Centre for Internet and Society

CIS Seminar Series

Cheshta Arora — 2022-04-11T15:19:11Z

The CIS seminar series will be a venue for researchers to share works-in-progress, exchange ideas, identify avenues for collaboration, and curate research. We also seek to mitigate the impact of Covid-19 on research exchange, and foster collaborations among researchers and academics from diverse geographies. Every quarter we will be hosting a remote seminar with presentations, discussions and debate on a thematic area.

The first seminar series was held on 7th and 8th October on the theme of ‘Information Disorder: Mis-, Dis- and Malinformation’,

Theme for the Second Seminar (to be held online)

Moderating Data, Moderating Lives: Debating visions of (automated) content moderation in the contemporary

Artificial Intelligence (AI) and Machine Learning (ML) based approaches have become increasingly popular as “solutions” to curb the extent of mis-, dis- mal-information, hate speech, online violence and harassment on social media. The pandemic and the ensuing work from home policy forced many platforms to shift to automated moderation which further highlighted the inefficacy of existing models (Gillespie, 2020) to deal with the surge in misinformation and harassment. These efforts, however, raise a range of interrelated concerns such as freedom and regulation of speech on the privately public sphere of social media platforms; algorithmic governance, censorship and surveillance; the relation between virality, hate, algorithmic design and profits; and social, political and cultural implications of ordering social relations through computational logics of AI/ML.

On one hand, large-scale content moderation approaches (that include automated AI/ML-based approaches) have been deemed “necessary” given the enormity of data generated (Gillespie, 2020), on the other hand, they have been regarded as “technological fixtures” offered by the Silicon Valley (Morozov, 2013), or “tyrannical” as they erode existing democratic measures (Harari, 2018). Alternatively, decolonial, feminist and postcolonial approaches insist on designing AI/ML models that centre voices of those excluded to sustain and further civic spaces on social media (Siapera, 2022).

From the global south perspective, issues around content moderation foreground the hierarchies inbuilt in the existing knowledge infrastructures. First, platforms remain unwilling to moderate content in under-resourced languages of the global south citing technological difficulties. Second, given the scale and reach of social media platforms and inefficient moderation models, the work is outsourced to workers in the global south who are meant to do the dirty work of scavenging content off these platforms for the global north. Such concerns allow us to interrogate the techno-solutionist approaches as well as their critiques situated in the global north. These realities demand that we articulate a different relationship with AI/ML while also being critical of AI/ML as an instrument of social empowerment for those at the “bottom of the pyramid” (Arora, 2016).

The seminar invites scholars interested in articulating nuanced responses to content moderation that take into account the harms perpetrated by algorithmic governance of social relations and irresponsible intermediaries while being cognizant of the harmful effects of mis-, dis- mal-information, hate speech, online violence and harassment on social media.

We invite abstract submissions that respond to these complexities vis-a-vis content moderation models or propose provocations regarding automated moderation models and their in/efficacy in furthering egalitarian relationships on social media, especially in the global south.

Submissions can reflect on the following themes using legal, policy, social, cultural and political approaches. Also, the list is not exhaustive and abstracts addressing other ancillary concerns are most welcome:

Metaphors of (content) moderation: mediating utopia, dystopia, scepticism surrounding AI/ML approaches to moderation.
From toxic to healthy, from purity to impurity: Interrogating gendered, racist, colonial tropes used to legitimize content moderation
Negotiating the link between content moderation, censorship and surveillance in the global south
Whose values decide what is and is not harmful?
Challenges of building moderation models for under resourced languages.
Content moderation, algorithmic governance and social relations.
Communicating algorithmic governance on social media to the not so “tech-savvy” among us.
Speculative horizons of content moderation and the future of social relations on the internet.
Scavenging abuse on social media: Immaterial/invisible labour for making for-profit platforms safer to use.
Do different platforms moderate differently? Interrogating content moderation on diverse social media platforms, and multimedia content.
What should and should not be automated? Understanding prevalence of irony, sarcasm, humour, explicit language as counterspeech.
Maybe we should not automate: Alternative, bottom-up approaches to content moderation

Seminar Format

We are happy to welcome abstracts for one of two tracks:

Working paper presentation

A working paper presentation would ideally involve a working draft that is presented for about 15 minutes followed by feedback from workshop participants. Abstracts for this track should be 600-800 words in length with clear research questions, methodology, and questions for discussion at the seminar. Ideally, for this track, authors should be able to submit a draft paper two weeks before the conference for circulation to participants.

Coffee-shop conversations

In contrast to the formal paper presentation format, the point of the coffee-shop conversations is to enable an informal space for presentation and discussion of ideas. Simply put, it is an opportunity for researchers to “think out loud” and get feedback on future research agendas. Provocations for this should be 100-150 words containing a short description of the idea you want to discuss.

We will try to accommodate as many abstracts as possible given time constraints. We welcome submissions from students and early career researchers, especially those from under-represented communities.

All discussions will be private and conducted under the Chatham House Rule. Drafts will only be circulated among registered participants.

Please send your abstracts to workshops@cis-india.org.

Timeline

Abstract Submission Deadline: 18th April
Results of Abstract review: 25th April
Full submissions (of draft papers): 25th May
Seminar date: Tentative 31st May

References

Arora, P. (2016). Bottom of the Data Pyramid: Big Data and the Global South. International Journal of Communication, 10 (0), 19.

Gillespie, T. (2020). Content moderation, AI, and the question of scale. Big Data & Society, 7 (2), 2053951720943234. https://doi.org/10.1177/2053951720943234

Harari, Y. N. (2018, August 30). Why Technology Favors Tyranny. The Atlantic. https://www.theatlantic.com/magazine/archive/2018/10/yuval-noah-harari-technology-tyranny/568330/

Morozov, E. (2013). To save everything, click here: The folly of technological solutionism (First edition). PublicAffairs.

Siapera, E. (2022). AI Content Moderation, Racism and (de)Coloniality. International Journal of Bullying Prevention, 4 (1), 55–65. https://doi.org/10.1007/s42380-021-00105-7

For more details visit https://cis-india.org/internet-governance/blog/cis-seminar-series

Big Data in India: Benefits, Harms, and Human Rights - Workshop Report

Vidushi Marda, Akash Deep Singh and Geethanjali Jujjavarapu — 2016-11-18T12:58:19Z

The Centre for Internet and Society held a one-day workshop on “Big Data in India: Benefits, Harms and Human Rights” at India Habitat Centre, New Delhi on the 1st of October, 2016. This report is a compilation of the the issues discussed, ideas exchanged and challenges recognized during the workshop. The objective of the workshop was to discuss aspects of big data technologies in terms of harms, opportunities and human rights. The discussion was designed around an extensive study of current and potential future uses of big data for governance in India, that CIS has undertaken over the last year with support from the MacArthur Foundation.

Contents

Big Data: Definitions and Global South Perspectives

Aadhaar as Big Data

Seeding

Aadhaar and Data Security

Aadhaar’s Relational Arrangement with Big Data Scheme

The Myths surrounding Aadhaar

IndiaStack and FinTech Apps

Problems with UID

Big Data: Definitions and Global South Perspectives

“Big Data” has been defined by multiple scholars till date. The first consideration at the workshop was to discuss various definitions of big data, and also to understand what could be considered Big Data in terms of governance, especially in the absence of academic consensus. One of the most basic ways to define it, as given by the National Institute of Standards and Technology, USA, is to take it to be the data that is beyond the computational capacity of current systems. This definition has been accepted by the UIDAI of India. Another participant pointed out that Big Data is not only indicative of size, but rather the nature of data which is unstructured, and continuously flowing. The Gartner definition of Big Data relies on the three Vs i.e. Volume (size), Velocity (infinite number of ways in which data is being continuously collected) and Variety (the number of ways in which data can be collected in rows and columns).

The presentation also looked at ways in which Big Data is different from traditional data. It was pointed out that it can accommodate diverse unstructured datasets, and it is ‘relational’ i.e. it needs the presence of common field(s) across datasets which allows these fields to be conjoined. For e.g., the UID in India is being linked to many different datasets, and they don’t constitute Big Data separately, but do so together. An increasingly popular definition is to define data as “Big Data” based on what can be achieved through it. It has been described by authors as the ability to harness new kinds of insight which can inform decision making. It was pointed out that CIS does not subscribe to any particular definition, and is still in the process of coming up with a comprehensive definition of Big Data.

Further, discussion touched upon the approach to Big Data in the Global South. It was pointed out that most discussions about Big Data in the Global South are about the kind of value that it can have, the ways in which it can change our society. The Global North, on the other hand, has moved on to discussing the ethics and privacy issues associated with Big Data.

After this, the presentation focussed on case studies surrounding key Central Government initiatives and projects like Aadhaar, Predictive Policing, and Financial Technology (FinTech).

