The Heart of the Machine is original
Interview: Wen Fei
The Heart of the Machine recently released the first part of the interview with Professor Xing Bo, the world's top CS/AI scholar, the following is his thinking on academic management and leadership, research taste, and the exploration and creation of a new and more in line with the current era of AI research and education environment as the founding president of the world's first AI university.
Heart of the Machine: Thank you for taking the time to conduct an interview on the School and Governance section. First of all, I would like to ask you how you went before and after you became the principal of MBZUAI. What are your reasons for deciding to accept this important appointment?
Professor Xing Bo: To make a long story short, it doesn't matter what happens before and after, but there are some ideas that are still worth sharing. Let me start with the background of the general environment and the state of the industry. In recent years, the development of AI has been quite rapid, challenging the current university research and development, teaching, research environment, and even management methods. For example, the competition between schools and enterprises for talents, the competition for equipment. There are also some policy and institutional problems. For example, due to the development of social media, the trend of showing off and comparing is popular, causing many researchers to float in the hearts of people. They can become influencers in some way, or choose to do scientific research very solidly, after all, the two reward mechanisms are not the same.
Coupled with the challenges of publication and publication mode, for example, some people can pour a lot of water in arXiv; they can use self-media or other means to create momentum for a certain work; resources and high voice can even influence anonymous peer review through public opinion, such as playing the edge of the anonymous review rule, holding a press conference during the review or promote on the Internet or even mainstream media to affect the judges' judgment and psychology, resulting in a fait accompli This situation becomes even more complicated when companies or schools are also behind them or are tied up. The phenomenon of publishing articles on concert, such as brushing the list, is not uncommon, and even becomes the mainstream.
These current situations make developers confused: whether to join this tide to do the kind of high-yielding, high-gloss, high-profile scholars, or to delve into some of the more difficult and time-consuming problems? Students face such challenges as well. However, the current school mechanism is actually a bit unable to keep up with these new situations. And because of a lot of political or social environment confusion, in the school's scientific research environment, many scholars now have some frustration, students also have some confusion, there is a trend of reverse elimination. I think everyone is very clear about this.
For example, now major Internet companies, and even traditional enterprises, have actively joined the competition for AI talents and research and development. The environment, resources, and treatment they provide are extremely tempting for senior AI scholars, young teachers, and students. How do the personnel systems, funding systems, intellectual property policies, etc. of universities face this new situation: is it to stick to a zero-sum competitive posture to force professors to choose sides, or to use flexible and open new laws to make professors, students, schools and enterprises win-win?
In the current environment, how to protect the freedom of speech (including freedom of silence) of teachers and students, academic freedom, and protect the reasonable rights and interests of teachers and students, especially in their innovation, entrepreneurship, cooperation, and exchanges, these are all problems that colleges and universities must face at present.
It has to be admitted that these problems have not been solved in most universities at present, and even attract attention. The service function, catalytic function and cultural function of universities are not in front of the challenges and opportunities of the current ever-changing scientific research and human environment. It is also common to see conformism, condescension, layman leadership and insiders. For this situation, blindly sighing helplessly will not help. Or do you need someone to stand up and do something to change.
Personally, there have always been many invitations for me to take on various positions. Because there is still a lot of work at hand, carnegie Mellon is still a pleasure to work at, and my own company Petuum also makes me feel very fulfilled, so I have been reluctant to consider taking an academic or technical management position. However, based on the above background, combined with my years of scientific research experience, research and development and management accumulation and experience, as well as some long-term thinking, I feel that it is necessary and very interested in making some new attempts on a good platform, that is, in the academic, educational, research and development, scientific research environment, try to create a new environmental system, to discuss and practice new methods. It's something worth doing.
The UAE's national environment is relatively simple, without many other aspects of interference and conflict, and can be more purely pursued in academic, educational, research and development excellence (in pursuit of excellence). MBZUAI is a brand new, completely zero-started university, so I think it is possible to try to create a new space for academics, education, technology transfer and business incubation: for example, in the academic aspect, how to achieve a balance between basic and applied research, how can we try a new balance between theory and practice in education? So my goal is to be able to explore how to create a new and more contemporary scientific research and education environment and the development of higher education in the new environment through this opportunity, on the one hand, to explore how to create a new and more suitable scientific research and education environment; on the other hand, through this new environment, we should make our contribution to the breakthrough of AI itself. At the same time, it can also make a certain contribution to this society and the community community and provide new ideas. For example, cultivating local talent, promoting technology research and development, and how to use knowledge to transform society with knowledge and innovation. Because the UAE or the Middle East, in the modern science and technology is still in the early stage of development of talent training, knowledge accumulation and landing deployment of the adaptation process, they have the layout, construction, accumulation of latecomer advantages, to have a better and more positive impact on society. So I hope that our universities will have these functional and social effects, and even provide new ideas or some new ideas for this pioneering culture.
Heart of the Machine: I think I just heard you say that if you can build such an environment and then promote the development of AI in this environment, it is really a very exciting and very valuable thing.
Prof. Bo Xing: Yes. In the conservative system, even small surgical changes are actually very difficult to make, because the fixed pattern of this establishment has its inertia, so it is difficult to change. But to do it on a brand new platform, not only scientific research innovation, but also in management innovation, ideological innovation, there are more opportunities.
Heart of the Machine: You mentioned the word "pure", can you explain what this more pure pursuit of EXCELLENCE in AI or technological development means?
Professor Xing Bo: The UAE is still a very young country, in a period of rise, seeking to actively integrate into the international community and play a greater role in the development of world economic and political scientific research. It is not a society of conformity, but an innovative and rising society. So its goal is specific, not to do everything, and to focus on prioritizing some important things. For example, I hope that the school can have a place in the world in the field of basic research, and I hope that it can have some impact on the development and transformation of the local economy and industry or the needs of society.
