Machine Learning: A Technical and Philosophical Standpoint

A curious mind may be surprised that the story of machine learning (ML) begins almost 200 years ago.

 

• Charles Babbage proposes the union of machine and mathematics in 1823.

• A machine is designed by Alan Turing in 1939 to decipher encrypted messages.

• In 1954, reinforcement learning (RL) computational models are designed by Marvin Minsky.

• John McCarthy introduces the term artificial intelligence (AI) to the world in 1956.

• The perceptron, or first artificial neural network, is developed in 1958 by Frank Rosenblatt.

 

By the 1960s the stage was set to join man-powered strategy and determination with machine-powered learning and inference.  Today, those tasked with making ethical, evidence-based, and at times critical decisions are supported by their reliable counterparts: ML models.

 

Types of ML Models

 

In ML, a model is similar to a function in mathematics.  Variables are input, operations are performed, and an output is produced.  Each model comes with strengths and weaknesses, and each is designed with a specific purpose in mind.

 

• Supervised, or trained on existing data, ML models use data that has already been collected to make classifications, or predictions.

• Deep learning (DL), or neural network-based, ML models make decisions based on a series of weighted values.

• Reinforcement learning (RL) models are created in a digital learning environment where programs exploit positive reinforcement.
  

 

DL and RL models can be classified as AI, as they are conceived with human neural processes in mind.

 

It was the work of Donald Hebb in 1949 that laid the foundation for understanding how data is analyzed by human neurons, inspiring others to wonder if biological processes could be replicated using math and machine.

 

 

 

Building a Model

 

The term algorithm in ML refers to the lines of programming code linked to mathematical computations.  For the vast collection of functions supporting each algorithm, libraries exist within high-level coding environments that allow users to directly import the computations through simple lines of code.

 

Once the programming language and coding environment are determined, the process of framing problems and stating objectives leads to the decision of which ML model(s) will be utilized for the needs of the project.

 

 

The Data

 

We've reached the element of ML upon which the whole operation turns: data.  It’s here where the team, project, collection methods, and purposes come under close inspection.  The who, what, when, where, why, and how determines the ethical standing of a ML model.

 

Questionable standards in data collection and ML models reaching conclusions during training yet lacking true predictions during testing have led to debates in the use of AI.  Unscrupulous parties using this technology to strip an individual’s ability to control their own likeness have brought this branch of engineering to a point where grey areas of governance are no longer suitable.

 

Organizations and individuals deploying AI ML models must be held up to the same federal and European Union (EU) standards as those collecting and storing data.  As companies ramp up their compliance efforts to include cybersecurity frameworks, standards, and models, language highlighting AI should also be included.

 

The road to building trust surrounding ML is paved with education, compassion, and an understanding that close inspection of the past is the only way to progress responsibly.

 

 

“They must consider that great responsibility is the inseparable consequence of great power.”

 

A translated excerpt from documents set forth by the Convention Nationale in 1792, an assembly inciting the abolishment of the French monarchy and the eventual Reign of Terror.

Dataiku. (n.d.). Charles Babbage, The Father of the Computer - History of Data Science.

Encyclopedia Britannica. (n.d.). National Convention | History, Definition, & Reign of Terror | Britannica.

Fremaux, N., Gerstner, W., Sprekeler, H. (2013, Apr.) Reinforcement Learning Using a Continuous Time Actor-Critic Framework with Spiking Neurons | PLOS Computational Biology.