Do you remember being an inquisitive child, learning life’s lessons the hard way? If the answer is yes, then you’ve most likely heard your mother yell “if you put your hand in the fire, it will get burned” five seconds before your pointer finger touched the gas range. Perhaps you were told that merely relaxing before your spelling test would propel you to the National Spelling Bee Championships? Maybe you played sports, and remember feeling slightly uneasy about your team’s chances of glorious victory upon learning of your rival school’s acquisition of an all-star player? In those instances, you made assumptions regarding the potential outcome of an event, or the results of an action, based on previously garnered limited information. This information may have been direct, or perhaps vague, but your mind connected this nugget of truth as being directly related to a particular outcome. In the most basic of terms, even as a child, you’ve always known that A + B = C.
Though basic human understanding, and our utilization of “sound logic”, dictates assumptions based on correlated information, we don’t often work to deduce answers in the opposite direction, or determine the difference between association and causality. By examining a particular result, one can begin to break down the various controls that have led to this result. When repeated, one can look for patterns, and streamline deductions based on these patterns. From there, one can determine what types of “A” and “B” factors will ultimately equal a “C” outcome. Essentially, these relationships are causal relationships, a matter of causal theory, which states that in order for an individual to have “knowledge” about a particular item, one must believe that the item is true, and there must be a causal relationship with that truth.
Since the birth of a philosophical approach to knowledge, many brilliant minds have contributed to the understanding of the vast relationships in existence between varying components of existence. Within the scope of modern science, one such influential scholar devotes his gifts toward academic research, and guiding the field of Bayesian Theory, and Causal Theories. Recently appointed as a Turing Fellow, Dr. Anthony Constantinou shapes the field of artificial intelligence, with a series of impressive professional accolades.
As the head of the Bayesian Artificial Intelligence Research Lab, a Professor in Machine Learning and Data Mining at Queen Mary University of London, and independent consultant, Dr. Constantinou’s breadth of knowledge constantly propels the field forward. His consulting work, which often utilizes causal theories in order to create plans of action for large companies, impacts the daily lives of thousands of employees. Additionally, he is widely involved within the arena of sports betting, parlaying his knowledge of Bayesian Networks, causal relationships, and rankings, in order to accurately predict sporting outcomes on a professional level.
From an academic standpoint, Dr. Constantinou’s published research, along with his role as a peer advisor for other academics seeking publishing, allows him to spread understanding about causal relationships. His expansive list of published research collateral is not only impressive, but continuously growing, with several titles currently under peer review. He actively shapes the future of the field, and oversees the research, development, and assessment of various funded research projects pertaining closely to the concept of causal theory.
Throughout his many published works, Dr. Constantinou seeks to discover causal relationships between two or more variables, and to parlay those relationships into measurable predictors of a particular outcome, event, or future change. In this year’s published research collateral for Queen Mary University of London, “Bayesian Artificial Intelligence For Decision Making Under Uncertainty”, Dr. Constantinou states that “scientific research is heavily driven by interest in discovering, assessing, and modeling cause-and-effect relationships as guides for action. Much of the research in discovering relationships between information is based on methods which focus on maximizing the predictive accuracy of a target factor of interest from a set of other related factors.”
Essentially, this statement suggests that the basic interest of science pertains to discovering meaningful cause-and-effect relationships between variables, and utilizing this repeatable knowledge to make scientifically sound decisions. For those not involved professionally in a scientific field, how is this knowledge useful throughout everyday life? By understanding why certain results occur time and time again, within a retail business, owners can gain understanding of the variables that lead to the desired results, and emulate those variables on an appropriate frequency. In a sports betting scenario, understanding the types of events that precede a certain match can result in desired monetary gain, increasing odds of desired outcomes.