In the international arena, the United States, Europe, and China are all highly developed societies that need to face problems such as aging, infrastructure upgrading, and so on. But in the UAE, you don't have to face these problems. In addition, environmental issues and DEI (diversity, equity and inclusion) issues are very important, but combined with research and development, it will become very complicated, involving the whole body, so that leaders and decision-makers sometimes have to lose sight of one or the other. As a new school, we do not have the problems and baggage accumulated from history in these aspects, and we can go lighter and explore a healthier and more reasonable development model.
Usually when we do scientific experiments, we can define the parameters, and then mobilize and adjust at certain points. Therefore, whether it is a social experiment or a management experiment, it is also necessary to have a controllable environment. In a new university like MBZUAI, the environment will be relatively simple; the goal will be more humble, not large and complete. Our school itself is a graduate university focused on AI. In this way, there are specific goals for the overall layout of the discipline and for the different stages of education. Although the goal is high, the breadth of the goal is a certain range, which is convenient for exploring some new scientific research and teaching ideas.
Professor Xing Bo is in the MBZUAI teaching building. Photo by Khushnum Bhandari, Image credit: thenationalnews.com
I. About running schools:
Pay more attention to the university's academic,
Empowerment of social and cultural spiritual connotations
Heart of the Machine: MBZUAI is the world's first AI research university, how do you understand its positioning and philosophy?
Professor Xing Bo: I actually don't like the title of "first", the first and second are actually meaningless, sometimes this is a way of writing to capture eyeballs. It's a university, it's an AI-focused university, that's all. This in itself is already an immutable fact, and as to whether it is the first or the second, I am not very interested. I am more concerned about whether the university can achieve its academic and social development goals.
Back to the university itself, as a research university focused on AI, or more broadly, computer and information science, MBZUAI has such a vision: (i) First, it is committed to becoming an academic center leading AI/CS research, application, and innovation, not only in this field, but also in related science, engineering, and humanities, to expand new frontiers, create new knowledge, and provide new capabilities; (ii) it wants to become the cradle of future technology, management, business and education talents in the UAE. To cultivate leading talents and thought leaders for the country and society, while sending the backbone and doers of all walks of life, that is to say, our graduates must have both leaders, thinkers, and doers; (3) to become the engine that drives the development of the economy and productive forces, become the cradle of incubation enterprises, technologies, and products, solve problems in the development of the country and communities, and drive the continuous progress of culture and ideas.
I hope that MBZUAI will not only become the best and most creative polytechnic university in the UAE and the Middle East, but also a leading scientific research institution that cannot be underestimated in the global AI arena, with global influence, attracting, cultivating and exporting top academic elites. Under this vision, we formulate specific short-term, medium-term and long-term plans and goals for the construction and development of the school.
Heart of the Machine: I looked at the school's official website and had a very detailed curriculum, CV, NLP, and machine learning, but I didn't see an introduction to the project or the topic. What are the key topics or projects in the school now?
Professor Xing Bo: At the level of school functions, there should actually be room for the choice and definition of topics. Back to the idea of running a university. When we mention Peking University, Tsinghua University, or Harvard, MIT, CMU, and talk about Cambridge and Oxford, what do we remember in the end? It's not which subject it does, or even which problem it solves — there's no doubt that these schools have solved and done a lot of problems over the years. Usually what we remember is not what its campus looks like, Tsinghua University's former president Mei Yiqi once said that the university is a master' word, not a building, in fact, it is not a subject. I think that such as projects should be positioned as a specific internal operation, rather than the focus of external publicity.
The first thing that universities should stay is to have masters and talents. Specifically, it is necessary to have a lofty academic status, exquisite academic attainments, and at the same time be able to continue to pioneer and lead, and forward-thinking leaders to sit and define the direction of majors and disciplines. Such anchors form a basic foundation of famous schools. On the other hand, I think that in the modern environment it is not enough to have a master, but also to be able to leave tradition and spirit behind. When we mention many famous universities, in fact, we often think of the spirit of Harvard, or the spirit of Southwest United University, which has a certain ideological and cultural connotation in it. This is something that a university should pay more attention to, and it is also something that a university president should pay attention to.
The subject and the specific project are actually only a means, and it is also one of the functions and responsibilities of the university. But the responsibilities of universities are very, very many, and the topics are dynamic. We now have a lot of work on topics such as healthcare, the local oil industry, and smart cities. These are all natural and natural things that any university will do when it lands in the local area, but this is not the essence and core of the university's function. And the final result is not to see what to do, but to see whether things can be done well, and then from the results of doing well, whether it can continue to improve and develop, and promote the continuous progress of the school's own core functions - scientific research, education, and development. For example, by doing these projects, our teachers and students gain experience, or receive honors, or the school receives financial benefits or reputational rewards. How do we build on that to create the next thing.
So university projects and projects are actually a dynamic process. Specifically, based on the actual local needs and the strong direction of our current faculty, MBZUAI strives to make breakthroughs or make groundbreaking work in the following directions:
1) Large natural language intelligence models, especially Arabic language models, can be trained continuously for life and high-order reasoning systems. Different from several popular super-large-scale NLP deep learning models, our focus is on practical applications, such as Arabic for other major languages in the world, rather than the display or brush type of text generation system; second, our focus is on the natural advanced reasoning of text, and even multi-media data (such as text + image), including common sense reasoning, symbolic reasoning, logical reasoning, counterfactual reasoning, etc. It does not stop at memory and associative text annotation or text generation based on big data.
2) LEGO-style composable, auto-optimizable, scalable machine learning backend platforms and systems. Our focus is on generalization, standardization, modularity, freely composable, highly integrated back-office systems that support a wide range of advanced features such as machine learning training, meta-learning, compositionality, provisioning, automatic distribution, and even package deployment, rather than just over-optimization of a single function.
3) Personalized medical diagnosis and health consultation system based on artificial intelligence and pan-data form. Use data from medical images, medical records, genomes, physiological measurements of wearable devices, and more to predict health risks, understand diseases, manage diets, and optimize lifestyles.
4) Continue to delve into the previous research on the "Standard Model of Machine Learning" and will not be repeated here.
