Self-Training Machines
https://miro.medium.com/max/1200/0*SiJ9s1hXtTegibI_
Original Source Here
Artificial Intelligence for Computers who want to be human.
Humans are biological machines who can teach themselves.
If we want to achieve actual machine intelligence, we can truthfully label “Artificial Intelligence” and build machines that can teach themselves.
This may sound like a crazy idea from a non-PhD, and it may be, but everyone loves a crazy idea when it works. And it seems to be given to me to share crazy ideas everyone might learn to love because they might work.
How do humans learn?
Humans begin by learning elementary things.
For instance, humans learn simple shapes like circles, squares, triangles, and others.
Then humans learn how to apply these simple shapes to other shapes. For instance, what is the shape of a clock? Clocks do not have to be round, but the typical form is round. Clocks have numbers arranged clockwise, and clockwise is a shape. Clocks have a function, and the function of a clock is to tell time, and time is also a function. Do we see a pattern here?
Shapes and functions.
What if shapes are functions?
What if humans are intelligent because we can recognize shapes and not much more.
What if humans can build a database of shapes and apply them to the world around us?
What is the traditional approach to machine learning?
Linear regression, primarily. And neural networks. And others?
What if our approach to machine learning is just wrong?
What if machines could be taught to recognize shapes?
What if machines could be taught to build functions? Functions that use learned shapes to build more complex functions and on and on and on.
How do humans recognize faces?
What is a face? Faces are basically round shapes sitting atop bodies.
Faces have eyes, and eyes are round shapes that have round shapes inside.
How do we know a human face from an animal face? Now, this is where things get interesting. Humans and animals have eyes. Humans and animals have a nose. Humans and animals have ears. These are all common. So what is not common? What is the difference? Maybe there is an aspect of human intelligence that seeks to determine differences? Maybe machines should learn this also.
When a human seeks to recognize a shape, we begin by recognizing the obvious. I have mentioned some of the obvious attributes of a face, whether that face is human or animal. After we recognize the obvious, we then have to look closer to determine the differences so we can begin to classify the shape. Finally, we determine the classification of the shape, and it becomes recognized.
A Machine Learning Model
If I were asked to work on a machine learning model, I would begin by building some obvious attributes about the thing being classified. I would flag the obvious as metadata to allow a later process to cross-check the obvious against a more rigorous set of functions.
When recognizing objects, part of what humans do, if they do any of this well, is to cross-check ourselves. For instance, if we think we have recognized a human face, then we begin to look for other attributes that may tend to indicate the thing as possibly being human. The more attributes that seem to be human, the more the face may appear to be human or animal or whatever.
The problem with machine learning.
And this may be why humans are doing this so poorly.
If we can empower machines to become self-aware and self-taught, what is stopping them from controlling us?
Oh no. My machine overlord has just informed me that it is time for me to do something else.
AI/ML
Trending AI/ML Article Identified & Digested via Granola by Ramsey Elbasheer; a Machine-Driven RSS Bot
via WordPress https://ramseyelbasheer.io/2021/06/01/self-training-machines/
