news

author: winnie

edited by guan ju

li feifei is raresocial media active members, who can be called a “big shot in the ai ​​world”, updates her twitter almost every day.

one of the latest contents is a forwarding of the latest research breakthroughs on robots from stanford university’s ai lab:

in the video, two robotic arms smoothly cooperate with each other to fold clothes, pour tea, and pack shoes, with perfect fluency.

these smooth movements are achieved through a process calledrelationship keypoint constraint (rekep)achieved with new technology.

this technology solves the complex constraint simulation when the robot interacts with the world.spatial intelligencethere is great potential in this area, and this is exactly the direction of fei-fei li’s latest entrepreneurial venture.

when reposting the post, fei-fei li wrote:

new work from our lab demonstrates deeper integration of vision and machine learning!

at the end there isstarry eyes and cheer upthe small icon of the camera overflows the screen.

fei-fei li’s twitter profile is a series of hashtags without a single adjective:

professor of computer science at stanford, co-director of the stanford institute for human-centered artificial intelligence (hai), co-founder and chairman of ai4all, and researcher (ai, computer vision, machine learning, ai healthcare).

although she was awarded the title of tenured associate professor at stanford university at the age of 33, was the first female director of the stanford university artificial intelligence laboratory, the first sequoia chair professor at stanford university, and was elected as a member of three american academies, she created the imagenet data set, a key catalyst for modern artificial intelligence...

these groundbreaking achievements are undoubtedly dazzling in the global science and technology world which is almost dominated by men, but she has never boasted about them, as if they were just small footnotes on the road to pursuing her scientific dreams.

in addition to the latest ai progress and some articles on ai regulation, fei-fei li also tweeted about her latest autobiographical work.《The Worlds I see》, the chinese version was published in april this year, entitledfei-fei li: the world i see。

unlike her high profile in ai, fei-fei li was very open when promoting her new book.humbleandlow profile。

this contrast also runs through the entire book "the world i see":fei-fei li spent a lot of time and effort to describe the vast history of the development of artificial intelligence and her personal dream-chasing history, as well as those who have given her even a small help in her life, but only briefly mentioned her own achievements.

but who else can, like her, write her personal history into a magnificent and eventful history of the development of artificial intelligence?

“migrant”“female"the combination of the two disadvantages further increases the difficulty of breaking through.

from chengdu to new jersey, why did fei-fei li, who was obsessed with physics, turn to ai? what lit up her north star? how did she achieve a life against all odds in the technology industry? why did she create ai4all? what is her view on artificial intelligence? what problems is she solving now?

the answers are all in the book.

feifei and curiosity

the origin of li feifei's name is quite dramatic.

in 1976, on the day li feifei was born, her father was late, not because of traffic jams or other accidents, but because he was in a whim and ran to the park to watch birds and forgot the time.

naming"fei fei", which is also what i think about when bird watching.

the mother was furious, but she also thought it was a good name.

li feifei was infected with her father's curiosity. when she was a child, her father often took her to the park to watch birds, to the fields to see buffaloes or catch insects, which stimulated her strong desire to explore the world.

middle school contactphysicslater, this became her most obsessed subject, and she even thought about the changes in acceleration and angular momentum when turning a bicycle.

she was also interested in one great figure after another in the history of physics, and she was obsessed with events such as archimedes discovering the law of buoyancy while taking a bath, and newton hiding in his hometown to write "mathematical principles of natural philosophy" when the plague was raging.

in order to pursue fei-fei li's scientific dream, after weighing the pros and cons, her parents and their family moved to the united states.

when she was in high school in the united states, she was particularly fascinated by the theory of relativity, but she couldn't understand it and found it very difficult. but suddenly one day, she seemed to have understood the theory of relativity in her dream. she was very excited and got up in the middle of the night to write.

following her curiosity about physics, fei-fei li found her north star.

when she was a sophomore, she began to read some books by physicists. coincidentally, the physicists she admired most, including einstein, in the second half of their lives, not only thought about physical problems, photons, atoms or the origin of the universe, but began to think about problems of life.

li feifei also began to be curious about life. what made her most curious about the story of life was intelligence. why is life intelligent? what exactly is intelligence? is it only human life that can be intelligent?