A corner of the MBZUAI campus. Image source: wam.ae
Heart of the Machine: I have two questions next, the first is what kind of spirit do you want MBZUAI to leave behind? Secondly, our research, including water resources, oil and renewable energy, feels that it also needs to be landed, will there be some engineering needs or assessment indicators? Hopefully, you'll talk a little bit about the spirit of college and then talk about how to measure that result.
Professor Xing Bo: The spirit of the university covers a wide range of depths, and it is difficult for me to define it specifically to MBZUAI, and we are still exploring. However, there are several general directions that are clearly stated in the opening few words of the ancient text of "University": the way of the university is in Mingmingde, in being close to the people, and in stopping at the highest good. "Matilda" refers to the cultivation of oneself, which can even be expanded into knowing oneself and knowing the world, which is a cognitive need. When we do learning, we must first recognize ourselves and the world around us. "Being close to the people" means to influence society and to have a positive impact on society. "Stop at the best" represents the continuous pursuit of the ultimate ideal and mission of educational research. I hope that our university can achieve such a pursuit. In the specific connotation of Mingmingde and close to the people, of course, we must combine with the local culture and environment, and at this level, we are still continuing to explore. In short, in the big framework, we pursue knowledge, pursue the unknown, and can also positively influence society. How to do or do it at a specific level needs to be further defined, such as the aforementioned MBZUAI's school positioning and some current topic directions.
We have some specific goals in the concept of running a school, which can be summed up in the hope that after it reaches a certain scale and level, it can at least achieve the following functions: One is that it can serve as a beacon of knowledge and ideas, which can lead our community and even the entire academic community, and have a beacon effect of demonstration, not only in direction and level, but also in style and taste. That's the first point. The second point should be a source of knowledge that can produce knowledge and be able to make more discoveries. Especially in the field of AI. In fact, there are many problems and opportunities at the level of AI research, and I think this university can provide some specific content and can also generate new ideas. The third point is that we need to train talents, which is a very important function, not only to output instrumental talents, but also to have thoughtful talents, they (they) can innovate, can continue to carry forward the inheritance. In addition, it can cultivate and export comprehensive talents who have entered various fields such as academia, politics, industry, and business.
It should be pointed out that in modern society, such talents must have sufficient knowledge of artificial intelligence. Originally, we talked about literacy basic cultural literacy, usually literacy in the early days refers to literacy, being able to read and write, and later being able to understand a little math. I think the third layer of literacy should be to understand AI, which is a contemporary required course, not an elective course. Because AI is a very integrated combination of technology, its boundaries, functions, future potential, and even the way to achieve it are actually a very complex and esoteric system. If you don't understand this at all, you will make some ridiculous mistakes. For example, we will think that AI will not surpass humans, whether AI will kill all people, and whether we can enter the singularity, in fact, it is a bit like the idea or delusion that will arise under the condition of AI-illiteracy (AI blindness). If the whole people or leaders are AI-illiteracy, there will be wrong judgments in many decisions, just as they will make wrong judgments if they don't understand mathematics or don't understand words. In this regard, I feel that it is necessary to provide such knowledge and training at universities and in the cultivation of talents.
An important meaning of the so-called "people-friendly" is to think deeply about and empower culture. Culture culture is not just culture in the narrow sense, such as art. I think culture is a way of thinking, a mindset, a mindset, a mindset. Through education, through technology, through innovation, it also allows the whole people to enrich its cultural connotation. So I guess that's a goal for the university. The most important legacy of a university should be knowledge, tradition and spirit. I want MBZUAI to make a useful and unique contribution at the national level.
When it comes to how to evaluate, to be honest I don't think universities are contractors, and their value and good or bad is assessed by contracting several projects and how well they are done. The university's true reputation comes from people's hearts and word of mouth. The reputation of the university is nurtured and disseminated by all people, and it is everywhere in the community environment, both in the local secular community and in the environment of scientific research, so the best evaluation is the praise of peers, students are willing to come to study, teachers are willing to join our team, and masters are willing to join, which in itself is a good evaluation. When Tsinghua University founded its studies in the early days, it did not undergo any evaluation. I just know that everyone is willing to go and be proud to go there to study and do things. From the government to the people, there is enough respect for the teachers of the university, and they feel that the teachers and students of this school have standards - the teachers are exemplary and the teachers are moral; the students are practical and diligent, and they can be the pillars.
And now it seems that on many occasions the word "professor" has been stigmatized. University presidents are also often consumed by everyone, and occasionally ridiculed or teased by the public, which in itself reflects the lack of respect for universities by the whole society, from officials to the media, to the public, which is actually a disastrous result, much more tragic than the university may not be able to complete several projects. Of course, there are many factors inside and outside the university, even including the problem of the cultivation level of the university leaders and teachers themselves, as well as the overall social atmosphere.
My goal is to enable the public to have a kind of expectation or approval of this school, and justice is in the hearts of the people, in fact, it is enough. As for how many projects are completed and how to complete them, these are subordinate products. Moreover, hard indicators alone are not enough. Even if the indicator is set, if it is defined too dogmatically, it will alienate or narrow people's work. Because it is entirely possible to water through shortcuts, to reach the number of articles, or to concentrate on doing a few projects, to recruit a few bids to become contractors, these so-called low hanging fruit, shortcuts, or a very cheap applause to obtain, does not have much impact on the overall reputation and quality value of the university. Running a university is a long-distance run, there is no trick to overtake in the corners, or to compete and do things in an upright manner, and to do things solidly. I think university presidents should actually think about what makes a good university.
II. About The Governance of Schools:
I want to lay a framework,
Let future generations continue to move forward on this basis
Heart of the Machine: As a principal, what goals have you set for yourself?
Professor Xing Bo: First of all, the goals you just mentioned about setting goals for myself are the work goals I set for the school, or the personal goals that I set for myself? From a work point of view, I need to achieve the goals and tasks that I have just talked about, which are professional goals. The work of the school is achieved by a team, so I will set goals for the team, and I am a member of the team. As a personal person, I have some self-esteem about myself.
