Seeing the world without eyes

You can make an AI much smarter in a simulation by not having it literally “look” with cameras or pixels. Instead of analyzing images, we focus on the essence: what is an object, where is it and what can you do with it? By removing all that visual noise, the AI ​​becomes a lot more efficient and we better understand why it makes certain choices.

Logic instead of pixels

Most AI models try to mimic the brain by processing millions of colored pixels. With my method I want to do that differently. By directly feeding the AI ​​structured information. Instead of a picture of a chair, the AI ​​sees a list of facts: “this is a chair, it is near the table and there is something on it.” This allows the AI ​​to quickly scan the environment and logically determine what is most important at that moment.

Understand immediately

For example, when the system sees a cup of coffee, it does not look at the shape or color, but immediately sees the facts: how hot is it and is it fragile? This way, the AI ​​can immediately respond to danger (such as heat) without first making complicated calculations.

To make the AI ​​behave more humanly, we deliberately give it a bit of “blinders”. For example, he cannot see through walls and only sees what is right in front of him. This is sometimes difficult because he misses hidden dangers, but it also ensures that he does not become overstimulated. He learns to solve problems much more creatively and robustly, just like us.

Setting priorities

Humans are very good at ignoring distractions, and we also teach this AI this through ‘importance scores’. The AI ​​rates each object in the room. Is something moving unexpectedly? Then the score goes up. Does he need a specific item for his task? Then that will take priority. This way, only the most relevant information remains in his working memory and he does not get distracted by side issues.

Self-awareness

Finally, the AI is aware of itself. He ‘feels’ his own virtual arms and legs, knows how he moves and how much energy it takes. This self-awareness is crucial for making plans.

We want to test this theory by giving the AI ​​difficult tasks, such as navigating through a difficult space or recovering from an ‘injury’ (simulated damage). We think that an AI that knows who and what it is, learns much faster, makes fewer mistakes and adapts more easily to new situations.

Frequently Asked Questions

16 questions

Leave a Reply

Your email address will not be published. Required fields are marked *