A recent neuroscience report explains that today’s advanced AI systems still lack something humans naturally possess: internal embodiment, or an awareness of their own internal state. Humans constantly monitor signals like tiredness, uncertainty, hunger, or stress, and these signals influence how we behave and make decisions. According to the UCLA-led study, current AI models can process information and generate responses, but they do not have any internal mechanism that tells them when they are uncertain or overloaded.
The article makes an important distinction between external embodiment and internal embodiment. External embodiment means interacting with the outside world through perception and action, such as a robot moving or responding to its surroundings. Internal embodiment, however, refers to continuously tracking internal states like confidence, processing load, or risk. Researchers argue that this missing layer is one reason AI can make highly confident mistakes, because it has no built-in “self-check” system similar to the human body and brain.
One example used in the study was a point-light display test, where only a few moving dots are shown in the shape of a walking person. Human infants can often recognize this immediately, but several leading AI models failed and even described it as a group of stars. This suggests that human understanding is deeply connected to lived bodily experience, while AI is still mostly relying on statistical pattern matching.
Overall, the study concludes that future AI systems may need internal feedback loops that act like synthetic feelings or self-regulators—for example, signals for uncertainty, capacity limits, and confidence levels. These would not be human emotions, but they could help AI behave more safely, avoid overconfident errors, and become more aligned with human needs in real-world settings.
