Sense with Cents

In my mind, AI is not artificial. It is the sum of the knowledge the tool has collected, with means to "boil down" that knowledge to answer what you ask the tool to give. The current tools are the next step past web search (which is what the AI tools are somewhat doing, searching the web and feeding you back the best most pertinent information). Think of it as a single tool replacing Siri, Google search, encyclopedias, and more — all in one place.

If you are not knowingly using AI, you still are using AI. Embrace the change, and take advantage. How? By experimentation you will learn how to best ask your AI tool in a way to draw out the best reply. Use voice commands instead of typing, be lengthy but direct. Talk to the tool like you are asking a human for help, not "what restaurants are near me", but "I really want a good fresh fish and chips meal, what is open within a 15 minute drive, and is under $20 for the meal".

But, be wary, AI results can be wrong, especially the more narrow the subject is. For instance, if you ask what is the oldest operating shareware company, you may or may not get Medlin as the answer (which is the correct answer). The sources of the answer are relatively limited, as many of the old shareware sites are long gone, so you have to add things like shareware companies prior to 1985, same ownership, still offering same or similar products, and so on. To further complicate the search, Medlin is considered an early second wave shareware company, because we self-describe our start as late 1984, when the first payment was received, not 1983 when Medlin first started sharing software with others without asking for payment, or in 1977 when Medlin's first application PC-GL (PC General Ledger) was created.

For payroll purposes, AI is a blessing and a curse. Ask AI something about a finer point of payroll processing, and you have maybe a 50/50 chance of getting an accurate reply. Why? Because the information is so detailed, it depends on how the question is asked, and how the AI tool interprets the source. Verify first by checking several sources, then the actual rule or law text too. Or better yet, ask someone you trust to give you an honest answer, and who will "show their work". Same with AI, ask the tool to prove their answer using different sources. Ask a second AI tool the same question and compare results.

A perfect example are some of the current complications in payroll, such as explaining how OBBBA might alter OT and tip reporting in a W-2 (see FLSA OT (QOC) and Agricultural Employees for one such complication). The rules are not even a year old, are very complicated in execution, and there is still plenty of misinformation online claiming to be accurate.

Ask AI something like "with these W-4 settings, for a weekly gross of X, what is the federal withholding," and you will likely get a perfect reply. There is one perfect source to refer to, it is a public document, and AI should find and use it.

Again, verify, as my test on AI telling me federal withholding has a flaw. It failed to use common sense, and IRS preferred, round to nearest dollar. Yes, calculating to the cent is allowed, but it is not common sense, as the IRS Form 1040 does not allow claiming anything other than whole dollars. AI is great at following specific direction. You have to be the one to add the human nudging — the nuances, if you will — such as reminding it to round federal withholding to the nearest dollar.

Remember, AI is filled with existing knowledge. It can organize it, retrieve it, and connect it in ways that save you enormous time. But humans are still obligated to create new knowledge. Every new law, every edge case, every real-world problem that hasn't been written about yet — that starts with someone doing the work and figuring it out. AI can help you get there faster, but it cannot get there first.

Use AI. Don't fear it. But bring your own knowledge to the table and always verify. The tool is only as good as the person asking the questions.