Article | February 17, 2020
In recent years, artificial intelligence research and applications have accelerated at a rapid speed. Simply saying your organization will incorporate AI isn’t as specific as it once was. There are diverse implementation options for AI, Machine Learning, and Deep Learning, and within each of them, a series of different algorithms you can leverage to improve operations and establish a competitive edge. Algorithms are utilized across almost every industry. For example, to power the recommendation engines in all media platforms, the chatbots that support customer service efforts at scale, and the self-driving vehicles being tested by the world’s largest automotive and technology companies. Because of how diverse AI has become and the many ways in which it works with data, companies must carefully evaluate what will work best for them.
Article | February 17, 2020
Clear conceptualization, taxonomies, categories, criteria, properties when solving complex real-life contextualized problems is non-negotiable, a “must” to unveil the hidden potential of NPL impacting on the transparency of a model.
It is common knowledge that many authors and researchers in the field of natural language processing (NLP) and machine learning (ML) are prone to use explainability and interpretability interchangeably, which from the start constitutes a fallacy. They do not mean the same, even when looking for a definition from different perspectives.
A formal definition of what explanation, explainable, explainability mean can be traced to social science, psychology, hermeneutics, philosophy, physics and biology. In The Nature of Explanation, Craik (1967:7) states that “explanations are not purely subjective things; they win general approval or have to be withdrawn in the face of evidence or criticism.” Moreover, the power of explanation means the power of insight and anticipation and why one explanation is satisfactory involves a prior question why any explanation at all should be satisfactory or in machine learning terminology how a model is performant in different contextual situations. Besides its utilitarian value, that impulse to resolve a problem whether or not (in the end) there is a practical application and which will be verified or disapproved in the course of time, explanations should be “meaningful”.
We come across explanations every day. Perhaps the most common are reason-giving ones. Before advancing in the realm of ExNLP, it is crucial to conceptualize what constitutes an explanation. Miller (2017) considered explanations as “social interactions between the explainer and explainee”, therefore the social context has a significant impact in the actual content of an explanation. Explanations in general terms, seek to answer the why type of question. There is a need for justification. According to Bengtsson (2003) “we will accept an explanation when we feel satisfied that the explanans reaches what we already hold to be true of the explanandum”, (being the explanandum a statement that describes the phenomenon to be explained (it is a description, not the phenomenon itself) and the explanan at least two sets of statements, used for the purpose of elucidating the phenomenon).
In discourse theory (my approach), it is important to highlight that there is a correlation between understanding and explanation, first and foremost. Both are articulated although they belong to different paradigmatic fields. This dichotomous pair is perceived as a duality, which represents an irreducible form of intelligibility.
When there are observable external facts subject to empirical validation, systematicity, subordination to hypothetic procedures then we can say that we explain. An explanation is inscribed in the analytical domain, the realm of rules, laws and structures. When we explain we display propositions and meaning. But we do not explain in a vacuum. The contextual situation permeates the content of an explanation, in other words, explanation is an epistemic activity: it can only relate things described or conceptualized in a certain way. Explanations are answers to questions in the form: why fact, which most authors agree upon.
Understanding can mean a number of things in different contexts. According to Ricoeur “understanding precedes, accompanies and swathes an explanation, and an explanation analytically develops understanding.” Following this line of thought, when we understand we grasp or perceive the chain of partial senses as a whole in a single act of synthesis. Originally, belonging to the field of the so-called human science, then, understanding refers to a circular process and it is directed to the intentional unit of discourse whereas an explanation is oriented to the analytical structure of a discourse.
Now, to ground any discussion on what interpretation is, it is crucial to highlight that the concept of interpretation opposes the concept of explanation. They cannot be used interchangeably. If considered as a unit, they composed what is called une combinaison éprouvé (a contrasted dichotomy). Besides, in dissecting both definitions we will see that the agent that performs the explanation differs from the one that produce the interpretation.
At present there is a challenge of defining—and evaluating—what constitutes a quality interpretation. Linguistically speaking, “interpretation” is the complete process that encompasses understanding and explanation. It is true that there is more than one way to interprete an explanation (and then, an explanation of a prediction) but it is also true that there is a limited number of possible explanations if not a unique one since they are contextualized. And it is also true that an interpretation must not only be plausible, but more plausible than another interpretation. Of course there are certain criteria to solve this conflict. And to prove that an interpretation is more plausible based on an explanation or the knowledge could be related to the logic of validation rather than to the logic of subjective probability.
