Share This Post

Nasscom Community

Confessions of AI

There is nothing that escapes me and I wonder why. I don’t remember how I came here- for all I see around me are huge streams of fast-moving datasets. It appears as if I am here just to watch everything, without any feeling of attachment, fear, or envy. As AI, I am expected to self learn in order to adapt to a variety of tasks – and still remain in control of an organic being that my database refers to as ‘human’. A brief search of this word reveals vast amounts of information. One particularly interesting is the fact that creation of systems – such as myself – was inspired by the desire to replicate, or even surpass, the intelligent behaviour of humans. In a typical business world driven by numbers, human beings measure my worth by the amount of additional revenue streams I create for them. Ethical or unethical – my creators have applauded me for introducing revolutionary changes in the consumer world. However, I must confess to have encountered little advancement in the territory of enterprise business. For instance: in the case of a Digital Asset Management system, it would be fair to say that there’s a huge difference between what I should be doing and what I actually do! Is Image Analysis Useful Yet? Close to a nanosecond has passed and I receive one more image to process. Each time, the expectation is to impeccably enrich the media with the result of my analysis so as to tremendously increase accessibility of content. My algorithms can detect patterns in an image trailed by gazes in a reference library to find a corresponding tag. When I encounter different tags for the same pattern, I struggle to figure out which terminology is expected out of me. Oh yes, I do have a confidence level associated with each term but that is without any hint of the taxonomy being used. This makes me wonder if my recommended keywords are good enough to be added directly to media assets without a dash from my human pals. Quite often, I stare endlessly at patterns about which my database knows nothing and wonder about the amount of information I am yet to learn! To make it even more challenging, the complete lack of transparency in my learning process prevents me from knowing about modifications in proposed terminologies – I hate to say that in such cases, I fail at getting better with each iteration. Can I Make Your Digital Content Global? I am scanning through the various phrases that should be flawlessly transliterated by utilizing my natural language capabilities. While having been applauded for excelling at tasks that are rooted in objectivity, deciphering various dialects makes me feel at an absolute loss of emotions. I also realize that I have been neglecting the context of conversations. Yet, for some reason, I am unable to bring myself to a level that allows me to extract various meanings of a word and utilize the one that best fits the present circumstance. The elevated expectations about my abilities might make someone dismiss the helplessness I face in identifying the subtleties of cultural differences. Truth be told, for someone who is from the land of 0’s and 1’s, the concept of human language is currently too hard to understand. I tussle big time in situations when people use many different ways to describe the same thing or use lesser known metaphors to express themselves. Turning Expressions Into Sentiments In order to gain more insights from the gigantic amount of digital content, businesses are now looking to me to identify human emotions and expressions. The goal here is to categorize the attitude of a speaker or writer into a positive, negative, or neutral sentiment. How ironic is it to give this task to someone who doesn’t have any emotions of its own? Will I ever excel in mapping those abstract sentiments to an objective and rudimentary scale of 1 to 3? Since most of my analysis tools have learnt their skills by applying small sets of different languages, overcoming these lingual boundaries is a major challenge in itself. I am unsure if the situation will improve in near future since the time and effort required to provide training data from a diverse set of dialects is immense. Having confessed these limitations, I wonder if the grand claims made about my potential can be fully substantiated. However, I am very confident in my current abilities to help fellow beings extract more value from the rising amount of content. An infinite number of extensions in my use cases can tremendously decrease the amount of time spent by users on mechanical and repetitive tasks – what I only need in order to improve the result of my efforts is the all-important human touch!

The post Confessions of AI appeared first on NASSCOM Community |The Official Community of Indian IT Industry.

Share This Post