Aadhaar as Big Data

In presenting CIS’ case study on Aadhaar, it was pointed out that initially, Aadhaar, with its enrollment dataset was by itself being seen as Big Data. However, upon careful consideration in light of definitions discussed above, it can be seen as something that enables Big Data. The different e-governance projects within Digital India, along with Aadhaar, constitute Big Data. The case study discussed the Big Data implications of Aadhaar, and in particular looked at a ‘cradle to grave’ identity mapping through various e-government projects and the datafication of various transaction generated data.

Seeding

Any digital identity like Aadhaar typically has three features: 1. Identification i.e. a number or card used to identify yourself; 2. Authentication, which is based on your number or card and any other digital attributes that you might have; 3. Authorisation: As bearers of the digital identity, we can authorise the service providers to take some steps on our behalf. The case study discussed ‘seeding’ which enables the Big Data aspects of Digital India. In the process of seeding, different government databases can be seeded with the UID number using a platform called Ginger. Due to this, other databases can be connected to UIDAI, and through it, data from other databases can be queried by using your Aadhaar identity itself. This is an example of relationality, where fractured data is being brought together. At the moment, it is not clear whether this access by UIDAI means that an actual physical copy of such data from various sources will be transferred to UIDAI’s servers or if they will just access it through internet, but the data remains on the host government agency’s server. An example of even private parties becoming a part of this infrastructure was raised by a participant when it was pointed out that Reliance Jio is now asking for fingerprints. This can then be connected to the relational infrastructure being created by UIDAI. The discussion then focused on how such a structure will function, where it was mentioned that as of now, it cannot be said with certainty that UIDAI will be the agency managing this relational infrastructure in the long run, even though it is the one building it.

Aadhaar and Data Security

This case study also dealt with the sheer lack of data protection legislation in India except for S.43A of the IT Act. The section does not provide adequate protection as the constitutionality of the rules and regulations under S.43A is ambivalent. More importantly, it only refers to private bodies. Hence, any seeding which is being done by the government is outside the scope of data protection legislation. Thus, at the moment, no legal framework covers the processes and the structures being used for datasets. Due to the inapplicability of S.43A to public bodies, questions were raised as to the existence of a comprehensive data protection policy for government institutions. Participants answered the question in the negative. They pointed out that if any government department starts collecting data, they develop their own privacy policy. There are no set guidelines for such policies and they do not address concerns related to consent, data minimisation and purpose limitation at all. Questions were also raised about the access and control over Big Data with government institutions. A tentative answer from a participant was that such data will remain under the control of the domain specific government ministry or department, for e.g. MNREGA data with the Ministry of Rural Development, because the focus is not on data centralisation but rather on data linking. As long as such fractured data is linked and there is an agency that is responsible to link them, this data can be brought together. Such data is primarily for government agencies. But the government is opening up certain aspects of the data present with it for public consumption for research and entrepreneurial purposes.The UIDAI provides you access to your own data after paying a minimal fee. The procedure for such access is still developing.

Aadhaar’s Relational Arrangement with Big Data Scheme

The various Digital India schemes brought in by the government were elucidated during the workshop. It was pointed out that these schemes extend to myriad aspects of a citizen’s daily life and cover all the essential public services like health, education etc. This makes Aadhaar imperative even though the Supreme Court has observed that it is not mandatory for every citizen to have a unique identity number. The benefits of such identity mapping and the ecosystem being generated by it was also enumerated during the discourse. But the complete absence of any data ethics or data confidentiality principles make us unaware of the costs at which these benefits are being conferred on us. Apart from surveillance concerns, the knowledge gap being created between the citizens and the government was also flagged. Three main benefits touted to be provided by Aadhaar were then analysed. The first is the efficient delivery of services. This appears to be an overblown claim as the Aadhaar specific digitisation and automation does not affect the way in which employment will be provided to citizens through MNREGA or how wage payment delays will be overcome. These are administrative problems that Aadhaar and associated technologies cannot solve. The second is convenience to the citizens. The fallacies in this assertion were also brought out and identified. Before the Aadhaar scheme was rolled in, ration cards were issued based on certain exclusion and inclusion criteria.. The exclusion and inclusion criteria remain the same while another hurdle in the form of Aadhaar has been created. As India is still lacking in supporting infrastructure such as electricity, server connectivity among other things, Aadhaar is acting as a barrier rather than making it convenient for citizens to enroll in such schemes.The third benefit is fraud management. Here, a participant pointed out that this benefit was due to digitisation in the form of GPS chips in food delivery trucks and electronic payment and not the relational nature of Aadhaar. Aadhaar is only concerned with the linking up or relational part. About deduplication, it was pointed out how various government agencies have tackled it quite successfully by using technology different from biometrics which is unreliable at the best of times.

The Myths surrounding Aadhaar

The discussion also reflected on the fact that Aadhaar is often considered to be a panacea that subsumes all kinds of technologies to tackle leakages. However, this does not take into account the fact that leakages happen in many ways. A system should have been built to tackle those specific kinds of leakages, but the focus is solely on Aadhaar as the cure for all. Notably, participants who have been a part of the government pointed out how this myth is misleading and should instead be seen as the first step towards a more digitally enhanced country which is combining different technologies through one medium.

IndiaStack and FinTech Apps

What is India Stack?

The focus then shifted to another extremely important Big Data project, India Stack, being conceptualised and developed by a team of private developers called iStack, for the NPCI. It builds on the UID project, Jan Dhan Yojana and mobile services trinity to propagate and develop a cashless, presence-less, paperless and granular consent layer based on UID infrastructure to digitise India.

A participant pointed out that the idea of India Stack is to use UID as a platform and keep stacking things on it, such that more and more applications are developed. This in turn will help us to move from being a ‘data poor’ country to a ‘data rich’ one. The economic benefits of this data though as evidenced from the TAGUP report - a report about the creation of National Information Utilities to manage the data that is present with the government - is for the corporations and not the common man. The TAGUP report openly talks about privatisation of data.

Problems with India Stack

The granular consent layer of India Stack hasn’t been developed yet but they have proposed to base it on MIT Media Lab’s OpenPDS system. The idea being that, on the basis of the choices made by the concerned person, access to a person’s personal information may be granted to an agency like a bank. What is more revolutionary is that India Stack might even revoke this access if the concerned person expresses a wish to do so or the surrounding circumstances signal to India Stack that it will be prudent to do so. It should be pointed out that the the technology required for OpenPDS is extremely complex and is not available in India. Moreover, it’s not clear how this system would work. Apart from this, even the paperless layer has its faults and has been criticised by many since its inception, because an actual government signed and stamped paper has been the basis of a claim.. In the paperless system, you are provided a Digilocker in which all your papers are stored electronically, on the basis of your UID number. However, it was brought to light that this doesn’t take into account those who either do not want a Digilocker or UID number or cases where they do not have access to their digital records. How in such cases will people make claims?

A Digital Post-Dated Cheque: It’s Ramifications

A key change that FinTech apps and the surrounding ecosystem want to make is to create a digital post-dated cheque so as to allow individuals to get loans from their mobiles especially in remote areas. This will potentially cut out the need to construct new banks, thus reducing the capital expenditure , while at the same time allowing the credit services to grow. The direct transfer of money between UID numbers without the involvement of banks is a step to further help this ecosystem grow. Once an individual consents to such a system, however, automatic transfer of money from one’s bank accounts will be affected, regardless of the reason for payment. This is different from auto debt deductions done by banks presently, as in the present system banks have other forms of collateral as well. The automatic deduction now is only affected if these other forms are defaulted upon. There is no knowledge as to whether this consent will be reversible or irreversible. As Jan Dhan Yojana accounts are zero balance accounts, the account holder will be bled dry. The implication of schemes such as “Loan in under 8 minutes” were also discussed. The advantage of such schemes is that transaction costs are reduced.The financial institution can thus grant loans for the minimum amount without any additional enquiries. It was pointed out that this new system is based on living on future income much like the US housing bubble crash. Interestingly, in Public Distribution Systems, biometrics are insisted upon even though it disrupts the system. This can be seen as a part of the larger infrastructure to ensure that digital post-dated cheques become a success.

The Role of FinTech Apps

FinTech ‘apps’ are being presented with the aim of propagating financial inclusion. The Technology Advisory Group for Unique Projects report stated that as managing such information sources is a big task, just like electricity utilities, a National Information Utilities (NIU) should be set up for data sources. These NIUs as per the report will follow a fee based model where they will be charging for their services for government schemes. The report identified two key NIUs namely the National Payments Corporation of India (NPCI) and the Goods and Services Tax Network (GSTN). The key usage that FinTech applications will serve is credit scoring. The traditional credit scoring data sources only comprised a thin file of records for an individual, but the data that FinTech apps collect - a person’s UID number, mobile number. and bank account number all linked up, allow for a far more comprehensive credit rating. Government departments are willing to share this data with FinTech apps as they are getting analysis in return. Thus, by using UID and the varied data sources that have been linked together by UID, a ‘thick file’ is now being created by FinTech apps. Banking apps have not yet gone down the route of FinTech apps to utilise Big Data for credit scoring purposes.

