AI Hiring Needs Explainable Models, Human Review and Candidate-Controlled Data

AI can shorten hiring workflows and help employers learn from recruitment data, but its value depends on how organizations design and supervise it. Duri Chitayat, chief technology officer at workforce management company Safeguard Global, argues that automation should support recruiting teams while people remain accountable for decisions. His view puts practical questions ahead of predictions: whether a model recognizes varied career paths, whether its criteria can be examined, and whether HR staff understand the tools they use.

Chitayat spoke with Techopedia about AI in applications, candidate assessment, workforce planning and career data. He sees opportunities for both employers and applicants, alongside risks that require ongoing review. A system that processes information quickly can still overlook qualified people if its inputs or objectives fail to account for the range of skills and experiences found in real careers.

Speed is useful only when the system sees the candidate clearly

Employers have used AI for tasks such as sorting resumes and conducting early candidate assessments for years. Used with care, these tools can reduce time spent on repetitive steps and make the application process more efficient. Chitayat also expects AI to help organizations analyze earlier hiring outcomes, identify productive sourcing channels and find qualities associated with long-term success in a particular workplace.

Those gains rely on a model that can interpret more than a narrow list of credentials. People change roles, build skills in different settings and take career paths that may not resemble the examples in a training dataset. An algorithm that misses those patterns can filter out suitable applicants before a recruiter sees their applications. Human review remains necessary to spot those gaps and bring context to model outputs.

Applicants could benefit from tools that match them with relevant openings, provide useful information about employers and make it easier to compare compensation. Chitayat says candidates are also beginning to reuse parts of applications that take substantial effort, including work samples. Reusing verified materials could reduce repeated work, provided applicants can decide how their information is shared.

For employers, the aim is not simply to automate more steps. Chitayat’s comments point to a two-part approach: use systems where automation fits the task, then have recruiting staff interpret the data and take responsibility for decisions. That division matters because faster screening does not, by itself, show that a process is accurate or fair.

Fairness depends on design, data and the people operating the tools

AI can process large volumes of information and apply consistent criteria. In recruitment, a model could focus on skills and qualifications while excluding demographic details that have influenced human decisions. But the system’s objectives and data shape what it learns. If an organization’s past decisions reflect unequal access or biased practices, a model trained on those records may reproduce the same patterns.

Chitayat places responsibility on the people who choose and operate these systems. Teams need to decide what outcomes they want, examine how models behave, and monitor results over time. A tool does not produce inclusive hiring simply because its instructions tell it to prioritize qualifications. The organization must check whether its criteria are appropriate and whether the results align with its stated aims.

He identifies representative training data and explainability as key safeguards. A training set should reflect the population the system will assess. Explainable models can help specialists see which features shape a recommendation and adjust the model when those features are unsuitable. Without that visibility, teams may find it difficult to understand why applicants receive particular assessments or to identify a faulty selection criterion.

There is a skills problem as well. HR teams may buy or deploy software without enough technical knowledge to judge how it works. Chitayat calls that a significant risk as AI becomes more common in recruiting. He recommends building technical skills across business functions, encouraging employees to learn, and connecting technology leaders with HR and other teams. Curiosity and informed skepticism can help organizations ask better questions before and after adoption.

Keep testing as recruitment work changes

At Safeguard Global, Chitayat describes AI automation as one track within a broader recruiting process. The company is applying it at selected points, from an initial application through qualification checks, while recruiters use the resulting data in their analysis. That model leaves people involved as the tools take on tasks they can handle.

Chitayat urges teams not to accept each new system or update without checking its results. Backtesting—examining how a tool would have performed on earlier cases—can help reveal strengths and weaknesses. He also points to comparing outputs from different AI systems as a useful way to probe results. The point is to keep evaluating performance as tools and workflows change, rather than treating a vendor’s claims as sufficient evidence.

This approach also affects the shape of HR work. As systems take on more sourcing and recruiting tasks, teams may spend less time on repetitive operations and more on planning, risk review and workforce needs. Chitayat compares this shift with changes payroll teams experienced as their work evolved. Training can help HR staff adapt, but organizations also need to recognize that automation changes roles and responsibilities, not just the software used to perform them.

Assess effects beyond the immediate hiring decision

Chitayat sees overreliance on AI as an ethical concern because it could entrench existing inequities. Human oversight, model checks and backtesting can reduce that risk, but he also argues that organizations should consider who is affected by their decisions. A company’s view of its responsibilities depends in part on whom it treats as a customer.

That view can extend beyond investors and clients to current and prospective employees, business partners and the wider group affected by company choices. Chitayat argues that ethical review should consider effects on the broader social fabric, as well as on individual applicants. For recruitment teams, this means asking what an automated process does to people who encounter it, including those who are screened out or required to provide data.

Career records could give applicants more control

Data security is a basic requirement when HR teams introduce AI, particularly when organizations manage remote workforces. Chitayat says companies need cybersecurity plans and resources before, during and after implementation. He also sees a possible role for blockchain in letting individuals control how career records are protected and shared.

The Velocity Network Foundation launched an Internet of Careers platform several years ago to change how career information is shared using blockchain. Chitayat describes the broader idea as a self-sovereign model: applicants would control access to verified credentials and reuse them across applications. Examples could include code samples and test scores. If adopted widely, this approach could reduce repeated submissions and give candidates more control over their information.

That possibility depends on adoption and on secure systems for handling sensitive records. Blockchain alone does not establish that a credential is accurate or that an employer will accept it. Its proposed value here is control over sharing and reuse, while employers and applicants would still need reliable ways to verify the information.

What Chitayat expects next

For near-term changes, Chitayat expects AI to help applicants tailor resumes to specific roles and to help employers draft job descriptions through natural language processing. These uses target text-heavy tasks on both sides of recruitment. Their practical value will depend on whether generated material reflects actual skills and role requirements, rather than simply sounding relevant.

Over a longer horizon—perhaps within five years, in his estimate—he expects recruiters to use conversational interfaces that bring multiple actions into one place. Recruitment teams often work across many separate systems, and moving between them can delay processes and contracts. A unified chat-based interface could reduce that friction, though the interview does not specify how such a system would connect to existing tools or manage its data.

He also anticipates greater use of predictive analytics on career data, including estimates of how long a potential employee might stay. Such predictions could inform decisions, but they should be treated as uncertain signals rather than facts about an individual. The same concerns raised for screening apply: data quality, explainability and human judgment shape whether an output is useful.

Chitayat’s preparation advice is to make data a core organizational resource and build teams that can assess AI critically. He leads Safeguard Global’s technology team and has worked in engineering across advertising technology, medical technology, banking and finance. His central argument is practical: AI may accelerate hiring, but fair and useful adoption requires people who can inspect the systems, test their results and remain accountable for how those systems affect candidates.