Millennials Are Open To AI In HR—But Say There’s Still A Place For Humans

AI in HR

HR technology has integrated into the human resources department over the last decade—and as technology advances, the inclusion of artificial intelligence (AI) technology has grown with it. Many HR professionals are embracing this type of technology. However, to some, the notion of machines taking over a role that is based upon building very human relationships is a big concern.

But what about the employee perspective? With 40% of millennials making up the Australian workforce, we wondered if this generation of employees will embrace the involvement of AI-powered technology in HR.

AI in HR

The results indicate that millennials are open to AI technology handling HR tasks but they’d still prefer some activities to be owned by humans.

Other key highlights from the survey include:

  • 63% of millennials believe AI could make HR a fairer department.
  • The older the millennial, the less comfortable they feel with AI in HR.
  • Men are the most open to AI-driven HR technology, women are on the fence.

How is AI used in HR?

Artificial intelligence is already transforming the jobs of HR professionals in several ways. Here are just a few examples:

Administration

Artificial intelligence is well-suited to handling repetitive, low-value tasks. By automating these, HR professionals can focus on more strategic work as well as ones that require uniquely human abilities (such as the analysis of a specific context.)

To give some context, AI can relieve HR staff from administrative tasks like:

  • Payroll processing
  • Workforce analysis
  • Policymaking and implementation
  • Employee paperwork processing.

One example of deep learning AI within HR administration is chatbots. Through natural language processes (NLP), the AI learns to understand language that humans use rather than the language that a computer uses. As hinted by the name, NLP sets out to understand human language in its more natural form, in both written and spoken context. It takes into account tone, context and intent to make a judgement on what information the user is seeking.

By powering the chatbot with an algorithm that provides prompt and accurate responses to employee queries, it relieves HR staff from constantly relaying the same information to different people.

Recruitment

It’s difficult to know the true financial cost of hiring the wrong candidate, but a SmartCompany survey found 18% of small business owners felt their hires were worse or much worse than they had expected.

A great hire, on the other hand, can boost productivity, positively impact a company’s culture, and drive business revenue. For this reason, HR departments are using AI in their recruitment efforts to make smarter hiring choices.

The technology can automate candidate CV screening through to providing feedback to unsuccessful candidates. For candidates entering the interviewing stage, companies are using AI to ask interview questions. This includes the more basic questions, such as ‘How many years experience do you have?’ to more complex questions—like asking them to describe a professional difficulty they’ve had to overcome?’

Additionally, companies are using AI to make an assessment of a candidates personality type to help them decide whether they would fit in well with their company’s culture and values.

Onboarding new hires

AI-integrated systems can introduce new employees to relevant company information on their first day on the job. In a timely manner, an AI application will automatically send a new staff member information about their:

  • Duties
  • Benefits
  • Company policies and practices
  • Reporting authority figures
  • Team members
  • Training schedules
  • Meet-and-greets
  • Task assignments
  • Documentation to sign.

More than this, AI allows for customisation. The application will adapt the information included in the onboarding process based on the specific job role. For example, requesting specific devices that the employee will need or providing readily-prepared answers to frequently asked questions for that position.

Correctly onboarding new employees, especially in a virtual setting, is important because it sets them up for maximum success. It gives them a clear idea of what is expected of them and what they can achieve. For HR professionals, automating this process frees up a great deal of administration time.

Learning & development

HR departments are making use of personalised individual learning and development programs for their staff, which are powered by AI algorithms. Rather than providing a standardised program, the AI analyses which content and training need to be relayed to the employee to maximise their educational experience.

Often, this is based on the new employee’s job role, existing knowledge levels, skill sets and development plan. At the same time, the AI draws upon data and analytics from the business to factor in which skills the employee needs to develop to:

  • Positively impact the company’s bottom line
  • Drive business resiliency
  • Future-proof the business (by ensuring it has the right talent and expertise to meet long-term goals).

Tracking flight-risk employees

Some employers use AI to alert them to employees demonstrating behaviours that would suggest they’re unhappy in their job—putting them at risk of leaving the company. This includes tracking the way they express themselves on email and other company communication platforms to detect whether they change their overall tone. It also checks for keystrokes and internet browsing.

