“AutoML is enabling persons with a non-coding background to find their feet in the domain. It automates the so-called mundane aspects of the model-building process and allows for us to realize no-code machine learning. Automation in machine learning has also ensured we depend less on the mean only-we-can-do-machine-learning engineers”, Sridhar remarked, with pronounced scorn.
The class knew of his peculiar sarcastic habit and so did the professor. However, the underlying meaning was lost upon Sushmita, who took the sentence on its face, without interpreting it. She got enraged. “Engineers won’t be replaced! Your advocacy is just an excuse to paper over your inherent laziness to make sense of the learning algorithms!”, she retorted vehemently.
To come to terms with the above lines, let us delve into the lore’s setting.
Top-of-the-league multidisciplinary institutions are a rather uncommon sight in India. While it isn’t exactly intuitive to visualize individuals pursuing dissimilar career trajectories sitting together in the same class, colleges like Ashoka University have come up with impactful programs like the YIF to address this point.
From Shakespearean literature to polity, multifarious courses are taught. The above excerpt is from one such class on data science, where Professor Yousouf was instructing his class on Machine Learning. This was the second lecture in the series. He fervently believed in argument being a vehicle to achieve critical thinking while learning.
Sushmita was an engineering grad, who had been an ML practitioner before opting to study here at Ashoka University. She had a great technical acumen and was keen to explore more about liberal arts, in order to have a wholesome and well-rounded opinion of the society around her.
Sridhar was an economics postgraduate and a data science buff. His background and industry exposure had enabled him to build an intricate understanding of his domain. He freely indulged in sarcasm and that had rendered him (un)popular.
The two of them were fast friends and liked indulging in debate. These experiences, while being seemingly bittersweet to the passing glance, were enriching learning curves for them. The two were also among the most enthusiastic participants in Prof. Yousouf’s class, owing to their similar liking for data science.
At the same time, it is fallacious to believe that data analysts spend most of their time on visualization and deriving final insights in the quintessential corporate setup, for instance. Data cleansing is arguably the most essential part of the process, while the other stages follow later on. This one exercise drives the decision-making process largely and it is bound to get monotonous if the problem isn’t understood appropriately.
The lecture itself
The class on data science was about to start. Sridhar was seated in close proximity to Sushmita. Their professor entered the class. He had stipulated that they would continue deliberating on machine learning. He went on the podium and switched the microphone on.
“Feature selection and fine-tuning the model is an essential part of the model-building process”, he resumed where he had left last day, in his own idiosyncratic way. “Writing code is no small feat. But, it isn’t everything. In my opinion, it is the proverbial tip of the iceberg called the machine learning process.’
This is when he stopped abruptly and seemed to recall something important. He asserted, “In fact, we are hearing of a new development very often these days. Automated Machine learning, shortened to AutoML, is rising in popularity. While I’d love to explain what it is, I am seeing Sridhar bubbling with enthusiasm. So, I’d ask him to go first!”
While Sridhar was an ardent proponent of automation in machine learning, he also liked to mock Sushmita for her apparent snobbiness, owing to her background. The two had been locked in a tug-of-war over it, practically since their first interaction on the topic.
He took this opportunity to charge on the point. This is where the heated exchange stated above followed.
“Sushmita, what makes you believe that codeless, automated machine learning models are a deal breaker?” Prof. Yousouf questioned calmly, as if it were a reminder to Sushmita to be confident but not loud, even if one registers their disagreement.
“Sir, I do not think it is a deal breaker. Undoubtedly, machine learning is for everyone. However, it has accorded a comfort zone to novices and the notion that it shall be able to do away with the requirement of engineers is grossly misplaced in my opinion”, replied her relatively more composed self.
A debate was imminent. Although he considered it to be a healthy way to aid students’ quest for knowledge, the professor did not feel it would do justice to the subject. “Let me prevent you from making another comment, Sridhar”, he declared, as he stopped him in his tracks.
“Predictive analytics has made it big in the industrial practice of machine learning. This has also enabled firms to observe that certain tasks of the exercise are quite repetitive and can be automated. A textbook example would be regression, which has the same underlying methodology, notwithstanding a few modifications that can be efficiently realized using automation.”
“Early-stage startups are strapped for cash and do not have the luxury of hiring and compensating machine learning practitioners big from the word go. Now, it is also not very uncommon to find professionals who have had a start in knowing about the theory behind the practice. However, they lack the nuances encountered while implementing the same.”
The students were all ears. They were oblivious to the cold of December, which had failed all meteorological predictions that year. The sombre sky, coupled with the Westerlies, had rendered the month among the coldest in the last one century with uncharacteristically chilly conditions. Yet, he continued:
“Thus, automation in machine learning has revolutionized the prospective workforce for the world of ML. The barrier to their entry has been significantly lowered, which enables building a machine learning model without having deep expertise in machine learning. It has been shown that these AI-crafted models usually work better than handwritten code for relatively menial tasks.”
“No-code machine learning piques their interest further and therefore, plays a role in moulding new practitioners of the subject. Candidates with a non-technical background have traditionally been hesitant to get into the practice. This is where I must reprimand Sushmita’s views as the trend has led to an increase in the inclusivity of machine learning, opening it for everyone”, he quipped, eliciting giggles from the class and a change of mindset for an introspective Sushmita.
“Having said that, one might be enticed to believe that these are capable of eliminating the need of engineers completely, as Sushmita had observed during her rebuttal. However, we must understand that tools like Cloud AutoML, Azure, H2O AutoML etc. cannot completely substitute human expertise yet.”
“We must understand that building a machine learning model has never been the most difficult part of the data professional’s work. Identifying the right model, which caters to the unique challenges posed by the problem at hand, is much more pivotal”, he observed, in modest vivacity.
“This is where a human practitioner’s skill continues to be indispensable. AutoML models thus are quite rigid and cannot accommodate the multitude of scenarios faced, despite having state-of-the-art metadata embeddings.”
“This is exactly where Sridhar’s pain point lies. It isn’t a mere coincidence that engineers have generally been the dominant force in the ML domain; it’s their expertise that has been the enabler and rightfully so.”
“In conclusion, we must keep ourselves abreast with the latest trends. I sternly believe that keeping yourself open to positive changes is the best you can do for society. I recall one of my student’s words, ‘regression is an antithesis of regressiveness’.”
This made the class laugh out loudly and set the lesson as well as the rest of the day up nicely for everyone, including Sushmita and Sridhar, who were much more content now.