AI is here to eat the world! As with all functions, hiring and recruiting have been greatly benefitted with the addition of AI. Its importance has even greatly increased because of the COVID-19 pandemic and the subsequent social distancing and safety regulations. Despite its growing popularity, concerns around the ethical use of AI, privacy, and reliability of algorithms still exist. Understanding the tech can help recruiters address these concerns, optimize, and leverage it for the best outcomes. So, who better to provide a technology deep-dive than AI experts from Microsoft: Nalin Mujumdar, Cloud Solutions Architect - Advanced Analytics & AI and Shubham Arora, Solution Specialist, Data Platforms & AI!
Our Head of Engineering, Vishal Madan, sat down with Nalin, Shubham, and our CTO, Sujit Karpe, to talk AI in virtual interviewing tech and what this means for recruiters. Excerpts of the conversation have been included below.
You can receive your recording and well-researched e-kits to help you up your recruiting game here.
VM: I was going through a Gartner study wherein the study said that there will be around 2.8 million jobs that will be created. And around 1.5 million jobs, that would be eliminated. So, we’re seeing demand for social creativity and digital dexterity. So everyone has to be prepared for this change in role. What do you think this means for remote hiring?
SA: On the topic of jobs creation or jobs deletion, I think the only thing that can delete jobs is politicians and not the AI. There are two aspects to it. One is empirical analysis that depending on which analyst you like to read about, whether it is McKinsey or Gartner, every analyst will predict that the number of net jobs that AI will create is only going to increase. Will these jobs vanish? The answer is no.
How much will these jobs change? How much will these jobs morph? That’s something that we’re yet to see. In any automation, the only thing that is missing is the human element of automation. And that is where the jobs will be morphed. We've had four industrial revolutions. I would say the fourth is almost about to conclude. The only thing that changes in those industrial revolutions is the nature of jobs. The jobs don't get deleted. The jobs become much better.
In the space of recruiting, I would say AI is something that can act as a decision support system. AI cannot completely automate recruiting. AI can take the manual grunt work out of the recruiting process, but still the things that are completely driven by human creativity, emotional intelligence, social skills the job of selling my company to a potential candidate – those are still jobs that will be done by humans. The onus is on us whether we adopt or lag behind.
NM: I was actually a listening to this talk by Jeff Bezos the other day. I'd like to tie that into what Shubham just said. Everybody's job has certain competencies, which are automatable. But there are some other decisions you, as a human, take that a computer will probably not be able to decide based on the technology that we have now. As Jeff Bezos says, we are still solving for artificial narrow intelligence. These are specific tasks with a predefined intent. And we're nowhere close to building an artificial general intelligence, something that can set its own objectives.
The point is we like to be productive. And we will figure out and create tools to complement this. So the jobs that are mostly routine now, these will become more automated. And these jobs are, I'm not sure, but not necessarily someone's career calling. So, these routine tasks in a particular person's job will become automated and you can start focusing on higher tasks.
So you can just be taking decisions and running experiments to see what works better to improve our lives on a daily basis. The quality of work will go on significantly improving and the nature of jobs will essentially improve, rather change jobs getting removed from the markets.
SK: So earlier, open jobs saw recruiters sifting through thousands of applications. In recent years, with smartphones, increased connectivity, a fuller labor supply recruiters are flooded with ten times more applications. Yes, they can use online assessment tools like imocha. But where is the intelligence on top of that?
I was talking to a vice president of a large consulting company and he said, ‘I don't want my recruiters and hiring managers to go through those videos to check for malpractices. Can we use AI to flag: green flag means no cheating, orange flag means the candidate did cheat within tolerable limits and red flag shows out of thousand candidates that appeared for the assessment remotely, five cheated, highlight those candidates and flag that data! That’s what we can do with AI.
Roles have definitely increased. How to handle these roles? Through technology! And can the technology become more and more intelligent, ike your assistant that can help you do your job in an intelligent way. So will AI reduce jobs? I would say AI will make jobs and people more intelligent.
VM: There’s sentiment analysis. There is speech recognition. There is tone analysis. There is personality, insights, and everything. So Nalin, given your technical expertise, would you be able to explain in layman's terms what they mean?
NM: Sure, in today's world of computing, these tasks that you just mentioned: text, sentiment analysis, automatic speech recognition, tonal analysis of voices, these are attempts at using AI tasks to solve human cognition, understand how humans communicate using machine learning. And how machine learning works is instead of using rule-based programming on top of these human signals, it attempts to leverage the signals and patterns hidden in the last few months of existing data to predict the implications of new signals.
But what are the best practices or limitations when using these AI models for applications, for certain business scenarios? Currently, AI models perform well for a particular context only. The context is based on the data used to train the model: say your sentiment analysis model was trained using Amazon's or Flipkart’s marketplace item reviews dataset. In this context, this is a good data set but it's not necessary that the same sentiment analysis model would work equally well in another use case, say Twitter, YouTube comments, or Oyo or Make My Trip reviews.
