Gaming the AI Imaginary

With the increased and imposed usage of AI in all digital platforms in the last couple of years, it is hard to imagine a conversation amongst ourselves that does not involve AI. Tech startups and tech companies in India have had to adapt and shape their work towards a vocabulary that requires the presence of AI in their pitch decks, conversations, budgets, and their day-to-day operations. While much needed critical interventions on AI and its impact—on climate change, the race to extract most amount of data to train AI models with no regard to environmental impact, data privacy and dignity, displacement and eviction of communities to enable data center infrastructure to host the GPUs that run AI models, and the precarity of employment amidst fears of AI displacing labourers—all continue, tech and IT companies try to steer the conversation away from these issues and on to the improved productivity and efficiency, and in turn, profitability of using AI in their operations. Many recent massive layoffs from these companies have been undertaken while citing “impact of AI”, as if that was the only option.

Through secondary research, personal experiences and observations, this essay seeks to turn the magnifying lens towards the inner workings of these companies and question the very idea (or the myth) of productivity and efficiency in the age of AI. Are these companies and the people within really productive to begin with? What are the modalities of using AI and is it really enabling the aim of exponentially increasing one’s productivity, and in turn the profitability of the company? How does labor get shaped by AI and who is really affected?

Gaming the AI Imaginary

Akash, an artist who modeled 3D art assets of decorative rooms for a mobile game in a gaming startup, had spent hours on Youtube to understand and put into effect the latest Artificial Intelligence (AI) workflows to make the process of creating such assets more efficient. He was building a workflow in which one could generate decorative rooms for the game in 3-5 days instead of the 2 weeks of manual work it would take. In a meeting with an expert-AI consultant, the in-house AI engineering lead, and the CEO of the startup, he showcased his workflow and was appreciated for his efforts in coming up with solutions to speed up art-asset production using AI, despite having no training in hands-on code development.

This was at a time when the startup I was working at was actively trying to integrate AI in all of their workflows. The company had recently hired an expert AI-consultant who had worked on the intersections of art and AI, and an engineering lead whose expertise lied in understanding AI-models. Whenever the CEO visited the office (since he worked out of his home in another city for most of the year), it was an assured spectacle, he would walk around the office animatedly showcasing how he was able to create an AI-generated marketing video within minutes, even have his talking-head in it, and remark that we were living in times where, soon, all we would need to do is write a few prompts and an entire game would be created.

The buzzword of AI shaped all our conversations everyday. There was a future that the CEO envisioned in which AI would allow us to automate redundant/manual tasks, automatically find and fix issues in the game, analyse vast amounts of player data, predict player behaviour, and in essence chart a path towards exponentially improving productivity and, in-turn, profitability. What was invisible to Akash was the fact that he and 15-20 of his colleagues (artists, testers, game developers, game designers) would soon be laid off. The startup was pivoting to AI-based workflows primarily because of a shift in the landscape of VCs (Venture Capitalists) and investors who started to focus on startups that centered AI in some way in its work.

The VCs who had invested in this startup had already started nudging the CEO–who had failed to show any profitability or a path to profitability for the last 2-3 years–to focus on building an AI-based tech platform for gaming companies. Instead of the current business of launching mobile games and trying to make it profitable, a suite of AI services would be built and offered to other gaming companies. This suite included automated responses to player reviews, generation of marketing assets and videos, analysing market trends, creation of 2D and 3D assets for mobile games etc.

In a management study conducted in 2007, they found that entrepreneurs are more likely to acquire resources for their ventures if they perform “symbolic actions” which convey the entrepreneurs’ “personal credibility, professional organizing, organizational achievement, and the quality of stakeholder relationships”. While this is a part and parcel of any entrepreneur and investor relationship, the reality of the startups I have worked for has been that the actions have been majorly symbolic and rarely substantive - the storyline of the startup takes center stage over actual actions within the company. In fact most of the upper-rung of management I worked with, had barely any tech, gaming, or AI experience. They raised investment money mainly because of their “people’s skills” which included public-speaking and smooth communication skills, connections, and the know-how to narrativize the path to profitability and building an AI-first company that integrates AI in its game-development processes, marketing, strategy and operations. While on the company website, one could see that they have a panel of expert-consultants (game-design, product management, tech, AI) from all over the world with substantial experience; they barely spent 1-2 hours in a week with the team. The AI-expert consultant would spend very limited time, attending few meetings every week, providing high-level suggestions, never really getting his hands “dirty” to shape the ground-level reality of game development processes. While the view that was fictionalised and marketed to investors by the company leads was that the company is rapidly moving towards AI-driven workflows, internally, the brunt of the work of research, experimentation, and building the AI pipeline was borne by artists like Akash.

Separating Conception from Execution

Harry Braverman, an American political economist, who elaborates on the rise of Frederick Taylor’s scientific management in the early 20th Century, argued that there has been a systematic monopolisation of the knowledge of the labor processes to ensure capital’s control over labor. There was a separation of conception from execution, wherein management assigned to itself the work of science and deprived the worker of any planning capacity, essentially leaving the worker in “ignorance, incapacity, and thus a fitness for machine certitude”. With the introduction of machines in chemical processing, metal cutting, automobile manufacturing etc, he explains how, while there were practical advantages to a machinized process, labour was divided between workers so that there was a separation of “intellectual work from the work of execution” and that workers would not possess the know-how of the entire process. Instead, labor and the worker’s knowledge was degraded to “manage” worker’s time, productivity, and efficiency. Building on this argument, Matteo Pasquinelli in his book The Eye of the Master: A social history of AI, explains how AI can be understood as the automation of general intellect and “what is now rebranded as machine intelligence rests on long histories of knowledge extractivism facilitated by empire and capital”. AI, in essence, is another instrument through which capital can be accumulated and surplus value can be generated through the extraction of knowledge, skills, and data of the workers.

