Robot teachers arrive at ASU
Is ChatGPT the future of tutoring?
In the field of education, artificially intelligent chatbots are currently helping students master the age-old art of cheating. No matter what the question, ChatGPT can provide an answer.
But what if the chatbots were programmed to tutor students, instead of just providing answers?
This is the new frontier of educational technology.
Arizona State University recently announced a partnership with OpenAI, the first such educational partnership for the artificial intelligence research and deployment company.
Together, ASU and OpenAI are planning to build a chatbot to help English 101 students learn how to write better. They might also make a robot avatar to serve as a creative “study buddy” that could do things like write poems about biology concepts.
They are still brainstorming ways to use this new technology. From the ASU press release:
Starting in February, ASU will invite submissions from faculty and staff to implement the innovative uses of ChatGPT Enterprise. The three key areas of concentration include: enhancing student success, forging new avenues for innovative research and streamlining organizational processes.
The Khan Academy is also keen on artificially intelligent tutoring. In a TED Talk given last year, Sal Khan boasted that robot tutors will lead to the “biggest positive transformation that education has ever seen.”
The Khan Academy chatbot tutor is called “Khanmigo.”
Khanmigo will “unlock” literature and history for students, according to Khan, because it can imitate characters. If a student is reading The Great Gatsby, and the student wants to know the significance of the green light in the story, the student can ask Khanmigo to answer the question in a Gatsby voice.
In other words, Khanmigo will summarize Sparknotes when prompted.
Sal Khan also says that AI will serve as an academic counselor, a career coach, and a life coach.
Color me skeptical.
My first reaction to hype around a new educational technology is to think we don’t need it.
Here is an argument against the inevitable tide.
Has 1:1 technology improved the quality of teaching and learning? Are students better educated today than they were in the days of chalkboards and physical textbooks? If they are better educated, is it because of laptops and digital curriculum?
Socrates didn’t have an iPad, he was a pretty good teacher. The Enlightenment happened without the world wide web. Sure, there’s a vast array of information and resources available in our digital age, but even this factor is a double-edged sword, because students are far more likely to get lost watching TikTok videos than they are to get lost in the dialogues of Plato.
No doubt, there are useful tools out there that can make learning fun and bring concepts to life. I use digital tools all the time as a teacher. It’s convenient. I suppose every generation learns with the tools available to them.
Fancy tools don't always enhance the learning experience, though. ASU currently uses a computer program called ALEKS for certain math classes. The program requires students to correctly answer five questions in a row to move to the next skill. If you miss even a decimal point on the fifth question, it will erase your progress and direct you to review the digital lesson. This can be extremely frustrating for some students.
Even if the stats say this computer program is more efficient, there is a psychic cost to training humans as if they are machines.
The conceit of the AI entrepreneurs is that they can capture the effect of a Socrates and turn him into a handheld machine that everybody can experience. That’s the dream.
The reality is that students will know they are talking to a robot. They will be staring into a screen, typing questions, and a text avatar will respond instantly. The answers will be beneficial, but nothing extraordinary. It will seem fun and motivational at first, but the novelty will wear off.
The functionality of ChatGPT is undeniable. Robot writing tutors will be able to correct grammatical mistakes, offer suggestions to make writing more structurally sound, and provide feedback for stylistic improvement. ChatGPT does an excellent job with textual analysis — it can explain the meaning of complex texts, summarize points from longer texts, etc.
But AI suggestions are drawn from an algorithmic pattern, not from a human being with an aesthetic sense of what makes writing worth reading.
I have a Wordpress blog in addition to this Substack. I haven’t posted on Wordpress for a long time, but I wrote a post there recently and I noticed a button for AI feedback. Click the button, and a robot gives you feedback and analysis of your writing. I clicked the button probably 15 times, and each time it provided me with a different analysis, offering different suggestions. Sometimes a new suggestion would contradict the analysis of a previous suggestion.
A robot tutor won’t necessarily be coherent. And while it might help students write in a more technically sound manner, it won’t help students write better in the way human beings evaluate writing.
Maybe the goal isn’t to get students to write better in terms of human aesthetics. Maybe the goal is to help students get through school. In which case, why not let ChatGPT write the whole thing and be done with it?
I’m open to the possibility that robot tutors can help students who don’t have access to a human tutor. But it takes self-motivation to learn in an isolated online setting. For a student who has the self-motivation to use an online tutor, there already exists YouTube videos of human teachers explaining pretty much any concept there is to learn. Granted, robots can offer individualized feedback and answer specific questions.
I don’t think robot tutors will be effective or desirable at scale in the way the AI entrepreneurs imagine. If a college student is struggling with writing, or biology, the best remedy is a real tutor who is a grad assistant or something — the way it has been done in the past. Human tutoring in the traditional way can also build a sense of human community, if we still care about that.
Artificial intelligence will probably succeed in upending the field of education, one way or another.
As for me, I’ll be the Luddite arguing we’d be better off reading physical texts and sitting around in a group with other human beings discussing the texts. And when we write, we should get human feedback on our writing. (Otherwise, who are we writing for?)
One educational technology that I would definitely support: Give every student a reading tablet, similar to a Kindle. Something that is e-ink based, not back-lit like a computer screen. Create an open-source library of texts so that teachers can upload texts and assign them to student reading tablets. This would allow teachers to capitalize on the information age while minimizing distractions.
Education is a human endeavor. While we can’t turn our backs on technological progress, I hope we don’t outsource our minds for the sake of novelty and convenience.
Links and News
A story that is not about artificial intelligence, but it made me think about artificial intelligence: Kari Lake and the tale of the secretly tape-recorded conversation that was leaked to the news for political purposes. The audio turned out to be real, but when it was first circulating across the internet, part of me was wondering if it might be an AI hoax. Artificial intelligence is already capable of producing convincing fake audio. This fake audio tape of Joe Biden was used as a robocall during the New Hampshire primary to try to manipulate voters. Strange times.
What We Might Mean by "Liberal Bias": Freddie deBoer critiques an example of culture war reporting in the New York Times. In another piece of media criticism, Freddie writes a scathing but entertaining response to the Atlantic-inspired moral panic about Substack.
Parker Molloy’s fantasy football league newsfeed caused her to become worried about the potential of AI-generated journalism.
A case study for why open primaries would be good for moderates, by Abe Kwok in the Arizona Republic
What can Democrats do to build a broader coalition for 2024? A podcast episode and transcript with Ruy Teixeira for Persuasion. Teixeira continues to plead for Democrats to talk more plainly to working-class voters.
Is U.S. industrial policy learning from its mistakes? Lots of money has been allocated by the government for projects to build, but due to the slog of bureaucratic inefficiency, the building part is slow going, writes Noah Smith.
The Messy Fight Over the SAT: Interesting analysis by “The Daily” podcast. Some experts are second-guessing the backlash against the SAT.
A great book on this topic is "Failure to Disrupt" by Justin Reich. It summarizes how some ed tech worked really well in some cases, but there was no ed tech that radically changed the nature of education. The book summarizes a lot of hype cycles that fizzled out for various reasons, but a consistent problem in ed tech is the desire to make large-scale changes to teaching and learning (either removing teachers from the process and/or casting teachers as the enemy) while making tremendous profits.
The more charitable read of some ed tech development is the need to lower overall costs as state funding declines. ASU has a lot of money, but funding from the state is not a big part. Michael Crow, President of ASU, claims the university could still function if the state reduced its funding to nearly zero. ALEKS works, in part, to reduce the need to hire qualified math faculty and reduce costs. AI development is really expensive and not likely to be a cost-saver right away, but it could allow ASU another revenue stream and make it even less dependent on state funding.