The two main problems with such apps is that there is no uniform way of credit scoring. This distorts the rate at which a person has to pay interest. The consent layer adds another layer of complication as refusal to share mobile data with a FinTech app may lead to the app declaring one to be a risky investment thus, subjecting that individual to a higher rate of interest .

Regulation of FinTech Apps and the UID Infrastructure

India Stack and the applications that are being built on it, generate a lot of transaction metadata that is very intimate in nature. The privacy aspects of the UID legislation doesn't cover such data. The granular consent layer which has been touted to cover this still has to come into existence. Also, Big Data is based on sharing and linking of data. Here, privacy concerns and Big Data objectives clash. Big Data by its very nature challenges privacy principles like data minimisation and purpose limitation.The need for regulation to cover the various new apps and infrastructure which are being developed was pointed out.

Problems with UID

It has been observed that any problem present with Aadhaar is usually labelled as a teething problem, it’s claimed that it will be solved in the next 10 years. But, this begs the question - why is the system online right now?

Aadhaar is essentially a new data condition and a new exclusion or inclusion criteria. Data exclusion modalities as observed in Rajasthan after the introduction of biometric Point of Service (POS) machines at ration shops was found to be 45% of the population availing PDS services. This number also includes those who were excluded from the database by being included in the wrong dataset. There is no information present to tell us how many actual duplicates and how many genuine ration card holders were weeded out/excluded by POS.

It was also mentioned that any attempt to question Aadhaar is considered to be an attempt to go back to the manual system and this binary thinking needs to change. Big Data has the potential to benefit people, as has been evidenced by the scholarship and pension portals. However, Big Data’s problems arise in systems like PDS, where there is centralised exclusion at the level of the cloud. Moreover, the quantity problem present in the PDS and MNREGA systems persists. There is still the possibility of getting lesser grains and salary even with analysis of biometrics, hence proving that there are better technologies to tackle these problems. Presently, the accountability mechanisms are being weakened as the poor don’t know where to go to for redressal. Moreover, the mechanisms to check whether the people excluded are duplicates or not is not there. At the time of UID enrollment, out of 90 crores, 9 crore were rejected. There was no feedback or follow-up mechanism to figure out why are people being rejected. It was just assumed that they might have been duplicates.

Another problem is the rolling out of software without checking for inefficiencies or problems at a beta testing phase. The control of developers over this software, is so massive that it can be changed so easily without any accountability.. The decision making components of the software are all proprietary like in the the de-duplication algorithm being used by the UIDAI. Thus, this leads to a loss of accountability because the system itself is in flux, none of it is present in public domain and there are no means to analyse it in a transparent fashion..

These schemes are also being pushed through due to database politics. On a field study of NPR of citizens, another Big Data scheme, it was found that you are assumed to be an alien if you did not have the documents to prove that you are a citizen. Hence, unless you fulfill certain conditions of a database, you are excluded and are not eligible for the benefits that being on the database afford you.

Why is the private sector pushing for UIDAI and the surrounding ecosystem?

Financial institutions stand to gain from encouraging the UID as it encourages the credit culture and reduces transaction costs.. Another advantage for the private sector is perhaps the more obvious one, that is allows for efficient marketing of products and services..

The above mentioned fears and challenges were actually observed on the ground and the same was shown through the medium of a case study in West Bengal on the smart meters being installed there by the state electricity utility. While the data coming in from these smart meters is being used to ensure that a more efficient system is developed,it is also being used as a surrogate for income mapping on the basis of electricity bills being paid. This helps companies profile neighbourhoods. The technical officer who first receives that data has complete control over it and he can easily misuse the data. This case study again shows that instruments like Aadhaar and India Stack are limited in their application and aren’t the panacea that they are portrayed to be.

A participant pointed out that in the light of the above discussions, the aim appears to be to get all kinds of data, through any source, and once you have gotten the UID, you link all of this data to the UID number, and then use it in all the corporate schemes that are being started. Most of the problems associated with Big Data are being described as teething problems. The India Stack and FinTech scheme is coming in when we already know about the problems being faced by UID. The same problems will be faced by India Stack as well.

Can you opt out of the Aadhaar system and the surrounding ecosystem?

The discussion then turned towards whether there can be voluntary opting out from Aadhaar. It was pointed out that the government has stated that you cannot opt out of Aadhaar. Further, the privacy principles in the UIDAI bill are ambiguously worded where individuals only have recourse for basic things like correction of your personal information. The enforcement mechanism present in the UIDAI Act is also severely deficient. There is no notification procedure if a data breach occurs. . The appellate body ‘Cyber Appellate Tribunal’ has not been set up in three years.

CCTNS: Big Data and its Predictive Uses

What is Predictive Policing?

The next big Big Data case study was on the Crime and Criminal Tracking Network & Systems (CCTNS). Originally it was supposed to be a digitisation and interconnection scheme where police records would be digitised and police stations across the length and breadth of the country would be interconnected. But, in the last few years some police departments of states like Chandigarh, Delhi and Jharkhand have mooted the idea of moving on to predictive policing techniques. It envisages the use of existing statistical and actuarial techniques along with many other tropes of data to do so. It works in four ways: 1. By predicting the place and time where crimes might occur; 2. To predict potential future offenders; 3. To create profiles of past crimes in order to predict future crimes; 4. Predicting groups of individuals who are likely to be victims of future crimes.

How is Predictive Policing done?

To achieve this, the following process is followed: 1. Data collection from various sources which includes structured data like FIRs and unstructured data like call detail records, neighbourhood data, crime seasonal patterns etc. 2. Analysis by using theories like the near repeat theory, regression models on the basis of risk factors etc. 3. Intervention

Flaws in Predictive Policing and questions of bias

An obvious weak point in the system is that if the initial data going into the system is wrong or biased, the analysis will also be wrong. Efforts are being made to detect such biases. An important way to do so will be by building data collection practices into the system that protect its accuracy. The historical data being entered into the system is carrying on the prejudices inherited from the British Raj and biases based on religion, caste, socio-economic background etc.

One participant brought about the issue of data digitization in police stations, and the impact of this haphazard, unreliable data on a Big Data system. This coupled with paucity of data is bound to lead to arbitrary results. An effective example was that of black neighbourhoods in the USA. These are considered problematic and thus they are policed more, leading to a higher crime rate as they are arrested for doing things that white people in an affluent neighbourhood get away with. This in turn further perpetuates the crime rate and it becomes a self-fulfilling prophecy. In India, such a phenomenon might easily develop in the case of migrants, de-notified tribes, Muslims etc. A counter-view on bias and discrimination was offered here. One participant pointed out that problems with haphazard or poor quality of data is not a colossal issue as private companies are willing to fill this void and are actually doing so in exchange for access to this raw data. It was also pointed out how bias by itself is being used as an all encompassing term. There are multiplicities of biases and while analysing the data, care should be taken to keep it in mind that one person’s bias and analysis might and usually does differ from another. Even after a computer has analysed the data, the data still falls into human hands for implementation.

The issue of such databases being used to target particular communities on the basis of religion, race, caste, ethnicity among other parameters was raised. Questions about control and analysis of data were also discussed, i.e. whether it will be top-down with data analysis being done in state capitals or will this analysis be done at village and thana levels as well too. It was discussed as topointed out how this could play a major role in the success and possible persecutory treatment of citizens, as the policemen at both these levels will have different perceptions of what the data is saying. . It was further pointed out, that at the moment, there’s no clarity on the mode of implementation of Big Data policing systems. Police in the USA have been seen to rely on Big Data so much that they have been seen to become ‘data myopic’. For those who are on the bad side of Big Data, in the Indian context, laws like preventive detention can be heavily misused.There’s a very high chance that predictive policing due to the inherent biases in the system and the prejudices and inefficiency of the legal system will further suppress the already targeted sections of the society. A counterpoint was raised and it was suggested that contrary to our fears, CCTNS might lead to changes in our understanding and help us to overcome longstanding biases.

Open Knowledge Architecture as a solution to Big Data biases?

The conference then mulled over the use of ‘Open Knowledge’ architecture to see whether it can provide the solution to rid Big Data of its biases and inaccuracies if enough eyes are there. It was pointed out that Open Knowledge itself can’t provide foolproof protection against these biases as the people who make up the eyes themselves are predominantly male belonging to the affluent sections of the society and they themselves suffer from these biases.

Who exactly is Big Data supposed to serve?

