The concept of Artificial Intelligence and its astonishing impact on the global economy and human lives is nothing new. With several businesses predicting that it can boost productivity by as much as 40%, this technology has taken the world by storm since its inception and its immense potential is undeniable and recognized the world over. There are, however, several shortcomings and challenges that inhibit the large-scale adoption of AI.
A study by Deloitte reveals that approximately 94% of enterprises face various issues while trying to implement AI. It is, therefore, necessary to know the nitty-gritty of this technology and the potential hurdles that one might face when trying to adopt it. Let us take a look at some of the most common concerns facing AI today.
Data bias
AI systems, as explained by Forbes India, are only as good or as bad as the data that they are fed and trained on. Bad data is very often laced with racial, gender, ethnic, or communal prejudices which, when fed into an algorithm, can manifest themselves in unfair consequences. Since algorithms are used to make many vital decisions in the modern world, for instance, who is being recruited or whose loan is being sanctioned, ensuring that these systems are being trained on untainted data is key.
One of the worst cases of data bias was when the AI recruiter of Amazon turned out to be gender-biased. The workforce of the technical departments was primarily dominated by men, which resulted in the systems picking up that male applicants were favorable and consequently penalizing the resumes containing the word “women’s” and downgrading graduates of women’s colleges.
Legal concerns
The collection and storage of the data of millions of people is one of the main factors on which machine learning models and AI operate. A minuscule flaw in the algorithm has the potential to cause the leakage of sensitive information and violate data privacy and regulation laws.
The sheer volume of data collected for AI-driven purposes keeps privacy at the very forefront of the concerns related to AI. This gives rise to myriad questions about whether or not data can be sold, who owns the data shared between AI developers and users, and whether or not this data should be de-identified to address privacy concerns.
Lack of manpower
The integration and implementation of AI require expert knowledge of the concept along with all its benefits and deficiencies. However, due to the lack of mass-market use cases, this subject was not well-represented in industry-focused curricula until very recently. Consequently, the number of people with adequate technical know-how is significantly smaller than what is needed to enable businesses to implement their vision of AI-powered progress.
SMEs (Subject Matter Experts) such as data engineers and data scientists, some of the only people who can efficiently work with machines that can think and learn for themselves, are expensive and sparse in the current marketplace. Small and medium-sized organizations looking to deploy AI systems are therefore at a disadvantage when it comes to recruiting the necessary manpower.
Lack of computer power
AI technologies use a tremendous amount of power and high compute speeds, which is another hindrance to its large-scale adoption. Machine Learning and Deep Learning call for a steadily increasing number of GPUs and cores in order to function efficiently. These larger infrastructural requirements come for a price that many businesses, particularly startups, are unable to fund.
Although cloud computing and parallel processing have provided a short-term solution to this problem, exponentially growing data volumes and the automated creation of complex algorithms will soon render them inefficient.
Case-specific learning
While human beings have the gift of the transfer of learning, which is the ability to transfer our knowledge from one context to another similar context, AI does not. The vast majority of AI implementations today are highly specialized which means that they have difficulties carrying over their experiences from one set of circumstances to another similar one.
Specialized AI, often called “applied AI”, is designed to perform a specific task and continually improve its performance at that very task. Performing a different task would involve building another model from scratch. This inability to draw from resources outside of those which are immediately apparent poses a persistent challenge to the general deployment of AI.
Ethical concerns
According to the World Economic Forum, AI automation will take over more than 75 million jobs by 2022. A McKinsey report reveals that AI bots will replace 30% of the current global workforce. Labor displacement is, therefore, the most immediate concern of AI as it continues to become increasingly evident that it will soon take over entire categories of work, especially in retail, transportation, and customer service. Trucking, for example, currently employs millions in the United States. If the self-driving trucks promised by Tesla become a reality in the next decade, millions would lose their jobs.
There is another layer to the ethical concerns regarding AI today. In March 2018, when an autonomous vehicle struck and killed a pedestrian, questions arose about who was to be held liable—the machine or those supposedly in control of them? As these machines continue to be fine-tuned to simulate human emotions and act like human beings, the dilemma of whether they should be governed as humans, animals, or inanimate objects becomes more pronounced.
In conclusion
All that AI technology has achieved thus far was once only possible in works of fiction. The concerns mentioned in this article, though pressing, are certainly not insurmountable. Going forward, implementing solutions that address these issues will allow AI to reach its full potential.
