Current AI works using statistics, that is, it looks at past data and tries to PREDICT future events.
For a system that PREDICTS, it is never sure. Such a system cannot be used as a tool when we expect a guaranteed 100% outcome.
,This I believe. is as a result of the use of things like Large language models (LLMs). These are a group of words and sentences that the machine memorizes without understanding what they mean. When the machine talks about 'classroom', it does not even know what a 'classroom' is.
Some AI scientist have identified the problem. That is , AI at this moment is unable to see. Unless AI is given the ability to see objects, it will not be able to KNOW what it talks about.
AI needs to have senses like humans, so it can see, and hear and smell and taste, that way, it can use these senses to KNOW what is going on in its environment and what the knowledge it has means.
Talking about vision ( the ability of machine to see), I worked on a project that would enable AI to see just like humans.
My findings were that, it is possible for AI to see, just like humans but how pictures on a computer works, makes it difficult to do so.
For example, pictures on computer screens are made using PIXELS. A pixel is a rectangular box that contains a single color. putting many pixels together forms an image on the computer screen.
For a computer to see just like humans, pixels are not the best things for a picture to be made up of, because not all things human eyes see are rectangular, but for a computer, everything it sees is a group of rectangles.
Also, I noticed that LIGHT has an effect on how a computer can see an object. When light is thrown on an object, a human eye sees that the object is the same, just that more light has shined on it.
But for a computer, the color of the object completely changes as soon as more light is thrown on it. If you are trying to identify an object based on the color of pixels, more light on the object changes the pixels' values completely.
In conclusion, I know that what I am writing here will not be seen by many people nor will my research get any recognition.
But if we want to make machines that work as close to humans, they have to be able to see just like humans.
And if they can see just like humans, we have to reconsider 'pixels' - how objects are created on the computer again, and also how 'light' affects objects on a screen.
Top comments (0)