Original Luna bund education
highlight
Recently, Harvard University announced that starting from the fall semester, the self-developed language model CS50 Bot will be used as the teaching assistant of the popular course Introduction to Computer Science, so the new students will become the "first batch to eat crabs". This move has caused heated discussion on the Internet, and some people worry that AI will undermine students’ motivation to learn; Professor Harvard said that AI teaching assistants can alleviate the real teaching dilemma and improve the learning experience.
Wen, Luna, Jennifer
Since last year, the popularity of the AI topic has not diminished. However, it is still surprising to see that Harvard will "hire" AI in the ace elective course: "Has the application of AI evolved so fast?"
The news came from Harvard Crimson. David Malan, the lecturer of CS50 (Introduction to Computer Science), confirmed that this course will start from the autumn semester, and AI will serve students with 50 Course Assistant and Teaching Fellow (TF is a more advanced teaching assistant than CA).
![](http://21545.cn/wp-content/uploads/2022/07/fQA9s8n2.jpg)
![](http://21545.cn/wp-content/uploads/2022/07/8Gq3WG50.jpg)
The headline of Harvard Crimson &Malan’s Twitter response
The AI used this time is not the familiar GPT-3.5 or GPT-4, but a language model CS50 Bot developed by Harvard itself. It will be responsible for answering some common questions in the course, giving suggestions for revising students’ homework, and answering students’ questions at any time outside the working hours of other teaching assistants.
The CS50 course, in which AI is a teaching assistant, is a popular course at Harvard. In the past decade, the number of students enrolled in the course ranges from 500 to 800 each year. There are even students from other majors who have switched to computer science or engaged in related work in this major after finishing classes.
![](http://21545.cn/wp-content/uploads/2022/07/fCnDprF9.jpg)
CS50 Statistics on the number of students who choose courses every year
Such a popular course, when Harvard has not launched a complete educational application scheme for AI, directly introduces AI teaching assistants, which makes people sigh, "How dare Harvard!"
The hot discussion on the Internet also has different ideas:
Some people think that the function of AI is too powerful, and worry that this is the beginning of AI replacing human beings?
Some people also question the ability of AI, and ask Harvard that the annual tuition fee is as high as $50,000. Is it fooling students with AI?
Some people also worry that AI will make students no longer motivated to learn, and even completely subvert human cognitive development.
So, as one of the representatives of the top universities in the United States, what does Harvard applied AI assistant think? What kind of thinking does Harvard have in the face of all kinds of doubts about the AI crisis?
![](http://21545.cn/wp-content/uploads/2022/07/5va1gya3.jpg)
Harvard’s hottest CS50 also has bad reviews.
AI teaching assistants are here to "put out the fire"
Will AI replace human teachers? This is the first reaction of many people after seeing Harvard news.
However, if you look through the report in Harvard Crimson, you will find that CS50 Bot, the teaching assistant, is more like "fighting fire" than replacing human teaching assistants and professors.
On the one hand, Malan, the lecturer, has always positioned CS50 Bot as a learning tool for students, and never mentioned that Bot should be the lecturer. Even as a teaching assistant, Bot is "inferior", and its advice can only be considered "accurate" after being confirmed by other teaching assistants.
On the other hand, Malan urgently hopes that AI can improve students’ learning experience, so that this big class of nearly 1,000 students can also provide 24-hour one-on-one tutoring. In the past, this was an impossible task.
![](http://21545.cn/wp-content/uploads/2022/07/CTa7l943.jpg)
Professor Malan gave a lecture on Introduction to Computer Science.
Photography alana Steinberg
Malan is not only planning a large-scale AI test, but also considering the dilemma of real teaching.
In many people’s impression, Harvard’s ace course has its own aura-in the YouTube recording of CS50, David Malan stood on the spacious stage of Sanders Theatre and explained those complex computer concepts like a drama.
If you are lucky enough to study at Harvard, you will also know that this course also has a huge teaching assistant group to provide reference materials and homework guidance for students.
But in March this year, another report in Harvard Crimson, Undervalued, but Highly Utilized, revealed the bitterness behind the brilliance of these large-scale introductory courses.
