"Zhang San, graduated from a prestigious university and participated in research on a certain subject..." The graduation season is full of resumes like this. "Behind the dry and identical resumes, there are actually vivid individuals." Xue Yanbo, head of the professional science laboratory directly hired by BOSS, called the job applicants presented by this resume as "paper people", "from the data From a perspective, job seekers actually include N-dimensional information, such as time, personality, psychology, etc." Xue Yanbo said. However, in the current job hunting process, job seekers cannot be understood from multiple dimensions.
The emergence of artificial intelligence is expected to change the contradiction between "paper man" and "N-dimensional". On July 25, BOSS directly hired to announce the establishment of a science laboratory. Xue Yanbo said that the information state of the "paper man" when looking for a job can be "restored" to the greatest extent by artificial intelligence. In the world of "information maintenance", the position will also be "Restore" from a single job description and recruitment notice to practical "parts" in the operation of society.
Coincidentally, there have been media reports recently that IBM is already using AI (Watson) to predict the future work potential of employees, and its motivation is also that the traditional paper-based evaluation methods are difficult to get the correct results that match the job. As Nickle LaMoreaux, vice president of remuneration and benefits at IBM, said: "It is too restrictive to decide whether you can be promoted based only on historical performance."
Artificial intelligence "mind-reading" now has to "upgrade" the human-post matching program. How can it break through and help people and positions match to obtain the optimal solution?
Establish goals and design two-sided markets that "bite" and match
"The low degree of matching between people and positions has caused a lot of human time to be spent on waiting, mismatches, and negative actions." BOSS directly hired CEO Zhao Peng said that the current situation of mismatches and mismatches is actually unnecessary in the talent market. "Internal friction".
In the process of job hunting, the simplification of "people" is one of the reasons for "internal friction". AI has the ability to present job seekers in a three-dimensional and historical manner. "Not only is it multi-dimensional matching, but it also includes issues such as when to match and how to match optimally." Xue Yanbo explained. For example, if a job is offered to a certain applicant today and 7 days later, the results may be different.
"Robot assessment" is gradually being applied in some units, but the assessment system of the time dimension is not considered. For example, the robot will not recognize short-term fluctuations in the status of the applicant for the day, and may consider this to be the norm. The limitations of this type of system also cannot reflect the matching situation of the macro-level job market.
Xue Yanbo believes that the "occlusal" matching can be described as two solvable equations: one is that there is no preference between A and B but there is no pairing; the other is that A and B, and C and D are already paired, but there is a better combination. Make the whole market situation better. From the micro to the macro, through the matching of individual or part of the market, the entire professional market will show a "stable" state. The goal of occupational science using artificial intelligence is a stable market, and it is a scientific problem that can be planned and designed through module disassembly and algorithm modeling.
Unlike some scientific issues with clear issues, the issue of recruitment and employment also needs to consider human factors. Xue Yanbo said that the traditional use of big data analysis to solve the recruitment problem is mainly through computational science, data mining and other methods, which are regarded as engineering problems. We believe that the parameters that need to be added include psychology, sociology, economics, and labor relations. Disassembling the humanities into modules and introducing them into the parameters of AI and participating in the mapping relationship of deep learning neural networks will make it more likely to establish a "stable" bilateral market.
Based on big data, supplemented by generative machine learning
For artificial intelligence, no matter what professional field it is, data is always the basis for seeking optimal solutions. Which data to choose, from which dimension to choose, and how much to choose is the first step.
According to the data, the data used by IBM's "Watson" includes employee information, historical projects taken over, employee experience and performance, employee training and learning recorded in the internal training system, etc.
The accumulation of human resource data from different sources is always advancing. According to the relevant person in charge of the Ministry of Human Resources and Social Security at the end of last year, the talent quality assessment service launched by the Ministry of Human Resources and Social Security had already assessed nearly 500,000 people. The relevant data directly hired by BOSS shows that the current sample size of the data held by the platform is about 40 million.
"The existing data is enough to support us to do some initial scientific research projects." Xue Yanbo said, but for some sensitive or unobtained data, some machine learning methods can "fill in", such as generative machine learning methods. He added and explained: If there are Zhang San and Li Si on the platform, and a character between Zhang San and Li Si is needed in the data, a machine learning model can be trained to "derive" a model that conforms to Zhang San and Li Si. Characters with intermediate characteristics are used for research work such as job matching in the real world.
Algorithms and models are still being explored
"Machine learning has three important pillars, data, models, and computing power." Xue Yanbo said that data is increasing exponentially, and the development of models is relatively slow. There are not many models available for machine learning. The professional market is A brand-new market may require brand-new models to solve problems, and the biggest challenge may be model design.