Of course, it’s for the employee to decide whether they want to seek out an opportunity elsewhere. However, by being aware that they’re considering other options, the HR department can proactively engage with them, and hopefully, encourage them to stay.

Incident management

AI-powered technology is also evident in the management of ethics and compliance situations. One business offering this technology is TalkToSpot. It allows businesses to identify cases of harassment and discrimination, policy violations, and whistleblowing.

According to the company: ‘Research shows that talking to a human is one of the biggest obstacles to reporting inappropriate behaviour at work.’ By allowing employees to communicate with its chatbot, a complaint is made anonymously—and importantly, the AI automatically triggers the appropriate corrective and disciplinary actions.

This sort of technology helps champion a healthy and inclusive culture; as well as demonstrating a no-tolerance attitude toward any behaviours that work against it.

The employee perspective: Are they ready for the future?

We surveyed 515 Australian millennials to ask how they’d feel about employers using AI within their HR department—a full methodology of the survey is available at the bottom of this page.

We set out to discover where the boundaries lie, from onboarding through to disciplinary actioning.

63% of millennials believe machines could make HR fairer

All humans are prone to bias. Unconscious biases are outside of our control and happen without us realising or intending it to happen. Our brain will trigger a biased response to help us make a quick judgement of a situation or person, and these thoughts are often based on our background, environment and personal experiences.

This was given as one of the key drivers behind the belief that AI could make HR fairer, according to 63% of Capterra’s survey respondents. One respondent said:

‘Artificial intelligence is impersonal. It does not judge on race, gender, or culture. It purely bases its judgments looking at skill whereas people can be unintentionally influenced.’

 

Respondents also made points about the potential for conscious bias in humans too. One millennial suggested that ‘human emotions sometimes get in the way of processes.’ Another said they believe AI ‘eliminates the possibility of favouritism and the involvement of personal preference’

On the other hand, 57% of respondents said they believe there is still an opportunity for bias to occur within AI. Of this number, 39% said the potential for bias depends on the situation.

Bias in AI in HR

Unfortunately, machines are only as good as the programmers that train them. It’s possible for AIs to learn biases due to engineers unwittingly introducing their hidden biases into the training data they use. In this respect, there are real concerns that AI could make issues of human and societal bias in HR worse. Mckinsey summarises this well:

‘AI can help reduce bias, but it can also bake in and scale bias.’

 

For this reason, experts advise companies to apply innovative training techniques and involve human judgement to ensure AI-supported decision making is fair.

Millennials value the people in HR but feel comfortable with some AI involvement

Recruitment

78% of millennials believe AI could help tackle bias in recruitment. 65% of respondents said they think AI could make HR fairer in the hiring process in general. However, the majority want a hybrid of AI and human recruiters to be used in the industry.

Within the hiring process;

  • 61% of millennials would be happy with an AI screening their CV, as long as a human review it too.
  • 40% of millennials aged 30 years old or more said they wouldn’t want an AI looking at their social history, compared to 32% of people under 30.
  • 27% of millennials said they’d be okay with an AI asking them basic interview questions.
  • 49% would be comfortable with an AI asking them more complex questions, as long as a human also makes a judgement.
  • 37% were interested in their skills being tested via an AI-powered game.
Read more on Capterra’s research into the millennials perspective of AI in recruitment.

The verdict: Millennials prefer a combination of humans and AI in the recruitment process

The use of AI within the recruitment industry relies partly on Australia’s readiness for it. Millennials are already showing indications of interest—and in some areas, such as tackling unconscious bias, they’ve demonstrated enthusiasm. But on that topic, employers must also consider ways that an AI-powered program could work against diversity efforts.

Technology that uses a video link to carry out personality type testing, for example, often relies on measurements such as eye movement, choice of words and tone of voice. However, it doesn’t account for blind candidates, deafness or disabilities such as autism. For this reason (and many more), it’s important that businesses don’t solely rely on AI in recruitment.

Onboarding

62% of millennials were happy with the use of AI in their new employee onboarding for a new job. However, of this number, almost half (47%) would prefer for a human HR professional to be involved as well.

AI-powered employee onboarding the millennial opinion

In terms of gender, women would prefer that this HR task was left to human professionals. A third (33%) would not like an AI involved at all (compared to 24% of men.)

