“People Don’t Know What They Don’t Know”: Paul Meinshausen on AI, Technology, and How We Make Choices

What drives the choices people make — and how can technology learn to understand them?

In a new episode of EVOLVE Voices, Aampe Co-Founder and CEO Paul Meinshausen joined Edilsa Bueno to explore what sits at the heart of both human and machine intelligence: the ability to learn from signals we don’t yet understand.

Why We Don’t Know What We Don’t Know

In the episode, Paul dives into how humans and machines both grapple with limited information:

“People don’t know what they don’t know. The best systems — human or AI — are the ones that keep learning anyway.”

From the subconscious cues people rely on — what they wear, when they engage, what they notice — to the invisible biases that shape decisions, Paul explains how agentic systems can help companies uncover and respond to those underlying patterns in real time.

Beyond Data — Toward Understanding

Much of today’s technology focuses on collecting more data. But as Paul notes, more data doesn’t automatically mean more understanding. The real challenge is making systems adaptive — enabling them to respond dynamically to new information instead of just describing what happened in the past.

That’s the foundation of Aampe’s agentic infrastructure: software that doesn’t just predict user behavior, but experiments, learns, and evolves alongside users.

“We shouldn’t expect people to bend to static systems. We should build systems that flex to human complexity.”

A Different Kind of Conversation

The episode also touches on the human side of technology building — the honesty that happens behind conference stages, how founders can collaborate with investors as real teammates, and how curiosity, not certainty, is often the most valuable mindset in innovation.

For anyone thinking about the intersection of AI, decision science, and human behavior, this conversation offers an authentic, wide-ranging perspective on how to make technology more human — not by simplifying people, but by designing systems that can learn from them.

Table of Contents

Related Posts

See All Posts ›
Load More
FAQ

Your Questions,
Answered

Still have questions? Contact our experts ›
Do I need "the basics" set up before I use Aampe?

No. Aampe *is* the basics (plus a whole lot more)! For example, you don't need to have user journeys set up before we can optimize. In fact, our customers don't set up user journeys at all! We achieve individualized personalization through the actual messages you send, so you don't need any formal messaging structure to get started. You only need a CPaaS connection so we can send message and a CDP connection so we can track impact.

Set up is a snap: Add your CDP and CPaaS API keys, set your goals, and you're ready to start writing!

Does Aampe replace my current CRM platform or CPaaS provider?

It doesn't have to! Aampe can work right alongside your current CRM platform or CPaaS provider, making them function better by producing much better results more efficiently. With a simple API connection, you can enable new levels of messaging personalization without having to spend time and effort building rigid segments and static, over-generalized user journeys.

Aampe doesn't need a large number of users to operate. We have apps with fewer than 10,000 users already seeing impressive results. If you write quality messages, we can use them to engage an audience of any size.

Not at all. Aampe will pull the relevant event data from the service you've instrumented your app with so the effort on your engineering team is kept to minimum.

You can share data with Aampe using a dump into a storage bucket (AWS, GCP, Azure, etc.), via SFTP. For more details refer to our Integration Guide. Cannot find your preferred method for data sharing? Let us know and we'll be happy to figure it out with your team.

There are a couple of options here. The preferred option is that an Aampe engineer installs the app on their test device (either from the store or using an APK). In that case, Aampe does all testing needed. In the cases where this is not possible, the alternative option is that you provide Aampe with a few user IDs either from your team or your test devices, and testing will be done in collaboration with your team.

Does Aampe require static segmentation?

No. Aampe uses semantic content labels and per-user learning rather than relying on broad predefined buckets.

No. The internal materials position Aampe as learning across multiple horizons and using reward functions to connect short-term signals with longer-term outcomes such as retention, revenue timing, and sustained engagement.

Aampe learns from real user behavior using tagged content, reward signals, abstraction layers, social calibration, and next-best-action policy selection to update what each agent believes from real user behavior.

No. Aampe relies primarily on real-world experiments, then uses machine learning to collect, generalize, and apply findings faster. We also use AI to support or perform tasks like writing copy, organize, and draw insights from user behavior, and identify messaging impact. Our hybrid approach of using both experimentation and machine learning avoid many of the common pitfalls of a naive ML approach.

Aampe frees your data science team to work on more productive things. Instead of bogging them down with requests for reports or messaging A/B tests, they can spend their time running more valuable analyses, doing the kinds of work that only they can do.

Absolutely! Aampe runs many of the tests and reports that a typical data science team normally would. Now apps of any size can have the same capabilities as the big guys.

Absolutely not! Even with Aampe running, you're still able to add or remove messages, constrain messages to certain time windows, send ad hoc messages, or even stop sending messages entirely.

Of course! (Although most of our customers report needing to send fewer ad hoc messages after starting with Aampe.) While Aampe is running, you can still send individual messages from your current customer communications platform.

We sure can! Aampe is designed to help you build a messaging library of thousands of unique and instrumented messages in minutes in a Bring Your Own Model approach. Check out how it works.

Does Aampe just help with sends and clicks, or can it help with mid and bottom of the funnel actions too?

Aampe helps with much more than clicks! One of the first steps of using Aampe is setting your goals. These may be any user action that you're able to track — anything from clicks to sessions to conversions — and, once they're set, Aampe optimizes for these goals.

Why should I trust Aampe's product and technology?

At Aampe, our answer will never be "Just trust us." In fact, we hate the "black box" mentality so much, we've gone to great lengths to explain how our system works for non-technical users, and we've even written a detailed whitepaper to explain our system to technical users! If you have any questions, please reach out! We love talking about how we make personalization better.

Continuous intelligence for customer engagement

Move beyond campaigns.
Let agents optimize every
interaction.

Book a Demo ›