There's a comfortable story about AI in business. The company is messy, decisions are slow, knowledge is scattered across inboxes and people's heads. Then AI arrives, reads everything, connects the dots and tidies up.

Miklós Róth thinks this story gets cause and effect backwards. His view fits in seven words, which he used as the title of an essay published by the Roth Complexity Lab: AI does not fix confusion. It scales it.

What does "AI scales confusion" actually mean?

It means AI behaves less like a fixer and more like an amplifier. Feed it a clear process, with defined inputs, owners and quality standards, and you get leverage: the same good work, faster and at greater volume. Feed it an unclear process and you get the unclear process at greater volume. The disagreements, the duplicated work and the undocumented exceptions don't disappear. They multiply.

Picture a sales team that never agreed what a "qualified lead" is. Before AI, the inconsistency costs a few wasted calls a week. After AI, a scoring model learns from inconsistent labels, an assistant drafts thousands of emails based on that score, and a dashboard reports the results with total confidence. The confusion is now faster, more expensive and much harder to see, because it looks like data.

Why does it happen? The S·I·C·T explanation

Róth's S·I·C·T framework supplies the mechanism. A system stays stable while Structure plus Cohesion keep pace with Information plus Transformation: S + C ≥ I + T. The full framework is here.

AI hits the right-hand side of that inequality twice:

• It floods the system with Information. Summaries, drafts, analyses, predictions, variations. The volume of plausible-looking output explodes.
• It accelerates Transformation. Workflows change monthly, roles blur, and new tools arrive before the old ones are understood.

What AI does not do, by itself, is add Structure (who decides, what the rules are, which data counts) or Cohesion (shared definitions, trust between teams, a common sense of what good looks like). People have to build those. When they don't, the balance tips, and the organization grows more fragile precisely as it grows more "AI-powered".

What are the warning signs inside a company?

A handful of symptoms appear again and again when AI is amplifying confusion:

• Faster output, slower decisions. Teams produce more reports than ever, and leadership feels less sure what to do.
• Duelling truths. Two departments run AI on overlapping data, reach different numbers, and both look authoritative.
• Invisible errors. Mistakes no longer look like mistakes. They look like well-formatted answers.
• Tool sprawl. Every team solves its own problem with its own tool, and nobody owns the whole.
• Pilot purgatory. Demos impress, nothing reaches production, and the next pilot starts anyway.

None of these is a technology problem. They are structure and cohesion problems that AI made louder.

How do you make AI amplify the right things?

The answer is not to slow AI down for its own sake. It is to build the stabilizers before the amplifier, or at least alongside it. Róth's advisory work at Roth AI Consulting follows a version of this sequence:

1. Clarify before you automate. Write down the process you're about to accelerate: inputs, decision points, owner, definition of done. If you can't, that is the first project.
2. Agree on shared definitions. One meaning of "qualified lead", "resolved ticket" or "approved content", recorded where both people and models can use it.
3. Name an owner for every AI system. Someone who can explain what it does, judge its output and switch it off. For high-risk systems, the EU AI Act makes human oversight a legal requirement, not a nice-to-have.
4. Pace the change. Limit how many new tools and workflows a team absorbs at once. Transformation is a pressure, not a free resource.
5. Measure outcomes, not output. More drafts is not a result. Faster resolution, higher conversion or fewer errors is.

What does this look like in marketing?

Marketing is where the pattern is easiest to see. A team without clear positioning, a consistent voice or an agreed list of priority topics gets access to AI writing tools. Output jumps from four articles a month to forty. Six months later the site has more pages and less authority: the articles contradict each other, chase overlapping keywords and describe the company in five different ways. Search engines and AI assistants, which reward clear and consistent entities, now see a blurrier brand than before.

The same team, with positioning, voice and topic structure agreed first, can use exactly the same tools to publish fewer, better pieces that reinforce one another. Same amplifier, different input. That is the core of how Róth's agency approaches AI visibility.

Isn't this just "garbage in, garbage out"?

Partly, and the old phrase is a good start. But Róth's point goes further. "Garbage in, garbage out" is about data quality. Confusion is about organizational clarity: unclear ownership, conflicting goals, no shared language. You can feed a model perfectly clean data and still amplify confusion if nobody agrees what the output is for.

That is also why, in his view, the most valuable AI work in many companies looks unglamorous at first: process maps, agreed definitions, ownership charts. Only then does the amplifier start amplifying the right signal, which is the idea behind the title of his book, Signal Over Noise.