Is GPT-3 A Game-Changer For HR?

Like a lot of people on the internet in recent weeks, I've been enjoying experimenting with OpenAI's GPT-3 functionality. There are many different potential uses for such powerful technology across many businesses, sectors, and functions. I was particularly interested in playing around to get an indication of what the potential use cases and implications would be for the People function.

Let's start off with some more information about GPT-3.

What is GPT-3?

According to their website, "OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity". OpenAI has created a variety of artificial intelligence tools, including:

• Codex, which translates natural language (e.g. a description of an application that someone would write in a normal sentence, which is then converted into computer code)

• DALL.E, which turns a description into an image or piece of art

• ChatGTP, which interacts with humans to help answer questions

• Whisper, which is an audio transcription tool, and

• GTP-3 API

GPT-3 stands for Generative Pre-trained Transformer 3. It is a powerful artificial intelligence tool that has been trained on a massive data set, including billions of books, Wikipedia articles, and a large portion of the internet. It is given some initial text as a prompt and will produce humanlike text to answer or follow the prompt. Some examples include creating a blog outline, answering questions, and creating and summarising documents.

OpenAI has opened up its GTP-3 API more broadly and has removed its waitlist. It's possible to sign up on their website and receive a certain number of credits to use in their playground. There is no coding required to use the playground, so anybody can sign up to have a try.

Experimenting with GTP-3:

I tried a few different prompts, curious to see what the quality of the output would be in some HR use cases.

Use Case 1: Job Description Creation

Prompt:

Write a job description for a recruiter

Output:

Job Title: Recruiter

Job Summary:

We are looking for an experienced Recruiter to join our team. The successful candidate will be responsible for identifying and sourcing qualified candidates for a variety of open positions within the organization. The Recruiter will be expected to develop and implement recruiting strategies to attract and retain a talented and diverse workforce.

Responsibilities:

• Develop and implement recruiting strategies to source, identify and hire qualified candidates

• Use a variety of recruiting sources to attract and recruit top talent

• Screen and evaluate applicants to ensure they meet the necessary qualifications

• Conduct interviews with applicants and make hiring decisions

• Maintain recruiting databases and track progress of recruitment efforts

• Create job postings and post them to various job boards

• Develop relationships with colleges, universities, and other organizations to build a strong candidate pipeline

• Stay up to date with industry trends, talent acquisition strategies and best practices

• Participate in job fairs and other recruiting events

Qualifications:

• Proven experience as a Recruiter or similar role

• Familiarity with applicant tracking systems and databases

• Knowledge of labor laws and recruitment best practices

• Excellent communication and interpersonal skills

• Strong problem-solving and organizational skills

• Ability to work independently and as part of a team

• Bachelor’s degree in Human Resources or related field preferred

Reflections:

GPT-3 took under 10 seconds to create this, which is pretty remarkable.

In general, the responsibilities and qualifications are directionally in line with a recruiter's job description, showing that the model has been accurately trained on similar job descriptions. It's definitely on the short side and very generic, so most companies would be unlikely to use this out-the-box, but could use this as a baseline to expand and tailor.

One of the challenges with artificial intelligence is that it reflects the biases of the programmer or data on which it has been trained. GPT-3 has been trained on massive amounts of data and is probably the AI out there with the largest volumes of data used to train. In general, this is a good thing, however, we are also aware that there is also a large data gap for underrepresented groups, and many parts of the internet on which GPT-3 has been trained also have posts and opinions included that are deeply biased. As with all AI, there are tremendous advantages, and caution should be used. For those interested in finding out more about this, I would highly recommend reading Caroline Criado Perez's book "Invisible Women", or watching the Netflix show "Coded Bias".

With this in mind, I ran the outputs through eploy's "Check My Job" website, which is a tool that helps to detect issues or biases in job descriptions. It indicated that this was a masculine-coded job description:

"You’ve used too many masculine words in your job description. Based on academic research by Gaucher, Frisen, and Kay, women may be deterred from applying for jobs if masculine words are used in the job description. For women, masculine job descriptions mean a decreased sense of belonging. Women struggle to see themselves fitting in, irrespective of factors like qualifications and ability."

Here are the words it flagged as masculine-coded:

• independently

• independent

• force (although this was part of the word "workforce", so I'm not sure this one should count)

• decisions

• decision

It also guided that the job description was too short - according to research by Appcast.io, job descriptions of 4000-5000 characters are optimal and result in a click rate of 15%.

Finally, it also raised a concern that the job description didn't include an equal opportunities statement. While not a legal requirement in many geographies, candidates may feel more comfortable applying to jobs that contain information about this. This is a good callout and could be amended by changing the GPT-3 prompt to include this statement.

