The Illusion of Free Will in Modern Machine Learning
This article was made in collaboration with Leonard Ang, an undergraduate student pursuing a BS Computer Science degree at the University of the Philippines Diliman who has shared experiences across philosophy and computer science.
What is Machine Learning? By definition, it is when a machine can imitate human behavior and emulate how they think and act. A published paper written by logician Walter Pitts and neuroscientist Warren S. McCulloch entitled “A logical calculus of the ideas immanent in nervous activity” was regarded as a breakthrough in laying the first foundations of machine learning. It indicates the usage of mathematical principles to detail the science and psychology behind human decision-making.
However, in 1950, Alan Turing introduced what computer scientists now know of as the “Turing Test” to determine whether a machine can be considered intelligent or unintelligent. Soon after, at a conference sponsored by the Defense Advanced Research Projects Agency (DARPA), Turing firmly emphasized the questions “Can a machine think?” and “Can a machine be linguistically indistinguishable from a human?”
Consider an image of the “Turing Test” above; we can think of A as the computer, B as a human person, and C as an interrogator. The primary objective of the interrogator is to identify which among A or B is the computer and, at the same time, test the conversational skills of the humans themselves. This test has proven to be significant as the machine must come up with an appropriate response to the human, almost like an actual conversation.
And this association with humans is quite interesting. Because for over a millennium, humans have always been in control, and never in our history have we conceived that technology could do the same. As such, one should become acquainted with three primary forms of machine learning: supervised, unsupervised, and reinforcement learning.
Supervised learning is concerned with the relationship between the developer and the system as it constantly relies on algorithms and models to create the desired output. Commonly used algorithms include: neural networks, decision trees, and support vector machines, to name a few. As most “practical” machine learning relies on supervised learning, we can represent how this control flow might appear — with X as the input variable and Y as the output variable. The process can be denoted in the equation — Y = f(X).
There are many real-world applications of more effective forms of this input-output process, including, but not limited to, creating a machine learning model to perform land classification, and the project required the developers to label each picture with a class/subject (i.e., house, tree, land, mountain, desert).
Following this is unsupervised learning; what becomes slightly different is that this requires a dataset providing adequate information on the patterns and tendencies of the user. For example, some of these datasets can include how much a user has purchased a specific item or a set of bank details on local clients. As such, standard algorithms include k-means clustering, Gaussian mixture models, and hierarchical clustering, as shown below.
With predictive models, developers can create models that cater to specific properties, rules, and filters to ensure that unlabeled datasets are analyzed and clustered accordingly. For example, with online shopping platforms on the rise, these companies have also utilized predictive models (in the form of recommender systems) to create consumer profiles based on customers’ purchase activity and preferences.
Finally, as mentioned earlier, reinforcement learning is a branch of machine learning, where the goal is for a machine to implement optimal decision-making strategies. In a game-like scenario, a developer must employ a version of trial and error for the device to devise a solution for a particular problem they want to solve. For example, in 2019, researchers from Stanford Neuromuscular Biomechanics were able to construct a 3D human model capable of running and walking with minimum commands. This allowed the machine to learn how humans run and recognize these patterns, which can significantly help persons with disabilities.
Artificial Intelligence and Machine Learning have played significant roles in our lives today. More than ever, we are becoming more digitalized, and our technological prowess has only drastically improved. However, what about preserving the aspect of machine learning concerning philosophical ideals? To understand this more, it is crucial also to classify machine learning into two significant ideologies, one of which is general AI and the other, Narrow AI.
Artificial narrow intelligence or ANI pertains to an application or task-specific AI. By its definition, the form of machine learning is focused on making singular tasks that imitate humans, given a limited number of constraints. An example is Siri on Apple devices or Netflixs’ algorithm to generate personalized recommendations. In addition, this form of artificial intelligence includes Language and Speech recognition that have positively affected consumers worldwide.
Artificial General Intelligence AGI is used to understand humans’ emotions, beliefs, and intellectual thought processes. These machines are almost as intelligent as humans, and although previously impossible, China’s Post K Supercomputer, the world’s fastest supercomputer, can achieve a top speed of 8 petaflops. To illustrate this further, one petabyte equals one quadrillion operations per second. Known as “strong AI,” it can be said that this form of AI bears its consciousness with the capability of reason and performing hundreds of mathematical functions at once.
Artificial Super-Intelligence (ASI), which from the term itself signifies that AI has surpassed human capability and, although a hypothetical futuristic notion, for now, it causes one to conceive a future composed of human augmentation. ASI machines can become capable decision-makers in society and could one day cause even greater havoc on the structures of modern society. With these in mind, we must understand that humans have an innate psychological trait of anthropomorphizing non-human entities or, in simpler words, attributing human characteristics to nonliving things.