because of these thoughts, fei-fei li discovered ai when she was studying for her doctorate. at that time, no one knew about ai, and it was still the winter of ai.

curiosity is like a searchlight. when you turn on your curiosity, the light can shine on different places. maybe the first or second place you see is not what you like, but if you keep looking for it through curiosity, you will see something that makes you very excited and interesting. -- fei-fei li

"wasted winning lottery tickets" and north star

life was tough when we first arrived in the united states. a family of three lived in a one-bedroom apartment, and the furniture was picked up from garbage discarded on the street.

li feifei's father found a job repairing cameras in a chinese-owned store, and he got off work very late every day. her mother found a job as a cashier in a store, doing repetitive and mechanical operations, away from the literary works she loved.

in order to supplement the family income, fei-fei li would do odd jobs every day when she was not in school.

the most common job is to work as a handyman in a chinese restaurant.work from 11am to 11pm, a total of 12 hours, and the hourly wage is $2.

when she was working as a handyman in a restaurant, during her afternoon shift change, li feifei would use the rare free time to read the literary classics her mother shared with her, but the restaurant manager was dismissive of this. "he felt that for people like us, imagination is redundant in life," li feifei recalled.

for these chinese immigrants who can't even speak english, class transition is an unattainable dream.

the academic achievements that she was once proud of have also become a burden. fei-fei li can barely speak english, and every day when she goes home to do her homework, she has to prepare two dictionaries, one for english-chinese translation and one for chinese-english translation, otherwise she will not be able to do her homework.

despite having very little time to study, fei-fei li still passed the exam.sat 1250, full score in mathematicsresults.

she applied to three universities: mit, princeton, and rutgers university.

in the end, all three schools sent admission notices, and princeton university even offered an almost full scholarship.

the news that fei-fei li was going to study at princeton caused a sensation in the entire community and was even featured in local newspapers.

but it is a bit regrettable that fei-fei li did not choose medicine or finance, which are stepping stones to the wealthy class, but chose what she is obsessed with - physics.

"it's such a good lottery ticket, what a pity it was wasted." the neighbors couldn't understand.

in 1999, fei-fei li was about to finish her undergraduate studies at princeton university and was once again faced with a choice between her scientific ambitions and real life.

she was torn between the temptation of postgraduate study and the pressure of starting a professional career.

she was approached by a host of big names, including goldman sachs and merrill lynch, offering everything: benefits, opportunities for advancement, a lucrative starting salary, and, of course, health insurance. they promised to forgive her debt, end her dry-cleaning grind, and provide for her family as her mother's health deteriorated.

and the only requirement was that she give up science.

after thinking about it for most of a week, fei-fei li had the following conversation with her mother at the dry cleaner (the following is an excerpt from the original text):

"mom, i'm considering a few options. i interviewed with a few 'companies', or 'companies' in chinese, right? they're wall street giants. i have to admit, the offers they made were pretty tempting."

"a wall street tycoon?"

i realized that she was not familiar with these american cultural terms.

"it's stocks, trading and stuff. investing. of course, there's still a lot to learn, but if i really set my mind to it, i think i can learn it."

"well," she replied calmly, "is this what you wanted?"

"i mean... the salary alone would be life-changing, and—"

"fei fei, is this what you want?"

"you know what i want, mom. i want to be a scientist."

"what else is there to say?"

fei-fei li chose to continue her studies.

towards the end of her doctoral career at caltech, fei-fei li once again found herself at a crossroads in her life.

a partner from mckinsey, a world-renowned management consulting firm, came to recruit. they were looking for an intern-level analyst.

li feifei applied for an interview and even bought a suit that was far beyond her budget.carefully hide the tag under the collar, so you can return the item immediately after wearing it.

the interview went incredibly smoothly.

mckinsey immediately gave a positive invitation and decided to bring fei-fei li’sinternship opportunities turn into permanent positions.

the job seemed to free her from a burden she had been carrying for so long and was a shortcut to the career that every immigrant family hoped for their children.

but the price was giving up her north star - science.

li feifei returned home to share this "good news", telling her the starting salary and generous benefits (the following is an excerpt from the original article):

"do we really want to discuss this again?"

"mom, i know, but listen to me—"

“i know my daughter. she’s not a management consultant or anything like that. she’s a scientist.”

