True robot intelligence requires a digital twin of the entire world: WEF ‘Summer Davos’ in China

In order to achieve true robot intelligence, a digital twin of the entire world would be required, according to a presentation on social robots at the World Economic Forum (WEF) “Summer Davos” meeting in China.

The WEF Annual Meeting of the New Champions (AMNC), aka “Summer Davos,” in Tianjing, China kicked off today with several high-profile panels, discussions, and presentations on a wide range of topics including AI and robotics, geopolitics, the one health agenda, the global economy, and the energy transition.

Giving a presentation today on “Social Robots and I” University of Twente professor Vanessa Evers explained that it would be possible to train a robot like a Large Language Model (LLM), but it would require the creation of a complicated Large Behavioral Model that be like a digital twin of the entire world.

“For true robot intelligence, you need to build a model of the world, like a digital twin of the entire world”

Vanessa Evers, WEF Annual Meeting of the New Champions, June 2025

“We could add superhuman capabilities, listen to the heartbeat or watch the breath of a person to know stress, to know pain. We can detect dominance, aggression, creative flow — there’s all things you could detect in an automated way”

Vanessa Evers, WEF Annual Meeting of the New Champions, June 2025

Evers acknowledged that we still have no idea how to create a digital twin of the entire world, but the process could start in smaller, closed environments and make its way up.

“If we limit the world to a classroom or a house or a factory floor or maybe a hospital corridor, we are able to do that” she said.

“But even then, despite what all the tech bros might be telling you to get your money, we have no approach to that yet.”

Creating a digital twin of the entire world is an extremely complex undertaking that goes way beyond Large Language Models like ChatGPT.

In order to achieve true robot intelligence, a Large Behavioral Model would be needed to access and process enormous data from real world phenomena, which is far more complicated than creating and feeding Large Language Models.

“ChatGPT has to choose from 26 characters. What is the best next character? But what about the real world? Imagine vision alone and we don’t have hearing, touch, taste, everything like that — just seeing stuff — if you would freeze-frame and then predict what is the next thing that’s going to happen in the visual, it would be impossible to know at the pixel level”

Vanessa Evers, WEF Annual Meeting of the New Champions, June 2025

“In theory it would be possible to train a robot like a Large Language Model, but then a Large Behavioral Model — to use a lot of data from the real world and have it predict the next best action,” said Evers.

“A Large Language Model predicts what is the next best token, but of course that is super complicated in the real world“

This is because an LLM like ChatGPT “has to choose from 26 characters” but a Large Behavioral Model would have to have general knowledge of the entire physical world.

When it comes to a Large Behavioral Model predicting the next best action, as opposed to ChatGPT predicting the next best token or character, Evers says that it would have to have general knowledge of the world.

For example, it would have to know what humans know intuitively like how if something falls, it falls down and not up, or that a wallpaper doesn’t change from one second to the next like a hallucination.

But what are researchers actually trying to achieve in creating “true robot intelligence?”

Of course, there are many avenues to explore, but for the purpose of this WEF session, it’s all about creating social robots that can learn about us from our behaviors.

“We’ve built a robot that has many expressive capabilities. It can shape-change, it can taller and smaller, its head can grow, its body can shrink, it can move around, it has colors, facial expressions, gestures, sounds — you name it — and we let it run around for an extended period of time, just randomly combining all of its expressive capabilities, and then over time, maybe a year or maybe more, some behavior will emerge that turns out to be the best in that environment, and we have no idea what that will be, but the robot will learn socially from us, and it will look at our feedback, our facial expressions — if we hug it, if we touch it, if we smile at it, it will learn from that, and we will engage in active teaching”

Vanessa Evers, WEF Annual Meeting of the New Champions, June 2025

“My proposal is that we look at robots like aliens. It’s like an alien landed on earth and somehow it’s understandable. It has its own weird language and somehow it seems to be able to work around us in a seamless flow, and it does its thing, it’s very useful, but we don’t know that language; we just happen to understand it”

Vanessa Evers, WEF Annual Meeting of the New Champions, June 2025

For Evers, deploying social robots should be used as a solution to a problem, and sometimes the best solution is to not deploy robots at all, and she speaks passionately about wanting to do good while acknowledging that others may use the technology for bad.

But when you look back on what she said about digital twins being required before true robot intelligence could achieved, you may start to wonder what other kinds of robots could be developed and for what purposes.

At the same time, globalist organizations, big tech companies, industrial leaders, pharmaceutical companies, cybersecurity firms, defense departments, and so many more are all investing in research and development for digital twins for various purposes.

Although they are siloed at the moment, little-by-little, the entire world is indeed being modeled, which may one day contribute to an ultimate digital twin of everything on earth and thus lead to “true robot intelligence” that can be specialized and purposed to perform any task, for good or ill.

You can check out the full “Social Robots and I” session below, and The Sociable will be covering the second presenter, Future Explorer Society founder Chang Dong-Seon, in a follow-up story, where he dives into how children raised around robots may not be able to distinguish human emotions from machines, and how people may lose the ability socialize with other people in the age of social robots.

---

Image Source: Screenshot of Professor Vanessa Evers from the WEF Annual Meeting of the New Champions session on “Social Robots and I” June 24, 2025.