AI Report on Payroll Processing Issued by PayrollOrg

Defining AI

1. Artificial narrow intelligence is for goal-oriented processes that can only perform designated tasks, such as biometric and voice recognition systems. The processes
   are limited. Narrow AI includes bots that are software applications operating with specific instructions that do not require humans to turn them on manually. Bots are
   already in use for payroll processing.


2. Artificial general intelligence goes a step further with the capability to mimic human intelligence. This includes learning and problem-solving when performing repetitive tasks.


3. Artificial superintelligence is a hypothetical system that would be completely self-aware and could surpass human intelligence. The thinking droids in the “Star Wars”
   movies are one example. This type of AI has not been seen in the workplace.

Many other terms are used interchangeably with AI, such as “deep learning,” “data analytics,” and “robotization.” Generally, these terms fit into one of the above categories; however, some systems may fall between them. For example, “generative AI,” which fits between AI narrow and general intelligence, describes machine learning systems that can generate text, images, codes, and other types of content, usually as a response to a prompt entered by a user (i.e., asking a question or for instructions). In generative AI, the user will receive a human-like response. Many government bodies are piloting generative AI to answer constituents’ questions.

Areas of Concern

There are six areas of concern with developing and using AI. These concerns include the following:

1. Outcome bias in which a decision is based on the results of previous events and data regardless of how those past events developed. If the past information is prejudicial or discriminatory, even if unintentional, bad information will
   carry forward in the AI system.


2. Lack of transparency resulting from payroll and employer misperceptions and misunderstandings about how AI works and minimal understanding by AI developers
   regarding their systems or through unwillingness to share information, such as about the AI algorithms.


3. Inaccurate results from AI providing the wrong answer, omitting information by mistake, or making up completely fake people, events, and articles. If not designed correctly, AI can mix truth with fiction. For example, misidentifying an employee or creating a nonexistent employee.


4. Unauthorized and unknowing release of personally identifiable information (PII). While this is not a new problem for payroll management, information breaches can also occur when using AI. For example, if an employer hires a third party to help process payroll and the wrong data is shared.


5. Interference with operations can occur if cybercriminals attack an employer’s AI system or use AI to disrupt an employer’s systems with demands for money.


6. Fear of job loss when employees feel they will be displaced by an AI system, such as a robot. AI is not intended to replace payroll professionals but to improve their job functions and payroll efficiency.

Potential Payroll Impact

Perhaps the most important step for payroll professionals in understanding AI is to recognize that its potential impact is not just for information technology (IT) departments. Payroll professionals must remain vigilant in ensuring processing accuracy even when AI systems are used.

Questions payroll professionals can ask include the following:

• Which payroll tasks are performed by AI?


• Who has access to payroll data through the use of AI?


• How is payroll data transmitted?


• What laws and regulations apply to AI systems used
  by payroll?


• If an employer outsources some or all of payroll operations, is the service provider using AI?


• If payroll systems are purchased through a third party, is AI used by the system or used to develop the system?

Payroll professionals should audit the data used to train the AI systems and audit the determined results. Is the payroll data accurate and void of bias? Biases can occur inadvertently because of attitudes at the time the data was generated. Biases can also occur because of different cultural and environmental values without realizing the influence on an AI system.

Government Interest in AI

Government action has centered around creating opportunities for entrepreneurs to develop AI systems, to prevent discrimination, and to limit misuse. These measures have ranged from study groups who advise government on AI development and use, consumer bills of rights on use of individuals’ personal information, and limitations and bans on AI. The Biden Administration has issued many orders on how government agencies should address AI, and Congress has created several advisory groups. In Washington, D.C., the mayor is exploring a chatbot to answer constituent questions.

AI is here and will continue to influence payroll’s future. It’s important to understand its proper use, how it can help, and how AI can be a part of your payroll processes (read more articles in the AI Special Section of June 2024 PAYTECH.