Automating employee onboarding men vs women

The verdict: Millennials value people in onboarding, especially women

Automating the onboarding process has many benefits for HR professionals—a big benefit being that it saves a huge amount of time spent on highly administrative and repetitive tasks. An AI can also combine company data with the new employee profile to produce a personalised process much quicker than a human could.

At the same time, employees value human connection when joining a new company. For this reason, it’s important that new hires have a human point of contact during the process. This doesn’t have to be the HR professional, however. A buddy-system, where employers partner a new employee with an existing employee is a good alternative. However, it may be that an AI assigns a suitable buddy for every new hire.

Employee record management

A third of millennials said they’d feel uncomfortable with an AI handling their employee records. However, 65% were comfortable with it (with a quarter of the 65% saying they are happy for humans to step away from this activity completely.)

Importantly, there is a difference of opinion between ages. 45% of millennials aged 30 years old and older said they’re happy with AI handling their employee records. This is compared to 50% of millennials under 30 years old.  A third of over-30s would prefer humans to manage their records without any kind of AI assistance.

Age breakdown millennial opinion on employee record management AI-powered

A divide between genders is also apparent in the survey results. Almost a quarter (24%) of women are indifferent, compared to 18% of men. Half (50%) of men would be okay with artificial intelligence being the sole manager of their records.

Men vs women millennial opinions on AI driven employee record management

The verdict: The majority of millennials don’t mind an AI handling their employee records

Most millennials are open to AI operating this HR task, and a fifth are indifferent. This points to a clear automation opportunity for businesses. However, they should also bear in mind that around a third (32%) of respondents expressed feelings of uncomfortableness around this concept. With the task involving personal information, it’s worth employers asking staff to authorise and agree to the use of artificial intelligence here.

Career development

55% of the millennial respondents said they’d feel comfortable with an AI providing them with educational assistance to help them in their career. However, a quarter (25%) would prefer humans and 19% are indifferent.

Men were most certain of where they sat on the argument for AI-powered education or human educators. Just 16% said they were indifferent compared to 20% of women.

The verdict: Knowledge gap insights work well with human-led training sessions

Machines can automate a lot of the learning and development process for employees, such as choosing the most relevant programs and identifying what skills an employee needs developing. However, it’s trickier for a machine to teach softer skills, such as relationship building. With that in mind, the survey results present an opportunity for businesses to combine AI with human-to-human training sessions.

Performance analytics

Again, male respondents were more comfortable about the concept of an AI making decisions around promotion and salary increases. A fifth (20%) would be happy for an AI to make the judgement call compared to just 13% of women.

Both genders agreed that an AI application driving performance analysis decisions would only be workable in some situations (43% for men and 41% for women). The older the millennial, the more likely they were to agree with this statement.

Age and gender breakdown of payrise promotion decisions ai

The verdict: Humans should consider the context behind AI-driven performance analysis for employees

The discrepancies between genders and ages suggest a potential confidence gap exists amongst Australian millennials. If utilising artificial intelligence to make these kinds of decisions, businesses should ensure they’re transparent about the framework they’re using to make judgements.

Compensation and benefits

More than half (51%) of respondents said an AI could be effective at conducting analysis into the compensation and benefits that other companies are offering their employees. The aim of this exercise is to then create attractive and competitive employee compensation and benefits packages of their own. This opinion was unanimous across genders and age groups.

21% of employees were indifferent to who handles this task, while 29% would prefer humans to take ownership.

The verdict: AI can help pull relevant information, but humans understand company culture most

Using artificial intelligence to present important data around benefits to human HR professionals could be hugely beneficial to employers in the future. Combining this information with human knowledge about their company culture, businesses can pull together hard-to-rival packages for their workforce.

Disciplinaries

We asked respondents how they’d feel if an AI was involved in their employer’s disciplinary process (such as flagging if a person was frequently late.) Surprisingly, 65% said they were comfortable with it, with a quarter of these respondents saying they’d prefer humans not to be involved at all.

Importantly, however, 28% of millennials would prefer a human HR professional to handle disciplinary tasks without the assistance of AI.

The verdict: Machines can help identify problems, but humans should address them

The results indicate there is still a place for human agents when it comes to disciplinary practices. While machines can help to flag issues that need attention, a human-to-human meeting is likely to be more effective when it comes to taking action.

