'This is a training and trust problem, not an AI problem': Barely any workers are verifying information given to them by AI

• Just 24% of employees verify the results of an AI-generated answer before proceeding
• A worrying 39% of employees have acted upon AI-generated responses that turned out to be inaccurate or unsuitable
• The research was conducted by Censuswide on behalf of learning company O’Reilly

AI chatbots are preferred to colleagues and managers for work-related matters, according to a survey of 2,000 UK workers, with under a quarter (24%) bothering to confirm the results or directions from an AI-generated solution.

Meanwhile, with over a third of employees revealing that guidance from an AI has resulted in inaccurate and incomplete answers (if they were even suitable in the first place), almost as many have received no AI training or guidance in the past year.

The research from O’Reilly Media raises some serious questions, not just about how AI is being used in the workplace, but what controls, oversight, and training accompany its use.

An AI training gap?

The use of AI in the workplace has had some interesting outcomes, but one that few will have predicted is the use of AI chatbots as a substitute for advice from colleagues.

O’Reilly’s survey has unlocked not just the concerning 24% bothering to verify AI-generated answers, but also the paltry 16% that are provided with AI training that relates to their role, and 35% say no AI guidance or training has been given at all.

Without the full context of individual workplaces to consult, it appears that AI has been introduced and teams simply left to get on with it, while training attempts to catch up.

"We're seeing a workforce that has fully embraced AI as a first port of call, but the safety net hasn't caught up with the speed of adoption," said Alexia Pedersen, SVP International at O'Reilly. "People are moving faster than their organizations' training and governance.”

Pedersen describes that gap as a place where “mistakes happen.” But the obvious answer isn’t necessarily the right one. There is an alternative to putting the breaks on AI, she says. “The answer isn't to slow down AI adoption; it's to close the gap with the right training, verification habits, and trusted expertise to back it up."

Building a culture of verification

While the results of the survey raise questions about AI use, the statistics do point towards a challenge in the delivery of training, as well as trust in colleagues and organizational structures.

Pedersen observes, "This is a training and trust problem, not an AI problem." Some solutions will need to be found to ensure that AI training is as important as other work-based learning.

"Organizations that invest now in structured, role-specific AI training and in building a culture of verification rather than uncritical acceptance will be the ones who turn today's AI adoption into a genuine productivity advantage," Pedersen concludes.