For Dr. Constantinou, the notion of causal relationships, and ways in which these relationships may assist with increasing accurate predictions of sporting events, is a familiar topic. Throughout his previous academic endeavors, Dr. Constantinou created a model, aptly named “Dolores”, that aimed to predict soccer match outcomes by combining dynamic ratings, Hybrid Bayesian Networks, and causal relationships. The model was able to successfully predict outcomes at an impressive rate as a direct result of the utilization of these three theories together, working in tandem to minimize potential outliers, and avoiding dependence on only data spreads. Placing second (only behind another entry that based its’ model on Constantinou’s pi-ratings system), within an international competition, Machine Learning For Soccer, the model’s accuracy garnered international accolades.
Though many researchers rely predominantly on patterns, equations, and sheer datasets, the implementation of causal knowledge can provide additional accuracy that computing alone cannot account for. Dr. Constantinou categorizes this combination of data, along with the utilization of causal relationships, as “smart data”. In his published journal, called “Towards Smart-Data: Improving Predictive Accuracy In Long-Term Football Team Performance”, Constantinou states that smart-data is “a method which supports data engineering and knowledge engineering approaches that put greater emphasis on applying causal knowledge and real-world ‘facts’ to the process of model development, driven by what data are really required for prediction, rather than by what data are available.” Simply put, having all of the data in the world means nothing if one cannot attribute real-world utilization of such data, or apply real-world meaning to the data sets.
Since scientists often seek to discover the causes of various events, outcomes, and phenomenon, the discovery of concrete causal relationships is revered within the scientific community. The understanding of a particular outcome’s causality can steer scientists closer to discover a way in which to manipulate the cause, in order to garner the most favorable effect. For this reason, the quest for causality is paramount within the various medical fields, environmental sciences, engineering realms, and various other professional scopes.
As the definition of causal relationships suggests, everything is relative, and thus, there are adjoining scientific means that aide in the creation, assessment, and explanation of causal relationships. For example, the previously mentioned utilization of Bayesian Networks assists with creating tangible and scientific outlines for information gathered. According to Anthony Constantinou, these networks “offer a framework for modeling relationships between information under causal or influential assumptions, which makes them suitable for modeling real-world situations where we seek to simulate the impact of various interventions.” If causal theories supply the raw information regarding the relationship between cause and outcome, then Bayesian Networks supply an easily relatable manifestation of this raw data.
Throughout his published work “Bayesian Artificial Intelligence For Decision Making Under Uncertainty”, Dr. Constantinou warns about the difference between mere “association”, and “causation”. Essentially, items and situations that appear to be tied together are not necessarily involved in a cause-and-effect relationship. In this collateral, Dr. Constantinou explains that “unlike association, causal assumptions make claims about the effect of interventions. For example, though associated, an intervention on yellow teeth will have no effect on smoking nor on lung cancer, whereas an intervention on smoking will have an effect on both yellow teeth and lung cancer.” Though this concept appears to be simple enough to understand, there is often a deep discrepancy between association and causation, where mere association can be masked as causation, due sheerly to the mind filling in the blanks, and seeking further relationship between two variables.
Additionally, though popular studies sometimes suggest that students who feel anxious prior to taking an examination score poorly on said examination, one cannot confidently report that feeling anxiety causes students to perform poorly within the scope of an examination. In this particular example, the student may not have studied for the examination, leading to a decreased score. Additionally, there may have been other outlying factors that caused the student’s poor performance. Thus, even though there can be a relative association between anxiety and low test scores, one does not cause the other. In real-world situations, mere associations may not amount to any significant increase in desired outcomes, while causations will directly, and deliberately, predict a specific outcome.
Of course, though all things can be considered related, and causal relationships can be assessed between various factors, there are concisely unique types of causal patterns that may provide additional insight within a particular set of outlying variables. According to Harvard’s Graduate School Of Education, there are six existing causal patterns in existence throughout the universe, each uniquely applicable to real-world situations. These six unique types of causal relationships account for all types of cause and effect scenarios. By being able to categorize a particular set of causal relationships, one can predict further effects, or gain insight regarding the manipulation of variables.