Heart of the Machine: You just make it for yourself.
Professor Xing Bo: Personally, the goal I set for myself is to be specific and pragmatic. First of all, I want to do a good job, just like Professor Mei Yiqi, the old president of Tsinghua University, said, the role of the principal is to serve tea, pour water, and move chairs for the professors. I think he felt it very well and expressed it very well. It turned out that when Professor Tian Changlin was in Berkeley, he had to go to the laboratory every day, often walking around the campus and talking to students, he was either an official or an employee of a school. What we see now is that many people treat the principal as an official, or as a symbol of power, as a symbol of power. The goal I set for myself should be to do the best service to the teachers, students and staff of the whole school, so that everyone feels that I do practical things for them and solve their problems, which is the first point.
The second point I hope to be able to be a role model, to be a good teacher, and then to be a good researcher, let everyone remember that I am just a professor, a teacher, still doing research and teaching at the same time. For example, I gave a normal seminar lecture in the school two days ago, which was completely technical. Over the past few months, I have done nearly 10 different academic presentations in academic settings to exchange the latest results. So I set myself the goal of being a business model, encouraging each other with colleagues, and being able to play the role of an educator or a scientific researcher, which is the second point.
The third point, the so-called ten years of trees and a hundred years of tree people, the university is a tree person, it really takes a long time, even longer than a hundred years to do. So what can I do? If you can pave a little way, you can leave a system or blueprint, you can leave a framework and foundation, so that it has a lasting long-term impact impact. I would be happy if I could achieve even a little bit of this ambitious goal.
Heart of the Machine: Is it to leave legacy? Chinese translated as heritage.
Professor Xing Bo: It's not a legacy, I think the more accurate word is "framer", that is, the frame maker. The most important thing for a founding principal is to set and leave a framework for the future. Because after the framework is available, the next work will be very orderly, global, and then add specific content to improve the details. I think the overall situation is very important, at the beginning of the framework, the future work is actually around a complete idea to carry out, rather than tinkering, not headaches, foot pain. This is actually very similar to the creation of the country. The creation of the United States, Washington, Franklin, Jefferson is not called the founder, they are called framer, framer is the framer, in fact, they did not complete the specific details of the later, the specific environment of the constitution more than 200 years ago is definitely different from the present, and it can still be used today because it does not stipulate a lot, but a frame framework, there can be amendments behind, there can be more things to improve, but its thinking has not changed.
Schools also need such a framework, and as a founding principal, I may be able to help the school produce such a framework. In the future, there must be more capable principals and more capable teachers, who may be better in all aspects of the technical link, and they can certainly do better, which is also the right thing to do, because our goal is to make it better from generation to generation. But these pioneers in the beginning were not necessarily the smartest and most capable people. But if you have a big enough mind and a broad enough vision to set up the framework, I think a goal has been achieved.
Heart of the Machine: Do you think a good principal wants to do the best for students and teachers, to do something groundbreaking and lays the groundwork, and at the same time, can I understand that you never leave the front line of teaching?
Professor Xing Bo: It can be understood that I hope to stay in the front line of education, to always do scientific research, and to always have the opportunity to exchange scientific research, which is a small wish of my own. Of course, the premise is that my own energy and level can make me qualified for such a role, rather than lying in my seat and not walking in the way of the able. This requires me to be able to maintain my professional ability in scientific research and teaching, continue to make progress, continue to achieve new results, and be a leader who convinces the teachers and students of the whole school. Research is interesting. As the saying goes, the song does not leave the mouth, the fist does not leave the hand. Once you give up, or if you put it on hold for a while, it will become sluggish and difficult to recover. This is a terrible situation for researchers. So it does pose a challenge for every administrator, principal, or department chair engaged in education management, which is how to balance the two aspects of research and management administration. The requirement I made to myself was to strike a balance between administrative functions and scientific research functions, rather than abandoning scientific research and only engaging in management work.
For academic administration work, my understanding is that this is a service and reward for senior researchers to academia, and it is both an honor and a sacrifice for individuals, which is particularly reflected in the possible negative impact on the ability of scientific research business. I am referring to the influence of my own actual independent scientific research ability, rather than the so-called academic "promotion" that may be obtained due to rights, such as the common H-index promotion from "leadership auto-signature", the highlight treatment in various academic occasions, and so on. These things tend to further erode academic capacity, resulting in a degradation of academic leadership knowledge and a decline in judgment. Therefore, I will continue to write articles, guide graduate students and postdocs, continue to make specific academic reports, and also give lectures, hoping to continue to make cutting-edge and valuable research results, conduct professional and in-depth academic discussions, and gain the respect and recognition of peers.
Heart of the Machine: Next, I would like to ask you about your understanding of leadership – in addition to various functional responsibilities, the key to the "principal" is a leader, and you just mentioned the need to cultivate a group of young people with AI leadership who can enter various fields and play a pillar role. How do you understand leadership and how do you develop such people?
Professor Xing Bo: Leadership is not something that can be trained in a targeted way. It's not that a student, after being recruited, the goal is to cultivate to become a leader and teach you leadership, I think this can't be done, and it shouldn't be done. At the same time, it is not innate, and then it does not need to be cultivated. For example, there may be some professors who suddenly become deans, or who for some reason get this position, or are willing to take on this responsibility, and automatically become a qualified dean or principal, which is not true. Leadership is the result of a gradual accumulation of experience, learning, and experimentation, so it is important to continue to learn and accumulate.
The old school song of Tsinghua University has such a sentence: Instrumental knowledge is the first, literature and art are subordinate, from the "New Book of Tang Dynasty Pei Xingjian Biography": Shi Zhiyuan, first instrumental knowledge and then literature and art. It means that first we must cultivate character, insight, realm, and sentiment, and then we must cultivate talent. This is the step of ancient Chinese intellectuals to learn to cultivate themselves. I think this is especially true of leadership development. Leadership contains some more unique qualities or spirits, on the one hand, there are innate factors, there are external environmental factors, and a large part of the acquired cultivation.
