news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

"it is better to study computer science and mathematics than artificial intelligence as an undergraduate, so choose carefully!"

"artificial intelligence is a new pitfall, don't even touch it if you are below peking university or tsinghua university."

a few months ago, chen peng, a college entrance examination candidate from chongqing, wanted to apply for the artificial intelligence major at a non-key university in zhejiang. however, most of the responses he received on social media were suggestions to give up. many people who have experienced this said that the artificial intelligence major has just emerged, and it has its own hot spots and marketing attributes, and the quality of education in universities is uneven. if you don't make up your mind to pursue a doctorate, you may be unemployed after graduating from college, not to mention a "million-dollar annual salary."

since the first batch of 35 domestic universities opened artificial intelligence majors in 2018, artificial intelligence majors have begun to appear in large numbers. in 2019, 179 universities successfully applied, and 130 in 2020. in march this year, the ministry of education announced a new list of 38 universities that have registered artificial intelligence undergraduate majors, which will begin to enroll students in the fall of 2024. so far, more than 500 universities in china have been approved to open artificial intelligence majors. in addition to top universities, there are many "non-key" and technical colleges among these schools.

in august this year, the ministry of education's higher education department held a meeting in beijing to promote the "101 plan" in the field of artificial intelligence. the meeting established the "101 plan" construction committee in the field of artificial intelligence. the "101 plan" first identified 15 artificial intelligence core courses, 10 core course extensions, and 2 comprehensive experimental courses, aiming to promote excellent artificial intelligence undergraduate professional training programs and improve the quality of training for top innovative talents.

on the one hand, ai majors are hot in colleges and universities, but on the other hand, there are many problems faced in the training of students. what is the next step for ai majors? many experts interviewed said that what to learn in ai majors, how to teach, and what kind of talents to cultivate are still unanswered questions.

photo/visual china

popular among students

the enthusiasm for artificial intelligence in colleges and universities is obvious.

in 2022, xu li enrolled in a non-key first-class university in zhejiang province, majoring in computer science and technology. at that time, her college was called the school of computer and control engineering. just one year later, the college merged with the adjacent school of data science to become the current school of computer and big data. in the same year, xu li's school's undergraduate major in artificial intelligence was approved.

xu li told china newsweek that when she enrolled, except for students majoring in data science and big data technology, all other students, including herself, were enrolled in the computer major category, and the majors were not divided until the sophomore year. when she was in the sophomore year, the artificial intelligence major was opened, and the students around her signed up. xu li got excited and thought that artificial intelligence was an emerging major and sounded promising, so she "followed the crowd" and became the first student of the school's artificial intelligence undergraduate major.

for some non-computer science students, if they want to pursue ai wholeheartedly, they have to change their major. li wei is a 2020 undergraduate student at a non-first-class university in jiangxi. when he entered college, his major was metal materials engineering. in his sophomore year, he successfully applied to transfer to the major of electronic information engineering and chose the direction of artificial intelligence because he ranked high in his grade and participated in the national mathematical modeling competition for college students and was exposed to artificial intelligence algorithms.

wu fei is guiding graduate students in the laboratory of zhejiang university school of computer science and technology. photo provided by the interviewee

in 2020, li wei's school began to offer an undergraduate major in artificial intelligence, but only recruited students from the second-tier universities. when he changed his major, the number of transfers to the electronic information engineering major was 10. by the time he graduated this year, the number of transfers to the major had increased to 20, nearly doubling in size, and the vast majority of students transferred because of artificial intelligence.

chen peng finally chose the artificial intelligence major of his favorite school. he admitted that it was inevitable that he followed the trend. another important reason was that his college entrance examination score was just at the transfer line of the school's artificial intelligence major in previous years, and this score line was nearly 10 points higher than the transfer line of the school's computer major. "it would be a waste if the score was not used."

wu fei is a professor at the school of computer science and technology of zhejiang university and the director of the institute of artificial intelligence of the university. he noticed that in the past two years, the scores of artificial intelligence majors in many top universities in zhejiang province have exceeded those of traditional computer majors. the reason is simple. wu fei analyzed to china news weekly that the current consensus is that humans are entering the era of intelligent society, and "no ai, no intelligence." large language models represented by chatgpt demonstrate language interaction capabilities similar to those of humans. faced with a future where "if you don't know ai, you will be eliminated," students' "skill anxiety" is natural. more students with strong academic ability are flocking to artificial intelligence majors, and their scores are also rising.

behind the enthusiasm for studying is the country's high attention to artificial intelligence. in 2017, the state council issued the "new generation artificial intelligence development plan", which clarified the "three-step" strategic layout of artificial intelligence development by 2030. in 2021, the country's "14th five-year plan" listed artificial intelligence as the first of the eight frontier fields. local governments are also making plans around the intelligent application of key industries.

