As the benefits of AI take hold and social issues rise to our collective attention, the need for trustworthy AI principles and practices has gained warranted focus.
In the recent Global AI Adoption Index 2021, conducted by Morning Consult and commissioned by IBM, 91% of IT professionals surveyed say their organizations ability to explain how their AI made a decision is critical.
74% of IT professionals surveyed also report their companies are exploring or deploying AI. This number is likely to rise as businesses accelerate digital transformation, with IDC estimating that 90% of digital transformation projects include AI1.
A proactive approach to identify and counter potential risks is needed to scale AI to its full potential, turning principles into practice.
1911: Black and female employees included from IBM’s founding.
1935: T.J. Watson, Sr. stated women will do the “same kind of work for equal pay” as policy.
1984: Non-discrimination on the basis of sexual orientation included in IBM’s equal opportunity policy.
2021: Practicing radical transparency, IBM releases hiring, pay and promotion statistics for diverse populations at the company. At IBM, women earn $1 for every $1 earned by men for similar work. The same is true for underrepresented minorities in the US.
“I fundamentally believe that as we continue to uphold the values of diversity, inclusion, and equity, we will make IBM a better and stronger company.”
Dynamic employee experiences
Competitive skills sets are evolving rapidly in today’s environment. To keep up in the workplace, employees and candidates expect consumer-grade experiences and equitable access to opportunity. These consumer-grade experiences require personalization, making AI not a luxury but a business imperative.
IBM has built AI solutions for HR designed to foster trustworthiness and deliver an end-to-end employee experience across the entire employee lifecycle. It helps to redefine how we attract, develop and retain talent to transform the human experience.
Our solutions are designed to deliver personalized experiences, with skills as the silver thread across the employee journey, empowering IBMers to make fact-based decisions. Key experiences are explainable, they empower users to give feedback, and they help identify and mitigate bias.
Trustworthy AI in hiring
IBM’s tools help managers build a skilled, diverse and inclusive workforce. Reflected within are IBM’s ethical principles and those of trustworthy AI: that AI augments human intelligence without replacing human decision making; that data and insights belong to their creator; and that AI systems must be transparent and explainable.
Sourcing candidates: AI augments recruiters and hiring managers by actively guiding them through the process and ensuring that diverse pools of candidates are being considered for job openings.
Assessment suite: Assessments augment, but do not replace, the hiring manager’s decisions. For example, a game-based assessment measures fluid reasoning, which encompasses the candidate’s ability to solve new problems without prior knowledge. This ability influences the capacity to learn quickly on the job and succeed in a role. Scoring is calibrated and validated and an adverse impact analysis conducted to help ensure equal access to opportunity. The process is designed with human oversight, with the recruiters ultimately accountable for assigning candidates to the slate.
Interviewing: Hiring teams receive intensive guidance on how to avoid bias when selecting candidates, through earning a “License to Hire” certification required to recruit for IBM.
Selection: The system guides mangers to incorporate diverse perspectives in the interview process to help evaluate skills and sends hiring managers further best-practice reminders.
The result: Quality of hires rose 10% year-over-year in 2020, and the hiring of US underrepresented minorities rose 20% over the last three years.
Ultimately, managers know their employees best, so while AI provides recommendations, it does not automate the decision. The manager is the decision-maker. Managers view the explanations for these recommendations and may use them to help inform and shape the conversation with the employee.
Transparency: Employees have access to their pay relative to the market. This transparency fosters evidence-based conversations between the employee and manager. For the underlying AI system, we are creating AI FactSheets which, like food nutrition labels, provide a framework to document machine learning models and AI services and discuss how the model was created, tested, deployed and evaluated; how it should operate; and how it should, and should not, be used.
Explainability: The system provides managers with salary increase recommendations tailored for each of their employees – each has the reasons supporting and explaining the individual recommendation for managers’ consideration.
Fairness: IBM uses AI Fairness 360, part of our open-source toolkit, to help examine the machine learning model for potential bias identification and mitigation.
Robustness: In addition to upholding our data privacy commitments, IBM upholds operational rigor around design and use of AI for pay recommendations. We’ve created and deployed a foundational training module for HR Professionals that includes a Code of Conduct outlining what we do and what we don’t do when building and training models.
Privacy: Employee data is sacrosanct. Rigorous processes and accountability help ensure that the use cases covered by all projects adhere to employee and data privacy requirements.
The result: Attrition was reduced by one-third when managers followed the recommendation.
While it’s important to be thoughtful in how we design and use AI, the more profound impact comes from the experience of IBM employees and candidates. Hearing employee stories of how AI can open doors and help with access to an opportunity is humbling and inspiring.
The opportunity before us is tremendous.
The trustworthy AI journey is a marathon, not a sprint. We need to continue to build the expertise of practitioners and to push the envelope in the creating and fostering of trustworthy AI. For example, IBM is building AI FactSheets for our solutions to explain to end-users how each AI solution is created, tested, deployed, and evaluated. Together with other stakeholders in the ecosystem, we need to build and adopt industry-wide standards for AI in HR, so we can all make progress together.