Philosophy, Math Seen as Key to AI Development in Mexico

Acknowledging Mexico’s reputation as a haven of talented software developers, a self-professed 'data philanthropist' explains how education in the country needs to prioritize data analysis and mathematics to create robust AI.

Despite Mexico having gained a reputation abroad as a land of talented software developers, with a higher-than-global-average number of programmers graduating annually, and which has resulted in the country becoming an attractive outsourcing destination, its education system and corporate culture need to refocus, according to Jesús Ramos, CDO of the Mexican Association for Data Science.

Ramos laments the fact that the country’s educational model is producing programmers who lack an analytical approach, something that must be fostered if that talent is to be put to better use, such as in the development of artificial intelligence (AI).

“Mexico has many great programmers, and this has been noted by US companies, and which have their HQ there but which operate development centers in Chennai, Mumbai or Guadalajara,” he told Nearshore Americas on the sidelines of the Inteligencia México Conference in Mexico City this week, where Ramos was a speaker.

“This phenomenon of the talented programmer, but who is more of a craftsman than an engineer, and who works for a US company, is very similar to the Mexican migrant that crosses the border to work in the US, the ‘wetback’, and we have the opportunity to rescue that term, take out its offensive and derogatory connotations, and convert it into a symbol of Mexican talent in software development,” Ramos, who is also director of the AI for Social Wealth Lab, said.

“Programming is now a transferable, secondary skill, and can be commoditized, because the languages and libraries do a lot of the software engineering work, and which means that a programmer does not have to be very experienced,” he said.

“Programming is more like a craft than engineering, due to the fact that technological infrastructure, languages and libraries are more and more intelligent, and which allows the programmer to focus on solving a problem instead of writing code to interact with the infrastructure, such as connections to databases.

This, coupled with the fact that programming is a task with immediate error feedback, makes it a mechanical job, and which led to it being commoditized,” he said.

Jesús Ramos

“Due to the high demand for programmers all over the world, and which is still not being satisfied, and seeing this as a profitable career path, with so much information available, such as tutorials and content that transmits this knowledge second hand, it has become a career option for people who don’t need to study for four years in a college or university, and they can still satisfy company demands quickly.”

“This is also evidenced by the infinite number of programming bootcamps. However, the mathematics required for machine learning cannot be commoditized, at least yet.”

A good education

Ramos also challenges the assumption that tech education in Mexico is best acquired at the ITESM, the prestigious technical and business school founded in 1947 in Monterrey and which now has campuses country-wide.

Interestingly, Ramos is himself a computer systems engineering graduate of the ITESM. He also holds a master’s in computational finance from the University of Nottingham in the UK.

“The most quantitative universities in Mexico are the Centro de Investigación en Matemáticas (CIMAT), the Instituto Tecnológico Autónomo de México (ITAM), and the Universidad Anáhuac,” Ramos said.

“The ITAM has a long tradition, since the 1970s, that began with economics, econometrics, and then mathematics and statistics, and which were then joined by computer science, which is not the same as programming,” he explains. The university launched the first postgraduate course in data science in Latin America, in 2014, and on whose board of directors Ramos sits.

“The education these institutes offer gives priority to analytical philosophy, in order to reach the right question without fallacies of reasoning, and mathematics, in order to faithfully model reality. These elements are crucial for an ethical machine learning, and which is really rooted in reality.”

“In contrast, bootcamps and other universities, above all the ITESM, unfortunately give priority to technical training, and which operate under a paradigm that is the opposite to the optimum one, with software first and context last, with no math. The ideal paradigm is context first, math second and software last.”

Those schools, he said, effectively train technicians or data engineers, defining the difference between a scientist and a data engineer as somebody who creates machine learning models, and somebody who creates a lot of machine learning models, respectively.

Toward a culture of data

As the founder and director of The Data Pub, a think-tank, consultancy and “community of drinks and data”, Ramos organizes courses and workshops, but which, he says, are different to bootcamps.

“The  courses and workshops we organize attend to two main gaps: the training of directors in data science, analytics, artificial intelligence or machine learning, as these roles are frequently called CAO, CDAO or CDO, and which are relatively new within companies. They group together the responsibilities of maturing a company’s effective use of data, fostering a culture of data, the execution of tech projects for the storage, concentration and distribution of data, and finally, forecasting and machine learning,” Ramos said.

“To be effective in these roles, companies need to make a good hire, build culture and intelligently exercise organizational capital for the transformation of the company into a data-driven culture.”

In addition, the courses and workshops Ramos champions aim to grow math skills focused on programming, such as mathematical and statistical thinking, as well as probabilities and statistics for developers.

“The aim of our courses and mentoring workshops is to close the gaps in mathematical knowledge that have opened up.”

“The big difference is the narrative, or storytelling, that we use in these courses, where we seek to debunk, eliminate hype and give priority to stories of failure, instead of feeding an irrational exuberance regarding the discipline.”

Ramos says that, the Mexican Data Science Society, “acts as guardians of the machine learning discipline, and that means we do not sell these courses as if we were training machine learning experts in four weeks, and which is wrong and unethical, and could even be immoral in the light of the possible social costs of a badly made model on a national scale”.

Gartner estimates that data analysis will become a key tool in business and industry, but that, by 2020, more than 40% of data science will be automated, feeding a need for more specialized data scientists. Earlier this year two other universities in Mexico, the UNAM in Mexico City and the ITESO in Guadalajara, announced they are introducing degrees in data science.

Sign up for our Nearshore Americas newsletter:

In an article published in Mexican newspaper El Economista in August 2018, Ramos bemoaned how technical education in Mexico prioritizes the training of software engineers to maintain costly equipment over the formation of engineers able to solve problems.

“What is lost when advanced mathematics and philosophy are excluded from education?” Ramos asked in the article.

“The fundamentals for creating knowledge. With philosophy, the argumentative ability and the capacity to formulate the correct question are strengthened, while mathematics provides the power to mold parts of reality and the mental structures to effectively manipulate them,” he wrote.

As a result of the prevailing educational model, Mexico lacks the mathematical and philosophical grounding to triumph in AI, he added.

“The ability to create judgmental and ethical knowledge is indispensable for the creation of robust AI. The algorithms and models that create AI are fed by data, and much data is fallible because it is produced by human processes. To achieve better AI practices, mental structures are required that prioritize asking ‘why?’, like a detective, and not ‘how?’, like programmers.”