By Shammy Narayanan
Larger Corporations with coffers brimming in the billions are wickedly smart in downplaying their monopoly status to avoid scrutiny, while smaller competitive firms mindlessly exaggerate their uniqueness to gain market share. Google, for example, is only a search engine, or a consumer electronics company? Or a product development company? Or a technology powerhouse? We will never get a clear answer, as Alphabet will undoubtedly employ its finest sleight of hand to keep those pesky regulators at bay. Whereas mom-and-pop stores nearby will beam with pride in differentiating their burgers from the ones sold in the adjacent store. This masterly game has been played for ages. A commoner, clocking in over two hours a day on social media, is the unsuspecting audience, nodding along in blissful ignorance. “AI Democratization” is one such marketing spiel intentionally spread to avoid the detection of the concentration of power on the greatest technological advancement of our generation.
The Computational Demands:
AI, at its core, is a realm of formidable complexity and voracious resource appetite. Crafting cutting-edge AI models demands not just know-how, but an exorbitant wealth of computational might. Consider this: a modestly scaled Language Learning Model (LLM) requires a training corpus of no less than a staggering trillion words, a feat made possible only through a legion of GPUs dispersed across vast arrays in a data sanctum. This colossal endeavour, let’s be honest, is a privilege afforded only by a scant handful of global behemoths. In this crucible of high-stakes, high-cost ventures, the notion of true democratization wavers, giving way instead to a consolidation of unbridled power
The Data Dilemma:
Data, the lifeblood of AI, is not evenly distributed. The most powerful AI models are trained on vast, diverse datasets. However, these datasets are predominantly controlled by large corporations and organizations with the resources to amass and curate them. This leads to a power imbalance, where those with control over the data hold the keys to advancing AI capabilities. This oligarchy of data ownership stifles the prospects of genuine democratization.
Illustrative Cases:
Consider the case of Google Translate, one of the most widely used machine translation services in the world. Google has access to an enormous amount of multilingual text data from its search engine, Chrome browser, and various other services. This vast and diverse dataset allows Google to train its translation models to convert text from one language to another accurately, whereas smaller startups in this space, irrespective of their technology prowess, will either get acquired or shut down due to the lack of access to quality data.
Yet another example is Waymo, a subsidiary of Alphabet, which has been at the forefront of developing self-driving car technology. The sheer scale and quality of the dataset give companies like Waymo a significant advantage in the development of autonomous driving technology. Smaller or newer players in the field simply don’t have access to the same volume of high-quality data, ultimately leading to their untimely demise.
The Challenge of Open-Source Initiatives:
Even open-source AI initiatives, while commendable, struggle to level the playing field. The development and maintenance of open-source AI projects require considerable effort and resources. It’s not sufficient to make the code available; there must be a robust community of contributors and a support system in place. With the average salary for an ML engineer trending above $170,000 per annum, maintaining a High-quality team is no child’s play; It’s like comparing a Formula- 1 car to a tricycle in a race, and the result is a no-brainer.
Ethical and Legal Imperatives:
The very data that fuels these models often mirrors the prejudices ingrained in our society. Without unwavering vigilance and deliberate efforts to rectify these biases, AI runs the risk of inadvertently perpetuating and, alarmingly, worsening existing disparities rather than ameliorating them. The stark reality is exemplified in glaring instances such as the discriminatory lending practices of 2019, which favoured White applicants over people of colour with similar financial profiles. Likewise, skewed recruitment algorithms, born from training data skewed towards male candidates, further underscore the potential for bias to run rampant. The labyrinth of regulatory hoops and ethical tightropes, coupled with the looming threat of substantial backlash and financial penalties, undoubtedly serve as formidable barriers for smaller enterprises seeking to step into the AI arena.
The Power Dynamics:
In conclusion, while the concept of AI democratization is alluring, the reality is far from it. The barriers of Deep wallets, technical complexity, resource constraints, data ownership, and ethical concerns remain formidable. What needs to be seen is how this concentration of power is going to play out. Some of the dangerous impacts can be an increase in cyber policing, weaponization of AI, willful manipulation of information, autonomous weapons, deep-engineering discrimination/bias to gain power, a fully controlled surveillance society and much more…Are alarm bells ringing? They should be! But fear not, for this is a saga that demands its own chronicle. Stay tuned for the next riveting chapter!
Shammy Narayanan is a Practice Head for Data and Analytics in a Healthcare organization, with 9x cloud certified he is deeply passionate about extracting value from the data and driving actionable insights. He can be reached at shammy45@gmail.com