Fitting Unstructured to Structured!

Forcing something less-defined to well-defined

The world we live in is seldom well-defined. Why is that the case? Because we communicate in languages and alphabets mostly as compared to using numbers.

Languages provides a lot of room for interpretation, interpretations that are usually modified by many things such as lessons from school, experiences, perspectives and many more. We do not have to go too far and just look these two words, “Fantastic” and “Good”. All of us, regardless of background and experience, will definitely say that “Fantastic” and “Good” if we are to put them on a positive-ness spectrum, “Fantastic” will lead “Good” i.e. “Fantastic” is definitely more positive than “Good”.

This is Human Intelligence. Our world model is partially made up of language through conversation and discussion, which comes with room for interpretation.

For those that has done data analytics, data science or perhaps more specifically supervised learning in machine learning, you know that the machine requires something very different. What is that?

Let’s use an example to illustrate. For instance, the bank needs to build a credit scorecard to determine the probability of default for existing credit card loans/line. Converting it as a supervised learning project, we will not have to define what is “default” in order to label the individual loans correctly, so as to meet the business requirement.

So what is “default” here? To set the label, we need it to be more definite than just the word “default”. We need details such as, when is a loan considered default or what does a default credit loan look like. Is it when it is 1 day past due or 30 day past due? We need the definition of “default” to be very well-defined, because the computer needs it, so that it can get a good labelled data, labelled data that can train a model suitable for the use cases.

This is Machine Intelligence. Because it is a computing machine, it only works with numbers and algorithms.

Conclusion

Herewith, we see the difference between Human Intelligence and Machine Intelligence. It is not so straightforward to say that machine intelligence (numbers and algorithms) can solve all human problems. Machine intelligence requires very well-defined boundaries or logic to work with. This means that it can solve or partially solve problems that Human Intelligence can convert into numbers, logic and algorithms.

Unless machine can define the boundaries on its own, converting a problem statement into numbers, symbols and algorithms, chances are good that we will get into an Augmented Intelligence era rather, where human and machine using their own individual intelligence to work together.

Your thoughts?

If you have benefited much from my posts, consider making a “book" donation. Link here. :)

To support the newsletter, please share, like and subscribe! :)

Recommended Content

What is the difference between programming for your bootcamp project vs doing it as a job in a firm? <Blog>

How much maths do you need for Data Science and AI? <Blog>