Training is the key to overcoming AI's illusion of competence

Access to AI is exceedingly easy. People can simply log into a website or download an app and are instantly connected to generative and agentic AI tools that can answer almost any question and carry out myriad tasks. But that level of access does not always translate to proficiency.

Many organisations are deploying AI without establishing the foundations for effective use. They are not considering important matters such as storage, prompting, integration and governance.

Mia Bowyer an AI Corporate Trainer and the Chief Executive Officer of Automate Social. She sees these issues almost every time she steps in front of a group of people when delivering AI training.

"AI is a really easy tool to use, and a lot of people can jump right into using it because that's the way the tool is designed," she said. "People think that they're proficient in AI because they can access some information from it quite quickly and they get a bit of a dopamine rush, but they haven't learned any of the basics."

As a result, Bowyer said organisations end up with messy AI systems. They don't know where work is stored, and they don't know how to instruct AI correctly. She said they waste more time using AI than they would do things manually

"AI is marketed to us as time saving, set and forget tools. But what ends up happening is people completely forget the fundamentals," said Bowyer.

The first thing organisations need to do is define what they think a good deployment looks like. They need to have a clear idea about what they would accept through marketing, or if bookkeeping needs to be done at a specific point of time, or if HR contracts need specific attributes.

Bowyer suggested a specific approach when starting with AI.

"Start by training the leadership team and set the goal and the vision for the AI systems. Have all your specific tools connected in one place.  Engage the technology team to set up the security and privacy of the system.  And then ensure all users follow the same prompt structure," she said.

While getting the set up right and ensuring people receive context-sensitive training are important, Bowyer said tackling AI reticence requires a deft hand. She recommended giving people some quick wins instead of lots of technical detail.

"You need to help them elevate what they're doing already."

For example, an executive assistant Bowyer recently worked with had been manually preparing meeting notes for several years. Bowyer equipped them with a meeting recorder and a Claude system that she could send the audio to. Once they saw how this could save time and enable them to focus on more interesting and important work, their AI reticence started to fade.

Bowyer said that once someone sees how AI can be relevant in their role they are more likely to engage. In the training Bowyer conducts, the focus is not just on showing what is possible, but how that impacts people in their role.

While most people would never grant an employee full access to their bank accounts, private emails, or personal schedules, many routinely hand those digital keys over to AI.

"People are willingly handing the keys of the castle over to AI, which is a million times worse than giving that to one staff member," Bowyer said.

The risk of important data being accidentally shared can be mitigated by auditing LLM settings, deploying data-redaction tools and using secure network protocols like VPNs. Bowyer said we'll see organisations moving away from entering data into public platforms. Businesses and individuals are increasingly looking toward private, independently operated AI models to safely harness the technology.

Organisations like control. When a new application or service is introduced, project and change management processes are employed to ensure people are prepared and that all risks are identified and mitigated. The pace of AI adoption often exceeds that capability. And this is exacerbated by the pace of change in how AI tools work. In just three years we have moved from generative systems that were high-powered question-answer tools to agentic systems that can monitor activity and automate complex processes.

"I think the gap is going to get wider and wider, and the tools are going to get better and better. Companies are going to get slower unless they get trained."

Businesses don't need more powerful AI tools. They need to become better at using, managing and securing the tools they already have.