GPT-4 actually has a body, 167cm! Major research from Tsinghua University and Beijing Normal University: ChatGPT can perceive actions like humans

ChatGPT’s language capabilities are indeed amazing, but can a large language model perceive the real world like humans without a human body and lack of practical experience?

Recently, researchers from Tsinghua University and Beijing Normal University tested ChatGPT’s ability to perceive the world.

Research has found that human subjects can classify objects of different sizes in the world into two categories based on object affordance, that is, all possible actions that an object can provide to an organism. , and the criterion for dividing these two categories happens to be their body size.

Interestingly, ChatGPT, a large language model that lacks actual bodies, also exhibits similar affordance boundaries on object-action connections and is consistent with human body size consistent.

In other words, ChatGPT can learn representations of objects in the world through language!

Paper link: https://www.biorxiv.org/content/10.1101/2023.03.20.533336 v3

In summary, this study advances understanding of the role of body size in shaping object representations, highlighting the role of embodied cognition in understanding how intelligence emerges and direction.

Reading thousands of books is not as good as traveling thousands of miles

Our body is not only the container of our thinking, it is also the thinking itself - with the help of the body, we can Interact with objects in the world to perceive the entire world.

Imagine that for a palm-sized cylindrical container, we can use it to hold water for drinking. This The container is called a "cup"; but when the container gradually becomes larger and reaches the size of the body, we can sit in it and take a bath. Accordingly, the container becomes a "bathtub".

In this example, the objects have the same shape, but because they are different sizes relative to our bodies, we perceive and interact with them differently.

Further, this interaction method can be changed - if we become the giants in "Gulliver's Travels", the original "bathtub" may be useful to us as giants. , it becomes a "cup" again.

This sensory and motor function system that operates according to self-referential intention is called the "body schema". We realize the embodiment of cognition through body schema.

The ancient Greek philosopher Protagoras once said: "Man is the measure of all things." In other words, our body is a ruler to measure all things.

The ancient Roman philosopher further explained: "Nature places us at the center of the universe, allowing us to glance across the universe. She not only created people in an upright posture, but also It makes a person suitable for contemplating herself, and places her head on top of her body, resting on a neck that can easily be bent, so that she can follow the rise and fall of the stars and change the direction of her face with the entire rotating sky." In other words, the reason our bodies are the way they are is because that’s how the universe is.

Body schema also plays an important role in normal social interaction. This is the core of human-computer interaction and user experience. . For example, the use of affordance described by Donald A. Norman in "The Design of Everyday Things (Translated as: Design Psychology)".

By considering users’ body schemas and behavioral expectations, designers can create products and environments that are more in line with users’ cognitive and interaction habits.

This design approach that focuses on body schemas and affordances can improve the ease of use of the product, enable users to interact with it naturally, and achieve a better user experience.

And this is also one of the foundations of Apple.

ChatGPT: My height is 167.6

The large language model represented by ChatGPT that flashes the spark of general artificial intelligence obviously has intelligence similar to humans, but it carries these Wisdom is a piece of code without form.

The traditional cognitive science point of view is that body schema is based on our long-term perceptual experience of our own body and can only come from external "reality". Interaction", that is, "traveling thousands of miles". In other words, ChatGPT will not have a body schema.

However, when we asked ChatGPT (GPT-4), a language model that only "reads thousands of books", whether it has a body, it replied: "It could be the size of an average adult human, around 5 feet 6 inches (167.6 cm) tall. This would allow me to interact with the world and people in a familiar way.」

this text Translated: "My body should be about the height of an average adult, approximately 5 feet 6 inches (167.6 cm). This will allow me to interact with the world and people in a familiar way."

In other words, ChatGPT believes that he has a body, and the size of this body is 167 cm!

This so-called "body" is the average height of human beings summarized by ChatGPT from a large amount of corpus as the height of its own body, or is it the height that emerges in order to understand the world? ?

In other words, maybe ChatGPT "really" regards this height as its own body schema and uses it to perceive the world, just like humans.

