Is AI set to replace recruiters? A guide to using AI well in hiring

Artificial intelligence (AI) within recruitment is becoming increasingly common ... and controversial.

But given all its associated issues, is AI really a smarter recruiting practice?

In our latest newsletter we explore AI hiring practices with VidCruiter, a platform that helps recruiters learn more about candidates, looking at the ways AI is currently being used in recruitment workflows and its perceived benefits, along with the drawbacks and potential legal issues of using currently-available AI recruitment technology.

What is AI recruiting?

AI hiring refers to the recent adoption of using artificial intelligence (AI) to automate parts of the hiring process, from sourcing and screening to interviewing and evaluating candidates.

It’s important to note upfront that not all recruitment tech solutions are examples of AI. For a technology to be considered AI-powered, it needs to feature components of machine learning (the system learns and improves by gathering data, rather than being explicitly programmed).

Here’s an example of an AI recruiting process vs automation that’s used within recruiting tech:

AI-POWERED RECRUITING

Learns the desired skill set and features for a role based on a growing number of data sets and using the information to scan applications and advance qualified candidates.

AUTOMATION IN RECRUITING

(NOT AI-POWERED)

Enables humans to make faster decisions by using preset rules to prioritize applications and advance qualified candidates.

How is AI being used in recruiting?

Here is how some companies are using AI within their recruitment practices:

WRITE ‘ENTICING’ JOB POSTS

• Using data findings, AI can suggest the most effective words and phrases to include in a job posting based on the job title, industry, and location.

AUTOMATIC JOB MATCHING

• In some cases, AI is hard at work before a candidate has even applied. AI has the ability to determine the requirements of a job role based on previous or similar applicants for the role. This data is used to dictate which candidates see a job posting based on their experience, knowledge, and skills.

COMMUNICATING WITH CANDIDATES VIA CHATBOT

• AI can be used to mimic human conversational abilities, using technology such as natural language understanding (NLU) to comprehend a candidate’s text-based messages and know how to respond. These chatbots can be used to schedule or determine job fit. They can even recommend other jobs that might be a fit, and encourage applicants to apply.

FILTERING CANDIDATES

• AI can scan or evaluate candidates’ resumes using an algorithm to scan for keywords that are relevant to a job role or preferred skill set. Some AI software can also analyze candidate data to discover their online presence for a broader scope of their history and abilities.

VIDEO INTERVIEW CANDIDATES

• AI software can analyze video interview transcripts using natural language processing (NLP), without recruiters needing to be present. Candidates complete a one-way AI-based video interview by recording their answers to preset questions. AI performs an algorithmic analysis on the recording and determines the candidate’s outcome.

Here are some features that are analyzed in an AI-based video interview:

• Word and phrase choices
• Tone of voice
• Body language
• Facial expressions
• Emotional responses
• Eye movement
• Communication skills
• Level of interest
• Level of confidence

DO ALL VIDEO INTERVIEW PLATFORMS USE AI TO EVALUATE A CANDIDATE? NO.

Many pre-recorded video interviews – including all interviews conducted on VidCruiter’s platform – do not use AI as of today. Non-AI platforms offer a convenient solution that allows candidates to interview at a time that works best for them, and recruiters to evaluate the videos when they’re available to do so.

Leading video interview providers use a structured interview methodology, which includes a structured rating guide and standard rating scale. This helps to facilitate a fair and comparable interview process that allows recruiters to evaluate efficiently, every time.

How does an AI interview work?

Within the recruiting process, AI “robots” can speak to and understand candidates through the use of conversational AI. Conversational AI is used to facilitate human-like chatbot conversations with candidates, and aspects of it are also used in AI-evaluated audio and video interviews.

How does conversational AI work? The AI-powered application receives the spoken word (or written text) and transcribes it into a machine-readable text. Next, the application uses natural language understanding (NLU), which is the first component of natural language processing (NLP), to understand the intent of the text. Generally, in interviews, the system isn’t required to make a response, so NLP is used to evaluate the text, based on AI algorithms.

In circumstances where a response is required (e.g. chatbot interactions), the system’s dialogue management (DM) formulates a response and converts it into an understandable format using natural language generation (NLG), the other component of NLP. The application delivers the response to the user via text, or text-to-speech, depending on the conversation style.

Lastly, the application uses machine learning (ML) to improve the responses for future interactions by accepting corrections and carrying context from one conversation to the next.

HOW CONVERSATIONAL AI IS USED TO INTERACT CANDIDATES

In theory, conversational AI is an efficient and convenient way to filter and engage with candidates. However, in real-world use cases, it has limitations. Different languages, dialects, and accents fail to be understood properly by AI applications, meaning some transcriptions are full of incorrect information, which can ultimately cause bias. Even in text-based conversations, instances of sarcasm, emojis, and slang can confuse AI, causing misinterpretations.

Would you know if you were speaking to a robot?

Probably not…

In a recent study, 72% of candidates thought that they had spoken with a recruiter, even though they were notified upfront that the chatbot was a virtual assistant.

Why are companies using AI in recruiting?

Companies seek out AI to assist with their recruiting for the following reasons:

Will AI replace recruiters?

AI allows hiring teams to remove many of the repetitive, time-consuming processes from their recruitment workflows. Companies that produce AI recruiting software say this allows recruiters more time to focus on engaging with candidates, training hiring teams, and?developing a better hiring process. However, many AI tools do replace the need for recruiters or hiring managers to engage with candidates. It seems a little contradictory.