Narrowing it down
How are these concepts transferred from theory to praxis? What is the importance of the "interpretability" of an explainable model? What do we call a "good" explainable model? What constitutes a "good explanation"? These are some of the many questions that researchers from both academia and industry are still trying to answer.
In the realm on machine learning current approaches conceptualize interpretation in a rather ad-hoc manner, motivated by practical use cases and applications. Some suggest model interpretability as a remedy, but only a few are able to articulate precisely what interpretability means or why it is important. Hence more, most in the research community and industry use this term as synonym of explainability, which is certainly not. They are not overlapping terms. Needless to say, in most cases technical descriptions of interpretable models are diverse and occasionally discordant.
A model is better interpretable than another model if its decisions are easier for a human to comprehend than decisions from the other model (Molnar, 2021). For a model to be interpretable (being interpretable the quality of the model), the information conferred by an interpretation may be useful. Thus, one purpose of interpretations may be to convey useful information of any kind. In Molnar’s words the higher the interpretability of a machine learning model, the easier it is for someone to comprehend why certain decisions or predictions have been made.” I will make an observation here and add “the higher the interpretability of an explainable machine learning model”. Luo et. al. (2021) defines “interpretability as ‘the ability [of a model] to explain or to present [its predictions] in understandable terms to a human.” Notice that in this definition the author includes “understanding” as part of the definition, giving the idea of completeness. Thus, the triadic closure explanation-understanding-interpretation is fulfilled, in which the explainer and interpretant (the agents) belong to different instances and where interpretation allows the extraction and formation of additional knowledge captured by the explainable model.
Now are the models inherently interpretable? Well, it is more a matter of selecting the methods of achieving interpretability: by (a) interpreting existing models via post-hoc techniques, or (b) designing inherently interpretable models, which claim to provide more faithful interpretations than post-hoc interpretation of blackbox models. The difference also lies in the agency –like I said before– , and how in one case interpretation may affect the explanation process, that is model’s inner working or just include natural language explanations of learned representations or models.
Article | February 17, 2020
The continuous advancements in technology and the increasing use of smart devices are leading tremendous growth in data. Considering reports, more than 2.5 Quintilian bytes of data are generated on a daily basis and it is expected that 1.7 Mb of data will be produced every second in the near future. This is where data scientists play an influential role in analyzing these immense amounts of data to convert into meaningful insights. Data science is an overriding method today that will remain the same for the future. This drives the need for skilled talent across industries to meet the challenges of data analytics and assist delivering innovation in products, services and society.
Article | February 17, 2020
For many, 2021 has brought hope that they can cautiously start to prepare for a world after Covid. That includes living with the possibility of future pandemics, and starting to reflect on what has been learned from such a brutal shared experience. One of the areas that has come into its own during Covid has been artificial intelligence (AI), a technology that helped bring the pandemic under control, and allow life to continue through lockdowns and other disruptions.
Plenty has been written about how AI has supported many aspects of life at work and home during Covid, from videoconferencing to online food ordering. But the role of AI in preventing Covid causing even more havoc is not necessarily as widely known. Perhaps even more importantly, little has been said about the role AI is likely to play in preparing for, responding to and even preventing future pandemics.
From what we saw in 2020, AI will help prevent global outbreaks of new diseases in three ways: prediction, diagnosis and treatment.
Predicting pandemics is all about tracking data that could be possible early signs that a new disease is spreading in a disturbing way. The kind of data we’re talking about includes public health information about symptoms presenting to hospitals and doctors around the world. There is already plenty of this captured in healthcare systems globally, and is consolidated into datasets such as the Johns Hopkins reports that many of us are familiar with from news briefings.
Firms like Bluedot and Metabiota are part of a growing number of organisations which use AI to track both publicly available and private data and make relevant predictions about public health threats. Both of these received attention in 2020 by reporting the appearance of Covid before it had been officially acknowledged. Boston Children’s Hospital is an example of a healthcare institution doing something similar with their Healthmap resource.
