There has been an increase in warnings against Artificial Intelligence (AI) by technology leaders, researchers and experts, including Geoffrey Hinton (artificial intelligence pioneer), Elon Musk (Tesla co-founder and CEO), Imad Mostak (British AI expert). Cathy O’Neill (American mathematician, data scientist and author), Stuart J. Russell (British computer scientist), and many others. Recently, the World Health Organization called for caution in using AI for public health care. But there have been very few voices in India to echo NITI Aayog’s initiative for responsible AI; Some journal papers and articles warn of impending danger. Instead of waiting for more industry leaders in India to add their names to this list, we need to realize that the time to pay attention and prepare is now.
Discourse, discussion required
The prospect of a dystopian world where machines outnumber humans is a more remote potential threat, the prospect of which requires more discourse and deliberation. The more immediate danger that algorithms and artificial intelligence pose is that of undermining bias, emphasizing representation, and trivializing diversity. With the backdrop of Indian states, their multiple languages, colours, cultures and traditions, this threatens to spread greater inequality and lead to further exclusion for marginalized groups and minorities, who are at a disadvantage in terms of livelihoods, opportunities, well-being. May translate into higher cost. , and life. A recent report by the United Nations Development Program and Apti Institute highlights that ‘algorithmic bias is [the] The biggest impact on the financial services, healthcare, retail and gig worker industries. The report also pointed out that AI-generated credit scoring showed a tendency to assign lower scores to women than men, even when they had similar financial backgrounds. Research on the application in health care diagnostics has identified significant biases against people of color because most of the data used to train the model was from the United States and China, not representing all communities even within these countries. .
With this backdrop, it is imperative that we stop and think before jumping on the AI roller coaster. Both commercial and government players in India are already evolving from Big Data to AI for better targeting, efficiency, profit as well as ‘perceived and (mostly) intended’ social welfare. ChatGPT’s sensational popularity is on the rise as social impact organizations and government agencies use it to optimize their public information systems and communication mechanisms. The Indian government continues to increase its budget allocation for emerging technologies, including AI, every year, and has established teams and an agenda for the deployment and use of AI. While the most populous country with the second largest internet user base is leaning towards this innovation and modernization, there needs to be more discussion about the data being used to train and build these algorithms. It is necessary to redirect our attention to these systems and the underlying foundations of their operation, governance and ethical considerations before they affect our daily lives.
machine learning world
AI relies on Big Data or programmatic rules to learn from and simulate human intelligence, communication and (potentially) actions. Machines learn from massive amounts of data created and provided by humans and are programmed to recognize patterns and learn to repeat them to make decisions. This translates into the replication of human languages, communication styles, competencies, abilities, and reasoning, as well as stereotypes, prejudices, belief systems, and preferences. While these algorithms are designed to improve through feedback loops and programmatic improvements, they lack an ethical compass; And unlike humans, they do not question conventions, norms, culture or traditions. Machines do not have the sense of fairness or empathy upon which a society leans, especially with minorities and disabled communities. Furthermore, machines generalize learned patterns to the population without knowing that the data used to train them is diverse and complete, with adequate representation of all communities and groups affected by their application.
For example, algorithms could potentially sift through the resumes, career trajectories, and performance of past employees in an organization and learn that men are in general more productive than women, much less long vacations. and are more suited to leadership roles. Why? Because, historically, there have been disproportionately high numbers of men in the leadership of organizations, not offering paternity leave to women to take a break from their careers to raise children, and often more so than their female colleagues. Performance is highly ranked. male supervisor. The problem can have deep layers of race, caste, geography, educational background etc. Whatever biases we may have had in the past can be learned as a rule and insight, and applied as a generalization.
Other examples of this include potential drug abuse in a specific geographic area, potential criminal activity for a specific community, an (illusory) intellectual supremacy for a fraternity, or an assumption for a profession based on past data, patterns, and historical records. Stereotypic gender characteristics may be involved. , In a way, algorithms learn from the past and project into the future.
What is the need of AI deployment in India
The poor state of administrative statistics across all states in India adds to the problem. There is currently no good quality data available to train the algorithms with the Indian context and background. Marginalized groups and communities are even worse represented in privately collected and held data.
This challenge is exacerbated by low awareness and understanding of algorithms and their functioning among government personnel and the general population affected by these algorithmic applications. Algorithms and systems usually appear as a black box to their users, with little or no transparency into their input variables or the data used for training. This can translate into over-reliance and exaggerated trust in algorithmic suggestions and outputs, furthering issues of discrimination and exclusion of minorities and disadvantaged groups.
AI deployment is not a leap we can afford without its development in India based on our data and context. We need clean, organized, digitized and well-governed public data to build algorithms and models that benefit our people. Both industry and governments should exercise caution before adopting this innovation and invest substantially in their research, development and testing. While the challenge of available data volumes can be addressed due to the scale of our administration and services operations, it is imperative that we prioritize the development of AI in a responsible and informed manner.
The private sector and practitioners must work hand in hand with governments in this journey. It is important to emphasize that the goal should be to build intelligent machines that are inclusive and reflect the diversity and heterogeneity of the country. Embracing this innovation should not hinder our progress towards equity and equality for all; It should support our efforts for reforms and positive change.
Tulika Avani Sinha is a public policy professional who works with state governments and multilateral institutions in India on the design, adoption and implementation of data policies and systems. He is a former Mason Fellow at the Harvard Kennedy School