![](http://21545.cn/wp-content/uploads/2022/07/nwxUL8R1.jpg)
The Harvard Crimson Report on Teaching Assistants
"Professor Malan gave a wonderful lecture on the stage, but other studies depend on teaching assistants." Fernandes, a former CS50 teaching assistant, expressed his opinion. "In essence, it is the teaching assistant who propped up this course."
Managing the professor’s office hours, leading small tutoring and attending weekly staff meetings are still reasonable jobs for teaching assistants. Nowadays, they have to answer questions for a long time after class and prepare most of the course materials.
These burdens have far exceeded the provisions of the school agreement-teaching assistants can be paid up to 10 hours a week, and course assistants can only calculate up to 4 hours. According to this standard, Fernandes leaves only three hours for homework review every week. Even if there are dozens of people in the teaching assistant group, each person is too busy for three hours.
Leggett, the current teaching assistant of CS50, added a "slot": "The teaching assistant has to support the exhibition of CS50′ s works, but often the day after tomorrow is our own final exam of advanced computer science. The agreement never mentioned this time conflict, and the consequences can only be borne by ourselves."
In this case, CS50 is no exception. Introduction to probability and comprehensive introduction to life science are all the same.
![](http://21545.cn/wp-content/uploads/2022/07/Rg8zbxbw.jpg)
In addition to the dissatisfaction of the student teaching assistants, the students of elective courses also suffered a lot of "dumb losses".
Harvard’s admissions website reads, "As world-class educators, professors work side by side with students to encourage them to look at problems with new eyes." However, many Harvard students feel that they are just the audience who listen to the professor’s lectures. If you really want to say that students work side by side, it is still the teaching assistant.
"Although it is indeed possible for students to have direct contact with some of the greatest scholars in the world, this opportunity is much smaller than it seems." "Harvard Crimson" commented.
What’s more, teaching assistants are students after all. No matter how much they love their subjects, there is a real gap between them and professors in knowledge, thinking and teaching.
Some students who have taken the introductory biology course complain that the teaching assistant knows a lot, but the level of counseling is worrying, and she can’t paraphrase complex concepts into expressions that she can understand. Finally, she is left to recite and memorize.
There are also teaching assistants who are limited by knowledge storage and understanding, and their opinions are contradictory to those of professors, which is simply worse for students who have doubts.
![](http://21545.cn/wp-content/uploads/2022/07/yN9hFZ65.jpg)
Kyle Waldman, who is studying for a doctorate, said that unless it is a small class or the students themselves are extroverted, it is "difficult for professors to interact with students continuously". Compared with smaller schools, Harvard students are "disadvantaged".
Students want more counseling, teaching assistants call for less work, and professors find it difficult to find a balance between research and teaching. The contradiction between students’ needs and the cost of manpower and time is becoming more and more prominent.
Since students cannot be banned from using AI,
It’s better to "turn an enemy into a friend" and take advantage of AI.
Malan asked AI teaching assistants to "put out the fire", so the way of using AI is also more "radical" among Harvard professors. But in fact, this is already slow.
Students’ hands and feet are faster than professors’ countermeasures. At the end of last year, just after ChatGPT came out of the circle, there were Harvard students driving ChatGPT and taking online open-book exams, and it was not a case.
It was not until late January that the email from the undergraduate education office arrived late, asking teachers to pay attention to the influence of ChatGPT, but no specific regulations were given, and the decision-making power was handed over to teachers.
Most professors in Harvard have a negative attitude towards the influence of AI technology, so many classes are forbidden to use ChatGPT, such as religion, Spanish and physics.
![](http://21545.cn/wp-content/uploads/2022/07/N86gkpQQ.jpg)
More teachers realize that it is impossible to ban it completely, so they try their best to reduce students’ chances of using ChatGPT and push them to learn by themselves.
Previously, the main assessment of French classes was done in the classroom, which restricted students from using Google Translate as a tool. After the popularity of ChatGPT, Amanda Gann, a lecturer, changed the final evaluation to "paper+dialogue", asking questions at any time to prevent students from writing a draft in advance with ChatGPT.