The final verdict: AI won’t replace humans in HR

With the HR tech solutions market coming in at around $148 billion according to global research by PWC, its role in the department will continue to increase. But what does this mean for HR professionals?

The reality of HR technology adoption increasing is that the role of HR professionals will change. It’s likely that uniquely human skills will grow in importance as a consequence. According to World Economic Forum predictions,  75 million current jobs will be displaced as the role of AI increases in human resources. However, in the same vein, the forum forecast that 133 million new jobs will be created by 2022.

Skills in both emotional intelligence and technical intelligence, like technology design and programming, will be important while analytical skills and the ability to operationalise change will be less important.

Looking for HR software? Check out our catalogue.

*Survey methodology

To understand the opinions of millennials in Australia with regards to artificial intelligence in HR, we conducted an online survey between 3rd March – 8th March. 

We surveyed people living in Australia from the generational group (aged between 24 to 39 in 2020) who work full-time, part-time, or are actively job-seeking. To mitigate the potential for bias, we screened out survey participants that worked within HR and recruitment. This left us with our final number of respondents for the survey: 516.

5 SaaS-toolkits om explainable AI te maken

explainable AI

Met explainable AI kunnen techteams bias voorkomen en eerlijke, nauwkeurige algoritmen maken. Deze SaaS-toolkits helpen je daarbij.

explainable AI

Steeds meer zakelijke beslissingen worden genomen op basis van zelflerende algoritmes. Hoe een AI-systeem tot zijn besluit komt is vaak onduidelijk, waardoor de vraag naar transparantere algoritmes toeneemt.  Explainable AI (XAI) kan in die behoefte voorzien. Ontwikkelaars van machine learning (ML) kunnen hiermee tijdens het hele trainingsproces van algoritmen zien wat de mogelijkheden en beperkingen van die algoritmen zijn. Het risico op bias wordt hierdoor kleiner.

Bias in algoritmen kan grote gevolgen hebben voor eindgebruikers. De consequenties van niet weten hoe AI werkt, zijn uitgebreid gedocumenteerd. Denk bijvoorbeeld aan creditcards met een lagere bestedingslimiet voor vrouwen of het koppelen van recidive aan ras. XAI is dan ook niet langer optioneel.

Als jij AI wilt maken en toepassen in jouw bedrijf, verwacht iedereen, van senior stakeholders tot klanten, dat je kunt uitleggen hoe het werkt. Het goede nieuws? Doordat de vraag naar XAI stijgt, komen er ook steeds meer tools en technieken die je technische team kan gebruiken om aan de slag te gaan met explainable AI. 

Bij het evalueren van de verschillende tools en toolkits:

  • Documenteer eerst de vereisten voor je algoritmes. Bepaal van te voren welke methoden voor eerlijkheid je gaat gebruiken en hoe je deze wilt prioriteren.
  • Probeer niet de perfecte toolkit te vinden voor explainable AI, maar ga voor incrementele implementatie met meerdere tools.
  • Bekijk hoe elke toolkit aansluit bij het zakelijke probleem dat jouw algoritme moet oplossen. De verschillende toolkits ondersteunen verschillende vereisten voor verklaarbaarheid. Als je bijvoorbeeld een algoritme wilt maken om klantbeslissingen te verklaren, moet je geen toolkit gebruiken die bedoeld is voor vereisten op het gebied van wet- en regelgeving.

5 SaaS tools die jij kunt gebruiken om explainable AI te maken

Deze vijf Saas-tools (Software-as-a-Service) kunnen je technische team helpen bij explainable AI. (Ze werden in een recent Gartner-artikel genoemd als voorbeeldtechnieken en -methodologieën die kunnen helpen bij het bouwen van XAI. Het volledige onderzoek is beschikbaar voor klanten van Gartner.)

Let bij de toolkits hieronder met name op de functionaliteit die gericht is op verklaarbaarheid en interpreteerbaarheid. Met het eerste kun je je model controleren tijdens de algoritmetraining en met het tweede kun je de resultaten van het model uitleggen aan stakeholders en klanten. (De producten worden in alfabetische volgorde weergegeven.) 

