. The main selling point of chatbots for education is the personalization that one can get. You can have a conversation that is tailored to your specific questions. However, this really only works if we know students unique needs and communication styles. The potential for adaptation is huge, as has been seen in ed tech groups like Khan Academy and Knewton. But, as chatbots get made for education, it is important to make sure chatbots align with the same communications tools that students already use, or else they risk being put to the side of more favored apps. We should also make sure that the adaptive principles that are used in these technologies align with modern learning theories. Educators also need to make sure that adaptive chatbots for learning avoid lumping students into categories or clusters that may disadvantage them later on by preventing future learning — such as any kind of profiling a student as “low performing,” which could then lead to chatbots delivering students content that is geared toward this bias in the design. To maintain a commitment to equity in education, adaptive chatbots should maintain an openness in their design that allows students to grow and progress, and that avoids any kind of response pattern from the chatbot based on their performance or use of the chatbot.
  2. Monitoring and nudging. Although we lack good chatbots that are intended for education, some chatbots out there now show some of the possibilities for chatbots for learning. Even though a user might know that they are interacting with machines, recent chatbots have been shown to be effective monitors of behavior and can even nudge users in positive directions. By using language, users of chatbots can be reminded of goals and tasks in a more natural way than phone notifications or emails, and can be prompted to reflect on everyday life through the seemingly basic conversation with the chatbot. A couple of good examples of this have grown in popularity recently. A chatbot called Fitmeal tracks meals and food intake and has conversations with users to improve physical health. Another chatbot called Joy interacts with people to help improve mental health. Joy chats with users daily to help monitor how you feel, and works with you to help you make decisions that are beneficial to your mental well-being.
  3. A more human vibe. Even if we know it is a machine, a more natural dialogue style can foster personality when interacting with a machine. I’m just hypothesizing here as more studies in this area are certainly needed, but dialogue with machines in human ways can make the whole experience more pleasurable. Instead of the impersonal interaction one has with library catalogs, book indexes, and even course syllabi and frequently asked question lists, a personal, conversant machine may be more effective at communicating information and helping learners. It may also foster students’ personal inquiry by prompting students to ask questions and encourage engagement by using one of our species’ most original interfaces: interactive language.
  4. Instant help. Functionally, chatbots might be most helpful at providing information on demand, as they are always ready to talk. The robot is always at the other end of the line. Recently, a great example of this was demonstrated when a Georgia Tech professor “hired” Jill Watson, a chatbot, as his teaching assistant. Throughout the whole semester, students didn’t know their TA was a chatbot, but found her quite helpful nonetheless. She fielded hundreds of questions and responded with useful feedback. In fact, many students fell just short of nominating her for a TA teaching award before finding out her digital identity. Asking questions and getting help from a chatbot can be beneficial in other ways too, as some students may have anxiety about asking questions to teachers directly. In addition, instructors can review the chat history and sift through the more meaningful questions that the virtual TA can’t answer and address these questions with students. As such, it can be a source of questions for teachers to answer — questions that they may not have ever received in class.

Next steps for chatbots and education

Unfortunately, there aren’t a lot of chatbots for specific educational purposes to try yet. Despite all of the hype in the headlines, educators are still left wanting some rich interactive chat experiences that actually work for learning.

One reason for the dearth of educational chatbots is that programming a chatbot is tough. The most advanced chatbots today are specialized around certain purposes with complicated approaches like machine learning, and significant investment in development talent and content is made. We can see this in some of the more business and organizational chatbot applications. However, for the technology layperson educator, there are some tools out there to get started if you’d really like to. For this, I recommend Chat Fuel to create a chatbot — no coding necessary! It may not do everything you want it to, but it would be a good first step to testing the chatbot waters. The Facebook Developer resources also offers a bunch of resources for those who would like to set up a chat bot in its Messenger chat app, but, like most chat apps, require significant coding.

So, it appears educators must continue to play a waiting game for developers to find useful applications for chatbots for learning. It will also be important to maintain strong commitments to learning theories and design principles that are known to foster learning instead of designing experiences that are not beneficial — like pure rote memorization and drilling activities. These apps have their value in certain circumstances, but rote memorization apps and games have a long, ineffectual history that continues to plague the field of educational technologies. Like most technology in education, the field would be best served if educators themselves took to programming the apps we’d like to use. If you have an idea, give it a try!

What do you think?

Have you found and tried out any chatbots for learning? What are your thoughts on chatbots being used in education? — tweet me at @jeremyriel and let me know what’s on your mind or what’s worked for you. I’m also always on the lookout for cool apps and resources, so if you’d like to drop me a line about what you use related to this post, let me know.