Gann admits that if it weren’t for ChatGPT, oral evaluation wouldn’t have such an important position as it is now. This also inspired her to reflect on her homework: "Is this something that only humans can do?" "How to make students not want to use AI to complete it from the beginning?"
As heroes think, so thought Bruce. Boaz Barak, a professor of computer science, also made homework and exams more difficult.
In the past, writing a simple neural network code was enough to complete an assignment; But now, students should reproduce a more complex code inspired by cutting-edge research. "Since students have access to more advanced tools, I naturally have to ask for more."
Professor Maria Dikcis of "English: The Dark Side of Big Data" went one step further and encouraged students to use ChatGPT critically. In her homework, she tried to make students compete with machine intelligence: the students wrote a short analysis article first, then generated a paper on the same topic with ChatGPT, and then compared the two articles. She hopes that students can see the shortcomings behind the brilliant appearance of AI and think about the significance and value of these shortcomings.
![](http://21545.cn/wp-content/uploads/2022/07/4B7vOS4J.jpg)
In contrast, the AI teaching assistants of CS50 have a feeling of "turning enemies into friends". Teachers are worried that students use AI because ChatGPT is so easy to use, so it is better to use this in turn and make AI a 24-hour online teaching assistant.
This idea is also in line with the personalized learning experience that Malan has been exploring. In order to keep students with different levels interested in CS50 all the time, Malan and teaching assistants designed assignments with different difficulties.
![](http://21545.cn/wp-content/uploads/2022/07/F0m71FeT.jpg)
Statistics of Students’ Learning Feeling in CS50 Course & Hierarchical Homework Design
For example, students with weak ability to design filters for pictures can write a black-and-white filter, while students with strong ability can challenge the filter of film negative+reverse color.
![](http://21545.cn/wp-content/uploads/2022/07/01WNjb4A.jpg)
Students of different levels can freely choose the difficulty of homework.
But the problem of students has been solved, and the time contradiction of teaching assistants still exists. They spend a lot of time commenting on homework. It takes less than 14 seconds for students to read the comments. After revision, it takes a long time to get a reply. "This is not an efficient way for teaching assistants and students to use their time." Malan said.
And the convenience of AI is like the missing piece of puzzle. Therefore, Malan makes AI teaching assistants undertake a lot of procedural and repetitive work. Although the revised opinions of AI teaching assistants still smell like robots, they hit the nail on the head to find the problem and give a detailed explanation, but they also got the evaluation of "weirdly informative".
![](http://21545.cn/wp-content/uploads/2022/07/pi51N9h2.jpg)
CS50 Bot under test
AI teaching assistant is a large-scale technical test, and it also shows students how to use AI reasonably. After that, Malan also hopes to integrate AI technology into other STEM classes and even humanities classes, bringing some changes to teaching.
Be the master of learning or the puppet of technology?
AI is a sword of Damocles.
However, the change of teaching methods still seems to be unable to solve a question-AI is coming, can human beings hold their ground and not be replaced?
The experience of Steven Pinker, a psychology professor at Harvard University, is a very intuitive answer:
As the speaker of the introductory psychology course, he was once worried that ChatGPT was strong enough to take his place until he discovered the cognitive difference between AI and human beings.
"I asked ChatGPT,’ If Mabel is alive at 9 am and 5 pm, is she still alive at noon?’ We all know that once a person dies, he can’t be resurrected, but ChatGPT replied,’ There is no information to clearly indicate whether Mabel is still alive at noon.’ It can be seen that AI does not grasp the basic facts of the world operation and lacks common sense and logic on which human beings depend. "
"Human memory and computing power are limited, and it must be admitted that AI is stronger in retrieving a large amount of information." Pinker said, but AI’s "cognition" comes from the collection of massive information, and it has not developed logical ability, which is completely different from human beings.
![](http://21545.cn/wp-content/uploads/2022/07/5q2L23Ii.jpg)
Steven Pinker said in an interview with Harvard Gazette.