Why do we feel this special connection with artificially intelligent beings? Can we classify them as such, and do they have free will?
To understand this further, we must provide a preliminary definition of what precisely free will is the ability of a person/individual to act upon a set of choices and do so independently. In his book, Discourses on the Method (1637), French Philosopher — Rene Descartes says, “Cogito ergo sum” or “I think therefore I am.” Now in the context of humans, this point can be applied given that we can think outside of our premises and, in effect, become capable of creativity.
But, as humans, we also have a set of limitations and parameters on said “free will,” which is quite an interesting counteract to this popular notion of free will, German Philosopher Arthur Schopenhauer says that “Man can will what he wants but cannot will what he wills.” We do not have free will, but we all carry out a particular program while living in an “illusion” of making independent, free choices. You may ask why? In quantum mechanics or any scientific experiment, there are a random set of outcomes; however, we do not have control over these outcomes at all. We, this universe’s physical and conscious beings, are subject to the tangible laws of quantum physics.
“We may regard the present state of the universe as the effect of its past and the cause of its future. At a certain moment, an intellect would know all forces that set nature in motion and all positions of all items composed of nature. If this intellect were also vast enough to submit these data to analysis, it would embrace in a single formula the movements of the greatest bodies of the universe and those of the tiniest atom; for such an intellect, nothing would be uncertain, and the future just like the past would be present before its eyes.”
— Pierre Simon Laplace, A Philosophical Essay on Probabilities
Therefore, Artificial Intelligent machines cannot possess “free will” as the parameters defined in the program can only permit the device to do a particular task. It cannot change the way it has been programmed unless a human explicitly does so. However, taking the stature of our current universe into account. It cannot and should not be possible.
At the onset of the pandemic, while the world was amid unprecedented times, the emergence of AI and ML only seemed to flourish. The pandemic brought numerous opportunities for utilizing AI and ML in different facets of life, especially in the healthcare sector. As airplanes were grounded worldwide, millions of citizens hoped to fly back into their respective countries.
In the Philippines, AI facilitated faster and easier processing of identification documents in airports to monitor the spread of COVID-19 and pursue the necessary health precautions or measures (e.g., quarantining, mandatory testing, etc.). In addition, in collaboration with the Philippine Red Cross, the world’s first DIY lab software company, Dashlabs.ai, automated identification document processing — precisely passports — through computer vision.
If you have ever seen a passport, you will notice two lines of text at the bottom of the passport data page. This is called the machine-readable zone (MRZ). The International Civil Aviation Organization (ICAO) defines the purpose of an MRZ as something which “may be used to capture data for registration of arrival and departure or simply to point to an existing record in a database.” Using an algorithm known as optical character recognition (OCR), a computer can extract different identifying pieces of information.
Let me briefly provide some details on how information from MRZs is removed. First, an algorithm known as the bounding-boxes algorithm is implemented. This algorithm works by looking for regions of text throughout the photo and obtaining the coordinates of the “invisible text box” surrounding each. Specifically, these coordinates refer to the coordinates of the corners of the said “text box.” Note that this algorithm is viable since the MRZ is consistently found in roughly the exact location within the passport.
After a few intermediary steps under the hood, the computer can crop the picture of a passport and effectively isolate the MRZ region. Second, using Google Cloud’s Vision AI, respective pieces of information — name, nationality, date of birth, and the like — are extracted. Lastly, this information is then displayed on the Electronic Case Investigation Form, which enables all passengers coming into the Philippines to enter their personal and travel details in advance, have an expedited testing process when they land, track their specimen, get their test results, and receive a clearance certificate from both the Philippine Red Cross and the Department of Health’s Bureau of Quarantine.
Although AI can automate manual processes such as form-filling, since we all know that prevention is better than cure, AI and ML have also been used in detecting and diagnosing COVID-19 in patients based on chest X-rays. In a 2020 paper published by Ozturk et al. entitled “Automated detection of COVID-19 cases using deep neural networks with X-ray images”, the researchers discussed their proposed 19-layer machine learning model “DarkCovidNet” which utilizes a combination of layers and network architecture similar to those found in typical deep learning neural networks: convolutional neural networks (CNN), pooling layers (also called max pooling), and Leaky ReLu as an activation function. Using the researcher’s proposed model, they noted: “that the proposed model has achieved an average accuracy of 98.08% in detecting COVID-19.”
The AI and ML fields are developing and ever-changing, and it is our responsibility to use discoveries or findings for the betterment of the world. Although AI and ML have reshaped and reformed our world in ways we never thought were possible, it is always worth noting and reflecting on the potential side effects a novel machine learning model or AI technology has on every person in the world — rich or poor, young or old — from an ethical and philosophical perspective.