1. DataRobot

Datarobot AI-voorspelling
Verklaring van voorspellingen door DataRobot (bron)

Met de software van DataRobot kunnen teams intern hun eigen AI-modellen maken en implementeren, zonder hierbij expliciet te programmeren. Als je nog geen data scientist hebt (of je bedrijf zich die niet kan veroorloven), kan DataRobot hiervoor als vervanging fungeren. De tool automatiseert standaard data science taken en helpt klanten specifieke zakelijke problemen op te lossen. United Airlines gebruikte DataRobot bijvoorbeeld om te voorspellen welke klanten de meeste kans hebben op een tassencontrole bij de gate.

Omdat DataRobot machine learning automatiseert, ondersteunt het ook interpreteerbare modellen. De tool beschikt over een modelblauwdruk die laat zien welke verwerkingsstappen elk model vooraf gebruikt om tot de conclusies te komen. Hierdoor is DataRobot met name een goede optie voor teams die modellen maken die moeten voldoen aan regelgeving.

Daarnaast biedt DataRobot verklaringen van voorspellingen, waarin de belangrijkste variabelen worden weergegeven die van invloed zijn op het resultaat van het model voor elke record. Dit is cruciaal, omdat algoritmen tijdens het hele trainingsproces verschillende gewichten toewijzen aan verschillende datapunten, wat dan weer van invloed is op de aanbevelingen. De verklaring van voorspellingen voorkomt mogelijke bias doordat wordt uitgelegd hoe de conclusies van elk algoritme tot stand zijn gekomen.

Prijs: DataRobot biedt aangepaste contracten voor drie jaar op basis van je zakelijke doelen. Bij de keuze voor jouw specifieke softwareconfiguratie word je persoonlijk geholpen door iemand van het data science team van DataRobot.

Meer informatie

2. Google Cloud Platform

transparant AI on Google Cloud Platform
Visualisatie van functiewaarden en interferentiescores in What-If tool (bron)

De Google Cloud-platformservices worden door een miljard mensen gebruikt en zijn dus qua grootte en reikwijdte moeilijk te evenaren. Het platform beschikt ook over een robuuste suite van  tools voor AI en machine learning. In november 2019 is er een explainable AI service bijgekomen, die algoritmische modellen tijdens de hele levenscyclus van het product evalueert.

Functies zoals AutoML Tables en AI Platform bieden transparantie voor de gebruikers, die zo kunnen zien of ze de datasets en/of architectuur van hun modellen moeten verbeteren. Met AutoML Tables of AI Platform krijg je realtime scores voor je modellen, die aangeven hoe bepaalde factoren van invloed zijn op de uiteindelijke resultaten. Als je dit combineert met de functie voor continue feedback van Google Cloud, kun je modelvoorspellingen vergelijken en de prestaties optimaliseren.

De What-If tool van Google Cloud geeft interactieve dashboards weer waarmee gebruikers de voorspellingsmodellen van het AI-platform kunnen bekijken. Deze tool werkt samen met Jupyter en Colab notebooks, en is vooraf geïnstalleerd op AI Platform Notebooks Tensorflow-exemplaren. Als de outputs van je modellen niet overeenkomen met de vereisten van de What-If tool, kun je aanpassingsfuncties in je code definiëren.

Prijs: De explainable AI tools van Google Cloud zijn gratis voor mensen die AutoML Tables of AI Platform al gebruiken.

Meer informatie

3. H2O Driverless AI

Machine learning Driverless
Interpreteerbaarheidsdashboard voor machine learning van H2O Driverless AI (bron)

H20 Driverless AI automatiseert verschillende aspecten van de ML-workflow, zoals validering, tuning, selectie en implementatie van modellen. Het werkt op standaardhardware en is ontworpen voor gebruik met grafische processing units (GPU’s). Dit is een belangrijk verkooppunt, omdat GPU’s een cruciale rol spelen bij deep learning.

Verder biedt H20 Driverless AI machine learning interpretability (MLI) als kernfunctionaliteit. De applicatie biedt het volgende:

  • Shapely (dat laat zien hoe functies rechtstreeks van invloed zijn op de unieke voorspelling van elke rij)
  • k-LIME (dat redencodes en uitleg in het Engels kan genereren voor complexere modellen)
  • Surrogaat beslisbomen (een stroomdiagram dat laat zien hoe een model beslissingen heeft genomen op basis van de oorspronkelijke functies)
  • Partial Dependence plots (met gemiddelde modelvoorspellingen en standaardafwijkingen voor de eerste waarden van functies)

H20 Driverless AI biedt ook de functie ‘disparate impact analysis’. Als een model negatieve effecten oplevert voor specifieke groepen gebruikers, kun je dat model met deze analysefunctie testen op een mogelijke bias. Dit een essentiële functie, omdat bias op verschillende punten tijdens de training van algoritmen in modellen kan binnensluipen.

