Accountability Is Human-Only
Another important consideration is that AI models will never be perfect. And they are, unfortunately, immune to consequences.
Consider the relationship between AI and news information. So far, there’s reason to suspect AI may be struggling to get it right. BBC research suggests more than half of news queries by major AI tools have major errors of some kind. One can only presume what that ratio looks like on other issues, particularly more niche or technical questions, where datasets are smaller and even minor errors can be hugely significant.
This can, of course, be improved, and will be (in theory). At an industry level, work continues to make better models. At the organization level, improving data management and breaking down information silos will improve the volume and quality of data available. But that is a slow, complex, and expensive process.
The reality is that even the most sophisticated AI models lack accountability. LLMs don’t analyze meaning, only correlation. If an AI tool gets the correlation correct, based on its analysis of existing material, that will be its output, whether the words themselves are meaningful or not.
AI cannot be relied upon for truth, insight, or alignment with your organization’s objectives — it simply reflects patterns in existing data.
Giving employees AI tools without training leaves them unprepared and unmotivated to validate results. Relying blindly on AI is a recipe for disaster. As a result, everyone in the organization, from the C-suite to entry-level employees, needs skills to complement AI’s strengths, compensate for its weaknesses, and cover its gaps.