It can be said that thinking ability is a unique advantage of human beings, and logical ability is only one of them. Whether students can exercise various thinking abilities in their study determines the result of "confrontation" with AI. On this issue, Harvard professors also emphasized the weaknesses of AI thinking ability.
Lack of critical thinking is another obvious weakness of AI at present.
The MIT Science and Technology Review issued an early reminder that because there are a lot of biased and hateful languages on the Internet, AI as a training set will further amplify these prejudices.
In addition, generating false content has always been an obvious problem for AI. In March of this year, experts from NewsGuard found that GPT-4 responded to false news 100%, and GPT-3.5 could reply to at least 20% of false news that "these statements are groundless".
![](http://21545.cn/wp-content/uploads/2022/07/yDK6MPLY.jpg)
NewsGuard’s report on GPT-4
In addition, GPT-4 is less "good" at programming than before, and the introduction of AI teaching assistants on CS50 has also caused some doubts. Malan himself predicted that AI teaching assistants might "occasionally perform poorly or even make mistakes" at the beginning.
However, he believes that this is precisely the opportunity for students to exercise their thinking ability. He tells students that "when receiving information, whether it is from AI or human beings, we must always think critically. Identify the authenticity of the information, evaluate the credibility, and then decide whether to accept the other party’s information. "
The ability of reasoning and thinking is also the weakness of AI, although it is very confusing.
Some students use ChatGPT to summarize the general idea of reading materials, and its word-for-word output makes people feel that it is really thinking. But is this really the case?
Soroush Saghafian, an associate professor of public policy at Harvard’s Kenny School of Politics, called this a flaw in AI. "People think that AI can think, reason, and even think that it is omnipotent, which is the result of commercial speculation."
Frankle, the chief scientist of MosaicML, an artificial intelligence startup, also verified this statement: "The purpose of developing AI by startups has always been to build a great language model product, rather than publishing and sharing knowledge, which is only an incidental effect."
Saghafian believes that students should actively interact with ChatGPT, but also understand the disadvantages of AI itself. "At that time in 2000, teachers taught students to understand the shortcomings of search engines like Google in thinking, and now they need to treat ChatGPT as well. Students should understand the difference between reasoning and simply outputting content. "
Where does human thinking ability come from? It must be in the process of learning. However, when high scores and GPA are especially needed, students will inevitably use ChatGPT to be lazy. Even Harvard professors have limited means to use, and the decision to study is ultimately in the hands of students.
![](http://21545.cn/wp-content/uploads/2022/07/Jss03cMX.jpg)
Nevertheless, Harvard University and professors try their best to inspire students:
Harvard calls the meaning of learning the value of intellectual transformation.
"It will be difficult to change yourself. To learn the principles of quantum mechanics or to understand the society before the industrial revolution, we need to deconstruct and reconstruct our world view. Therefore, Harvard University not only emphasizes academic ethics, but also expects students to experience this difficult process. "
John Hamilton, a professor of German and Comparative Literature at Harvard, said something more profound:
Technology has broken all kinds of restrictions and made our life easier and more convenient. We can fly across the world, scan literacy maps, or search keywords in Google to learn anything. However, what can give us meaning in life is limitation.
If there is enough gold, we won’t care about it; If we can live forever, we won’t care too much about life. Our concern comes from our limited ability. Life without limits will eventually be worthless.
References:
1.CS50 Will Integrate Artificial Intelligence Into Course Instruction
2.‘Undervalued, but Highly Utilized’: CAs, TFs, and the Quality of a Harvard Education
3.ChatGPT, Cheating, and the Future of Education
4.SIGCSE 2023 – Differentiating for Comfort with Computer Science
5.Harvard Taps AI to Help Teach Computer Science Course
6.How to make a chatbot that isn’t racist or sexist
7.Despite OpenAI’s Promises, the Company’s New AI Tool Produces
8.Misinformation More Frequently, and More Persuasively,than its Predecessor
9.Will ChatGPT supplant us as writers, thinkers?
Original title: "The ace course asks AI to be a teaching assistant. After reading this wave of operations at Harvard, I found that AI is a blessing or a curse, only in a moment."
Read the original text