Prijs: H20 is een open source ML-platform en het gebruik ervan is dus gratis. Driverless AI is een standalone product voor ondernemingen en hiervoor moet wel worden betaald. Het gebruik van dit systeem op Google Cloud kost ongeveer $ 2.281 (circa € 2025) per maand.

Meer informatie

4. IBM Watson OpenScale

AI explainablility
AI explainability 360 open source toolkit van IBM (bron)

Watson is een vraag-en-antwoordsysteem dat door IBM is ontwikkeld. Het schokte de wereld toen het jaren geleden twee ervaren menselijke deelnemers aan het spelprogramma Jeopardy versloeg en daarmee een prijs van $ 1 miljoen won. Tegenwoordig gebruiken bedrijven IBM Watson OpenScale om modellen te bouwen voor het voorspellen van kredietrisico, defecten in activa, claimsverwerking enzovoort.

Watson OpenScale biedt verschillende beheerfuncties voor modellen. Gebruikers worden bijvoorbeeld gewaarschuwd wanneer ‘drift’ wordt aangetroffen in een AI-model. Dat gebeurt wanneer modellen in de productiefase data tegenkomen die verschillen van de data waarop ze zijn getraind. Dergelijke waarschuwingen zijn van groot belang, omdat je bij ‘model drift’ het risico loopt dat er bias wordt geïntroduceerd in een model.

Watson OpenScale biedt ook contrastieve verklaringen voor de classificatiemodellen die je bouwt. Dat betekent dat er pertinente positieven en pertinente negatieven worden weergegeven, die allebei nuttig zijn om het gedrag van elk model te verklaren. Twee typen de-biasing (passief en actief) zijn ook beschikbaar.

Prijs: Watson OpenScale biedt twee soorten abonnementen, op basis van het aantal modellen dat je wilt implementeren en monitoren. De Lite-service is gratis, maar wordt na dertig dagen inactiviteit verwijderd.

Meer informatie

5. Microsoft Azure

explainable AI Microsoft Azure
Explainability framework SDK dashboard van Microsoft Azure (bron)

Microsoft Azure is een cloud computing service waarmee gebruikers applicaties kunnen bouwen, testen, implementeren en beheren. Het ondersteunt een breed scala aan programmeertalen, tools en frameworks binnen en buiten het Microsoft-ecosysteem. Azure bestaat al tien jaar en ondersteunt inmiddels meer dan 600 services.

Gebruikers van de Basic en Enterprise editie van Azure hebben toegang tot de modelinterpreteerbaarheid van het platform. Volgens de documentatie van Azure heeft dit de volgende drie belangrijke voordelen voor gebruikers:

  • Waarden voor het belang van de functie voor ruwe en speciale functies
  • Interpreteerbaarheid op schaal, op echte datasets in de praktijk, tijdens de trainings- en inferentiefasen
  • Interactieve visualisaties om patronen te ontdekken binnen data

Met Azure kun je deze functies globaal op alle data toepassen of alleen op specifieke lokale datapunten. Ook kun je interpreteerbaarheidsmethoden toepassen op globaal gedrag of specifieke voorspellingen.

De modelinterpreteerbaarheid van Azure biedt negen verklaringstechnieken waaruit je kunt kiezen. Zo kun je de verklaringstechniek afstemmen op de techniek die jouw team heeft gebruikt om je model te trainen. Als je bijvoorbeeld de deep learning techniek hebt gebruikt, kun je de SHAP Deep Explainer gebruiken om te kijken wat er precies gebeurt.

Prijs: Azure heeft abonnementen op basis van gebruik, afhankelijk van de regio, het type gebruik, soort facturering enzovoort. Op de website van Azure vind je prijscalculators, prijzen per product en meer. Er is een apart tabblad voor de AI- en machine learning producten van Azure.

Meer informatie

Wil je meer weten? Bekijk dan onze online catalogus met een uitgebreid aanbod aan AI-software.