AI & Machine Learning
Usually we get our information from other humans, or books written by humans, or look at graphs made by (preferably trusted) humans, point being, there’s always a person behind the mask (Scooby Doo, am I right!).
And that’s the first bridge I want to cross, stating the fact that AI is backed by human intelligence.
The whole idea is computers are our prefect computing tools. The time it would take me to calculate the distance between Earth and the Sun, and then use that to estimate how much fuel would be used to get to the Sun from Earth, based off average spaceship fuel consumption rates, would be done MUCH faster by a computer. But the computer has to be told what to calculate and how to calculate.
Artificial Intelligence can be summed up as utilizing powerful computers to run some human defined logic on some sort of data to give some sort of output. Utilizing the speed of the computer, we can have a view of a large group of results in record time, then use that to run more computations.
As we analyze results, we use the data to refine our computer programs and make them more efficient
Let’s try and humanise this for a second:
Say you wanted to write a computer program that, given any sentence in English, would then predict the sentiment and emotion contained within the words.
The first step is always to train the computer program to understand emotions. To do this, we need a training data set, which is a set of data that has been generated or collated by a human. One of the most popular data sets used for such semantic analysis, is a list of movie reviews, where each review is tagged with a 0 or 1 for Negative or Positive review respectively.
Applying some mathematics, algorithms, a little bit of magic and a hint of alchemy, the computer would come up with a model for what words it finds are most utilized in the Positive and Negative reviews. One of the most basic algorithms used creates a “bag of words” model. This means, it converts each review into a list of words contained in it. It then counts the number of appearances of each of these words in one of the classes. So for example, from the training data set, the word “fantastic” might have appeared 59 times out of 200 positive reviews and 0 times out of 200 negative reviews. This causes the computer model to associate the word fantastic with positivity, and as such, when it encounters that word in a sentence out in the wild, it would be inclined to believe that the sentence is a Positive one. And voila, we have programmed the logic behind a super basic semantic analyzer.
I often like to think of computing systems as little children. We teach children our native languages, and the meaning of words, then show them examples of how those words are used together and what they mean and convey.
Over months and years, children learn more and add more words to their personal dictionary. As such, they can then express more complex ideas and also comprehend deeper sentences. Computers are very similar, except they can do the learning a child does in years, in a few minutes. With both children and computers, they will make mistakes as they learn and we will supervise them along the way, until the point where we have impacted all our knowledge on them and can trust them to operate on their own (mostly). In a nutshell, this is the underlying concept of Machine Learning. Taking an input data set, applying algorithms to it to form a model, and then using that model to make predictions.
This is just where the world of AI opens up. Using meaningful logic in a meaningful way, to give meaningful results to people, and making that happen in minimal time and at minimal cost. Then continuously refining models and algorithms to improve our systems gradually but assuredly.
This has been a severe simplification of the process of Sentiment Analysis, Machine Learning and Artificial Intelligence, but it sheds some light on some of the steps involved in making Artificial Systems Intelligent. Hopefully I have been able to convince you and not confuse you with all the tech lingo. I think we’ll call this Part 0, cause the artificial rabbit hole doth go very deep, and stay tuned to find out more on the next episode (of Dragon Ball Z!)
