Mind-reading technology aims to make AI truly smart

Despite all his intelligence – and can be extremely intelligent – Giving instructions to AI can sometimes feel like talking to a toddler. You provide a detailed message explaining exactly what you want and the AI ​​does something that may be related to it, but it’s definitely not what you wanted to say. Scientists have now developed a system that can detect that unspoken “That’s not what I meant” response directly from your brain waves, potentially allowing an AI to realize that it has misunderstood you and attempt to correct itself in real time.

Researchers at the Korea Advanced Institute of Science and Technology (KAIST), in collaboration with Microsoft Research Asia, created a brain-computer interface technique called Neural Value Alignment (NVA). The system uses electroencephalography (EEG) to detect different brain responses that occur when a person sees an AI pursuing the wrong target or performing an unexpected action. Those signals can then be used as feedback to help the AI ​​figure out exactly where it went wrong and adjust its behavior accordingly.

For humans and AI to work together seamlessly, AI must understand exactly what a person is actually aiming to achieve. Current AI systems infer this primarily from things they can observe, such as directions, speech, actions, and gestures. The problem is that human behavior can be quite ambiguous, even to other humans, let alone an algorithm.

As the researchers explain, the same action can serve several completely different objectives, while the same objective can also be achieved through several different actions. If you pick up a cup, for example, you might be planning to drink from it, wash it, move it to another location, or give it to someone else. On the other hand, if your real goal is simply to quench your thirst, you can grab the cup, grab a bottle of water, or ask someone to bring you a drink. This creates what the team calls goal-action ambiguity, where observing the action alone does not necessarily reveal either the true goal or the preferred way of achieving it.

The researchers’ solution was to stop relying solely on what the person physically asked, whether by action or direct prompting, and instead receive confirmation from the brain itself. Our brains continually predict what should happen next, and when the outcome differs from that prediction, characteristic neural responses can appear. The team focused on two of these responses: reward prediction error (RPE) and state prediction error (SPE).

Reward prediction error reflects a discrepancy in the outcome or goal and, in the NVA system, may indicate that the AI ​​did not understand what the person ultimately wanted. For example, you say to a home robot: “Bring me something to drink.” You actually want water, but he brings you coffee. The goal or outcome itself is incorrect, so your brain generates a reward prediction error.

The state prediction error is slightly different. It reflects an unexpected state transition, meaning that the goal may still be correct while the action, intermediate step, or method chosen by the AI ​​differs from what the person expected. Therefore, an SPE can effectively tell the system: “Yes, that’s where I wanted you to go, but not like this.”

So, for example, you want water and the robot understands it correctly. But instead of picking up the bottle of water next to you, walk into the kitchen and fill a glass from the tap. You still get water, so the objective is correct, but the action or sequence of states differs from what you expected. That can lead to a state prediction error.

The researchers conducted experiments that produced different brain wave patterns depending on whether there was a reward prediction error, a state prediction error, or both errors simultaneously. They then applied deep learning to these EEG signals, allowing the system to classify how the person responded to the AI’s behavior. This allows AI to detect and distinguish when the target is wrong or the method is wrong without a person saying anything.

The team implemented the system through a “human-AI synergy algorithm based on neural value alignment” that uses the decoded neural signals as feedback for AI decision-making. In addition to being able to detect that something is wrong and thatThe system can use this feedback to change its course of action in real time depending on which of the signals it detects.

The importance of the work goes beyond eliminating the occasional frustrating conversation with the chatbot. A system capable of detecting this type of unconscious disagreement could eventually prove useful anywhere humans and AI-controlled machines need to work closely together. A home robot, for example, could realize, from its user’s neural response, that it has misunderstood an instruction and change course without waiting to be verbally corrected.

Similar feedback could eventually be useful in industrial robots, autonomous vehicles that respond to driver judgment, medical or rehabilitation robots for people who have difficulty speaking or moving, and educational systems that adapt to a student’s cognitive state.

“This research is significant because it shows that AI can go beyond inferring human intention from visible behavioral outcomes alone and instead directly use cognitive signals generated in the brain during collaboration with AI,” said Professor Sang Wan Lee, who led the research.

Obviously, there are at least a few steps between this research and being able to stare at a robot and stop it in its tracks. For one thing, the system was demonstrated in simulations, not a full real-world demonstration in which a freely operating robot continually reads someone’s EEG and corrects their behavior around the house.

Additionally, EEG requires electrodes to measure small electrical signals produced by the brain, and those signals can be noisy and considerably less convenient for acquiring externally controlled experiments. It will therefore be important to establish exactly how reliably the NVA works during natural interactions, with people moving around and dealing with situations much more complicated than laboratory tasks.

The research was published in the journal. IEEE Transactions on Cybernetics.

Source: KAIST