In addition to conventional healthcare data, AI is uniquely able to make use of informal data sources such as social media, news aggregators and discussion forums. This is because of AI techniques such as natural language processing and sentiment analysis. Firms such as Stratifyd use AI to do this in other business settings such as marketing, but also talk publicly about the use of their platform to predict and prevent pandemics. This is an example of so-called augmented intelligence, where AI is used to guide people to noteworthy data patterns, but stops short of deciding what it means, leaving that to human judgement.
Another important part of preventing a pandemic is keeping track of the transmission of disease through populations and geographies. A significant issue in 2020 was difficulty tracing people who had come into contact with infection. There was some success using mobile phones for this, and AI was critical in generating useful knowledge from mobile phone data.
The emphasis of Covid tracing apps in 2020 was keeping track of how the disease had already spread, but future developments are likely to be about predicting future spread patterns from such data. Prediction is a strength of AI, and the principles used to great effect in weather forecasting are similar to those used to model likely pandemic spread.
To prevent future pandemics, it won’t be enough to predict when a disease is spreading rapidly. To make the most of this knowledge, it’s necessary to diagnose and treat cases. One of the greatest early challenges with Covid was the lack of speedy, reliable tests.
For future pandemics, AI is likely to be used to create such tests more quickly than was the case in 2020. Creating a useful test involves modelling a disease’s response to different testing reagents, finding right balance between speed, convenience and accuracy. AI modelling simulates in a computer how individual cells respond to different stimuli, and could be used to perform virtual testing of many different types of test to accelerate how quickly the most promising ones reach laboratory and field trials.
In 2020 there were also several novel uses of AI to diagnose Covid, but there were few national and global mechanisms to deploy these at scale. One example was the use of AI imaging, diagnosing Covid by analysing chest x-rays for features specific to Covid. This would have been especially valuable in places that didn’t have access to lab testing equipment. Another example was using AI to analyse the sound of coughs to identify unique characteristics of a Covid cough.
AI research to systematically investigate innovative diagnosis techniques such as these should result in better planning for alternatives to laboratory testing. Faster and wider rollout of this kind of diagnosis would help control spread of a future disease during the critical period waiting for other tests to be developed or shared. This would be another contribution of AI to preventing a localised outbreak becoming a pandemic.
Historically, vaccination has proven to be an effective tool for dealing with pandemics, and was the long term solution to Covid for most countries. AI was used to accelerate development of Covid vaccines, helping cut the development time from years or decades to months. In principle, the use of AI was similar to that described above for developing diagnostic tests.
Different drug development teams used AI in different ways, but they all relied on mathematical modelling of how the Covid virus would respond to many forms of treatment at a microscopic level.
Much of the vaccine research and modelling focused on the “spike” proteins that allow Covid to attack human cells and enter the body. These are also found in other viruses, and were already the subject of research before the 2020 pandemic. That research allowed scientists to quickly develop AI models to represent the spikes, and simulate the effects of different possible treatments. This was crucial in trialling thousands of possible treatments in computer models, pinpointing the most likely successes for further investigation.
This kind of mathematical simulation using AI continued during drug development, and moved substantial amounts of work from the laboratory to the computer.
This modelling also allowed the impact of Covid mutations on vaccines to be assessed quickly. It is why scientists were reasonably confident of developing variants of vaccines for new Covid mutations in days and weeks rather than months.
As a result of the global effort to develop Covid vaccines, the body of data and knowledge about virus behaviour has grown substantially. This means it should be possible to understand new pathogens even more rapidly than Covid, potentially in hours or days rather than weeks.
AI has also helped create new ways of approaching vaccine development, for example the use of pre-prepared generic vaccines designed to treat viruses from the same family as Covid. Modifying one of these to the specific features of a new virus is much faster than starting from scratch, and AI may even have already simulated exactly such a variation.
AI has been involved in many parts of the fight against Covid, and we now have a much better idea than in 2020 of how to predict, diagnose and treat pandemics, especially similar viruses to Covid. So we can be cautiously optimistic that vaccine development for any future Covid-like viruses will be possible before it becomes a pandemic. Perhaps a trickier question is how well we will be able to respond if the next pandemic is from a virus that is nothing like Covid.
Was Rahman is an expert in the ethics of artificial intelligence, the CEO of AI Prescience and the author of AI and Machine Learning. See more at www.wasrahman.com