"think about your health, mother! think about our expenses. what can academics bring us?"

"fei fei, we haven't come this far for you to give up now."

"this isn't giving up! this is my dream job, a career that will get us out of our current predicament. look at what we're living in now! three adults living in one dorm!"

the mother paused for a moment, perhaps thinking about these words, and then responded, "fei fei, you keep saying that the path you're taking is 'selfish,' as if you're pursuing science at our expense."

"how could i not feel that? i could provide for my family now, and—"

"you don't understand what i mean. this has never been your path alone. from the beginning, this has been the path of our entire family. whether you are destined to become a scientist, a researcher, or some other profession that i can't imagine, and whether you can make money from it or not, our whole family has been working towards this goal since the moment our plane left shanghai."

i don't know what to say.

“i’ll say this one last time: we didn’t get this far for you to give up now.” she was right.

she was always right. this time, for some reason, i finally listened to her. i would never question my path again.

“i’ve heard enough dissuasion to last me a lifetime” and imagenet

in 2006, computer vision research was still a discipline that was underfunded and received little attention from the outside world.

many researchers are focused on building better algorithms.

they firmly believe that algorithms are at the core of computer vision. if we compare machine intelligence to biological intelligence, then algorithms are equivalent to the synapses of the machine, or the intricate neural circuits in the brain.

but fei-fei li doesn’t think so. while pursuing her doctorate, fei-fei li realized the limitations of this research approach:

even the best algorithm won’t do its job well if the data it’s trained on doesn’t reflect the real world well.

fei-fei li's idea is to build a data set that can fully reflect the real world.

teaching machines to recognize objects in pictures like humans does is a major challenge that the field of artificial intelligence research has always hoped to overcome.

this is also fei-fei li’s most important contribution - the creation of the imagenet database, which has been commented on by practitioners in the field of artificial intelligence."without imagenet, there would be no deep learning revolution today."

imagenet’s ambitious goal is to collect 1,000 unique images for each category, covering 22,000 categories, from violins to german shepherds to pillows, which requires a total of about 20 million images.

but at the beginning, fei-fei li's idea of ​​imagenet was unanimously opposed by almost everyone."i've heard enough dissuading voices to last me a lifetime (and probably the next)."fei-fei li said.

finally she met her first supporter, professor li kai, a leading figure in the field of microprocessor architecture.

microprocessor architecture is the art of arranging millions of nanometer-scale transistors into the world’s most complex devices, so he understands the power of exponential thinking better than most. he believes li is on the right track.

professor li kai

he donated a workstation to fei-fei li's research without hesitation, solving the computing power problem that the project urgently needed.

at the same time, because he was about to leave his post, he also recommended his student deng jia to fei-fei li.

fei-fei li and jia deng formed a two-person team and started this difficult project.

at first, li's strategy was to pay undergraduates $10 an hour to manually search and add images to the database.

"i can understand it in theory, but the workload is too huge. it's an astronomical figure and it can't be completed by just a few google searches." deng jia looked doubtful.

at the established pace, it will take a full 19 years to complete imagenet.

at the suggestion of a graduate student named sun min, fei-fei li learned about the turkish robotic project developed by amazon, which uses human crowdsourcing to complete detailed, precise and intelligent tasks that computers are unable to perform.

it was this robot that transformed imagenet’s team of college student annotators into an international team of dozens, hundreds, and thousands of people.

as support continued to grow, dunja's estimated completion time was dramatically shortened, first to 15 years, then 10 years, 5 years, 2 years, and finally less than 1 year.

however, the funding needs were approaching the limit of what the team could afford. in 2009, fei-fei li jumped to stanford, which provided new research funding for her research.

in june 2009, the initial version of imagenet was finally completed: a collection of 15 million images covering 22,000 different categories, selected from nearly 1 billion candidate images and annotated by more than 48,000 global contributors from 167 countries.

imagenet not only reaches the level of scale and diversity that fei-fei li has dreamed of for many years, but also maintains consistent accuracy: each image has been manually annotated, organized in a hierarchy, and triple-validated.

in terms of quantity, fei-fei li has achieved her set goal and established the largest manually edited data set in the history of artificial intelligence at that time.