I think leadership is first and foremost the courage to be the first, to excel. Because you are the leader, you have to lead the way for everyone. In ancient times, the general was at the forefront. It is common to see a military battle in which the only person killed or wounded is their commander, or their commander. This can be seen in Israel and in many examples of warfare in Europe. Leaders' thoughts and behaviors reflect the courage to be the first and stand out. But the other part, which we don't see often, is that leadership is also reflected in the humble valley and the sea. He can carry many opinions, even aggregate or combine different opinions, not to reconcile different opinions and mud, but to seek an optimal solution from many different and even opposing opinions. And then you can convince everyone to do it together, so the ability to unite is also very important. In short, I think leadership is manifested as daring to be the first, standing out, being humble, and being full of rivers, and at the same time there is this spirit of unity in the core.
Finally, leadership also embodies a way of behaving or an idea, that is, it embodies an independent spirit, a free thought. This is very important and cannot be found in textbooks. Especially after entering the no-man's land and entering the unknown territory, it is actually difficult to have a playbook ready-made manual or a trick for you to follow the script. That's when creativity is needed. Creativity requires free thought and an independent spirit, and it allows you to explore and find solutions in a free space without boundaries. It is essential for school leaders, academic leaders or academic administrative leaders, especially the president of the university. Because innovation and exploration of basic and applied research, especially basic research, is to explore the unknown. Exploring the unknown is actually a habit of mind, a way of life, which must be rooted in some kind of soil, with its independent and free environment for growth. If the university is such an environment, students or teachers, they will naturally pull the topic and the goal to this aspect. As the old saying goes: If you can't think of your place, you are abandoning yourself to shallow learning. If we always question whether we have touched a certain boundary, or whether we have done a topic that is not very acceptable, the result is that we die without starting to think, so that kind of thinking will certainly not be able to go out of the boundary to innovate. I think leadership is also reflected in the fact that leaders should dare to advocate and promote this concept and practice.
Heart of the Machine: It sounds like you don't set limits on your team. I still want to ask you what are your requirements for faculty members?
Professor Xing Bo: Faculty members need to be evaluated, and this is necessary. But the assessment is not simply a quantitative tick box form filling option assessment. Now MBZUAI's assessment is from the general direction, no different from other universities, one is scientific research, one is teaching, one is service, these three. However, the function and effect of the assessment mainly depend on the specific rules and implementation, otherwise the best rules will be out of place or out of place in the wrong hands.
There will be some more detailed specific indicators in scientific research. For example, publishing excellent results in this journal or conference, being able to obtain funds, writing good project reports, awards, and so on. But the most important one is to be recognized by peers, and the definition of peer recognition should also be rigorous. It's not about going to a few acquaintances to write reference letters. Peer recognition definitions should be specific, one is that first of all, it must be an expert, have the ability to make a specific and accurate assessment of the work of the assessee, and secondly, there is no interest in the assessee, and it can be objective and fair. For example, he wants to get a number of letters of recommendation, usually 15 or 20, some of which can be nominated by himself, and let some people who know him better write. But the other part is that the assessment committee collects it independently, and it asks people who have not worked with teachers to write articles of recommendation. Of course, we can't fully discover all the relationships, and the academic community is very close and everyone will know. But there is still a certain distance, and we want to judge as independently and unrelatedly as possible. This is such a requirement for academics.
The second point is the requirements in teaching. It should be pointed out that the teaching requirements are as important as academics, not the second important, but equally important. This includes the preparation of classroom courses, which courses are opened, and the quality of teaching. Generally, our lecture videos, lecture notes and other courseware will be placed on the public homepage, not only the students of the school can continue to review and evaluate after the class, but the entire public can use and evaluate. Including whether they like this course, whether teachers in other schools have chosen this course, and so on. For example, the Graphical Models courseware that I teach at CMU has been updated every year for more than a decade and has become a public resource. Probably now there are many school teachers in the United States are using, basically either change it, or enrich it to use it. This indicates a certain degree of acceptance, which is also an important assessment criterion. This then includes the teacher's teaching in meetings, in various academic activities, the degree of activity and active participation of seminars, which is to meet certain requirements in the field of teaching.
Service is also very important, how to promote the development of the academic community on and off campus, such as being a conference or a journal reviewer, this is a very demanding and arduous service, because it takes a lot of time, sometimes it is very hard, and doing this can not be famous, is an unsung hero. But we still have to count it in the work of teachers. Then organize a meeting, or organize a workshop, and the organization itself is actually very hard. Attending admissions in schools, participating in interviews for teacher candidates, and even being in charge of equipment, all kinds of chores inside and outside the school must be done by someone. We still hope that the professor can feel the master, treat this school as a home, and you will find that there is no thing that is not important. You can't just pick important things and push others on others, these are your own business.
Therefore, we will have such a comprehensive indicator in the evaluation of teachers to encourage the development of teachers who are sound rather than deformed. That's why the old system may now need some adjustments, and how it can promote the development of a sound teacher or researcher, rather than a deformed worker. Of course, the school is an inclusive and diverse environment, and there will certainly be a variety of styles that can exist, but there needs to be a balance and not one-sided. Unless he's doing so well on this side that he's putting other things off when it comes to weighted averages. Some people may have a 0 score in Chinese, but their mathematics is extremely high, and maybe he can also be admitted, which may exist, but this is not the norm. After all, when we take the exam, we still have to take the math, chinese, physics, and chemistry exams. So this is my vision for teacher evaluation, and now I'm communicating with our faculty and implementing it.
Heart of the Machine: According to what you just said, both the ability to study and teach needs to meet a minimum standard, and on this basis can it be considered a difference, that is, professors must be comprehensive talents?
Professor Xing Bo: Yes, this is the ideal state. Of course, we also have Research Track and Teaching Track, and there will be such in the future, especially after increasing different needs, and even now we have some professors who are research-oriented. But this cannot be used as a mainstream idea, but only as an option. If professors are more inclined to actively do teaching tracks, or do research tracks, this is ok. But his requirements will have a slightly different emphasis, such as if the research track, can it not teach at all? Maybe yes, but in the U.S. research track and teaching track are not tenure, that is, there are no tenure. There are different tier levels and different expectation expectations. It is more transparent in advance, so that both sides understand and expect the same.
