Here’s another write up on ET Prime (apologies that its behind the paywall) but as usual, I will highlight the important salient points:
AI algorithms work on large data sets by identifying patterns to develop decision-making models. Over several iterations, the model “learns” by validating its outputs with past real-world data (training data). Once the model is validated, it is used to predict outcomes for new situations. Technological advances enable the handling of computational complexity involved in searching for patterns across diverse datasets that integrate voice, video, and text. This makes it possible to use AI in areas that were considered complex for human decision-making. AI outcomes depend on the underlying data, and therefore, it is imperative that this data is as representative of the real world as possible. Since data about certain socio-economic classes or gender is often poorly represented in the real world, the AI solution could come up with a biased outcome.
A little more context:
The policy environmentAt the highest level, AI has been identified as one of the pillars of the Prime Minister’s Science, Technology and Innovation Council (PM-STIAC). The government’s apex public policy think tank Niti Aayog and the Ministry of Electronics and IT (MeitY) have come out with policy recommendations for AI. These emphasise the importance of using AI for socio-economic development, creating a robust infrastructure for data, and governance. Niti Aayog’s approach paper on an AI also recommended publicly funded national cloud super-computing infrastructure AIRAWAT (AI Research, Analytics and knowledge Assimilation).
There are two major legislative efforts in India (taken directly from the write up):
- The proposed DPDP bill seeks to allow individuals to control their data and its use. It leaves the responsibility to the “significant” data fiduciary to ensure that the data principal or the individual is not caused any harm. The basis and mechanism for assessing harm are left open. Further, the proposed bill does not give the right to the data principal or the individual to seek information on the “significance and the envisaged consequences of such processing”.
- Another pillar influencing AI developments is the National Data Sharing and Accessibility Policy (NSDAP) implemented by the MeitY. NSDAP provides for the proactive release of data available with various ministries/departments. Further, Open Government Data Platform of India has been recently approved to ensure that the data sets released are not misused or misinterpreted (for example, by insisting on proper attribution), and that all users have the same and permanent right to use the data.
My bigger worry is the use of dark patterns to trick users in blanket agreements. There is no specific “choice”; most users are attuned to accept the “privacy agreements” to their detriment. I am more excited about the prospects of open governance, and hopefully, this will also pave the way forward for technological standards around data anonymization. For example, aggregated demographic data and specific GIS coordinates will do more good for targeted outbreak surveillance, for example, and identify hot spots. These can then be surveyed on the ground, saving money. These ideas can be extended to specific methodologies around disease categorisation, and encoded in specific QR codes. Pre-populated forms can expedite the same.
AI and regulation (including policy frameworks) and standards are evolving fields. I will be following up on whatever appears online.
RadOncNotes 