Is AI interview very popular now? Is it really reliable to use this for interview recruitment?

According to the previous job-seeking experience shared by job seekers in Guipin.com,

Since last year, many gold-lettered signboards in the school recruitment season have also joined the AI interview/digital interview in the interview process. Many recruitment platforms regard intelligent interview system as their business growth point and play the role of technology disseminator.

As for the applicant, knowing that he has to face the tireless, heartless and observant "roadblock" of AI, I'm afraid many people will feel Alexander. Freshmen, in particular, used to be guided by the words and deeds of their seniors and sisters, but there is really not much effective experience for reference in the market to face the new thing of "AI interviewer".

Today, let's talk about where the boundary of AI interview is and how to "attack" it.

Behind the proliferation of AI, the conspiracy of employers

If you see a company or position you like, you need to go through AI first. Don't sell yourself short. Strategic contempt and tactical attention are often prerequisites for AI interviewers under the "Raiders".

The reason why I say this is because there are too many "golden daddies" on the market to promote the truth in the name of AI. Among them, the field of fast-moving consumer goods closest to the mass market is the hardest hit area. Most of the first people who put up "AI interview" signs in school recruitment and other links are like this, such as food and beverage (Coca-Cola), daily necessities (Procter & Gamble, Unilever) and so on.

On the one hand, most of these enterprises have no special professional restrictions, bearing the aura of top 500 multinational companies, and the pressure of resume screening is very great; At the same time, school recruitment is also a very good opportunity to "attract attention" among young people. Many companies will play the slogan of "looking for future managers", and the competition is extremely fierce. Creative recruitment methods such as open-ended questioning, AI interview and game evaluation can also effectively help them expand their influence and strengthen their brand image.

This also determines that its AI interview system needs to meet two core demands:

First, whether the timing of the launch is fast enough to "take the lead" in publicity; Secondly, whether the diversity of data and the robustness of the algorithm are high enough to avoid detonating the negative emotions of candidates due to discrimination and passing rate.

Therefore, the AI interview solutions adopted by these brand enterprises are mostly polished by third-party algorithm companies combined with some mature AI technology applications on the market. This means that the AI interview system can only play a certain auxiliary role in the initial rough screening, and it is difficult to fundamentally decide whether to get an offer. On the contrary, it can avoid the deviation caused by the interviewer's perceptual judgment (such as face value, accent, graduate school, etc.). At the beginning, let more candidates with job matching have opportunities; On the other hand, the training logic based on the deep neural network model, as well as the current technical ceilings such as NLP, face recognition, and emotional algorithms, also give candidates the opportunity to take a look at the inspection points of the AI ? ? system and break through them one by one.

Let's meet these special interviewers with different abilities.

Raiders difficulty one star: Q&A AI interview

This product often combines NLP natural language understanding technology with speech recognition algorithm to obtain some key information related to job matching in the form of question and answer.

Japanese talent company En Japan asked the graduating college students to have an hour-long question-and-answer interview with a mobile phone that can only ask questions, including 126 questions. I was speechless when I saw that the job seeker was finally abused. ...

This kind of "interviewer" first analyzes the reaction speed and psychological emotion of the candidate by means of phonetic algorithm and phonetic intonation statistics. In addition, with the help of NLP algorithm, the key words and semantics of the answers are analyzed, and the preliminary judgment of matching degree is made by combining the problems related to the business and job requirements of enterprises, such as the "Eight Questions of Procter & Gamble" that FMCG must ask and its variants.

In the face of these interviewers who only select people based on their voice and language characteristics, because their parameters are artificially set and some hidden perceptual preferences are excluded in advance, their judgments will be relatively fair and their strategies will be targeted.

First of all, we know that most of these interview softwares are pre-set standard questions for questions and answers. For example, the En Japan exam software mentioned above is a collection of 15 pre-interview question-and-answer training, which mainly involves basic information, work skills, personality characteristics and so on. Unlike human interviewers, they sometimes talk about relaxing the atmosphere at home, or put pressure on them by asking some tough questions. Artificial intelligence interview systems usually only interact and ask questions in a one-on-one way. As long as they do more homework before attending the interview, understand the corporate culture and employment concept of the employer, whether it is fast-paced, creative or global, reasonably adjust and consciously train their pronunciation, they can avoid many unexpected situations.

At the same time, just as the teacher repeatedly reminded before the college entrance examination that "beautiful writing can get more impression points", it is best to do what you like when answering AI questions, and try to use some languages that conform to machine logic to organize the language. Because AI is mainly based on real-time voice transmission, keyword extraction and matching, and semantic understanding to judge whether a job seeker wants to match the job description. Therefore, it will be more helpful to find out some necessary hard indicators and skills and point out the key words in some sentences, such as leadership, national projects, conversion rate, team awareness and so on.

As long as you follow the picture, you may find AI easier than on-site interview.

Raiders difficulty two stars: video AI interview

If the company you are interested in is stronger or knows more about technology, then the probability of meeting a senior interviewer like video AI is even greater.

In short, the video AI interview is based on intelligent question and answer. AI will also analyze the facial expressions and muscle movements of candidates in real time to judge the authenticity and personality tendency of candidates, and multi-dimensionally examine whether candidates are close to the ideal candidates of enterprises.