"our research is bold and forward-looking. although it is not complete, it is thought-provoking, and many of them are conceptually simple. but it was not until the emergence of imagenet that everything became feasible," said fei-fei li.

for the first time in history, machines have the ability to "see" like humans.

more importantly, her idea of ​​using big data to train multi-layer neural networks, expanding from images to other fields such as voice, text, and video, ignited the ai ​​revolution that continues to this day.

ai4all (ai for all), eliminating bias

yahoo flickr was plagued by troubles soon after its launch in 2015, with one blunder after another, describing a monochrome portrait of a 56-year-old black man as an "ape," labeling a photo of the gate of the dachau concentration camp as a climbing frame, and labeling a white woman with colored powder on her face as an "ape."

google was also embroiled in a similar controversy after its google photos service incorrectly labeled two black teenagers as "gorillas."

artificial intelligence is a "sea of ​​men". the lack of diversity in datasets including imagenet has led to algorithms being unintentionally biased and performing poorly on non-white, non-male users; untested algorithms and questionable decisions have further exacerbated the negative impact.

when the internet presents a picture of daily life that is predominantly white, western, and male, it becomes difficult for technology to understand other groups of people.

in addition to the imbalanced dataset, is there a problem with the model itself? are there undiscovered weaknesses hidden in the algorithmic architecture that relies on all the data? is there a problem with the learning technique that can facilitate the training process?

to combat bias, fei-fei li and her student olga opened their stanford ai lab course to ninth- and tenth-grade girls.

the idea of ​​inviting a small group of people to take an ai course was so popular that their project quickly snowballed into a national nonprofit, with a presence on campuses across north america and an expanding mission.

soon, they began offering similar programs to marginalized groups, such as students of color and economically disadvantaged students.

just a few years later, the project was officially named ai4all.melinda french gatesmelinda french gates’ pivotal ventures andnvidia founder jen-hsun huanga round of funding was provided.

ai4all continues to impact the world.

it focuses on students who are currently underrepresented in the ai ​​field, such as girls, african americans, latino students, or children from low-income backgrounds, with the goal of inspiring more young people to pursue ai jobs to increase diversity in artificial intelligence and offset evidence of bias and discrimination embedded in machine learning algorithms.

as the industry pursues the future of artificial intelligence, it often acts recklessly and lacks self-reflection, but ai4all's efforts show that at least a small number of people are moving in the opposite direction.

“human-centered ai”

in 2018, after returning to stanford university, fei-fei li officially announced the launch of the human-centered artificial intelligence project and established the stanford human-centered ai institute (hai), with the goal of putting humanity at the center of artificial intelligence.

hai co-directors: john etchemendy and fei-fei li

in her proposal, she mentioned that "human-centered ai" comes from three simple but powerful ideas:

1. in order for artificial intelligence to better serve our needs, it must contain the diversity, nuance, and depth of human intelligence.

3. the ultimate goal of artificial intelligence should be to enhance our humanity, not weaken or replace it.

this means that artificial intelligence needs to work with neuroscience, psychology, and other disciplines to create algorithms with more human sensitivity, ensuring that ai helps people get their jobs done rather than replacing them.

fei-fei li and arnie milstein, a professor at stanford university school of medicine, collaborated on“ambient intelligence"technology is a good embodiment of people-oriented.

this is a study that starts from the trivial matter of “washing hands”.

hand washing is the most important means of preventing the spread of infection, but to this day, medical staff not washing their hands or washing their hands improperly remains an important factor in the spread of disease in the medical environment.

li and arnie envisioned a technology designed to fill space with intelligent and reliable perception, without being noticeable.

unlike human monitors, their technology will blend quietly into the background, silently watching and only sounding the alarm when it senses danger.

“whether it’s addressing bias in data or protecting patients in hospitals, the common point is how our technology treats people, especially how to protect individual dignity. ‘dignity’ is the key word i have always emphasized. the most important question is, how can artificial intelligence respect human dignity? this question is the starting point of all research work.”

“human-centered ai.” i’ve been thinking about this phrase for months, and now i finally say it. “i’ve always described my philosophy this way. i hope this phrase can appropriately describe my future career. i hope that in the years to come, ‘human-centered ai’ will have some meaning for all of you.” - fei-fei li