Similarly, these AI models may not solve a single aspect of human cognition robustly, say automatic speech recognition today might work well for a young English speaking adult with an American accent but ASR research is still trying to reach human parity across various issues and in variations in speech, across ambient noise, age groups, accents, language, multilingual scenarios, like a code mixing code switching, etc.
And, just to reemphasize, we’re still solving for artificial narrow intelligence. We're nowhere near artificial general intelligence that will help us manage the entire recruitment or the interview process. Today's AI models are programmed to perform a single task well, and don't do so well when the broader context needs to be taken into account.
Now recruitment is a complex process. For today's discussion, I'd like to take a look at the recruitment through the traditional Application Tracking system life cycle. So from when the job is created to finally an applicant being hired, there are typical applications of AI across this whole life cycle. So things like job description creation itself, ensuring the job is advertised to the audience – this is another aspect where AI could make sure the relevant target audience is being reached. The third part would be actually filtering the applicants itself where gamified models could be used – hackathons, etc.
So, this would help in the sense that you're getting more quality candidates because they are filtered through the set parameters. So how do you use the awesome amounts of data that you already have over the years to match a potential candidate with a target job description? There are multiple factors to take into account things like bias in data sets, but this is something where AI can potentially help in filtering out or mapping that alignment better.
In the interview itself, you have companies like HireVue analyzing micro expressions. They are quantifying human behavior. They're quantifying human expression, human voices, essentially a bunch of different variables from video interviews. Having said that, what these applications do during the whole hiring process is something that an AI developer has to take into account, this includes things like bias that permeates unconsciously as well. For instance, Amazon had a case where they were only recruiting male candidates because of an AI-based training data that was fed into the system. This is why, the onus is on the developer to make sure the practices are bias free.
You also need to take into account privacy of candidates and legal considerations and the regulations around AI that may or may not exist today, but may be implemented tomorrow for more ethical AI applications.
VM: So my next question on similar lines is whether AI can replicate the human bias? What do you think can be done by the AI developers or by the recruiters themselves to ensure that the bias does not come into picture while using this AI?
SA: The news is completely flooded with AI systems that are withholding opportunities, finances, resources in the criminal justice domain, in the recruitment domain, in hiring, in finance, anything and everything and not to forget about cultural repercussions as well.
There are two things that people should focus on: the first being fairness. This needs to be brought into action via our development practices. So this fairness has to be present in the AI developer as well. And I believe fairness is a socio-technical challenge rather than just being a technical challenge. Or rather, in our context, this is a technical challenge viewed in the social context.
Just to give you an example, there is no single definition of fairness that we can quantify today. It is still an unsolved problem but not one that cannot be solved. The only thing that we, as developers, can do is to build a system intelligent enough or robust enough to be able to account for that social diversity that this system will go through.
The second is inclusiveness. For this, I believe if we can build for the 3%, that is, minority and dis-included groups and communities, we can build for the rest 97%.
VM: What is the future of video interviewing tech?
NM: Especially during COVID, we have identified that video interviewing is definitely useful for interviewing at scale. But how do we address some of the concerns that arise after video interviewing?
There have been cases of fraud associated with instances of candidates lip syncing while having a knowledgeable person prompt answers away from the camera's field of view. How do we detect scenarios of fraud, cheating in interviewing? That's one thing we could do.
The other thing is, and again, this is slightly further down the line in terms of timeline, but how do we approach the artificial general intelligence scenario as well? Where we are taking multiple signals from the video itself. So the micro expressions from my face, how do I tie that back to the content that I'm also speaking, my conversation? And this has to be done across different people, across skillsets, across age groups, across experiences, levels of pay or different kinds of jobs. So how do we solve something more related to artificial general intelligence for hiring?
SA: The world has moved away from text - blogs seems to be like something from 20 years ago, people are making videos or vlogs, instead of blogs, the resume is also going to see a fundamental shift. So, we may move to video resumes.
Probably once people attend an assessment on our platform, as we learn better as we expose ourselves to more data, we'll be in a position where the AI can itself recommend better jobs to the same candidate who's passed on one assessment. There will be technologies for pronunciation assessment for let's say high tech jobs, like probably sales or probably, customer architects, and similar stuff.
I came across this the other day: there is this company known as Rephrase, they call themselves MailChimp but for video. So let's say a candidate applies for a job at Microsoft, they receive a video that has a human impersonation, with lip sync that goes -
Hey, thanks for applying to Microsoft and this is how you go about it.
So this is essentially a machine generated video, which lip syncs according to whatever is fed into it, which makes the experience of that candidate extremely personalized. So, the candidate is much more likely to respond to this application mail than just probably dump it in the junk. I feel this is one more area that recruiting could really benefit from.
That’s my take on how recruiting is going to change in the very near future. When I say near future, it could be somewhere between three months to probably less than a year.