In the Indian context, caste as a social structure intersects with this extraction. As I argue in my previous article, there is a “division of labourers” within the gaming industry where, in my experience, most of the testers (Quality Assurance) and some of the artists/game designers have happened to be from SC, ST, OBC, or Muslim backgrounds, while most of the CEOs, COOs, managers, product and project managers, game developers have mostly been from twice-borne (mainly Brahmin, Baniya) castes. This results in accumulation of capital, decision making, skills, reinforcing a hierarchical relationship, whereas those in non-decision making positions rarely get to move up the corporate ladder or change their designations to another role.

Almost no product manager or startup founder I have reported to in my career has made the effort to understand day to day realities and work that developers, artists, designers do to build a game. Many don’t even have time to play the games that they are building. While they usually build their expertise in market research, data analysis, and figuring out the best way to chart a path to a profitable product, they are rarely ever concerned about the working conditions of those who actually build the game. The role of a product manager or a startup founder has always been centered around “conception” of an idea and knowing the best way to market that to the public, or their investors or higher-ups, while the people who report to them execute and build different aspects of a game. Many of them end up in such roles primarily because of their credentials of having worked at well-known companies, scaled revenue-generation of a game with the right measures, studied at well-known business-schools, having the right connections. It is rarely ever from having had “hands-on” experience with a game like game-testers, designers, or developers do. Hence, when it comes to building AI workflows, it is inevitable for them to rely on the appropriation of the knowledge of the workers and finding the best way to convert every step that the worker does in building a game into a document or a data-point that can be processed by the AI models.

Developing the AI Imaginary

I now work at a highly profitable company which has the resources to provide access to the latest AI models, subscriptions, and AI related courses to its employees. For instance, the monthly cap for an entire company’s usage for one of the AI models is upwards of 1 million dollars and this cap has been utilised to its fullest in some of the recent months because of extensive AI research, building AI features in the games, and general day-to-day usage. The company ensures it is seen in the public eye as an AI-first company and has, in fact, established a separate tech team to build AI workflows that can be used by everyone to improve productivity and efficiency. The social composition of the teams around me is no different than the previous workplaces I have been at - core-teams of product managers and game designers who mainly come from twice-borne castes (brahmins/baniyas) provide direction on different games. The primary difference between the startup and this established company is that most of the work of building the game is actually outsourced to different game-development companies, perhaps those who don’t have access to these latest AI models and subscriptions. Usually, and especially if it is an Indian company to whom work is outsourced, while the company may charge higher rates, the outsourced workers themselves live precarious lives, often paid quite low wages, work longer hours, and are easily disposable or replaceable.

With unfettered access to AI subscriptions, what I see happening is that almost all colleagues around me use the most energy-intensive latest AI model for their everyday tasks, from market research and writing game-design documents, to generating good-looking weekly reports and decks through different sources of data (player data, chat/email conversations, marketing results), and analysing vast amounts of player data and behaviour, with obvious disregard to the impact of AI on the environment, land, and communities. While on the surface it might look like AI is improving productivity, what is hidden is that the work of understanding and cleaning up the AI-Generated documents/decks to make it usable, fixing issues in the game-code or the manual work of polishing the art-assets created by AI to make it launch-ready, is passed on to the lower rung of outsourced workers, who have the additional burden of doing more work in lesser time through enforced AI usage. While one can argue that this can be fixed with better and more productive AI-workflows, the gaze is to be turned towards the product managers, project managers, and CEOs/COOs, questioning their role in perpetuating an AI imaginary within these companies, built on the bedrock of existing inequalities of caste, class, and gender.

Science and Technology Studies scholars, Sheila Jasanoff and Sang-Hyun Kim, define socio-technical imaginaries as “collectively held, institutionally stabilized, and publicly performed visions of desirable futures, animated by shared understandings of forms of social life and social order attainable through, and supportive of, advances in science and technology”. In the companies I have worked at, the decision-makers endlessly desire and actively push for a future that involves replacing (fully/partially) each role that is involved in the development of a game with an AI-workflow. These roles include research, data analysis, game design, coding/development, testing, art creation, and understanding player behaviour. They expect AI to take in vast amounts of documents, data, email/chat conversations, understand all of the context, and provide directions to an ideal path that can make the game successful and in turn the company profitable. Everything that a worker does becomes a data point to train AI workflows. This imaginary is oriented towards finding the shortest path to profitability and supplements the imaginary perpetuated by the company of being AI-first to its investors/shareholders, justifying the costs of the latest AI model subscriptions. It does not matter whether this imaginary is achievable. To reach this imaginary, what is kept aside are the questions on the environmental toll of data centers, extraction and appropriation of knowledge and skills of workers, and the resulting precarity in employment.

The startup I worked at shut down since the idea of building an AI platform suite didn’t pan out. While Akash and many of his colleagues took 5-6 months to find another job, the CEO took a break and is now trying to raise investment for other AI and tech-startups and trying out ideas for other business ventures, the AI expert-consultant continues to market his own AI startup, and the product managers, whose resumes have been amped up with added AI focused terminologies, easily found jobs in other companies–all striving to build an AI Imaginary, for themselves and for Capital.