Many students are afraid that their hearts are half cold. Is it possible to "betray" yourself even by rolling your eyes and looking around? Can you save yourself by watching Lie to Me (an American drama with the theme of micro-expression psychology)?

An interviewer who wants to know about reading and listening, in addition to the necessary internet search and gfd (see human interviewers should also pay attention), I am afraid he has to fight a prepared battle from the technical cognition.

To be sure, there is not enough scientific basis in the algorithm for emotion recognition of facial expressions using AI. Even the AI giants such as Microsoft, Google and IBM, their emotion recognition algorithms are not rigorous, and it is likely to be seriously misleading when applied to recruitment scenarios. Therefore, it is also considered inappropriate for job-hunting software to monitor emotions and screen job seekers.

For example, frowning occasionally does not mean "angry" emotions, nor does it mean that the interviewer must have the characteristics of being difficult to cooperate. In fact, video interviews are more helpful to find those who perform best.

Take the example of HireVue or Sonru adopted by large groups such as Goldman Sachs, JPMorgan Chase, KPMG, Unilever and L 'Oreal. Its principle is to recognize 65,438+05,000 features, including the language chosen, the vocabulary used, the expression of eyes, the size of voice and so on. Then, combining these characteristics, according to the feature database of previous "successful" candidates, judge a person's reaction, emotion and cognitive ability. Finally, through the ranking algorithm, a certain proportion of the best candidates enter the next round.

It is understood that Hilton Group has used its algorithm to interview more than 43,000 positions. Globally, the HireVue system provides 65,438+0,000 interviews and more than 65,438+0.5 million pre-employment assessments every quarter.

As its technical director said, "the extreme complexity of human language, body language and expression requires great attention to algorithm deviation and potential harmful effects." If customers screen out 90% job seekers in some topics, it means that "the scope of the inspection point is too small" and will make changes.

After understanding the basic principle of this video AI interview, you will find that you don't need to be cautious about AI. Of course, we can find some tips that are good for the algorithm, but we suggest you play as usual, because every habit may affect the fun of your future work.

For example, only 10%-30% of the score weight is determined by facial expressions, and most of the others depend on the interviewer's speech performance. In terms of vocabulary, use more characteristic words that meet the preference of the target enterprise. Whether job seekers like to say passive verbs or active words, whether they often use "I" or "we" and whether they often use technical vocabulary will affect the evaluation of the matching degree by the system.

Another example is the tone of voice. If some people speak really slowly, it may not be suitable for telephone consultation. If they speak too fast, users will not understand. Use empathy to find the state that best suits your favorite position. Perhaps "feeling" is more reliable than data.

Difficulty of Raiders Samsung: AI who can read minds

Have you already felt the loneliness of job seekers when you hear this? Don't be sad yet. If you are "unlucky" to be interviewed by the AI publishing system that reads your social network, you can only buy a lottery ticket to comfort yourself after "streaking" under the algorithm.

At this level, AI often analyzes the daily behavior of job seekers based on some complex data, and then deduces its matching degree with the position. Not long ago, California startup Predictim used NLP technology and computer vision technology to scan the Facebook, Instagram and Twitter history of nanny candidates, and then predicted whether they might bully or harass others, abuse children and so on.

Of course, such an interviewer was quickly resisted by the industry. Facebook believes that the company violated the ban that prohibits developers from using this information to review job seekers, thus greatly limiting the company's access to user data on Facebook and Instagram. Twitter also interrupted Predictim's access to its API on the grounds that it was forbidden to use Twitter data for monitoring purposes.

Similar algorithmic risks also occur in the job search platform LinkedIn, because the third-party website HiQ collects LinkedIn data to predict when employees may leave.

The reason why technology giants use such an AI interview system to clarify the boundaries is because machine learning cannot reliably explain the subtle differences in intonation and speech, such as irony or jokes, and its application in interview scenarios is very unstable; At the same time, this algorithm is unsupervised, that is, it has a black box, and some employees who might have been very dedicated are likely to lose their job opportunities without knowing the reasons or explaining them.

In addition, it is obviously not in line with technical ethics to hand over the decisive factors of interview success to AI. If enterprises directly use small-scale and single data sets for training out of the preference/prejudice of decision makers, the fairness of AI interview will disappear, and may even aggravate the discrimination and exclusion of enterprises in age and race. As Anna Cox, a professor of human-computer interaction at UCL, said, "Any data set will be biased and people who are really good at this job will be excluded."

At present, it seems that AI's analysis of the complex elements in the recruitment interview process is still a controversial future.

However, the wheel of technology is accelerating. IBM announced that it will use Watson to actively search the data of the internal training system, understand the training and learning situation of employees, and judge whether they have promotion potential, so as to conduct internal evaluation. From such a relatively structured data to the interview and recruitment process, it may also be directed at the majority of job seekers.

It can be said that in the past, we saw some fields with high mechanization and structured data, such as documents, translation and identification. Have been replaced by AI. Nowadays, interview, an emotional communication field, has not been spared. Fortunately, the magic of technology is fading, and human beings holding the sword of knowledge will finally find the best way to "* * *" with AI. Iterative optimization in repeated contests and confrontation, and finally get out of the pain of man-machine running-in.