The Mathematical Road to AI

Experts feel, one major reason why AI talent is in short supply may be the lack of mathematics training in undergraduate IT courses

Artificial Intelligence (AI) is now the big craze as AI-based applications are infiltrating every industry. Their scope is as varied as innovation can think of, and the goal is always to improve business outcomes. No one doubts the phenomenal potential that AI offers. Every day, we wake up to the news of yet another exciting AI-enabled innovation being pressed into service. The time has arrived when these innovations are no longer considered as niche concepts; rather, most of them are reaping rich rewards. However, despite demand, there is a pitiful dearth of trained AI engineers all over the globe. What may be the reason behind it? Industry experts have pointed out to a possibility that sounds pretty straightforward; but can have far-reaching consequences – both to the AI industry and to future curricula in tech institutes.

It is now common knowledge that the requirement for AI engineers is astonishingly huge. Going by ballpark estimates, the required figures are close to several millions the world over, based on the current and projected scope of Artificial Intelligence across industries. And there is need for candidates from every wake of the AI spectrum – from AI theory, to writing AI codes, developing Machine Learning algorithms and implementing them and even building AI-compatible hardware products on which these algorithms would learn. Connect to this the fact that it is not a localised requirement, and every technology aspiring nation need each one of these skills – and the demand is only going to spike in the coming years. However, the actual supply is pitifully low – perhaps a few hundred thousand AI professionals all over the globe!

Machine Learning is at the heart of AI. It involves a lot of parameters, statistics, calculus, linear algebra, and other related domains; in short it involves mathematics of a high order. And experts fear this, precisely, is where the problem lies. But first, let us dive deeper to understand the backdrop.

Usually, software engineers are trained in the basics of general coding in their academic days after which they go on picking up skills in specific programming languages and platforms, as and when required, throughout their professional life. This lifelong learning is based on project requirements, and such upskilling is in sync with their intuitional training as software programmers. This is effective and has been the typical operating procedure in resourcing for IT projects for all these years. IT professionals are always ready for this, because technologies and programming languages keep on evolving – and what is standard today becomes extinct a few years later when an upgrade appears. Any average programmer with formal undergraduate training can develop working knowledge of the required language in a few months.

This is where things turn different for AI skills. It is not something that can be picked up with a crash course in the relevant programming language. It is about mathematical acumen to a great extent, for which there is really no shortcut. Undergraduate computer science curricula focus on system designing, coding and algorithm formulation – but understandably not on mathematics. They don’t need to and students who aspire to become IT professionals are mostly not the ones who would be interested in profound mathematical studies. However, without mathematics there can be no serious Machine Language programming. This gap is gradually dawning upon industry experts, and they mostly agree that the lack of AI talent is directly related to the level of mathematics skills required for it – and there will not be any quick fix for this.

Tech institutions are globally waking up to this lacuna as more and more computer science students want to learn AI and the industry demand for AI professionals skyrocket each passing day. Undergraduates now realize that mathematics is integral to machine learning and they want the requisite training. It is now up to the educators to make the necessary changes in their course curricula. For example, Columbia University has started a data science institute where there is a judicious mix of mathematics, programming and software applications to build AI products. Data science programs with very specific focus on training for Ai and Machine Language are gaining popularity.

Another approach that some experts suggest is splitting programming courses to cater to student interests. As Sameer Maskey – adjunct Assistant Professor at Columbia University and the Founder of FuseMachines, an advanced machine learning company that builds software robots for automated customer servicing – recently explained in a media interview: “we are starting to see… mini-programs where it might not be a two year master’s program but a one year program to sort of do hyper-focused courses on machine learning, deep learning, and computer vision, natural language processing….[T]his kind of a mix of full-on master’s programs and data science… would be good to create more talent, to build more AI applications.”

Let us hope that this trend soon picks up in the Indian academic scenario too – where, till now, we only have a handful of private institutions that offer extensive courses in Data Science.