Allen Frances, chair emeritus of psychiatry at Duke University, wrote a piece in Psychiatric Times recently on the human brain versus the chatbot brain.
Most of it is the comparison you would expect. 86 billion neurons. Trillions of synapses that powered Alexander the Great (on the image), Isaac Newton, Rick Rubin and people like you and me. 20 watts of power. Versus thousands of watts and brute-force computation in dense data centers.
In the article, Frances draws a clean line between how we learn and how chatbots learn.
Humans learn from lived experience, encoded into synapses through plasticity. Chatbots learn from large datasets, optimized in training phases separate from their operation.
In other words, a machine has a knowledge basis that is statistical rather than experiential.
Read that twice if you work in B2B marketing. If you have 15 years in this job like I do, your value is less what you remember, more what you have lived through: the launch that flopped at the beginning. The campaign that won the award and lost the customer. The client who only reads headlines and then asks questions answered in the first bullet on the slide. The CEO who hates a certain phrase. All the pain was worth it :)
No LLM has any of that experience. Models have statistics about marketing. You have scars.
Why humans have an advantage
Humans are in good shape competing against bots because two things are true at the same time. AI keeps getting better at the statistical work. Drafting, summarizing, comparing, translating. Frances calls it “a flame that continues to grow hotter the more it is fed.”
And the human experiential layer does not transfer any better to models just because the model is bigger. A larger model with no exposure to your brand, your sector, or your last three years of internal politics is just a more confident generalist. Not a senior practitioner.
Most AI rollouts in B2B marketing are stuck at this sh*tty seam. The team has a powerful model. The model is statistical. The team’s value is experiential. Without a bridge, the team gets faster at producing work that does not sound like them.
The output goes up. The pride goes down. That is the slop.
And it is what Frances is pointing at when he warns we “may become even less efficient as we are deskilled with increasing dependence on chatbots.” He’s saying that if you use a tool that does not carry your judgement, over time, you’ll stop bringing your judgement to the work.
You do not solve this by using less AI. You solve it by getting your team’s judgement inside the AI they already use.
Three moves are needed, from my point of view
- Find the senior judgement already in your team. This is the two or three people whose review notes everyone trusts, the template that works because someone fought for every line, and the unwritten rules protecting the brand. Most of this context lives in heads, not docs.
- Encode this judgement into workflows. Putting it into prompts is not enough for a measurable output and a quality standard you can audit. Bring in the senior person to build the workflow and let it carry their name.
- Run the encoded workflow where the work already happens, whether that is Copilot, Claude, ChatGPT, or Gemini. Don’t make your people log in or migrate to another app. Make senior judgement the default.
Done well, the results should come in immediately. Junior briefs will come back at senior quality the first time. Regional teams can produce work that sounds like the headquarters but in a local way. Senior people will get their afternoons back and have less pingpong, less tool-hopping, and less frustration.