The discussion also looked at questions such as who is this data for? Janata Information System (JIS), is a concept developed by MKSS where the data collected and generated by the government is taken to be for the common citizens. For e.g. MNREGA data should be used to serve the purposes of the labourers. The raw data as is available at the moment, usually cannot be used by the common man as it is so vast and full of information that is not useful for them at all. It was pointed out that while using Big Data for policy planning purposes, the actual string of information that turned out to be needed was very little but the task of unravelling this data for civil society purposes is humongous. By presenting the data in the right manner, the individual can be empowered. The importance of data presentation was also flagged. It was agreed upon that the content of the data should be for the labourer and not a MNC, as the MNC has the capability to utilise the raw data on it’s own regardless.

Concerns about Big Data usage

Participants pointed out that privacy concerns are usually brushed under the table due to a belief that the law is sufficient or that the privacy battle has already been lost.
In the absence of knowledge of domain and context, Big Data analysis is quite limited. Big Data’s accuracy and potential to solve problems needs to be factually backed.
The narrative of Big Data often rests on the assumption that descriptive statistics take over inferential statistics, thus eliminating the need for domain specific knowledge. It is claimed that the data is so big that it will describe everything that we need to know.
Big Data is creating a shift from a deductive model of scientific rigour to an inductive one. In response to this, a participant offered the idea that troves of good data allow us to make informed questions on the basis of which the deductive model will be formed. A hybrid approach combining both deductive and inductive might serve us best.
The need to collect the right data in the correct format, in the right place was also expressed.

Potential Research Questions & Participants’ Areas of Research

Following this discussion, participants brainstormed to come up with potential areas of research and research questions. They have been captured below:

Big Data, Aadhaar and India Stack:

Has Aadhaar been able to tackle illegal ways of claiming services or are local negotiations and other methods still prevalent?
Is the consent layer of India Stack being developed in a way that provides an opportunity to the UID user to give informed consent? The OpenPDS and its counterpart in the EU i.e. the My Data Structure were designed for countries with strong privacy laws. Importantly, they were meant for information shared on social media and not for an individual’s health or credit history. India is using it in a completely different sphere without strong data protection laws. What were the granular consent layer structures present in the West designed for and what were they supposed to protect?
The question of ownership of data needs to be studied especially in context of a globalised world where MNCs are collecting copious amounts of data of Indian citizens. What is the interaction of private parties in this regard?

Big Data and Predictive Policing:

How are inequalities being created through the Big Data systems? Lessons should be taken from the Western experience with the advent of predictive policing and other big data techniques - they tend to lead to perpetuation of the current biases which are already ingrained in the system.
It was also pointed out how while studying these topics and anything related to technology generally, we become aware of a divide that is present between the computational sciences and social sciences. This divide needs to be erased if Big Data or any kind of data is to be used efficiently. There should be a cross-pollination between different groups of academics. An example of this can be seen to be the ‘computational social sciences departments’ that have been coming up in the last 3-4 years.
Why are so many interim promises made by Big Data failing? A study of this phenomenon needs to be done from a social science perspective. This will allow one to look at it from a different angle.

Studying Big Data:

What is the historical context of the terms of reference being used for Big Data? The current Big Data debate in India is based on parameters set by the West. For better understanding of Big Data, it was suggested that P.C. Mahalanobis’ experience while conducting the Indian census, (which was the Big Data of that time) can be looked at to get a historical perspective on Big Data. This comparison might allow us to discover questions that are important in the Indian context. It was also suggested that rather than using ‘Big Data’ as a catchphrase to describe these new technological innovations, we need to be more discerning.
What are the ideological aspects that must be considered while studying Big Data? What does the dialectical promise of technology mean? It was contended that every time there is a shift in technology, the zeitgeist of that period is extremely excited and there are claims that it will solve everything. There’s a need to study this dialectical promise and the social promise surrounding it.
Apart from the legitimate fears that Big Data might lead to exclusion, what are the possibilities in which it improve inclusion too?
The diminishing barrier between the public and private self, which is a tangent to the larger public-private debate was mentioned.
How does one distinguish between technology failure and process failure while studying Big Data?

Big Data: A Friend?

In the concluding session, the fact that the Big Data moment cannot be wished away was acknowledged. The use of analytics and predictive modelling by the private sector is now commonplace and India has made a move towards a database state through UID and Digital India. The need for a nuanced debate, that does away with the false equivalence of being either a Big Data enthusiast or a luddite is crucial.

A participant offered two approaches to solving a Big Data problem. The first was the Big Data due process framework which states that if a decision has been taken that impacts the rights of a citizen, it needs to be cross examined. The efficacy and practicality of such an approach is still not clear. The second, slightly paternalistic in nature, was the approach where Big Data problems would be solved at the data science level itself. This is much like the affirmative algorithmic approach which says that if in a particular dataset, the data for the minority community is not available then it should be artificially introduced in the dataset. It was also suggested that carefully calibrated free market competition can be used to regulate Big Data. For e.g. a private personal wallet company that charges higher, but does not share your data at all can be an example of such competition.

Another important observation was the need to understand Big Data in a Global South context and account for unique challenges that arise. While the convenience of Big Data is promising, its actual manifestation depends on externalities like connectivity, accurate and adequate data etc that must be studied in the Global South.

While the promises of Big Data are encouraging, it is also important to examine its impacts and its interaction with people's rights. Regulatory solutions to mitigate the harms of big data while also reaping its benefits need to evolve.

For more details visit https://cis-india.org/internet-governance/big-data-in-india-benefits-harms-and-human-rights-a-report

Banking on artificial intelligence: In hiring drive, Bots are calling the shots now

Anjali Venugopalan — 2019-07-02T05:38:26Z

Algorithms analyse expressions, tone to check for traits such as confidence, anger in video interviews.

The article by Anjali Venugopalan was published in Economic Times on June 4, 2019, Sunil Abraham was quoted. Also mirrored on ET Tech.com.

The future of hiring is already upon us. Algorithms are analysing people’s expressions and tone of voice to check for traits such as “confidence” and “happiness” during video interviews. The robotic video assessment software is then used to hire candidates — customer service operators and assistant vice presidents alike — though the process comes with its own set of problems.

Axis Bank used algorithm-based video interviews — along with aptitude tests — to hire around 2,000 customer service officers from a pool of more than 40,000 applicants this year, said Rajkamal Vempati, HR head of the private sector bank, adding it could standardise and scale up the process of hiring.

HR managers only gave offer letters, he said.

Nirmal Singh, CEO of Wheebox, a division of PeopleStrong which carried out the hiring, said it trained the face-indexing software — sourced from Microsoft — using around 50,000 candidates who had applied to Axis Bank in 2017. The software picked up emotional states such as “nervousness” and “happiness” based on eye movements, expressions and tone of voice and marked the candidates, Singh said. Scores from candidates who were shortlisted were used to come up with the “cutoff ” for these traits. Nirmal Singh, CEO of Wheebox, a division of PeopleStrong which carried out the hiring, said it trained the face-indexing software — sourced from Microsoft — using around50,000 candidates who had applied to Axis Bank in 2017. The software picked up emotional states such as “nervousness” and “happiness” based on eye movements,expressions and tone of voice and marked the candidates, Singh said. Scores from candidates who were shortlisted were used to come up with the “cutoff ” for these traits.

Insurance provider Bajaj Allianz has hired more than 1,600 people, including underwriters and assistant vice presidents, with the help of robotic video assessments that analysed behaviour, said Vikramjeet Singh, chief HR officer, adding it could help reduce human bias. Insurance provider Bajaj Allianz has hired more than 1,600 people, including underwriters and assistant vice presidents, with the help of robotic video assessments that analysedbehaviour, said Vikramjeet Singh, chief HR officer, adding it could help reduce human bias.

Concerns over Software's Biases

Talview, a Palo Alto-headquartered company with operations in Singapore and the United States, provided the assessment for the insurer.

The software, sourced from Microsoft and IBM, can analyse states such as “anger” and “happiness” from expressions, “confidence” from voice tone and traits like “ability to work ina team” and “decisiveness” from text analysis, according to Rajeev Menon, chief product officer, Talview.

Candidates may be able to beat questionnaires by giving expected answers to questions like “Can you work in a team?”, but video assessments pick up on subtleties in expression and vocabulary, and cannot be gamed, Menon said.Be that as it may, Amazon.com scrapped its artificial intelligence-based recruiting system after it found the AI system biased against women, according to an October 2018 report by Reuters.

The AI system was drawing on data from the past, where more men had made it into the company than women.“If you can fool a human, you can fool a computer,” said Sunil Abraham, executive director of Centre for Internet and Society.Recruitment algorithms could “homogenise the emotional economy” by forcing people to act a certain way, he said.

Since the software is based on expressions and tone of voice, it could disadvantage less expressive people, like those who are autistic, said Wheebox’s Singh.

Facial recognition by companies such as IBM, Microsoft and Amazon got the gender of a dark-skinned woman wrong one out of three times (20-35% error rate), a 2018 study by MIT researcher Joy Buolamwini found. For white males, the error was 0.8%.