Linear causality highlights a direct link between cause and effect, where one particular effect can be traced back directly to one particular cause. There are no additional effects in this type of causality, and no chances of outlying effects. In essence, this type of causality is by far the most straightforward, and easiest to comprehend. By adding additional steps to this simple linear causality, one would effectively create what is known as a domino causality, which takes the theory of linear causality, and adds the concept of time in a sequential unfolding. In this domino causality situation, it is important to trace effects back to the initial first effect, and first cause, in order to obtain the most accurate information, and for the sake of creating an informed prediction. By not following those general guidelines, and researching an effect other than the initial effect in a domino causality situation, one would potentially run the risk of only evaluating a completely altered effect. Thus, returning to the root effect is crucial toward maintaining integrity, and effective understanding of the causal relationship.
Perhaps the most frustrating of the six types of causality, cyclic causality imparts no concise beginning, and no concise end. In theory, it is a perpetual question of cause and effect, which leads to more cause and effect, and so forth, in a repeating pattern. Which came first; the chicken, or the egg? The old adage is actually a prime example of cyclic causality, and ponders one of life’s greatest mysteries for all mankind. In this example, both answers could potentially have merit, as one choice creates the second, and vice versa, creating a seemingly never ending circle of cause and effect. Much like cyclic causality, the effects of one thing spearhead another action with its’ own effects, and so forth, within the scope of spiraling causality. However, in the cause of spiraling causality, there is eventually an escalation, or de-escalation of effects. By manipulating one of the eventual effects, or adding other variables, the change in effects can begin to occur, thus ending the spiral effect in the so-called spiraling causality.
In a relational causality example, two particular variables work together in order to produce one effect, often implying relation between the two variables. As is the case for Dr. Constantinou’s arguments, everything is related, and thus, any changes in relationship between the two variables in a relational causality atmosphere will alter the effect of said changing relationship. For this reason, relational causality equations may often be less rigid than their counterparts, and may be harder to predict with grave certainty. Finally, in what is perhaps the most straightforward categorization of causality other than linear causality, the principle of mutual causality states that two things directly impact each other, either simultaneously, or throughout a long-term relationship.
Why are these classifications of causality important in everyday life, and what purpose do they serve for professionals who utilize them? How can categorizing the concept of causality allow for increased positive effects? Garnering concise insight regarding the specific relationships between variables allows researchers, and other individuals, to successfully quantify those relationships. By being able to further categorize the causality of several factors, one can look for patterns directly seen by the particular type of causality. As with most complex theories throughout the world, they become easier to work with once they are quantified in a repeatable, straightforward, and categorized manner.
In real-world situations, including environmental occurrences, categorizing causality provides the ability for more accurate predictions. In the event of the ocean being contaminated via an oil spill, there will be many overt effects of this event, as well as many secondary effects that are caused by the direct effects of this one particular variable. For example, an oil spill in the ocean may cause the direct effect of killing native fish populations within the affected area. This effect, in turn, may impact the migratory and predatory pattern of local birds, who previously relied on the fish for their sustenance. This effect, in turn, could create a tertiary effect of those birds relocating to another region, depleting its’ potentially smaller fish population, throwing off the entire local ecosystem.
In this particular example, cyclic causality is most likely the culprit of the seemingly never-ending chain of cause and effects. Due to the sensitive and ever-evolving nature of environmental changes, cyclic causality often affects the effects continuously occur as the result of an original change in the natural order. By understanding the nature in which these changes occur, conservationists, scientists, and other professionals can work to diminish the secondary and tertiary effects of natural occurrences. Conversely, if the effects are positive, they can also utilize this notion to continue the scope of the cyclical effects.
Throughout Dr. Constantinou’s professional career, he has sought to streamline artificial intelligence, the processes involved in predictive sciences, and machine learning. He has developed many useful applications that allow researchers to combine these scientific processes with the exploration of causal relationships, and increase their predictive accuracy, garnering success within their desired problem-solving niche. He continues to make great strides within the field, and maintains his position at the forefront of modern artificial intelligence. By understanding the limitations of only relying on datasets, and in turn, utilizing causal relationships within his algorithms, he develops more well-rounded results, and propels the field forward. After all, it’s all relative, and Dr. Constantinou can prove it.