in 2022, the ministry of education will release the "catalogue of disciplines and majors for postgraduate education", and intelligent science and technology will officially become a first-level discipline in the category of interdisciplinary disciplines. at present, artificial intelligence has not yet been listed as a first-level discipline, but is listed under the catalogue of electronic information majors, and has many overlaps with the major of intelligent science and technology. many experts interviewed said that this is only temporary, and there is a possibility of integration between the two in the future. liu zhiyuan, associate professor of the department of computer science and technology at tsinghua university, told china news weekly that previously, majors such as software engineering and cyberspace security had been separated from computer majors and became first-level disciplines, and it was "only a matter of time" for artificial intelligence majors to become first-level disciplines. this can be seen in the application of projects. in 2018, the national natural science foundation of china established an independent application direction for the artificial intelligence major, with a first-level application code and 10 second-level codes, including different research branches.

the major is not yet completely "independent", which gives many students a feeling of "not knowing what they are learning". xu li regretted talking about her major. after studying for a year, she found that students majoring in artificial intelligence, information security, internet of things and big data basically learned the same things, covering basic computer courses, as well as a small number of advanced artificial intelligence courses such as machine learning and deep learning. most of the courses in the artificial intelligence major are pieced together from courses in automation, computer, and electrical engineering, and some of them have not even changed their course names. as a first-year student, xu li felt that she was "perfunctorily treated."

in july 2024, students in the southwest university artificial intelligence undergraduate summer course were listening to a foreign teacher teaching the introduction to signal processing course. photo/provided by the interviewee

china newsweek searched the training programs of many second- and third-tier colleges and universities and found that the training programs for their newly opened artificial intelligence majors are extremely similar to those of existing majors such as computer science, electronic information, and data science, with most of them differing by only one or two courses.

"computer science covers a wider range of areas. classic fields include system architecture, graphics drawing, etc. it does not simulate human intelligence, but rather presents and depicts real-world scenes. artificial intelligence requires the simulation of human behavior. although there are differences between the two, there is no need to divide them very clearly." wu fei said that artificial intelligence is an ultimate application form of computers and is closely related to computer science.

the training objectives of artificial intelligence majors vary with the positioning of different schools. wu fei described the training objectives of "double first-class" universities as "daring to enter the no-man's land", that is, to train first-class scientific and technological innovation talents. some universities will open artificial intelligence majors in existing colleges based on the application or technological innovation in a certain field. wu fei believes that students will "vote" for the major with their own performance. if students give poor feedback and their career paths are limited, the major will naturally die out. after the emergence of the internet, the network major also went through a similar process.

however, the question of "what to learn" in the field of artificial intelligence is not a perfunctory one. do more than 500 universities need a general solution for their training programs?

a variety of training programs

"the pressure is too great." xu li sighed.

since control engineering is a traditional advantage major in her school, the training program for artificial intelligence has also been tinged with the color of "control". in her sophomore year, she needed to complete traditional control engineering courses such as digital electronics, analog electronics, and automatic control, and a lot of time was taken up by different experiments every week. as a student majoring in artificial intelligence, she wants to go the route of algorithm engineering and development in the future, but she has no time to learn software knowledge. her confusion is that while taking into account professional courses, she also has to learn software to a professional level, otherwise it will be difficult to have a career. is such a training program really reasonable?

zhao yan has a completely different experience. she is a 2020 undergraduate student majoring in artificial intelligence at a "double first-class" university in sichuan, and is also the first student of the major. during her four years in college, she studied almost only algorithms, such as machine learning, image processing and analysis, computer vision, natural language processing, etc., and never took any hardware courses.

"if you want to make a breakthrough in the field of artificial intelligence, you have to focus on both software and hardware." chen feng, vice dean of the school of artificial intelligence at southwest university, has taught undergraduate courses such as information theory, machine learning, and computer vision. he told china news weekly that the school established the institute of artificial intelligence as early as 1993 and established the school of artificial intelligence in 2018. before the official enrollment of undergraduate students in artificial intelligence in 2022, the school's majors such as electronic information, data science and big data technology have many intersections with artificial intelligence in professional courses. in addition to classic computer basic courses and artificial intelligence algorithm courses, the college also offers courses such as robot control and analog electronic technology, striving to apply algorithms to practice and combine software and hardware.