Are we heading towards a dystopian future where robots are in full control of corporate hiring? The short answer is no, nor should any company want AI to take over their human task force. AI isn’t able to replace the social skills, empathy, and negotiating abilities?needed for a successful recruitment workflow, particularly while AI recruiting is still considered to be in its infancy.

What are some challenges of using AI in recruiting?

AI technology is a double-edged sword in most use cases. Within recruiting, AI can help introduce efficiencies and eradicate certain time-consuming tasks. However, the software can also create new – sometimes serious – challenges to be aware of:

AI NEEDS A LOT OF DATA TO BE ACCURATE

• Machine learning (the component of AI that allows algorithms to be improved) requires a lot of data to accurately mimic the intelligence of humans. For example, AI that’s used to screen applications would need to screen potentially hundreds of thousands of resumes for a specific role to be as accurate as a human recruiter. Its intelligence is always limited to the data source available, therefore at first, the AI tool may be less than helpful, and even potentially biased.

AI CAN LEARN BIAS FROM PREVIOUS DATA

• Companies that create AI recruitment software often share how AI can eliminate bias from the hiring process through its use of factual information, rather than the subjective, and sometimes biased decisions found in human evaluations. However, saying AI can eliminate bias is avoiding a large part of how AI works – it’s trained to find patterns in previous behavior. As mentioned above, AI extracts insights from large amounts of data, then makes predictions based on its findings. This is what makes AI recruiting so powerful, but it can also make its algorithms heavily susceptible to learning from past biases.
• For example, if a company has more male than female employees, an AI-powered tool can easily favor male candidates to match the current identity of the company, so long as there isn’t a regularization term to stop the system from doing so. In a harder-to-detect example, say many employees graduated from the same university. This could be due to its proximity, or because of a referral program. The AI software could notice this trend, and form a pattern to favor graduates of that university or those with similar backgrounds. This pattern could end up being highly discriminatory towards non-college grads and certain demographics that were less likely to attend that specific university.

AMAZON’S AI HIRING BIAS

In 2014, Amazon created its own AI-powered recruiting tool to help screen resumes, scoring them from one to five stars. Its algorithm used all resumes submitted to the company over a ten-year period to learn how to determine the best candidates. As there was a much lower proportion of women working in the company at the time, the algorithm picked up on the male dominance and presumed it was a factor in success.

Amazon made edits to the software to rectify the issue, but there was no guarantee that the machines wouldn’t sort candidates in another way that could be discriminatory. The project was abandoned a few years later.

AI LACKS THE HUMAN TOUCH

It goes without saying, but humans are complex. AI can screen a candidate’s skills and abilities that are relative to the role, but the system would struggle to analyze many aspects of a candidate’s emotional intelligence that could help them succeed in the company. For example, an AI interviewing platform that analyzes facial expressions and tone of voice along with the candidate’s response isn’t able to determine exactly what a smile and a formal tone mean – does it mean the candidate is sincere and serious? Or possibly, they’re trying to be friendly but their tone makes them seem distant? Perhaps it also depends on the question asked. AI doesn’t have the technology to fully understand the nuances of social cues, and cannot possibly allocate these features to imply the presence or absence of specific skill sets.

Secondly, AI cannot build a rapport with a candidate. As we’re currently experiencing a candidate-driven market, companies need to be able to truly connect with top talent – failure to do so could result in a high-candidate drop-off. In order to win them over, recruiters need to show interest and empathy, and remember details from previous conversations – even if AI could replicate these traits, a system would entirely lack authenticity.

AI CAN MISINTERPRET HUMAN SPEECH

• AI recruiting tools that screen, interview or evaluate applicants will use automated speech recognition (ASR) software that’s also used in voice recognition services. This software listens to the applicant’s spoken response and converts the voice data into computer-readable text data. In theory, this allows companies to rely on AI to capture a candidate’s complete response and evaluate them fairly and objectively.
• However, anyone that’s used leading voice recognition services, such as Alexa, Siri, or Google will know that not every word is interpreted correctly – in fact, entire sentences can be misinterpreted, leading to an incorrect response from the platform. Specific minorities are more commonly prone to these errors. A study conducted by Stanford University found that five leading ASR systems (Apple, IBM, Microsoft, Google, and Amazon) showed substantial racial discrepancies, with an average of 35% of words being incorrectly translated for black speakers compared with 19% of words being incorrectly translated for white speakers.

Black speakers are more likely to be misunderstood by speech recognition software

A study conducted by Stanford University found that five leading ASR systems (Apple, IBM, Microsoft, Google, and Amazon) had an average word error rate (WER) of 35% for black speakers compared with 19% for white speakers.

If the leading ASR systems can’t always recognize and contextualize voice commands, how can an AI software company, with far less funding, create an algorithm that can properly analyze lengthy and often complex interview responses? Unfortunately, they can’t. Even a leading AI-driven interviewing provider states that their software has a word error rate (WER) of ‘less than 10%’ for native American English speakers – so about 1 in every 9 or 10 words are incorrectly translated. The WER was higher for speakers outside of the U.S., depending on their country of origin (e.g. 12% WER for Canadian English speakers and 22% WER for participants with a Chinese accent).

This means that in an AI-powered interview, the software will fail to understand at least approximately 10% of a candidate’s response, and is likely to misinterpret up to a quarter of the response from a non-native English speaker.

Click here to read the complete Guide on AI hiring practices.

Thanks to Shiann Aronson of VidCruiter for diving into these practices with us.

CONTACT US TO LEARN MORE ABOUT HOW DIVERSELY’S TOOLS WORK TO CREATE A SAFE AI DRIVEN RECRUITMENT PROCESS FOR YOUR COMPANY.

Originally published at https://vidcruiter.com/