"The work we are currently working on is to classify people with similar career plans through collaborative filtering, so as to understand the real job hunting intentions; and to try to reconstruct three-dimensional work scenes through collaborative methods." Xue Yanbo said, This will make it possible to solve the problem of unknown preference lists in the theory of occupational science.
"Collaborative filtering, also known as collaborative filtering, is a common recommendation algorithm. It first appeared on Amazon. For example, what users who bought this product generally bought." Smarter CTO Mo Yu explained that the algorithm can be used by buyers The similarity of the group is used to evaluate the similarity of different commodities; at the same time, the similarity between different people is evaluated according to the similarity between the sets of commodities purchased by different people. Achieve "things gather together, people are divided into groups", and then match through different lists, and then make recommendations between people and things.
In the early stage of professional scientific research, Xue Yanbo said that the algorithm will be used to subdivide the list of people and positions. The reason for establishing such a preference list stems from an important assumption of microeconomics. “Only when both parties know what the other party’s preferences are, can a stable matching market be formed. For example, large companies know that well-known university graduates are willing to come, while entrepreneurial applicants are more inclined to enter small start-up companies. The preference list will help to form a perfect market match."
Through deep learning, the preference list can be further improved. Try to do some actual person-post matching, and the matching results will in turn influence the preference list and make corrections.
Zhao Peng said that there are nearly 600 million people working in tens of millions of companies in China, but they are concerned about the sense of accomplishment, happiness, and security of people in the workplace, and the competitiveness, insights, and the matching of both parties in the competition for talents. There is a lack of systematic research on such issues. It is hoped that through the opening of "professional scientific research", from a scientific point of view, using rigorous methods and introducing new technical means such as artificial intelligence, a systematic study of the science of "professionalism" will lead to industry-level research. s concern.
Words of a Family
Can assist recruitment, but some responsibilities cannot be undertaken by TA
Zhang Galun
IBM began to use its own AI Watson to "decide" employees to stay. According to the scientific and technological media "qubit", Watson is changing the working state of HR. It will retrieve employee information and their historical project performance, understand the training and learning situation of employees, and comprehensively judge whether it is suitable for promotion and salary increase, and whether it is possible to reach the peak of life.
sounds good. However, one thing needs to be clarified. Judging from the information currently disclosed, AI will not play a decisive role in this entire evaluation process. Those colleagues in the human resources department are still your "gentle killers" on your promotion path.
A few years ago, people began to discuss the possibility of introducing AI into human resources. AI can solve the problem of fast matching, which is indeed a liberation for HR work. In some large companies, tens of thousands of resumes are received during school recruitment, and keywords must be set for rough screening. After the coarse screening, a fine screening is required to see the matching degree between the applicant's ability and the position. AI is handy in this regard. It can even collect other information about candidates, draw portraits of job applicants, and determine whether or not to enter the next round of recruitment based on the data it has accumulated.
However, to continue to apply AI to other areas of human resources, I am afraid we need to be more cautious.
One thing to keep in mind is that AI does not have any "mysterious power from the world of data", what it can give is only a reference. If you are superstitious and blindly follow AI's judgments, it is not only irresponsible, but even immoral.
Matters involving people are mostly complicated, and there is no standard answer. The industry and academia generally agree that machine learning is a "black box." You feed the artificial intelligence data and adjust the algorithm model again and again. You know that the accuracy of its judgments is getting higher and higher, but you don't know why.
The "heart" of artificial intelligence is also a needle on the bottom of the sea. You don't know what it has learned. Similarly, you don't know how much bias is contained in a company's human resource algorithm model.
AI can evaluate its potential based on the employee information it has, but how does it evaluate it? Are there any ethical and moral risks in the evaluation process, and where is the boundary of its data collection? Zeynep Tufekci, a social technologist, gave a public lecture. She gave an example: a machine can infer things you haven’t made public—for example, it thinks you have a high probability of getting depression, or it thinks you have There is a high probability that you will get pregnant within three months. As a result, it cut off your way "intimately" in advance, and until this time, you are still in the dark, not knowing why you were drawn into the "pending zone" by the machine.
So, is this a precise prediction or a blatant prejudice? Prejudice may be everywhere, and when it is imposed on the group in the name of a machine, it will become more hidden.
No matter how far the development of artificial intelligence is, people must bear their responsibilities openly and resolutely. HR should understand and examine their own prejudices in rounds of assessments of employees, and reflect on their own corporate culture.
This involves complicated judgments, and one cannot "snap the pot", let alone be absent.