Schools need this diversity and flexibility in the form of employment. The question you just asked, I just answered a more mainstream common way, but also contains a variety of other ways, such as various forms of adjunct professors, remote professors, and so on. Our focus is not on mechanically stipulating the types of contracts and work forms of teaching and research personnel, but on how to produce the best scientific research and teaching results, maximize the release of professors' energy, benefit from their services, and also take into account their objective reality and needs, and strive for a win-win situation for schools, teachers and students.
Heart of the Machine: In a 2014 article you interviewed at Microsoft's Asian Research Institute, you mentioned that when you recruit students, you value his value as an ordinary person not only as a researcher, but also as an ordinary person. For example, have an independent spirit and the ability to think independently, be knowledgeable, be honest, and be open-minded. As a principal, how do you think about selecting and assembling faculty and research teams?
Professor Xing Bo: In fact, teachers are no different from students, they are all human beings, and teachers should have the same or even higher requirements for personality, cultivation, and morality. This is not only an expectation of my own, but also a requirement that the teachers must have for themselves, which is the so-called "first self-cultivation". When my students graduate, they go looking for jobs, especially interview faculty, and often ask me how to participate in interviews. I'll make them think about it, what kind of mentality will you have as a teacher at the school to recruit a new colleague? New colleagues may have to work with you for 30 years in the future, do you want to recruit a monster, a walking supercomputer, or do you want to recruit a living person who can cooperate, communicate, can grow with each other, have friendship and other important things. Therefore, I think that as a requirement for teachers, we should pay more attention to the literacy of humanities and self-cultivation. He must be a living, healthy person, and he should be more exemplary in character and character, more important than students. This is what I expect from teachers. Students are like when they come in, when they graduate and go out, they start to work and teach, and they should be more enhanced in this regard.
For teachers, honesty, knowing how to respect, including mutual respect for their own students, others, work, and mutual respect with everyone, is a very important quality, because in this way can produce reciprocity, health education is possible. In academia, we will often disagree, because scholarship pursues truth, and sometimes there is only one fact or truth, and other things are wrong, so there must be competition and debate. At this time, if there is no respect, and then take personal feelings, it is completely impossible to work and communicate. Therefore, the cultivation and manner of man himself are very important. Honesty, not to mention, directly determines the quality and credibility of the results, and the trust between colleagues in work cooperation.
I admire a quality called "humbition," humble on the one hand, and ambition on the other, which I think is important for teachers. Excellent research and excellent teaching and research are a necessary condition for teachers, which is the minimum requirement. Only when the ability to research and teach reaches a certain level can you enter a good university. But I don't think that's enough. That's why at every school, at CMU, at Berkeley, we don't just look at a PPT presentation when we interview, but we also have to look at everyone he interviews with, including teachers and students, face-to-face, talk about various academic topics, even various non-academic topics, and watch his behavior and demeanor. It's all part of the interview. Some people may be very depressed, saying that my paper is so many times, several times more overwhelming than the other party, why I didn't get the offer, is not hacked. In fact, he did not understand, because we are not recruiting a thesis machine, we are recruiting a person, a colleague. After thinking about this point, in fact, all the doubts are solved and become very simple.
III. About Governance:
Good research can provide a long-term perspective,
The topic, style, and means have all reached a certain level
In order to count as tasteful research
Heart of the Machine: From your earlier and past interviews, you have mentioned that research is the pursuit of the unknown and perfection. First of all, please define what good research is and how to make good research?
Prof. Bo Xing: What is good research? I'll make this a little more concrete, and research can be divided into basic research and applied research, and of course, intermediate. At least from my personal understanding, basic research should have some characteristics: there is a desire to exhaust, or a goal to grasp the core essence and law to do research, rather than to show off how strong my mathematics is, how smart I am. In music performance, we often see the showmanship school, and even more valuable is a kind of fundamentalist school or composer school, which will play according to or restore the original idea of the composer as much as possible, rather than playing for the display of their own technology. I think basic research still has to understand the problem itself, which is a good study, not how many formulas, terms, and derivations are piled up. In fact, the most valuable basic research in history, such as the uncertainty principle of quantum mechanics and the discovery of the DNA double helix, has only a few pages of papers, and some of them have not yet been published in prestigious journals. There may be some differences in applied research, and I think we should pay more attention to solving practical needs, be able to face users, and be able to communicate with insiders, rather than just brushing the list.
The most taboo way to do scientific research is a so-called asymmetrical way of playing: to show the results of experimental applications in front of people who do theory and foundation, to brush a few formulas in front of people who do applications, to show the so-called theoretical results, and to use the audience's unfamiliarity with the essential level of the onshore field to get this asymmetrical applause. I think this is the most undesirable kind of taste. This taste is still quite common, especially among some young professors and students. I used to advise my students that if you really think you've done a particularly advanced theoretical work, go to the Math Yearbook or the Statistical Yearbook, instead of going to an ML or CV conference proceeding and scaring people with your own derivation work paper. If you think you're doing an application, then write the code well, make the prototype complete, and implement it into the real scene, without just brushing up on a few untrue standard data sets. For myself, I have always positioned myself as an engineer, not a mathematician, and although I have been publishing articles on the editorial boards of several core mathematical journals, in those tasks, I have honestly answered the basic questions well, and have not mixed in with problems that mathematicians are not familiar with to make things up. Vice versa, in the application work, we strive to be intuitive, rigorous, transparent, in the experimental design and display of the detailed and sufficient, not to mix in those inexplicable and reluctant to challenge the theorems to make the mystery. In short, if you really feel that you have done a good job of research, whether the theory is applied or not, you should try to show it to experts as much as possible. Making money on social networks is a waste of your own time and everyone else's time.
