Who Does Make It a Rule? Human? or Machine?

hello world machine learning

“Rule-based algorithms have instructions written by humans, they’re easy to comprehend. (…) Machine-learning algorithms, by contrast, have recently proved to be remarkably good at tackling problems where writing a list of instructions won’t work.”

[Hello World: Being Human in the Age of Algorithm, Hannah Fry]

Nowadays we often hear the word “algorithm” on the news and social networks. By the way, what is the algorithm? The “algorithm” is a (mathematical) recipe to accomplish a certain task. So, your grandma’s recipe for chicken soup is, in some ways, an established algorithm. But when we say about the algorithm recently, it usually refers to a computer algorithm, a series of computer languages to solve a certain problem. There are two different types of algorithms: (1) a rule-based algorithm that follows the prescribed details by humans and (2) a machine-learning algorithm that makes its own rule by machine (computer) itself.

Who does make it a rule for a new task in the future? Humans can make a crystal clear algorithm so that anybody can check the inherent bias or errors of the new rule. Machines, on the other hand, can make a high-performance algorithm without any prior knowledge and deep understanding of the new system. In the age of AI, the power of machine-learning algorithms is no way negligible and the use of this power in various fields is inevitable. However, “Great power comes great responsibility”. So, we, as humans, repeatedly scrutinize such black-box algorithms and prevent misuse of algorithms. We should always know that the final decision should come from humans because machines have no responsibility for their decision. Also, a human should provide some important rules to machine-learning algorithms such as consideration for others, tolerance, and sacrifice, which may lead to creating not only better performance algorithms but also impartial algorithms.

[Wrap up] Book Review: Weapons of Math Destruction

NOW, we live in the age of Big Data and nobody can stop the invasion of artificial intelligence (AI) into our world. Many books about data science have shown the effectiveness, accuracy, and efficiency of the data-driven models in various fields such as economics, social science, and engineering. Do you agree with that data-driven models via mathematics/computer science/machine learning make a prosperous future?

THE author, Cathy O’Neil, based on her academical and industrial experiences, showed us the dark side of mathematical (or data-driven) models and called this ‘Weapons of Math Destruction’. The author said that WMDs have peculiar characteristics: (1) Opacity; (2) Scale; and (3) Damage. Through several examples that these characteristics of WMD have a negative effect on, the author called our attention to ‘fairness’ in Big Data and AI.

The following links are some quotations from the book with my thoughts.

(1) Make Your Crystal Ball Shine

(2) What Ingredients Do We Need for Yummy Data Soup?

(3) How to Make my Model Unspoiled?

(4) Am I the Same Person as I Was Yesterday?

(5) Some Numbers to Represent You

(6) The Whole is Different the Sum of its Parts

(7) Don’t put me in, Data

(8) Justice: What’s the Right Thing to Do in Data Science?

(9) As Human Beings, We are Flawed but We Learn

How to Make my Model Unspoiled?

“Without feedback, however, a statistical engine can continue spinning out faulty and damaging analysis while never learning mistakes.”

[Weapons of Math Destruction, Cathy O’neil]

To deal with a spoiled child effectively, parents (or teachers) should consistently observe their child. When they become rude, parents should give them feedback immediately and teach responsibility, appreciation, and respect about others.

Data-driven models also require the right feedback so that they update them in the right direction. To this end, data-driven models use their mistakes to adjust the model. However, this feedback loop commonly includes only one-side mistakes. For example, the AI HR-screening model has only data about the bad performance of chosen applicants. It cannot have data about the successful career of unchosen applicants.

What Ingredients Do We Need for Yummy Data Soup?

“To create a model, then, we make choices about what’s important enough to include, simplifying the world into a toy version that can be easily understood.”

[Weapons of Math Destruction, Cathy O’neil]

Imagine you make chicken soup for dinner. What ingredients do you need for delicious soup? Chicken absolutely and maybe celery and onion and more; it depends on your mother’s recipe. Organic ingredients will be much better for your health.

Successful data-driven models commonly require enough important data. However, we do not know which data is important so we just put all the data into the model. Fortunately, a data-driven model may know what data is salient among all these data (via feature selection or dimensionality reduction) and make the own recipe. Also, unbiased (like organic) data will be much better for the accurate model.

Make Your Crystal Ball Shine

“These mathematical models were opaque, their workings invisible to all but the highest priests in their domain: mathematicians and computer scientists. Their verdicts, even when wrong or harmful, were beyond dispute or appeal.”

[Weapons of Math Destruction, Cathy O’neil]

The fortune-tellers show our future with their crystal balls. We don’t know how this crystal ball works but we just believe (or not) its prediction. In fact, the fortune-tellers don’t know, too. The only thing they can do is to make their crystal balls shine.

Most of the predictive models based on machine learning are black-box models like crystal balls so we cannot know what happens inside. Someone worries about this opacity but this opacity may eliminate prejudice and bias. The one thing we can do is to feed them on unbiased and accurate data.

More Data, More Simple

“Single molecules are far too random. The balloon, along with the air it contains, follows a predictable pattern, but only when considered in aggregate.”

[Uncharted, Erez Aiden & Jean-Baptiste Michael]

A dot-to-dot has scattered points with the number and we connect these dots in numerical order. Finally, we get the hidden figure. But, if there is no number on each dot, we may not draw the right figure. What if we have more and more dots? Now, we realize the silhouette of the figure and draw it easily (now, no number required).

More data also provides the silhouette of the underlying pattern of Big Data effectively. After spotting the hidden patterns, we don’t need the detailed data (like a number in dot-to-dot). It can be a clear and simple model in large-scale data domain (intertwined with many features). Don’t worry about adding more features and data in your model. More data, more simple.

David Hume as a Data Scientist

“Hume believed that we can’t be absolutely certain about anything that is based only on traditional beliefs, testimony, habitual relationships, or cause and effect. In short, we can rely only on what we learn from experience.”

[The Theory That Would Not Die, McGrayne, Sharon B.]

Empiricism put forth by David Hume claims that observation/investigation is the correct way to extend our cognitive capacities. However, individual experiences often are incomplete and biased due to limited experiences. Hence, practically, it is skeptical to gain certainty based only on personal experience, observations, and investigation.

Nowadays, in the Big Data era, we have enormous data collected from all the people in the world. Integrated data can provide unbiased and common observation about human nature. That is, it is time to recall Hume’s empiricism. In fact, the fundamental philosophy of machine learning is totally based on Hume’s empiricism and this would bring us closer to “Truth”. If Hume was still alive, he would be a Googler.