the training program is also changing dynamically. chen feng said that the school will consider industry applications more in the course setting, and modify the training program and teaching outline according to the actual situation. in addition to experts and scholars, the participants also include senior engineers from well-known companies in the industry. since the artificial intelligence major was officially approved in 2019, the training program has been overhauled several times, reducing the total class hours and increasing the proportion of practical courses.

during the "tiangong cup" graduate student innovation experiment competition of nanjing university of aeronautics and astronautics, participating students demonstrated the "passive power exoskeleton robot" (left) and the "facial muscle rehabilitation training and evaluation system based on wearable sensors" (right). photo by yang bo, a reporter of this magazine

in 2022, the institute of cross-disciplinary information sciences of tsinghua university announced the merger of the computer science experimental class (the "yao class"), the artificial intelligence class (the "intelligent class") and the quantum information class opened by the dean of the institute, yao qizhi, and they are collectively referred to as the computer science experimental class. liu zhiyuan said that the merged classes basically use the core computer courses of the "yao class" in the freshman and sophomore years. although tsinghua university established the school of artificial intelligence in april this year, the school currently only recruits graduate students in artificial intelligence. "you need to lay a broad disciplinary foundation in the undergraduate stage, because you don't know in which field innovation will occur, especially in interdisciplinary subjects," said liu zhiyuan.

in order to enhance the basic scientific research and innovation capabilities in the field of computer science in china, at the end of 2021, the ministry of education officially launched the pilot work plan for undergraduate education and teaching reform in the field of computer science (the "101 plan"), aiming to build 12 first-class core courses such as "introduction to artificial intelligence" and "software engineering". all core courses are self-declared by 33 universities with outstanding undergraduate computer majors, and the corresponding responsible universities are finally determined. wu fei is the person in charge of the "introduction to artificial intelligence" course. he pointed out that the undergraduate majors in artificial intelligence in more than 500 universities across the country have different course outlines, and the quality of students trained is also uneven. therefore, there is an urgent need for a national-level teaching syllabus to unify the teaching content, which is the so-called "general solution."

when constructing the core course "introduction to artificial intelligence", wu fei and his team referred to the training system developed by academic organizations such as the american computer society. since purdue university established the world's first undergraduate computer major in 1962, the american computer society has updated its training system every 10 years. artificial intelligence has never been absent and has always been a core course. the team also combined the characteristics of the new generation of artificial intelligence and widely collected expert opinions, and finally formed a course system with 10 modules and 63 knowledge points, expanding from basic theory, technical means to system composition, and then to application and ethics.

chen feng sees the learning process of the major as a line, and the contents of related courses such as machine learning and deep learning are like small pearls. if students do not fully understand them from the beginning, it is difficult to make the pearls into beautiful handicrafts. only by having a sufficient grasp of each technical link can one naturally realize in practice which piece of knowledge can better solve the problem. how beautiful the final necklace is depends on how many pearls are truly learned, understood, and applied.

at present, all 12 courses of the "101 project" in the computer field have been completed. the core textbook "introduction to artificial intelligence" compiled by wu fei's team was also published in june this year and officially put into use, filling the gap in the previous systematic introductory textbooks on artificial intelligence. however, wu fei pointed out that not all universities have to teach according to this textbook. in addition, the so-called "general solution" regulates "what to learn", and as for the question of "how to teach", universities have the right to choose by themselves, and teachers can write their own handouts based on existing textbooks and their own understanding.

just as the introduction has set off a fire, “general solution” is only the first step towards standardized training. wu fei emphasized that other courses in the undergraduate major of artificial intelligence have not yet been unified and are still under exploration and construction. this needs to be completed by the aforementioned “101 plan” in the field of artificial intelligence.

teachers “learn and apply”

"colleges and universities have a wide variety of answers to the question 'how to teach,'" meng yu told china newsweek.

meng yu graduated from a "double first-class" university with a major in information engineering. he then went to australia to study computer vision. after graduating with a doctorate, he returned to china and has been teaching computer vision theory at the school of information of a "double non-first-class" university in fujian. computer vision is an application field of artificial intelligence. he frankly said that only the top few top computer schools have the supporting faculty and excellent students' academic ability to support the study of artificial intelligence undergraduate majors, and students should choose carefully when applying.

in his opinion, the core of ai is still computer science and mathematics. for ordinary students, it is very difficult to master one of them in four years of college. to master ai, students need to go beyond the mathematical foundation of ordinary engineering students and undergo repeated training in computer engineering, ai algorithms, application scenarios, etc. it is very difficult to complete the above training in undergraduate studies. it is easy to give up at a superficial level and end up with nothing. this is also the reason why most ai students are confused.

