Even with all the training and computing power in the world, artificial intelligence (AI) tools like ChatGPT still fall short in one surprising area: understanding a simple flower. A new study shows that AI can’t grasp the idea of a flower the way a human can.
The study, led by researchers at The Ohio State University, explains why. Most large language models (LLMs) that power AI are trained on language alone and sometimes images. That limits how they experience the world.
“A large language model can’t smell a rose, touch the petals of a daisy or walk through a field of wildflowers,” said Qihui Xu, the study’s lead author and postdoctoral researcher in psychology.
“Without those sensory and motor experiences, it can’t truly represent what a flower is in all its richness. The same is true of some other human concepts.”
Can AI capture emotion?
The team compared how humans and LLMs represent the meaning of 4,442 words. These words covered a wide range, from objects like “flower” and “hoof” to ideas like “humorous” and “swing.”
The researchers tested two families of models – GPT-3.5 and GPT-4 from OpenAI, and PaLM and Gemini from Google. Humans and AIs were compared on two main measures.
The first, called the Glasgow Norms, asked people to rate words based on nine qualities. These included how emotionally arousing a word is, how concrete it is, and how easy it is to imagine it. For example, how strongly does “flower” stir emotions? How clearly can you picture it?
The second measure, called the Lancaster Norms, focused on sensory and motor experiences. Participants rated how much each word involved senses like touch, smell, and sight, and motor actions like moving the hands or torso.
The individuals were asked how much the senses of smell and touch were involved in experiencing a flower.
The researchers then compared how closely human ratings aligned with AI ratings. They checked if AIs and humans agreed on which words were more emotionally charged, more imageable, or more tied to senses and actions.
AI lacks the human experience
The results were telling. AI models did quite well on words that didn’t involve sensory or motor information. Abstract concepts were easier for them to grasp.
But for words rooted in the physical world – those involving sights, smells, touches, and movements – AI fell behind.
“From the intense aroma of a flower, the vivid silky touch when we caress petals, to the profound joy evoked, human representation of ‘flower’ binds these diverse experiences and interactions into a coherent category,” noted the researchers.
Language alone, the study found, isn’t enough. “Language by itself can’t fully recover conceptual representation in all its richness,” Xu said.
Even though AIs can come close to human-like understanding for some concepts, the way they learn is very different.
“They obtain what they know by consuming vast amounts of text – orders of magnitude larger than what a human is exposed to in their entire lifetimes – and still can’t quite capture some concepts the way humans do,” Xu said. “The human experience is far richer than words alone can hold.”
Will AI catch up?
There’s hope that AI will improve. The study found that LLMs trained with images along with text were better at grasping visual concepts than text-only models.
Xu said future models might get even closer to human understanding when they can interact more with the physical world. “When future LLMs are augmented with sensor data and robotics, they may be able to actively make inferences about and act upon the physical world.”
For now, though, a flower remains something uniquely human that AI does not experience. “If AI construes the world in a fundamentally different way from humans, it could affect how it interacts with us,” said Xu.
Why human experience still matters
While machines are improving at handling language and images, they still lack something essential – the richness of real-world experience.
This deep connection to the senses shapes how people understand the world. It’s not just about describing things, but about living them. As the study shows, these experiences bind our concepts together in ways that text-based models cannot match.
Even as technology advances, these small but profound moments remind us of what it truly means to be human – and why the richness of lived experience remains beyond the reach of machines.
The full study was published in the journal Nature Human Behaviour.
—–
Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates.
Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.
—–
English (US) ·