Video Assessments

Facial recognition has nothing to do with video analytics, Wheebox’s Singh said. The two are, however, closely linked, said Animashree Anandkumar, professor of computing andmathematical science at California Institute of Technology.

She said such software was “deeply problematic”, as it could correlate wrong factors (likegender or skin colour) and show that as the cause for success. It is possible dark-skinned people would be disadvantaged, said Menon of Talview. Thecompany uses facial expression as just one input among many and gives it a low weightage, he said.The software they use is only 39% accurate, and will improve with more data, said and will improve with more data, said Ridhima Gauba, co-founder of Interview Air, a Navi Mumbai-based company that provides a similar service to companies and colleges.

Companies also say video assessments are a risky business.

Bajaj Allianz does not use video assessments for recruitments beyond middle management. It is “important to see a person physically” when hiring for senior positions, said Asha Sharma, manager (corporate HR) of Everest Industries.

The company, however, uses pre-recorded video interviews — where the computer asks questions — to hire juniors from campuses, she said.

For more details visit https://cis-india.org/internet-governance/news/economic-times-anjali-venugopalan-june-4-2019-banking-on-artificial-intelligence

Artificial Intelligence: a Full-Spectrum Regulatory Challenge [Working Draft]

sunil — 2020-08-04T06:10:13Z

Today, there are certain misconceptions regarding the regulation of AI. Some corporations would like us to believe that AI is being developed and used in a regulatory vacuum. Others in civil society organisations believe that AI is a regulatory circumvention strategy deployed by corporations. As a result, these organisations call for onerous regulations targeting corporations. However, some uses of AI by corporations can be completely benign and some uses AI by the state can result in the most egregious human rights violations. Therefore policy makers need to throw every regulatory tool from their arsenal to unlock the benefits of AI and mitigate its harms.

This policy brief proposes a granular, full spectrum approach to the regulation of AI depending on who is using AI, who is impacted by that use and what human rights are impacted. Everything from deregulation, to forbearance, to updated regulations, to absolute and blanket prohibitions needs to be considered depending on the specifics. This approach stands in contrast to approaches of ethics, omnibus law, homogeneous principles, and human rights, which will result in inappropriate under-regulation or over-regulation of the sector.

Find a copy of the working draft here.

For more details visit https://cis-india.org/internet-governance/artificial-intelligence-a-full-spectrum-regulatory-challenge-working-draft

Artificial Intelligence in India: A Compendium

Centre for Internet & Society — 2023-05-09T06:56:25Z

Artificial Intelligence (AI) is fast emerging as a key technological paradigm in different sectors across the globe including India.

Towards understanding the state of AI in India, challenges to the development and adoption of the same, and ethical concerns that arise out of the use of AI - CIS is undertaking research to understand and document national developments, discourse, and impact (actual and potential) to ethical and regulatory solutions and compare the same against global developments in the space. As part of this, CIS is creating a compendium of reports that dive into the use of AI across sectors including healthcare, manufacturing, governance, and finance.

Each report seeks to map the present state of AI in the respective sector. In doing so, it explores: Use: What is the present use of AI in the sector? What is the narrative and discourse around AI in the sector? Actors: Who are the key stakeholders involved in the development, implementation and regulation of AI in the sector? Impact: What is the potential and existing impact of AI in the sector? Regulation: What are the challenges faced in policy making around AI in the sector?

The reports are as follows:

AI and the Healthcare Industry in India
AI and the Manufacturing and Services Sector in India
AI and the Banking and Finance Industry in India: (19th June 2018 Update: This case study has been modified to remove interview quotes, which are in the process of being confirmed. The link above is the latest draft of the report.)
AI in the Governance Sector in India

The research is funded by Google India. Comments and feedback are welcome. The reports are drafts.

For more details visit https://cis-india.org/internet-governance/blog/artificial-intelligence-in-india-a-compendium

Artificial Intelligence in Governance: A Report of the Roundtable held in New Delhi

Saman Goudarzi and Natallia Khaniejo — 2018-05-03T15:49:40Z

This Report provides an overview of the proceedings of the Roundtable on Artificial Intelligence (AI) in Governance, conducted at the Indian Islamic Cultural Centre, in New Delhi on March 16, 2018. The main purpose of the Roundtable was to discuss the deployment and implementation of AI in various aspects of governance within the Indian context. This report summarises the discussions on the development and implementation of AI in various aspects of governance in India. The event was attended by participants from academia, civil society, the legal sector, the finance sector, and the government.

Event Report: Download (PDF)

This report provides a summary of the proceedings of the Roundtable on Artificial Intelligence (AI) in Governance (hereinafter referred to as ‘the Roundtable’). The Roundtable took place at the India Islamic Cultural Centre in New Delhi on March 16, 2018 and included participation from academia, civil society, law, finance, and government. The main purpose of the Roundtable was to discuss the deployment and implementation of AI in various aspects of governance within the Indian context.

The Roundtable began with a presentation by Amber Sinha (Centre for Internet and Society - CIS) providing an overview of the CIS’s research objectives and findings thus far. During this presentation, he defined both AI and the scope of CIS’s research, outlining the areas of law enforcement, defense, education, judicial decision making, and the discharging of administrative functions as the main areas of concerns for the study. The presentation then outlined the key AI deployments and implementations that have been identified by the research in each of these areas. Lastly, the presentation raised some of the ethical and legal concerns related to this phenomenon.

The presentation was followed by the Roundtable discussion that saw various topics in regards to the usages, challenges, ethical considerations and implications of AI in the sector being discussed. This report has identified a number of key themes of importance evident throughout these discussions.These themes include: (1) the meaning and scope of AI, (2) AI’s sectoral applications, (3) human involvement with automated decision making, (4) social and power relations surrounding AI, (5) regulatory approaches to AI and, (6) challenges to adopting AI. These themes in relation to the Roundtable are explored further below.

Meaning and Scope of AI

One of the first tasks recommended by the group of participants was to define the meaning and scope of AI and the way those terms are used and adopted today. These concerns included the need to establish a distinction between the use of algorithms, machine learning, automation and artificial intelligence. Several participants believed that establishing consensus around these terms was essential before proceeding towards a stage of developing regulatory frameworks around them.

The general fact agreed to was that AI as we understand it does not necessarily extend to complete independence in terms of automated decision making but it refers instead to the varying levels of machine learning (ML), and the automation of certain processes that has already been achieved. Several concerns that emerged during the course of the discussion centred around the question of autonomy and transparency in the process of ML and algorithmic processing. Stakeholders recommended that over and above the debates of humans in the loop [1] on the loop [2] and out of the loop, [3] there were several other gaps with respect to AI and its usage in the industry today which also need to be considered before building a roadmap for future usage. Key issues like information asymmetries, communication lags, a lack of transparency, the increased mystification of the coding process and the centralization of power all needed to be examined and analysed under the rubric of developing regulatory frameworks.

Takeaway Point: The group brought out the need for standardization of terminology as well as the establishment of globally replicable standards surrounding the usage, control and proliferation of AI. The discussion also brought up the problems with universal applicability of norms. One of the participants brought up an issue regarding the lack of normative frameworks around the usage and proliferation of AI. Another participant responded to the concern by alluding to the Asilomar AI principles.[4] The Asilomar AI principles are a set of 23 principles aimed at directing and shaping AI research in the future. The discussion brought out further issues regarding the enforceability as well universal applicability of the principles and their global relevance as well. Participants recommended the development of a shorter, more universally applicable regulatory framework that could address various contextual limitations as well.

AI Sectoral Applications

Participants mentioned a number of both current and potential applications of AI technologies, referencing the defence sector, the financial sector, and the agriculture sector. There are several developments taking place on the Indian military front with the Committee on AI and National Security being established by the Ministry of Defence. Through the course of the discussion it was also stated that the Indian Armed Forces were very interested in the possibilities of using AI for their own strategic and tactical purposes. From a technological standpoint, however, there has been limited progress in India in researching and developing AI.

While India does deploy some Unmanned Aerial Vehicles (UAVs), they are mostly bought from Israel, and often are not autonomous. It was also pointed out that contrary to reportage in the media, the defence establishment in India is extremely cautious about the adoption of autonomous weapons systems, and that the autonomous technology being rolled out by the CAIR is not yet considered trustworthy enough for deployment.

Discussions further revealed that the few technologies that have a relative degree of autonomy are primarily loitering ammunitions and are used to target radar insulations for reconnaissance purposes. One participant mentioned that while most militaries are interested in deploying AI, it is primarily from an Intelligence, Surveillance and Reconnaissance (ISR) perspective. The only exception to this generalization is China where the military ethos and command structure would work better with increased reliance on independent AI systems. One major AI system rolled out by the US is Project Maven which is primarily an ISR system. The aim of using these systems is to improve decision making and enhance data analysis particularly since battlefields generate a lot of data that isn’t used anywhere.