during the "tiangong cup" graduate student innovation experiment competition at nanjing university of aeronautics and astronautics, participating students demonstrated the functions of the "multi-joint continuum robot". photo by yang bo, our reporter

confusion was the norm for li wei during his undergraduate years. the machine learning course offered in his junior year was a discussion-based course with no practical experience, and li wei felt that he had no sense of achievement. he found that there was no course specifically teaching the programming language python in the training program, and only one class in the machine learning course taught some basic knowledge. without programming knowledge, subsequent courses such as deep learning and natural language processing will become increasingly difficult. so, he began to teach himself python in his junior year. "fortunately, knowledge in fields such as machine learning is highly open source. there are many open courses taught by professors from famous universities on the internet, which are much better than college courses." li wei said.

many students told china newsweek that the low level of teachers is an important factor that limits their sense of achievement in class. xu li said that since the teachers of artificial intelligence courses are basically part-time computer majors, many of them have not received professional training in artificial intelligence, they need to go to universities in beijing and other places for training every now and then. "if they have learned, they will come back to teach. if they can't learn, they will directly play teaching videos in class and let students learn on their own."

after chatgpt became popular, meng yu was asked by the college to add relevant content to the course. he had been exposed to knowledge of generative models during his graduate studies, but not much, and it was in two directions with computer vision. although he could understand the basic principles, he was not confident in teaching, just like "middle school teachers teaching advanced mathematics". it's not that the teachers themselves have never learned advanced mathematics knowledge, but that they can't know how to teach it immediately after understanding a new thing. "the latter requires a long process of digestion." meng yu said.

the field of artificial intelligence is changing with each passing day, and it is common for teachers to "learn and apply" at the same time. chen feng gave an example, saying that around 2010, after the advent of a wave of artificial intelligence brought about by deep learning, voice recognition and face recognition quickly occupied major mobile phone applications. however, this level of recognition cannot support complex semantic understanding. for example, navigation can understand "go to zhongguancun", but cannot understand "go to a restaurant that is most worth visiting in beijing." the latter became a reality after the rise of chatgpt in 2022. this means that compared with students, the artificial intelligence major poses greater challenges to teachers.

during the 2022 winter olympics curling man-machine competition, the intelligent curling robot independently developed by harbin institute of technology competed with professional curling athletes. photo by our reporter jiang hui

as big models quickly became the top priority for ai talent training at home and abroad, college teachers are also constantly adjusting course content. liu zhiyuan teaches a natural language processing course at tsinghua university's department of computer science, an english course for graduate students in the department. as soon as chatgpt became popular, liu zhiyuan began to tilt the entire course's teaching content toward big models. he pointed out that the professional core courses in the training program are relatively stable, but cutting-edge dynamics need to be quickly reflected in the course content.

in addition to the improvement of teachers themselves, the horizons of teachers should also be broadened. while introducing national talents and young teachers, the school of artificial intelligence of southwest university also introduced dual-qualified teachers with industrial experience to better face industrial applications and cultivate students' practical ability.

since artificial intelligence requires a combination of software and hardware, the practical platform provided by the college for students is also crucial. liu zhiyuan's computer department has a classic compulsory course called computer organization principles. the course assignment requires "work hard for 20 days and build a computer." this course has a special experimental table for students to build computers by hand. zhejiang university released a new generation of "zhihai platform" this year. wu fei introduced that this is a practical platform for various artificial intelligence learning scenarios, including many open source codes and open cases, which can enable students to independently implement scenario programming.

the same practice may become a mere show in another school. when li wei participated in the national college mathematical model competition, he did simple flower, cat, dog, and face recognition. the teacher in charge of the team explained the principle vaguely, saying "just run the code and it will be fine."

employment is far below expectations

li wei did not participate in campus recruitment in his senior year, and chose to take the postgraduate entrance examination directly. previous students told him that ai-related positions only recruit graduates with a master's degree or above, and "non-key" undergraduates "won't even look at them." although classmates in the same major found jobs relatively smoothly, their positions were not very related to ai. many classmates who found jobs in the province only had a monthly salary of 5,000-7,000 yuan, which was far below li wei's expectations.

li wei passed the postgraduate entrance examination this year and will go to a "double first-class" university in hubei to study for a master's degree in electronic information. he recalled that during the re-examination, he left a good impression on the interviewer because of his self-taught artificial intelligence algorithms and competition experience. after the master's degree, li wei did not want to continue his studies and hoped to find a job as an algorithm engineer because he heard that the salary was very high. in his hometown of anhui, the minimum annual salary for a master's degree graduate is 400,000 yuan.

