Problem-Solving with AI and Digital Tools

Modern tools can generate options, crunch data, draft text, and simulate outcomes faster than any person. Used well, they widen the range of solutions you consider and cut the drudgery out of analysis. That is a real gain, and refusing to use them is its own kind of error.

The difficulty is that these tools are confidently wrong in ways that are hard to see. A fluent, well-formatted answer feels trustworthy whether or not it is true. This lesson is about using the leverage without inheriting the failures — which comes down to knowing which step the tool is doing, and never letting it be the last word.

Where the tools help

AI and digital tools are strongest as an amplifier for a step you could do yourself but slowly. Generating a long list of options to prune. Summarising a document you will still skim. Crunching a dataset for patterns you will then test. Drafting a first version you will rewrite. In each case the tool expands throughput while a human keeps the judgement.

Where they fail

Three failures recur. Hallucination — the system produces plausible, specific, entirely invented content: a citation, a statistic, a quote that does not exist. Hidden assumptions — a model or spreadsheet encodes choices you never see, so a clean number rests on a guess buried in the method. False confidence — the output's fluency and formatting signal certainty the system does not have. None of these announces itself. A wrong answer looks exactly like a right one.

Keeping a human accountable

Because the failures are invisible from the output alone, verification cannot be optional. Someone must be able to say, of any answer that ships, "I checked this, and I stand behind it." The tool is not accountable — it cannot be. Accountability is a property of a person, and it does not transfer to software no matter how good the software gets.

You need to choose a supplier and want to compare five on cost, lead time, and reliability. A good use: ask a tool to draft a comparison table and surface trade-offs you might have missed. It works fast and catches a factor you had forgotten. Then you verify each figure against the actual quotes, because the tool may have filled a gap with an invented number. The tool did the structuring; you did the checking and the deciding. The recommendation is yours to defend.

A researcher asks a chatbot for sources supporting a claim and pastes the citations straight into a report. They look perfect — real-sounding journals, plausible authors, formatted correctly. None of them exist. The tool was used for the one thing it cannot be trusted to do unverified: assert specific facts about the world. The failure was not the tool. It was treating fluent output as checked output.

In the US federal case Mata v. Avianca, a lawyer submitted a court filing citing several judicial decisions to support his argument. He had used ChatGPT to find them. The cases did not exist — the tool had generated realistic-looking citations, complete with quotes and docket numbers, for decisions that were never written. When the court asked for copies, the fabrications came to light.

In June 2023, Judge P. Kevin Castel sanctioned the lawyers and imposed a $5,000 fine, noting they had abandoned their responsibility to verify. The tool was fluent and confident and wrong, and no one checked before it mattered. The episode is now a standard example of why verification stays essential — not a reason to avoid AI, but a reason to own what it produces.

• "It sounds authoritative, so it's probably right." Fluency and truth are independent; the tool optimises the first.
• "The tool is responsible for the answer." Accountability rests with the person who uses and ships it.
• "Newer models don't hallucinate." They hallucinate less in places, but confident fabrication has not been eliminated.
• "Verifying defeats the point." Verifying a fast draft is still far quicker than doing the whole task by hand.

Given an AI-produced output — a summary, a statistic, a list of sources, a recommendation — decide what you would accept as-is and what you would independently check before acting, and say why.

Think Like a Maester: Before you ship anything a tool produced, ask who will say "I checked this" — and make sure it is a person, not the tool.

1. Name three tasks where AI or digital tools genuinely amplify problem-solving.
2. What is hallucination, and why is it hard to detect from the output?
3. Why can fluent, well-formatted output be more dangerous than obviously rough output?
4. What actually went wrong in Mata v. Avianca — the tool, or how it was used?
5. Why does accountability stay with a human rather than transferring to the tool?

AI and digital tools are powerful amplifiers for generating options, crunching data, drafting, and simulating — but they fail through hallucination, hidden assumptions, and false confidence, none of which shows in the output. Use the leverage, verify what you ship, and keep a named human accountable for the result.

• Creative Problem Solving
• Wicked Problems and Complexity