Just now I talked about just a style, but good research is actually not enough to have a good style, the most important thing for good research is to set the right topic, to ask the right question, the right topic, this is the goal of the thing. Because there are many topics, and the topics are definitely good and bad, some topics are imaginary topics, you can put up a target there yourself, say it is a target, I go to fight, this is OK, but it does not really reflect the value and ability. Another kind of target is the natural target, which is real, which is some of the open problems in the world. Are you willing to try this kind of question.
Just now I talked about style, choosing a topic. Then there are the means, in fact, now we also see the rise and fall of different research methods. What is now more concerned is the so-called resource accumulation and arms race that emphasizes violent calculations. For example, if I choose a problem and do a "big model", my topic is to make a trillion parameter model, to do the trillion parameter first, as to why the trillion magnitude is secondary. This in itself changes the track, because others can't play, I can play, because I (my factory) has toys, money, data. Of course, this kind of topic is still valuable, there is engineering value, there are other values, but I don't think it is a good taste. Because this kind of violent thing, first of all, excludes healthy competition, or excludes the competition of the mind, it is a competition of resources and funds.
So why do many universities say they're pessimistic or frustrated that they can't do research, because they don't have so many computers, and then they don't have so much data, they can only do something else. So what about the results of this "scientific research"? In fact, in the original international political or military competition, we have seen some superweapons, such as the huge 50 million tons of TNT equivalent Tsar hydrogen bomb, and the former Soviet Union built a ground effect vehicle called the Caspian Sea Monster, which has ten engines on it, and can even carry large anti-ship missiles. I really can't judge its taste. The same weaponry, like the B52 bomber, came out for nearly 70 years and is still in service, which is a good taste product, can withstand various tests, everyone is willing to reuse, constantly updated. Therefore, I think that the means must have good taste, can not win the unarmed, should really go to a fair platform, the master can compete. In short, the topic, style, and means all need to reach a certain level in order to get a good and tasteful study.
Let's take a few examples of good research work with taste. I recently listened to a very good talk from Judea Pearl, a senior researcher, who put forward a good research topic on the limitations of these machine reasoning. Now work based on super big data, whether it's GPT-3, or other supermodels in the future, is largely a factual reasoning of data driven. It is useful to memorize a lot of things, store it as a form of expression, and then it can be retrieved quickly, but all it can achieve is a factual reasoning, factual reasoning. What about common sense reasoning common sense reasoning? For example, do the sun have eyes? Or just ask the sun how many eyes it has? Or can I fry an egg in the water? This is common sense reasoning, and you'll find that the big models of GPT-3 are hard to answer. Because these common-sense contents have not appeared in the literature, and even its transformation has not appeared. But the human mind is still very easy to do this kind of reasoning. Also, for the parts of the model that have memories, can we use counterfactual reasoning, counterfactual reasoning? For example, if Oswald had not assassinated President Kennedy, would President Kennedy still be alive? Then Judea Pearl pointed out that this is the flaw of our current popular model, which has nothing to do with size, it has to do with architecture, and can we design to introduce new architectures, introduce new parameters or representations of variables, and then define them with a new objective equation? So I think it's a very good and tasteful topic, because first he raised a deep problem, did not solve the problem, but also put forward some ideas to solve it, and there are many people who can continue to do it. And fairly, everyone can do it, and it doesn't need to be done with ten thousand computers, and only one computer can't do it. At the same time, it also has a good return on mathematical tools and the expansion of the way of thinking.
Another topic that I think also shows very good taste, is the variational inference, variational reasoning, which has achieved a lot of results in the past 20 years. This is a work that Michael Jordan and Martin Wainwright, along with many of his students in his group, including myself, have been involved in and played a role. It is to take the original heuristics of several probabilistic reasoning, including Belief Propagation, Mean Field to carry out a mathematical description of continuous evolution, and finally figure out what the nature of its mathematics is. By understanding the mathematical nature, the next approximation can be introduced, such as the current variational model, which was originally an approximation to the posterior distribution in the initial principle, and later became a variational network, which itself became a training object, and then the tools of the entire neural network can also be included, so that it is like the current autoencoder or other transformer , can be trained through a similar mathematical framework. But in essence (because of the original VI theoretical framework) we can know what it is doing, while at the same time being able to combine different mathematical tools to enrich the concrete implementation. Of course, the number of people doing research with their eyes open is now decreasing. In fact, the early work of variational inference was an earlier introduction of optimization theory and the whole convex optimization and the tools of pan-optimization theory into machine learning, which is quite beautiful. The article itself, though nearly 300 pages long, was very thorough, understandable, not difficult to read, and allowed to continue to work heuristically. Here I have to point out that in recent years, many researchers, especially many people engaged in deep learning, have not read enough literature, and even deliberately ignore the results of their predecessors, re-invent the wheel, and are keen on secondary packaging, naming, and creating new terms, which has caused a lot of unnecessary confusion and mystery in scientific research, resulting in narrowness and faults in the knowledge of many young students. I don't agree that these people are researchers, what they do is not scientific research, more like creating traffic, is the way internet celebrities and public figures are created. But this style is now popular and does a lot of harm to the group of serious researchers.
Not only theory, but also good high-grade work in application. For example, about a decade ago, there was a study of parameter servers. This is a source of current massively parallel models and computations. At that time, Hadoop was a distributed open source framework system based on database ideas. When we got to the parameter server, we really started to redesign the requirements of machine learning. It uses a large virtual memory hub to store and update all model parameters, and the hub is connected to a large number of substations through a "master-servant" network architecture, and each substation holds part of the data to undertake part of the computing task. In principle, the concrete implementation of the virtual memory hub can be physically central or distributed through the key value store, inheriting the Hadoop logic concepts and architectural design that are effective for large database operations. However, when used in machine learning training, the relationship between the hub and the substation is not a simple data mobilization and mapping, but a concentration of the hub's knowledge from all the substations, including integration, de-noise, compression, and so on. Therefore, it puts forward the requirements for synchronization and asynchrony in communication, various innovations in the calculation, coding, transportation, integration method of parameter delta delta, and even various theoretical analysis work, such as whether convergence can be achieved if asynchronous, and how many steps are required. In fact, our Petuum started from this job. We did some theoretical and some framework innovations in the early parameter server, which led to a lot of work in the system field, in the theoretical field, and even in the algorithm field, which is still continuing to this day.