Use Case 2: Manager Redundancy Script

Prompt:

Write a script for managers to use as part of a redundancy process due to a recession.

Output:

Good morning/afternoon everyone.

I'm here today to discuss the difficult situation we are facing as a result of the current recession. We have made the difficult decision to move forward with a redundancy process.

I know this news is not what any of us wanted to hear, and I want to thank you all for your hard work and dedication to the company. During this difficult time, I want to ensure that all of our employees are treated fairly and with respect.

That said, I want to be clear that this is not a personal decision. We are making this decision to ensure the long-term success of the company.

We are committed to taking all necessary steps to ensure a smooth transition for those affected by the redundancy process. We will provide all necessary support and resources, including access to outplacement services, so that employees can find new employment.

We are also committed to being open and transparent throughout the process. We will be available to answer any questions, and we will provide regular updates to all employees.

Again, I want to thank you all for your hard work and dedication to this company.

If you have any questions, please do not hesitate to reach out to me or any of the other managers.

Thank you.

Reflections:

GPT-3 again took under 10 seconds to create this.

On the positive side - the script comes across as human and compassionate, perhaps more so than some of the examples that have been shared more publicly on the internet in recent months.

Again, it is generic and would need to be further tailored to the situation of the company, and more information on the process that the company would be following (which will require local country knowledge and inputs), and companies should also check what is promised in this so they can meet any promises that are generated through this (e.g. In this scenario, it would be the offer of outplacement services that was included in the script). As a tool, I can see it having its uses to help to support HR. As a People Professional, however, given the consequences of getting a message or the redundancy process wrong, I would not want anything to be used by managers without it being checked and tailored by the HR/People team supporting the redundancy process.

Use Case 3: CV/Resume Creation

Prompt:

Write a resume for an experienced HR Business Partner.

Output:

Thomas J. Smith

HR Business Partner

Objective

An ambitious HR Business Partner seeking to leverage 10+ years of experience in Human Resources, Talent Acquisition, and Change Management to provide strategic direction and support the growth and development of an organization.

Experience

HR Business Partner, ABC Corporation, 2018 - Present

• Develop and implement HR programs, policies, and procedures to ensure compliance with labor laws and company regulations.

• Provide consultative services to senior management on all matters related to human resources.

• Design and execute talent acquisition strategies to recruit, hire and retain top talent.

• Collaborate with other departments to ensure alignment of company goals and objectives.

• Identify and resolve employee relations issues in a timely and effective manner.

HR Manager, XYZ Company, 2012 - 2018

• Developed and implemented HR programs, policies, and procedures.

• Recruited and managed a team of HR professionals.

• Conducted employee orientations and onboarding processes.

• Analyzed data to identify trends and develop action plans to address issues.

• Monitored and evaluated employee performance and provided feedback.

Education

Bachelor of Science in Business Administration, ABC University

Certified Professional in Human Resources (PHR)

Skills

• Leadership

• Change Management

• Communication

• Conflict Resolution

• Employee Engagement

• Performance Management

Reflections:

Another fast creation time for GPT-3. The information included was once again directionally accurate, with plausible skills, education, and experience for an HR Business Partner. It is, of course, far too short and generic and would be unlikely to get someone through the selection process as it stands.

Final Thoughts:

Of all three of the use cases I experimented with, the AI-generated CV was the one that concerned me the most. At the moment the quality is not yet sufficient to compete with a CV created by a human who has worked in the field. With AI still in its infancy, the question is really how long that remains the case. Will we in the not-too-distant future still be able to tell apart a resume written by an unqualified individual using an AI platform versus that of a great candidate? With legislation around AI lagging behind the application of such tools, will there ever be any controls that prevent specific use cases or flag up certain outputs? Can AI itself be used to flag when something looks close to what could have been generated by another AI (similar to a plagiarism checker used by many educational institutions today)? How do the roles of the People and Talent Acquisition teams evolve to both incorporate, challenge, and critically evaluate candidates, processes, and content in the new era of AI?

It is clear from these experiments and the outpouring of interest in GPT-3 in recent weeks that we are entering into a very exciting period with some extremely powerful platforms and applications. AI is here to stay and will become more and more embedded in our working lives and processes. AI has many advantages to help businesses and People functions become more efficient, and well-constructed and trained AI can also help us make better and less biased decisions. Skills will undoubtedly change in the future and we as a function need to be at the forefront of advances such as this so we can help guide our organisations through the transition to the future of work.

For those that are curious, I would highly recommend setting up an account with OpenAI and running your own use cases in GPT-3.