Latest news with #QihuiXu
DW
11-06-2025
- Science
- DW
AI art can't match human creativity, yet — researchers – DW – 06/11/2025
Generative AI models are bad at representing things that require human senses, like smell and touch. Their creativity is 'hollow and shallow,' say experts. Anyone can sit down with an artificial intelligence (AI) program, such as ChatGPT, to write a poem, a children's story, or a screenplay. It's uncanny: the results can seem quite "human" at first glance. But don't expect anything with much depth or sensory "richness", as researchers explain in a new study. They found that the Large Language Modes (LLMs) that currently power Generative AI tools are unable to represent the concept of a flower in the same way that humans do. In fact, the researchers suggest that LLMs aren't very good at representing any 'thing' that has a sensory or motor component — because they lack a body and any organic human experience. "A large language model can't smell a rose, touch the petals of a daisy or walk through a field of wildflowers. 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," said Qihui Xu, lead author of the study at Ohio State University, US. The study suggests that AI's poor ability to represent sensory concepts like flowers might also explain why they lack human-style creativity. "AI doesn't have rich sensory experiences, which is why AI frequently produces things that satisfy a kind of minimal definition of creativity, but it's hollow and shallow," said Mark Runco, a cognitive scientist at Southern Oregon University, US, who was not involved in the study. The study was published in the journal Nature Human Behaviour , June 4, 2025. What are the challenges to book preservation? To view this video please enable JavaScript, and consider upgrading to a web browser that supports HTML5 video AI poor at representing sensory concepts The more scientists probe the inner workings of AI models, the more they are finding just how different their 'thinking' is compared to that of humans. Some say AIs are so different that they are more like alien forms of intelligence. Yet objectively testing the conceptual understanding of AI is tricky. If computer scientists open up a LLM and look inside, they won't necessarily understand what the millions of numbers changing every second really mean. Xu and colleagues aimed to test how well LLMs can 'understand' things based on sensory characteristics. They did this by testing how well LLMs represent words with complex sensory meanings, measuring factors, such as how emotionally arousing a thing is or whether you can mentally visualize a thing, and movement or action-based representations. For example, they analyzed the extent to which humans experience flowers by smelling, or experience them using actions from the torso, such as reaching out to touch a petal. These ideas are easy for us to grasp, since we have intimate knowledge of our noses and bodies, but it's harder for LLMs, which lack a body. Overall, LLMs represent words well — but those words lack any connection to the senses or motor actions that we experience or feel as humans. But when it comes to words that have connections to things we see, taste or interact with using our body, that's where AI fails to convincingly capture human concepts. What's meant by 'AI art is hollow' AI creates representations of concepts and words by analyzing patterns from a dataset that is used to train it. This idea underlies every algorithm or task, from writing a poem, to predicting whether an image of a face is you or your neighbor. Most LLMs are trained on text data scraped from the internet, but some LLMs are also trained on visual learning, from still-images and videos. Xu and colleagues found that LLMs with visual learning exhibited some similarity with human representations in visual-related dimensions. Those LLMs beat other LLMs trained just on text. But this test was limited to visual learning — it excluded other human sensations, like touch or hearing. This suggests that the more sensory information an AI model receives as training data, the better it can represent sensory aspects. AI's impact on the working world To view this video please enable JavaScript, and consider upgrading to a web browser that supports HTML5 video AI keeps learning and improving The authors noted that LLMs are continually improving and said it was likely that AI will get better at capturing human concepts in the future. Xu said that 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. But independent experts DW spoke to suggested the future of sensory AI remained unclear. "It's possible an AI trained on multisensory information could deal with multimodal sensory aspects without any problem," said Mirco Musolesi, a computer scientist at University College London, UK, who was not involved in the study. However, Runco said even with more advanced sensory capabilities, AI will still understand things like flowers completely differently to humans. Our human experience and memory are tightly linked with our senses — it's a brain-body interaction that stretches beyond the moment. The smell of a rose or the silky feel of its petals, for example, can trigger joyous memories of your childhood or lustful excitement in adulthood. AI programs do not have a body, memories or a 'self'. They lack the ability to experience the world or interact with it as animals and human-animals do — which, said Runco, means "the creative output of AI will still be hollow and shallow." Edited by: Zulfikar Abbany
Gizmodo
05-06-2025
- General
- Gizmodo
Things Humans Still Do Better Than AI: Understanding Flowers
While it might feel as though artificial intelligence is getting dangerously smart, there are still some basic concepts that AI doesn't comprehend as well as humans do. Back in March, we reported that popular large language models (LLMs) struggle to tell time and interpret calendars. Now, a study published earlier this week in Nature Human Behaviour reveals that AI tools like ChatGPT are also incapable of understanding familiar concepts, such as flowers, as well as humans do. According to the paper, accurately representing physical concepts is challenging for machine learning trained solely on text and sometimes images. 'A large language model can't smell a rose, touch the petals of a daisy or walk through a field of wildflowers,' Qihui Xu, lead author of the study and a postdoctoral researcher in psychology at Ohio State University, said in a university statement. '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.' The team tested humans and four AI models—OpenAI's GPT-3.5 and GPT-4, and Google's PaLM and Gemini—on their conceptual understanding of 4,442 words, including terms like flower, hoof, humorous, and swing. Xu and her colleagues compared the outcomes to two standard psycholinguistic ratings: the Glasgow Norms (the rating of words based on feelings such as arousal, dominance, familiarity, etc.) and the Lancaster Norms (the rating of words based on sensory perceptions and bodily actions). The Glasgow Norms approach saw the researchers asking questions like how emotionally arousing a flower is, and how easy it is to imagine one. The Lancaster Norms, on the other hand, involved questions including how much one can experience a flower through smell, and how much a person can experience a flower with their torso. In comparison to humans, LLMs demonstrated a strong understanding of words without sensorimotor associations (concepts like 'justice'), but they struggled with words linked to physical concepts (like 'flower,' which we can see, smell, touch, etc.). The reason for this is rather straightforward—ChatGPT doesn't have eyes, a nose, or sensory neurons (yet) and so it can't learn through those senses. The best it can do is approximate, despite the fact that they train on more text than a person experiences in an entire lifetime, Xu explained. 'From the intense aroma of a flower, the vivid silky touch when we caress petals, to the profound visual aesthetic sensation, human representation of 'flower' binds these diverse experiences and interactions into a coherent category,' the researchers wrote in the study. 'This type of associative perceptual learning, where a concept becomes a nexus of interconnected meanings and sensation strengths, may be difficult to achieve through language alone.' In fact, the LLMs trained on both text and images demonstrated a better understanding of visual concepts than their text-only counterparts. That's not to say, however, that AI will forever be limited to language and visual information. LLMs are constantly improving, and they might one day be able to better represent physical concepts via sensorimotor data and/or robotics, according to Xu. She and her colleagues' research carries important implications for AI-human interactions, which are becoming increasingly (and, let's be honest, worryingly) intimate. For now, however, one thing is certain: 'The human experience is far richer than words alone can hold,' Xu concluded.