So I think there is tasteful work, it has a timeless, long-term impact, and then it can trigger the integration of many different disciplines, and more importantly, it can drive a generation of researchers to get good all-round training. For example, it can be trained in mathematical techniques, programming techniques, and even research styles by doing such problems, which I think is good.
Some topics also have a wide range of applications, but they are not necessarily good taste or good topics, and the more researchers study, the more stupid they become. For example, grab a tool, and then directly try and make mistakes, or rely on a few algorithms to directly stack larger and larger models, and then there is a so-called leading thing that brushes the list, and now there are many studies that are produced in this way. So we sometimes say that many scholars in the field of machine learning research are not as good as they were ten years ago, because their skills are insufficient. It's like when you always play the electronic keyboard, your playing technique will actually decrease, and you won't be able to play the piano. There are many different touches, colors, and a lot of expressiveness in the piano, which cannot be achieved by the electronic keyboard. At least in my opinion, such research is not a good taste of research, although it may be useful, very space, and need, but I have to admit that it still has a lot of shortcomings in educating talents and forming a broader long lasting academic precipitation.
In a good university, we need to prevent the reverse elimination of excellent taste, results, and teachers, and we can't just look at articles, or even the number of citations. An Annals, JMLR, or OSDI, SOSP, which has been condensed for many years, is not the same as a so-called "top meeting" paper that comes out of a two- or three-month alchemy furnace. Especially now, these "top meetings" often have two or three thousand papers per year, why is there a "top"?
Heart of the Machine: According to what you just described, good research is bound to be less, and even one person may make one or two in a lifetime.
Professor Xing Bo: Yes. For example, there is a study in physics, which I think is extremely outstanding, and that is Yang Zhenning's Yang-Mills gauge field theory. This job did not win the Nobel Prize, but it has since given birth to multiple Nobel Prizes, and there will be more in the future, because it has been continued to be done. It provides the basic framework for the unified gauge theory of weak currents, as well as the possible gravitation gravitation gauge theory, and the mathematical problems, the existence of Mass Gap under non-Abel gauge fields and standard equations, which can be solved by someone who can later win the Fields Medal in Mathematics or the Millennium Prize, and also promote the development of mathematics. I think this is a good and tasteful topic. He asked questions, came up with ideas at the beginning, and then allowed many future generations to be studied from generation to generation, produce more new results, and develop mathematical and physical tools. There are also such topics in AI, but there are still not many people who are interested in doing such a problem. MBZUAI, as a university, hopes to provide an environment in which people who do such research can be recognized, honored, and rewarded accordingly.
Heart of the Machine: You just talked about what a good research taste is, but how do you cultivate this research taste?
Professor Xing Bo: Cultivating taste is actually more difficult than using good taste. It's a much more upstream problem. I don't actually have a complete answer, but I can share some observations. I have to admit that the social environment we are in now, or the overall technical environment, is not as easy to cultivate good taste as in ancient times or modern times, which is my personal observation. Because of the cultivation of taste or vision or vision, it cannot be quickly achieved, and then there needs to be enough independence. Today's social networks actually make it harder not to rush and not be affected. Therefore, "prudence" is very important, not to mean solitude, but to have a certain ability to allow oneself to be independent of the noisy secular world.
In the natural society, you can imagine two scenes: one is a scenic area with high mountains and long waters, connected to and isolated from people, like Zhangjiajie, Yosemite National Park or the Alps, the clouds are clear, the leaves are cold, you will find it interesting, and then you have an inspiration, willing to go to a high place, go to the mountain to see the scenery, and are willing to go down to the river to see the end. Because nature's non-connectivity stimulates your desire to explore and appreciate; you can also choose to sit there meditate, get along with yourself, and think. But you can imagine another local scene, that is, a horse flat river, a thousand miles of red land, and even mud and sand, which is difficult to make people feel the same as before.
The current social network is a bit like a kind of mud and sand in people's social life, because everything is connected together, it is difficult to isolate, and all kinds of information will even take the initiative to pounce on you. If the reservoir is not closed and the water is all connected, it is impossible to accumulate potential energy to generate electricity. Now it is very difficult to quit the internet, most people can't stand it for half an hour, don't say two days. It's hard to calm down and cultivate a taste, or cultivate a deeper level of thinking, or even think about the same problem for more than 48 or 72 hours. But in ancient times, Shakyamuni would sit under the Bodhi tree and meditate, archimedes were still buried in calculus when the enemy approached, or many sages and monks would have such a process, high concentration to face themselves, or face nature, or face God, there will be a kind of introspection or a kind of introspection.
Zeng Zi said, "I have three provinces and my body", and it is difficult for me to imagine how the three provinces and my body are in the current environment, and none of the provinces can do it. So I don't know how to cultivate my taste now, if I want to find an easy way, maybe give myself a little bit of this space to face myself, and then face my own problems to think about. Of course, it is also very important to read more good books, it is some more heuristic, not necessarily completely thesis, can be humanistic, can be historical biographical, to enrich their inner world, enrich their spiritual world. This taste may be able to brew up slowly, which is a feeling for me. Of course, diligence is also very important, try to contact as much as possible, so that you can keep your thinking sensitive, have the ability to get inspiration or explore.
Heart of the Machine: Do you take time to meditate, or do you keep yourself in a quieter environment like you just did?
Professor Xing Bo: I am trying, this is very difficult to do, because this is not completely controlled by yourself, once the network is closed or shut down, it will also make people who want to contact you because they can't get your feedback instructions and are anxious. Although it is not easy to do, I try to give myself time, even the determination, to shut up for a few days or a few hours. It's getting harder and harder to do, but it's also becoming more and more necessary, and it's actually a very rewarding experience. No matter how busy you are, you should still give yourself time to think.