Getting Real About Artificial Intelligence

Whatever abstract visions we do have in mind, the truth is that artificial intelligence and machine learning are no longer the future, they are already part of our present, our everyday lived reality.

For many of us, the phrases “Artificial Intelligence” and “Machine Learning” are likely to conjure up abstract visions of an imagined future. Perhaps a gleaming city: all shiny metal and squeaky clean buildings, a world where virtual assistants are at our beck and call. For others the vision is more dystopian, where each human thought is a monetizable commodity for corporate overlords who are hand in glove with an autocratic regime.

Whatever abstract visions we do have in mind, the truth is that artificial intelligence and machine learning are no longer the future, they are already part of our present, our everyday lived reality.

Yet there is a mystery associated with these terms, reminiscent of the black box symbolism erstwhile associated with “Rocket Science”. But unlike rocket science, which has little practical value in our day to day lives, an understanding of Artificial Intelligence and Machine Learning can help us navigate the world, reflect on modern politics and even become better citizens.

AI refers to the ability of a machine to mimic human-level intelligence in specific tasks, and ML refers to a set of algorithms developed by computer scientists to achieve this ability using large amounts of data. Simply put, ML is a means to the end that is AI.

But then what do we mean by “learning” and “intelligence” in the first place? Suppose a child is in a kitchen and accidentally touches a knife which causes her to yelp ― if the kitchen is a particularly dangerous one with lots of knives lying about, the accident repeats itself a number of times. The child soon spots a pattern and figures out that knives cause pain and should be avoided.

Figuring out that knives are a class of objects and that these objects can cause pain is “learning” while avoiding the knife in the future is “intelligence”. The implicit tool being used by the child to learn is pattern recognition, while the essential raw materials for learning are data (the pain caused by the knife) and a processing unit (the child’s brain).

Most of machine learning is essentially prediction models driven by sophisticated pattern recognition algorithms. The ever-growing data being generated by internet users and the ever-more efficient computer processors have led to massive improvements in the performance of these algorithms in the last decade, which has largely fueled the major advances in AI technology that we see today.

“ML methods are also not a silver bullet solution to all types of problems.”

These advancements in ML techniques in recent years has led to a proliferation of prediction models across a broad range of contexts. Application areas vary from personalized entertainment to medical diagnosis, from dating recommendations to legal decisions, from weather predictions to regulating agricultural markets ― the list goes on.

The data for each of these applications may be very different, for example: the data for a music recommendation app may come from how many times you play a particular song, while the data for a medical diagnosis may come from an x-ray. The common thread running through the applications is the ability to make accurate predictions, powered by large datasets and algorithms.

That said, the label is getting applied pretty freely as it’s a popular buzzword. Applied statisticians, econometricians and computer scientists have long been in the business of predicting trends with often very successful results. There is a tendency in the data community to brand related methods from these domains as machine learning, possibly to capitalize on the hype.

ML methods are also not a silver bullet solution to all types of problems. While modern ML and AI systems are becoming increasingly good at predictions (or answers to questions beginning with “what” or “who”), they are not yet proficient in answering questions around the “how” and “why” of things. In fact, theoretically grounded models of the natural and social world still dominate when the objective is to understand mechanisms or find explanations.

Leaving aside the hyperbole and white noise though, think of some of the application areas where ML is being successfully employed. Entertainment, dating, healthcare, ecology ― these are all inherently human concerns with societal repercussions being driven by algorithms.

Our lives are becoming increasingly and inevitably intertwined with these algorithms. Not surprisingly then, social scientists, anthropologists and ethicists are starting to critically analyze how our world is changing as a result of this entwinement.

The emerging field of “AI and Ethics” is centered around prickly questions like polarization, misinformation, surveillance and inequity resulting from algorithmic bias. Books like “Weapons of Math Destruction” and groups like the “Algorithmic Justice League” which shed light on the negative impacts of unregulated AI are gaining ground in the popular discourse in USA. Stanford also recently established a “Human Centered AI Institute”. Incidentally the fledgling institute is already facing criticism for its predominantly white, male and technocrat heavy advisory council!

How this cocktail of algorithms and society will play out in the Indian context should be of essential concern for policy makers. Keep in mind that the cutting edge research in AI and ML is predominantly done in the US and a handful of other developed countries, whereas the impacts of these technologies is global. Some institutes and think-tanks in India are starting to focus on questions of the opportunities and risks of AI and ML in India.

For instance “Wadhwani AI”, “Center for Internet and Society” and IIT Delhi’s new School of Public Policy are amongst the few that are venturing into building awareness and research expertise in social science aspects of AI and ML. The Central Government’s planning agency, NITI Aayog has also dipped its toes into the AI and policy waters ― in 2018 they published a discussion paper on a “National Strategy for Artificial Intelligence”.

All these efforts are still quite nascent though, and there is an urgent need for a public dialogue on how lives of ordinary citizens are going to inevitably change, for both good and bad, as a result of the AI revolution.


What's Hot