ThoughtStorms Wiki

More or less most ArtificialIntelligence these days is done through Machine Learning. Ie. throwing huge data-sets at number crunching algorithms. And getting the machine to fit patterns to it.

Machine learning efficiency is outpacing MooresLaw

Quora Answer : Is the belief that AIs can learn by themselves real or just a myth? If it's not true how close are we to making this reality?

Feb 7, 2019

Machine learning algorithms can be supervised or non-supervised.

Supervised means you give them training data that says "this is an example of a pattern you put in category 1, this is an example of a pattern you put in category 2 ..." etc.

Unsupervised learning, you say to the computer "here are a bunch of patterns, go and work out what categories they should be separated into". And usually the computer works out the categories based on some kind of cluster analysis or measurement of similarity between the example patterns.

99% of the time when we talk about AI "learning by itself" we really mean unsupervised learning.

Unsupervised learning is certainly real. We use it a lot, though we may prefer supervised learning for being less computationally expensive.

But, of course, all computers are built by humans. All machine learning is instigated by humans. Even if we just hook an unsupervised algorithm to a webcam, it's still the humans who are selecting that particular camera and therefore the training data that that camera implies.

There are no "wild" computers doing learning simply because of their human-independent nature.