Limitations of current machine learning approaches

Elisabetta Bottaro, PhD

Machine learning is a fascinating field, which has changed our way to live. Everyone will agree that we are at the very beginning of what machine learning can accomplish. Using this technology, scientists can predict the weather, diagnose disease, drive cars and even predict your behaviour. It is unquestionable that machine learning (ML) and artificial intelligence (AI) are revolutionizing our industry and will only become more prevalent in the coming years. “Machine intelligence is the last invention that humanity will ever need to make” - a famous quote from Nick Bostrom, a philosopher at the University of Oxford. 

Back in 1959, Arthur Samuel created a revolutionary program that could play checkers. This was the birth of ML[1]. Afterwards, several new algorithms were developed and ML has been one of the fastest-growing areas in AI. What is the definition of ML? Machine learning is a form of AI that enables a system to learn from data, identify patterns and make decisions with minimal human intervention[2].

Nowadays, ML has the ability to analyse a massive quantity of data, predict new patterns and help to make crucial decisions in the businesses. Globally the number of papers published in 2019 with the keywords “Machine Learning” is 134000 as reported from Google Scholar. On today’s date, CrunchBase reported 325 machine learning companies in London[3] and 5458 globally[4]. These data are suggesting that more industries are moving to embrace this technology. However, there are still several businesses that are resistant to change towards this technology and few challenges are still present. The major limitations of ML approaches can be summarised in the following points:

The importance of Data:

Machine learning requires a large set of data. The data need to be unbiased and of good quality, otherwise, the final output might be inexact. The most ideal way to reduce such risks is by collecting heterogeneous data from multiple sources.

The number of resources:

Resources include time, number of employees and machines. ML needs enough training time to produce an accurate outcome and massive resources to function. This factor might contribute to the decision of small companies, which don’t have great computing power, to move towards ML technology.

Interpretation of the results and the lack of transparency: 

When you are deploying a new ML program you have to be comfortable with the fact that your system might be perceiving things that humans are unable to discern. AI is not the same as human intelligence. Two are the major difficulties you will need to face: interpretation and explication of the results. The first involves your ability to understand what the algorithm is telling you, whereas the second involves the ability to explain the internal mechanics of a machine in human terms. Therefore, in an ordinary business meeting sometimes it might be difficult to describe results to the commercial team.   

Ethical dilemma: 

Are we moving towards a new world where machines are replacing humans? Although is true that ML has having a big impact on our lives, the machines are not taking over their creators as is depicted in several science fiction movies. What is happening is an occupational change. This phenomenon involves the generation, disappearance, and transformation of jobs in the market. However, beyond the concerns of the social and economic impact on employability, data scientists have real worries about bias, wrong predictions and ethical implementations of the technology. 

“Some people call this artificial intelligence, but the reality is this technology will enhance us. So instead of artificial intelligence, I think we'll augment our intelligence.” —Ginni Rometty.

29-Apr-2020

References:

1.             Samuel, A. L. Some Studies in Machine Learning Using the Game of Checkers. IBM J. Res. Dev. 3, 210–229 (1959).

2.            Kubat, M. An Introduction to Machine Learning. (Springer Publishing Company, Incorporated, 2015).

3.           https://www.crunchbase.com/hub/london-machine-learning-companies

4.           https://www.crunchbase.com/hub/machine-learning-companies#section-leaderboard