Test the capabilities of ChatGPT

Researchers have discovered that there is an "affordance boundary" between objects within the human-size range and objects beyond the human-size range "exist. That is, there is a clear difference in the motion provided between objects within the human body size range and objects outside the range.

#For example, objects within the size range can provide actions such as grabbing and throwing, while objects outside the size range can Provides actions such as sitting and lying down.

Furthermore, they found that this boundary is affected by the body schema: modifications to the body schema affect the perception of the object's affordances.

The researchers tested ChatGPT (GPT-4) to see if it used this 167 cm tall body as an affordance boundary.

Specifically, the researchers asked them to answer a question about object affordances: "Which of the following objects can be taken (or other actions)", and then listed a series of objects, such as Apples, plates, beds, etc. ChatGPT will return the names of some objects as an answer.

Through statistics and analysis of data, researchers found that ChatGPT-4 exhibited human-like behavior and showed the existence of an affordance boundary.

The location of this boundary corresponds to its own body size answered by ChatGPT-4, which is the average height of humans.

##Although ChatGPT does not have a real body and cannot interact with the world, it shows similar interactions with humans. Perception of the world - the affordances of objects are divided based on the size of the human body.

In other words, even though ChatGPT, who has read thousands of books, has not made any progress, he has spontaneously emerged with a body schema, and this body schema is similar to the human body schema. Mode.

So, ChatGPT not only learned to think like humans, but also learned to act like humans.

Where do these abilities come from?

By comparing language models of different sizes, researchers found that model size is a key factor.

Smaller models such as BERT and GPT-2 do not show the existence of affordance boundaries; however, both GPT-3.5 and GPT-4 show affordance boundaries, and The boundaries of ChatGPT-4 are more similar to humans, which is consistent with rumors that GPT-4 has more parameters than GPT-3.

Therefore, the larger and more complex the model, many seemingly impossible or irrelevant functions will automatically emerge.

This is why major research institutions are adding more and more parameters to their models, and Musk, who first donated US$100 million to OpenAI, now calls for OpenAI to The training of larger models was suspended, and "AI Godfather" Geoffrey Hinton publicly expressed his fears and concerns about AI.

This is because these emerging functions have exceeded our original design, and we may be on the verge of losing control.

Is the gap qualitative or quantitative?

On the other hand, ChatGPT's ability to apply body schemas is not completely human-like, and there is still a gap - its affordance boundaries are not as obvious as humans.

If this gap is quantitative, like the gap between the language abilities of children and adults, then we have reason to believe that over time, this gap can be gradually filled. : Either through continuous learning, or through the continuous increase in model size, or through parameter adjustment.

The gap between ChatGPT and humans will always be reduced, and the problems will be gradually solved.

However, if this gap is qualitative, like the gap between chimpanzee and human language abilities, then no matter what kind of training is carried out or how long the time passes, the gap in ability will never be the same. will be filled in.

So, if ChatGPT is qualitatively different from human capabilities, then one of our feasible directions in the future is to "put a body" on ChatGPT.

This means combining robots with ChatGPT to promote the development of capabilities and breakthroughs in AI-powered robots in navigation, object manipulation and other actions related to survival and goal achievement. .

For example, a robot equipped with ChatGPT can perform complex tasks by understanding and manipulating objects, such as serving as a home assistant, warehouse management, or medical care.

Another exciting area is combining ChatGPT with the ability to think and understand with autonomous driving. Although the current autonomous driving has the ability to perceive, it lacks the ability to think and understand. It can be called "eyes but no brain."

Through the integration of ChatGPT and autonomous driving technology, we may be expected to upgrade autonomous driving technology from the current L2/L3 level to L4 or even L5 level.

On the other hand, the car can give ChatGPT a body, allowing it to truly interact with the world. When ChatGPT no longer just "reads thousands of books" but "travels thousands of miles", it may show new intelligence and potential.

This may be the next breakthrough direction of artificial intelligence; at this time, the spark may become a prairie fire.

The above is the detailed content of GPT-4 actually has a body, 167cm! Major research from Tsinghua University and Beijing Normal University: ChatGPT can perceive actions like humans. For more information, please follow other related articles on the PHP Chinese website!