Another sector discussed was the securities market where algorithms were used from an analytical and data collection perspective. A participant referred to the fact that machine learning was being used for processes like credit and trade scoring -- all with humans on the loop. The participant further suggested that while trade scoring was increasingly automated, the overall predictive nature of such technologies remained within a self limiting capacity wherein statistical models, collected data and pattern analysis were used to predict future trends. The participant questioned whether these algorithms could be considered as AI in the truest sense of the term since they primarily performed statistical functions and data analysis.

One participant also recommended the application of AI to sectors like agriculture with the intention of gradually acclimatizing users to the technology itself. Respondents also stated that while AI technologies were being used in the agricultural space it was primarily from the standpoint of data collection and analysis as opposed to predictive methods. It was mentioned that a challenge to the broad adoption of AI in this sector is the core problem of adopting AI as a methodology – namely information asymmetries, excessive data collection, limited control/centralization and the obfuscatory nature of code – would not be addressed/modified. Lastly, participants also suggested that within the Indian framework not much was being done aside from addressing farmers’ queries and analysing the data from those concerns.

Takeaway Point: The discussion drew attention to the various sectors where AI was currently being used -- such as the military space, agricultural development and the securities market -- as well as potential spaces of application -- such as healthcare and manual scavenging. The key challenges that emerged were information asymmetries with respect to the usage of these technologies as well as limited capacity in terms of technological advancement.

Human Involvement with Automated Decision Making

Large parts of discussions throughout the Roundtable event were preoccupied with automated decision making and specifically, the involvement of humans (human on and in the loop) or lack thereof (human out of the loop) in this process. These discussions often took place with considerations of AI for prescriptive and descriptive uses.

Participants expressed that human involvement was not needed when AI was being used for descriptive uses, such as determining relationships between various variables in large data sets. Many agreed to the superior ability of ML and similar AI technologies in describing large and unorganized datasets. It was the prescriptive uses of AI where participants saw the need for human involvement, with many questioning the technology making more important decisions by itself.

The need for human involvement in automated decision making was further justified by references to various instances of algorithmic bias in the American context. One participant, for example, brought up the use of algorithmic decision making by a school board in the United States for human resource practices (hirings, firing, etc.) based on the standardized test scores of students. In this instance, such practices resulted in the termination of teachers primarily from low income neighbourhoods.[5] The main challenge participants identified in regards to human on the loop automated decision making is the issue of capacity, as significant training would have to be achieved for sectors to have employees actively involved in the automated decision making workflow.

An example in the context of the healthcare field was brought up by one participant arguing for human in the loop in regards to prescriptive scenarios. The participant suggested that AI technology, when given x-ray or MRI data for example, should only be limited to pointing out the correlations of diseases with patients’ scans/x-rays. Analysis of such correlations should be reserved for the medical expertise of doctors who would then determine if any instances of causality can be identified from this data and if it’s appropriate for diagnosing patients.

It was emphasized that, despite a preference for human on/in the loop in regards to automated decision making, there is a need to be cognisant of techno-solutionism due to the human tendency of over reliance on technology when making decisions. A need for command and control structures and protocols was emphasized for various governance sectors in order to avoid potentially disastrous results through a checks and balances system. It was noted that the defense sector has already developed such protocols, having established a chain of command due to its long history of algorithmic decision making (e.g. the Aegis Combat System being used by the US Navy in the 1980s).

One key reason why militaries prefer human in and on the loop systems as opposed to out of the loop systems is because of the protocol associated with human action on the battlefield. International Humanitarian Law has clear indicators of what constitutes a war crime and who is to be held responsible in the scenario but developing such a framework with AI systems would be challenging as it would be difficult to determine which party ought to be held accountable in the case of a transgression or a mistake.

Takeaway Point: It was reiterated by many participants that neither AI technology or India’s regulatory framework is at a point where AI can be trusted to make significant decisions alone -- especially when such decisions are evaluating humans directly. It was recommended that human out of the loop decision making should be reserved for descriptive practices whereas human on and in the loop decision making should be used for prescriptive practices. Lastly, it was also suggested that appropriate protocols be put in place to direct those involved in the automated decision making workflow. Particularly when the process involves judgements and complex decision making in sectors such as jurisprudence and the military.

The Social and Power Relations Surrounding AI

Some participants emphasized the need to contextualize discussions of AI and governance within larger themes of poverty, global capital and power/social relations. Their concerns were that the use of AI technologies would only create and reinforce existing power structures and should instead be utilized towards ameliorating such issues. Manual scavenging, for example, was identified as an area where AI could be used to good effect if coupled with larger socio-political policy changes. There are several hierarchies that could potentially be reinforced through this process and all these failings needed to be examined thoroughly before such a system was adopted and incorporated within the real world.

Furthermore the discussion also revealed that the objectivity attributed to AI and ML tends to gloss over the fact that there are nonetheless implicit biases that exist in the minds of the creators that might work themselves into the code. Fears regarding technology recreating a more exclusionary system were not entirely unfounded as participants pointed out the fact that the knowledge base of the user would determine whether technology was used as a tool of centralization or democratization.

One participant also questioned the concept of governance itself, contrasting the Indian government’s usage of the term in the 1950s (as it appears in the Directive Principle) with that of the World Bank in the 1990s.

Takeaway Point: Discussions of the implementation and deployment of AI within the governance landscape should attempt to take into consideration larger power relations and concepts of equity.

Regulatory Approaches to AI

Many recognized the need for AI-specific regulations across Indian sectors, including governance. These regulations, participants stated, should draw from notions of accountability, algorithmic transparency and efficiency. Furthermore, it was also stated that such regulations should consider the variations across the different legs of the governance sector, especially in regards to defence. One participant, pointing to the larger trends towards automation, recommended the establishment of certain fundamental guidelines aimed at directing the applicability of AI in general. The participant drew attention to the need for a robust evaluation system for various sectors (the criminal justice system, the securities market, etc.) as a way of providing checks on algorithmic biases. Another emphasized for the need of regulations for better quality data as to ensure machine readability and processiblity for various AI systems.

Another key point that emerged was the importance of examining how specific algorithms performed processes like identification or detection. A participant recommended the need to examine the ways in which machines identify humans and what categories/biases could infiltrate machine-judgement. They reiterated that if a new element was introduced in the system, the pre-existing variables would be impacted as well. The participant further recommended that it would be useful to look at these systems in terms of the couplings that get created in order to determine what kinds of relations are fostered within that system.

The roundtable saw some debate regarding the most appropriate approach to developing such regulations. Some participants argued for a harms-based approach, particularly in regards to determining if regulations are needed all together for specific sectors (as opposed to guidelines, best practices, etc.). The need to be cognisant of both individual and structural harms was emphasized, mindful of the possibility of algorithmic biases affecting traditionally marginalized groups.

Others only saw value in a harms based approach insomuch that it could help outline the appropriate penalties in an event of regulations being violated, arguing instead for a rights-based approach as it enabled greater room for technological changes. An approach that kept in mind emerging AI technologies was reiterated by a number of participants as being crucial to any regulatory framework. The need for a regulatory space that allowed for technological experimentation without the fear of constitutional violation was also communicated.

Takeaway Point: The need for a AI-specific regulatory framework cognisant of differentiations across sectors in India was emphasized. There is some debate about the most appropriate approach for such a framework, a harms-based approach being identified by many as providing the best perspective on regulatory need and penalties. Some identified the rights-based approach as providing the most flexibility for an rapidly evolving technological landscape.

Challenges to Adopting AI

Out of all the concerns regarding the adoption of algorithms, ML and AI, the two key points of resistance that emerged, centred around issues of accountability and transparency. Participants suggested that within an AI system, predictability would be a key concern, and in the absence of predictable outcomes, establishing redressal mechanisms would pose key challenges as well.

A discussion was also initiated regarding the problems involved in attributing responsibility within the AI chain as well as the need to demystify the process of using AI in daily life. While reiterating the current landscape, participants spoke about how the usage of AI is currently limited to the automation of certain tasks and processes in certain sectors where algorithmic processing is primarily used as a tool of data collection and analysis as opposed to an independent decision making tool.

One of the suggestions and thought points that emerged during the discussion was whether a gradual adoption of AI on a sectoral basis might be more beneficial as it would provide breathing room in the middle to test the system and establish trust between the developers, providers, and consumers. This prompted a debate about the controllers and the consumers of AI and how the gap between the two would need to be negotiated. The debate also brought up larger concerns regarding the mystification of AI as a process itself and the complications of translating the code into communicable points of intervention.

Another major issue that emerged was the question of attribution of responsibility in the case of mistakes. In the legal process as it currently exists, human imperfections notwithstanding, it would be possible to attribute the blame for decisions taken to certain actants undertaking the action. Similarly in the defence sector, it would be possible to trace the chain of command and identify key points of failure, but in the case of AI based judgements, it would be difficult to place responsibility or blame. This observation led to a debate regarding accountability in the AI chain. It was inconclusive whether the error should be attributed to the developer, the distributor or the consumer.