in november last year, the "2023 artificial intelligence talent insight report" released by the job search and recruitment platform maimai showed that from january to august 2023, the talent supply-demand ratio in the artificial intelligence industry was 0.39, equivalent to 5 positions competing for 2 talents, and algorithm talents were in the greatest demand. however, data from the job search and recruitment platform liepin showed that in the first quarter of this year, the talent supply-demand ratio for large model algorithm positions was only 0.17, equivalent to 6 positions competing for 1 talent.

in liu zhiyuan's view, artificial intelligence will definitely trigger the intelligent transformation of all walks of life in the future, so there will be a huge gap in artificial intelligence talents. from a national perspective, it is necessary to set up artificial intelligence majors. artificial intelligence is similar to computers. it requires both high-level innovation and practical talents who can quickly realize applications.

in addition, the addition of large models will widen the gap between talents. liu zhiyuan analyzed that under the same programming ability, people with strong learning ability and willingness to actively embrace ai tools will definitely be more competitive. large models make computers smarter, which means that high-end practitioners will increase in value, while low-end practitioners will depreciate.

in september 2020, 30 high-level underwater robot teams from domestic and foreign universities, scientific and technological institutions, and enterprises participated in the national underwater robot competition in dalian, liaoning. photo/visual china

a product manager of a beijing-based internet giant who did not want to be named told china newsweek that high-paying positions have extremely high requirements for academic qualifications, number of papers, and internship experience, and that those who are competing for them are graduate students from top universities. there are very few positions that can truly lead projects and produce results. in the general sense, the work of an algorithm engineer is likely to be just to call models or data packages and adjust parameters according to project requirements. the threshold is low and the substitutability is strong. this type of position will soon be saturated.

in the computer vision field that meng yu is familiar with, a shortage of high-end talents and an oversupply of ordinary engineers have become a reality. although there are computer vision positions in industries such as the internet, biomedicine, and automotive security, most undergraduate graduates can only be considered entry-level. in the case of many graduates, they have learned machine learning and deep learning courses at school or on their own, read a few articles on computer vision, learned about classic models, and then found open source code and data sets to run the results, and then they think they have entered the industry.

meng yu believes that in practical applications, computer vision pays great attention to the underlying information of images. if you want to truly master it, you cannot skip the basics and directly build models. the improvement of computer vision lies in the continuous optimization and improvement of models, and training in this area is generally lacking, at least at the undergraduate level.

when zhao yan was preparing for postgraduate study in her senior year, she chose to switch to digital economy, which is a traditional strong subject of the school. among her classmates majoring in artificial intelligence, except for a few who continued to study for postgraduate degrees in universities, more than a dozen of them worked in large internet companies. most of them chose to take the civil service exam directly, or work in small internet companies and prepare to resign to take postgraduate studies. zhao yan felt that the employment threshold of artificial intelligence has not yet been lowered to ordinary people, and it is difficult for "non-key" masters to find corresponding jobs. "the lower limit is not guaranteed, and the tolerance rate is low" is the positioning of many "experienced people" on undergraduate majors in artificial intelligence.

academic qualifications are still hard currency. at southwest university, about 50% of undergraduate students majoring in artificial intelligence choose to pursue further studies, and chen feng believes that this number will continue to increase. wu fei also hopes that undergraduates can delve deeper into the fields they are interested in. currently, the proportion of undergraduate students majoring in artificial intelligence at zhejiang university who pursue further studies has reached 70%. liu zhiyuan found that many undergraduate students in the computer department of tsinghua university have begun to contact research groups and participate in scientific research training in their sophomore year, and their desire for further studies is very strong. but this does not mean that undergraduates cannot find jobs. overall, those who explore theoretical breakthroughs are still a minority, and more students will use artificial intelligence to solve problems in application fields.

starting from september this year, artificial intelligence will become a general education course at zhejiang university. not long ago, the beijing municipal education commission also announced that starting from september, all municipal public universities will offer general education courses on artificial intelligence. the undergraduate major of artificial intelligence is still in the early stages of development and change. in wu fei's view, the training goals of the undergraduate major of artificial intelligence are three: "from 0 to 1, propose new theories and paradigms for artificial intelligence; from 1 to n, use artificial intelligence technology to leverage the application revolution in all walks of life; from 1 to x, change the research paradigm of other majors, and build new formats such as smart education, smart justice, and smart agriculture."

wu fei believes that there is no need to confine majors to specific areas. in the future, artificial intelligence will become a social necessity like water and electricity, and mastering artificial intelligence will become a necessary ability rather than an additional ability. undergraduate education in artificial intelligence should first solve the problem of textbooks, and then start to solve the problems of teachers, practical training environment and training methods.

(chen peng, xu li, li wei, zhao yan, and meng yu are pseudonyms in this article)