A suggestion that was offered in order to counter the information asymmetry as well as reduce the mystification of computational method was to make the algorithm and its processes transparent. This sparked a debate, however, as participants stated that while such a state of transparency ought to be sought after and aspired towards, it would be accompanied by certain threats to the system. A key challenge that was pointed out was the fact that if the algorithm was made transparent, and its details were shared, there would be several ways to manipulate it, translate it and misuse it.

Another question that emerged was the distribution of AI technologies and the centralization of the proliferation process particularly in terms of service provision. One participant suggested that given the limited nature of research being undertaken and the paucity of resources, a limited number of companies would end up holding the best tech, the best resources and the best people. They further suggested that these technologies might end up being rolled out as a service on a contractual basis. In which case it would be important to track how the service was being controlled and delivered. Models of transference would become central points of negotiation with alternations between procurement based, lease based, and ownership based models of service delivery. Participants suggested that this was going to be a key factor in determining how to approach these issues from a legal and policy standpoint.

Takeaway Point: The two key points of resistance that emerged during the course of discussion were accountability and transparency. Participants pointed out the various challenges involved in attributing blame within the AI chain and they also spoke about the complexities of opening up AI code, thereby leaving it vulnerable to manipulation. Certain other challenges that were briefly touched upon were the information asymmetry, excessive data collection, centralization of power in the hands of the controllers and complicated service distribution models.

Conclusion

The Roundtable provided some insight into larger debates regarding the deployment and applications of AI in the governance sector of India. The need for a regulatory framework as well as globally replicable standards surrounding AI was emphasized, particularly one mindful of the particular needs of differing fields of the governance sector (especially defence). Furthermore, a need for human on/in the loop practices with regards to automated decision making was highlighted for prescriptive instances, particularly when such decisions are responsible for directly evaluating humans. Contextualising AI within its sociopolitical parameters was another key recommendation as it would help filter out the biases that might work themselves into the code and affect the performance of the algorithm. Further, it is necessary to see the involvement and influence of the private sector in the deployment of AI for governance, it often translating into the delivery of technological services from private actors to public bodies towards discharge of public functions. This has clear implications for requirements of transparency and procedural fairness even in private sector delivery of these services. Defining the meaning and scope of AI while working to demystify algorithms themselves would serve to strengthen regulatory frameworks as well as make AI more accessible for the user / consumer.

[1]. Automated decision making model where final decisions are made by a human operator

[2]. Automated decision making model where decisions can be made without human involvement but a human can override the system.

[3]. A completely autonomous decision making model requiring no human involvement

[4]. https://futureoflife.org/ai-principles/

[5]. The participant was drawing this example from Cathy O’Neil’s Weapons of Math Destruction, (Penguin,2016), at 4-13.

For more details visit https://cis-india.org/internet-governance/blog/artificial-intelligence-in-governance-a-report-of-the-roundtable-held-in-new-delhi

Artificial Intelligence for India's Transformation

Admin — 2019-03-20T01:38:48Z

ASSOCHAM's 3rd International Conference was organized at Hotel Imperial in New Delhi. Amber Sinha a session on use, impact and ethics in AI.

Click to view the agenda.

For more details visit https://cis-india.org/internet-governance/news/artificial-intelligence-for-indias-transformation

Artificial Intelligence for Growth: Leveraging AI and Robotics for India's Economic Transformation

Admin — 2018-05-05T09:08:07Z

Amber Sinha took part in the second international conference organized by ASSOCHAM at Hotel Shangri-La in New Delhi on April 27, 2018.

Keynote Address

12.15 p.m. - 12.30 p.m.: Shri Gopalakrishnan S., Joint Secretary, Ministry of Electronics and IT, Government of India

Special Address

12.30 p.m. - 12.45 p.m.: Dr. Pushpak Bhattacharyya, Director and Professor, Computer Science and Engg, IIT Patna and Chairman, BIS Committee for Standardisation in Artificial Intelligence

Panel Discussion

Session Moderator

12.45 p.m. - 1.40 p.m.: Shri Sudipta Ghosh, India Leader, Data and Analytics, PwC

Panelists

Shri Amber Sinha, Senior Programme Manager, Centre for Internet and Society
Shri Utpal Chakraborty, Lead Architect - AI, L&T Infotech
Shri Atul Rai, CEO & Co-Founder, Staqu Technologies
Shri Prabhat Manocha, IBM

For more details visit https://cis-india.org/internet-governance/news/artificial-intelligence-for-growth-leveraging-ai-and-robotics-for-indias-economic-transformation

Artificial Intelligence and Data Initiative

Admin — 2019-05-14T15:06:02Z

On 3 May 2019 Arindrajit Basu attended a meeting of the Artificial Intelligence and Data Initiative held at IIC in Delhi. I am a member of the Working Group and co-authoring a report with Anindya Chaudhuri of Global Development Network on the prospect of collaborations in Public uses of AI.

The agenda can be viewed here.

For more details visit https://cis-india.org/internet-governance/news/artificial-intelligence-and-data-initiative

Artificial Intelligence - Literature Review

Shruthi Anand — 2017-12-18T15:12:52Z

With origins dating back to the 1950s Artificial Intelligence (AI) is not necessarily new. However, interest in AI has been rekindled over the last few years, in no small measure due to the rapid advancement of the technology and its applications to real- world scenarios. In order to create policy in the field, understanding the literature regarding existing legal and regulatory parameters is necessary. This Literature Review is the first in a series of reports that seeks to map the development of AI, both generally and in specific sectors, culminating in a stakeholder analysis and contributions to policy-making. This Review analyses literature on the historical development of the technology, its compositional makeup, sector- specific impacts and solutions and finally, overarching regulatory solutions.

Edited by Amber Sinha and Udbhav Tiwari; Research Assistance by Sidharth Ray

With origins dating back to the 1950s Artificial Intelligence (AI) is not necessarily new. With an increasing number of real-world implications over the last few years, however, interest in AI has been reignited over the last few years.

The rapid and dynamic pace of development of AI have made it difficult to predict its future path and is enabling it to alter our world in ways we have yet to comprehend. This has resulted in law and policy having stayed one step behind the development of the technology.

Understanding and analyzing existing literature on AI is a necessary precursor to subsequently recommending policy on the matter. By examining academic articles, policy papers, news articles, and position papers from across the globe, this literature review aims to provide an overview of AI from multiple perspectives.

The structure taken by the literature review is as follows:

Overview of historical development
Definitional and compositional analysis
Ethical & Social, Legal, Economic and Political impact and sector-specific solutions
The regulatory way forward

This literature review is a first step in understanding the existing paradigms and debates around AI before narrowing the focus to more specific applications and subsequently, policy-recommendations.

Download the full literature review

For more details visit https://cis-india.org/internet-governance/blog/artificial-intelligence-literature-review

Amazon launches Machine Learning-based platform for healthcare space

Admin — 2018-12-03T00:23:06Z

Amazon’s Comprehend Medical platform uses a new HIPAA-eligible machine learning service to process unstructured medical text and information such as dosages, symptoms and signs, and patient diagnosis.

The article by Kul Bhushan was published in the Hindustan Times on November 28, 2018.

With an objective to push deeper into the health space, Amazon has introduced a new Machine Learning (ML) software to analyse medical records for better treatments of patients and reduce overall expenditure.

Unveiled at the company’s re:Invent cloud conference in Las Vegas, Amazon’s Comprehend Medical platform uses a new “HIPAA-eligible machine learning service that allows developers to process unstructured medical text and identify information such as patient diagnosis, treatments, dosages, symptoms and signs, and more.”

“Comprehend Medical helps health care providers, insurers, researchers, and clinical trial investigators as well as health care IT, biotech, and pharmaceutical companies to improve clinical decision support, streamline revenue cycle and clinical trials management, and better address data privacy and protected health information (PHI) requirements,” explains the company on its website.

Amazon aims to mitigate the time spent on manually analysing medical data of a patient. The company hopes the software will ultimately empower users to make a more informed decision about their health and even things like scheduling care visits.

“Unlocking this information from medical language makes a variety of common medical use cases easier and cost-effective, including: clinical decision support (e.g., getting a historical snapshot of a patient’s medical history), revenue cycle management (e.g., simplifying the time-intensive manual process of data entry), clinical trial management (e.g., by identifying and recruiting patients with certain attributes into clinical trials), building population health platforms, and helping address (PHI) requirements (e.g., for privacy and security assurance.),” the company added.

Amazon also pointed out that some of the medical institutes such as Seattle’s Fred Hutchinson Cancer Research Center and Roche Diagnostics have already implemented the software.

Amazon’s expansion into the healthcare space comes after it acquired health-focused startup PillPack for $1 billion earlier this year. Apart from Amazon, other technology companies like Apple and Microsoft are investing into the healthcare space.

Apple is already offering HealthKit and CareKit platforms to develop apps focused on health. The company earlier this year launched Apple Watch Series 4 with ECG support. Microsoft, however, has deeper footprints in the health segment. The company is building a bunch of Artificial Intelligence-based tools for healthcare.

For instance, Microsoft’s Project InnerEye uses machine learning technology to build tools for automatic, quantitative analysis of three-dimensional radiological images.

According to various reports, Artificial Intelligence is going to make a big impact in the healthcare industry. An Accenture report in 2017 predicted that the AI apps can create $150 billion in annual savings for the United States alone.

Back in India, the adoption of AI in healthcare is growing. According to a report by the Centre for Internet and Society India, “the use of AI in healthcare in India is increasing with new startups and large ICT companies offering AI solutions for healthcare challenges in the country.”

Bengalure-based startup mfine has developed an AI-based healthcare platform which learns medical standards and protocols and diagnosis and treatment methods to further help the doctors with necessary data and analysis. The company earlier this year raised $4.2 million in funding.

For more details visit https://cis-india.org/internet-governance/news/hindustan-times-november-28-2018-kul-bhushan-amazon-launches-machine-learning-based-platform-for-healthcare-space

AI Opera- AI as a total work of art

Admin — 2019-10-14T14:30:56Z

On October 11, 2019, Shweta Mohandas and Mira were invited as panelists for the 'AI Opera- AI as a total work of art' event organized by Goethe as part of the India Week Hamburg 2019 held in Bangalore. CIS was an event partner.

The panel had to present different perspectives and possibilities of Artificial Intelligence (AI). The discussion was facilitated by German artist, performer and filmmaker Christoph Faulhaber. For more info, click here.

For more details visit https://cis-india.org/internet-governance/news/ai-opera-ai-as-a-total-work-of-art

AI in the Future of Work

ambika — 2021-12-07T01:51:42Z

Artificial Intelligence and allied technologies form part of what is being called the fourth Industrial Revolution.

Some analysts project the loss of jobs as AI replaces humans, especially in job roles that consist of repetitive tasks that are easier to automate. Another prediction is that AI, as preceding technologies, will enhance and complement human capability, rather than replacing it at large scales. AI at the workplace includes a wide range of technologies, from machine-to-machine interactions on the factory floor, to automated decision-making systems.

Studying the Platform Economy

The platform economy, in particular, is dependent on AI in the design of aggregator platforms that form a two-way market between customers and workers. Platforms deploy AI at a number of different stages, from recruitment to assignment of tasks to workers. AI systems often reflect existing social biases, as they are built using biased datasets, and by non-diverse teams that are not attuned to such biases. This has been the case in the platform economy as well, where biased systems impact the ability of marginalised workers to access opportunities. To take an example, Amazon’s algorithm to filter workers’ resumes was biased against women because it was trained on 10 years of hiring data, and ended up reflecting the underrepresentation of women in the tech industry. That is not to say that algorithms introduce biases where they didn’t exist earlier, but that they take existing biases and hard code them into systems in a systematic and predictable manner.

Biases are made even more explicit in marketplace platforms, that allow employers to review workers’ profiles and skills for a fee. In a study of platforms offering home-based services in India, we found that marketplace platforms offer filtering mechanisms which allow employers to filter workers by demographic characteristics such as gender, age, religion, and in one case, caste (the research publication is forthcoming). The design of the platform itself, in this case, encourages and enables discrimination of workers. One of the leading platforms in India had ‘Hindu maid’ and ‘Hindu cook’ as its top search term, reflecting the ways in which employers from the dominant religion are encouraged to discriminate against workers from minority religions in the Indian platform economy.

Another source of bias in the platform economy are rating and pricing systems, which can reduce the quality and quantum of work offered to marginalised workers. Rating systems exist across platform types - those that offer on-demand or location-based work, microwork platforms, and marketplace platforms. They allow customers and employers to rate workers on a scale, and are most often one-way feedback systems to review a worker’s performance (as our forthcoming research discusses, we found very few examples of feedback loops that also allow workers to rate employers). Rating systems have been found to be a source of anxiety for workers, as they can be rated poorly for unfair reasons, including their demographic characteristics. Most platforms penalise workers for poor ratings, and may even stop them from accessing any tasks at all if their ratings fall below a certain threshold. Without adequate grievance redressal mechanisms that allow workers to contest poor ratings, rating systems are prone to reflect customer biases while appearing neutral. It is difficult to assess the level of such bias without companies releasing data comparing ratings of workers by their demographic characteristics, but it has been argued that there is ample evidence to believe that demographic characteristics will inevitably impact workers ratings due to widespread biases.

Searching for a Solution

It is clear that platform companies need to be pushed into solving for biases and making their systems more fair and non-discriminatory. Some companies, such as Amazon in the example above, have responded by suspending algorithms that are proven to be biased. However, this is a temporary fix, as companies rarely seek to drop such projects indefinitely. In the platform economy, where algorithms are central to the business model of companies, complete suspension is near impossible. Amazon also tried another quick fix - it altered the algorithm to respond neutrally to terms such as ‘woman’. This is a process known as debiasing the model, through which any biased connections (such as between the word ‘woman’ and downgrading) being made by the algorithm are explicitly removed. Another solution is diversifying or debiasing datasets. In this example, the algorithm could be fed a larger sample of resumes and decision-making logics from industries that have a higher representation of women.

Another set of solutions could be drawn from anti-discrimination law, which prohibit discrimination at the workplace. In India, anti-discrimination laws protect against wage inequality, as well as discrimination at the stage of recruitment for protected groups such as transgender persons. While it can be argued that biased rating systems lead to wage inequality, there are several barriers to applying anti-discrimination law for workers in the platform economy. One, most jurisdictions, including India, protect only employees from discrimination, not self-employed contractors. Another challenge is the lack of data to prove that rating or recruitment algorithms are discriminatory, without which legal recourse is impossible. Rosenblat et al. (2016) discuss these challenges in the context of the US, suggesting solutions such as addressing employment misclassification or modifying pleading requirements to bring platform workers under the protection of the law.

Feminist principles point to structural shifts that are required to ensure robust protections for workers. Analysing algorithmic systems from a feminist lens indicates several points in the design at which interventions must be focused to ensure impact. The teams designing algorithms need to be made more diverse, along with integrating an explicit focus on assessing the impact of systems at the stage of design. Companies need to be more transparent with their data, and encourage independent audits of their systems. Corporate and government actors must be held to account to fix broken AI systems.

Ambika Tandon is a Senior Researcher at the Centre for Internet & Society (CIS) in India, where she studies the intersections of gender and technology. She focuses on women’s work in the digital economy, and the impact of emerging technologies on social inequality. She is also interested in developing feminist methods for technology research. Ambika tweets at @AmbikaTandon.

The blog was originally published in the Organization for Ethical Source

For more details visit https://cis-india.org/raw/oes-ambika-tandon-ai-in-the-future-of-work

AI in Healthcare

Admin — 2019-09-19T16:15:24Z

The Center for Information Technology and Public Policy (CITAPP) and the International Institute of Information Technology Bangalore (IIITB) invited Radhika Radhakrishnan for a talk at IIIT-Bangalore on September 13, 2019.

In her talk, she critically questioned the dominant narrative of “AI for social good” that has been widely adopted by various stakeholders in India (including the private sector, non-profits, and the Indian State) from a feminist standpoint. Specific to healthcare in India, such a narrative has been employed towards solving development challenges (such as a shortage of medical practitioners in remote regions of the country) through the introduction of AI applications targeted towards the sick-poor. Through her research and fieldwork, she analysed the layers of expropriation and experimentation that come into play when AI technologies become a method of using 'diverse' bodies and medical records of the sick-poor as ‘data’ to train proprietary AI algorithms at a low cost in the absence of effective State regulatory mechanisms. She argued that structural challenges (such as lack of incentives for medical practitioners to join public healthcare) get reframed into opportunities to substitute labour (people) by capital (technology) through innovation of “spectacular technologies” such as AI. Throughout the talk, she also highlighted the methodologies she used to conduct this research.

For more details visit https://cis-india.org/internet-governance/news/ai-in-healthcare

AI for Social Good Summit

Admin — 2018-12-25T01:02:01Z

Arindrajit Basu was a speaker at the event co-organized by Google AI and United Nations ESCAP on December 13, 2018 in Bangkok, Thailand.

Arindrajit spoke at the panel " How can governments use AI in Public Service Delivery" along with Malavika Jayaram, Jake Lucci,Punit Shukla,Simon Schmooly and Gal Oren. He presented CIS research on AI in agriculture in Karnataka-which will be published as part of a compendium documenting case studies worldwide soon.

Click to read more

For more details visit https://cis-india.org/internet-governance/news/unescap-and-google-ai-december-13-bangkok-ai-for-social-good-summit