With the need to manage efficiency, liquidity and financial risks, treasurers are always on the lookout for new solutions to meet such objectives. Banks now understand the power of automation and digitization to help treasurers attain these objectives; possessing a clear idea as to how it can help with their needs for control, visibility and centralization. Most Islamic banks, previously not known for their technological agility, are successfully adopting transformational programmes to become industry leaders in the digital era. Pioneering technology has allowed them to develop new propositions in the areas of payments, collection, liquidity management, FX, trade finance and other areas of the working capital cycle to name a few. The landscape has become extraordinarily demanding and competitive for banks with the digital revolution that is taking place. This has resulted in an increased number of solutions and unprecedented innovation, with FinTechs and companies seeking to benefit from new technologies such as blockchain, Big Data, application programming interfaces and process automation.
Use of digital services has become inevitable for the financial sector to enhance its coverage, especially when governments are promoting financial inclusion. Islamic social finance has the potential to serve the financial inclusion agenda in a complementing and effective manner. It is, therefore, important that potential applications of digital technology in Islamic social finance are considered and discussed with care – this chapter does the needful by focusing on potential applications of digital technology on zakat and awqaf, in particular.
Digital Technology for Zakat and Awqaf
One of the things that technology may address and resolve relates to the identity of the donors (muzakki/waqif) and beneficiaries. Zakat may be paid out by an individual muzakki (zakat payer) directly to an individual mustahiq (zakat beneficiary). This is the practice in countries where there is no organized or institutional mobilization and disbursement of zakat. Arguably, organized zakat management has many positives. Available evidence suggests that organized zakat management may be more efficient in identifying and distributing zakat among eligible beneficiaries. At the same time, institutional zakat management does leave room for some individual discretion and direct distribution. Moving from individual to institutional management of zakat brings in the first level of efficiency. Indeed, most Muslim societies either already have or seek a zakat infrastructure to ensure institutional zakat management. Moving from physical distribution of goods to cash may take us to the second and higher efficiency level.
1. Rationale for Digital Distribution
Paying in hard cash is simple. It appears familiar. However, dealing with hard cash has its downsides. For instance, it requires a system to handle it because of its size and volume. It is also unsafe. There is always a risk of robbery and internal leakages. It is also hard to account for since immoral insiders can be hard to monitor.
It is not hard to see that paying through digital format has decisive advantages over hard cash distribution. Digital payments involve user authentication. This makes it possible to ensure that cash reaches the intended beneficiary. A digital environment also ensures the integrity of data relating to the beneficiaries. Thus, there are built-in mechanisms that protects against theft and leakages. Every single payment can be accounted for or tracked, minimizing the possibility of corruption. Furthermore, every single payment can be properly authorized.
What are the operational requirements for a successful digital zakat payment program? Essentially, the program has to use infrastructure that already exists on the ground. For example, the required infrastructure would consist of bank servers for accounting, mobile phones for notifications, ATMs, or banking agents for cash withdrawal, so on and so forth. Assuming that such an infrastructure exists, a digital zakat payment program will have distinct benefits over a physical cash disbursement program.
However, there is another factor other than infrastructure that needs to be part of the equation. It is about mustahiqoon or beneficiaries themselves. Are beneficiaries familiar with and ready to take up digital cash? The bank needs to ensure that the beneficiaries, who are mostly illiterate and poor, are taken care of with the fulfillment of the following conditions:
- Do the mustahiqoon or beneficiaries understand the process?
- Is there a cost/fee for withdrawing their funds?
- Is the merchant charging a fair price when good are bought with the card?
- Do they have problems using the hardware?
- Can a beneficiary use the card/PIN by her/himself? Are there helpers around to explain the process to them?
- Are certain groups (e.g., disabled, or the elderly) particularly disadvantaged allowed to use the technology?
- Is there some help out there whom they can turn to if they have a problem and are they treated fairly?
- Is there an increase or decrease of security threats to participants, staff, and partners
- with the introduction of digital zakat transfers?
- Has control and access of zakat transfers improved or made worse by the introduction of new technologies?
- If the beneficiaries are not familiar with and ready to take up digital cash, the zakat body needs to factor in significant awareness and customer support efforts and related costs.
What if there is no appropriate digital infrastructure in the area such as financial service providers, ATM or agent in the ground, mobile phone coverage, and ownership. In that case, a digital zakat distribution program would factor in high setup costs.
The ultimate objective of the zakat distribution program is to enable the people in distress to buy goods and services they need. It makes little difference if liquidity takes the form of hard cash or digital cash. Notwithstanding the hype around digitization, the required infrastructure and the beneficiaries’ readiness determine the program’s success. Once technology requirements, beneficiary readiness, and operational implications are fulfilled digitization will undoubtedly enhance efficiency and make sense.
Another dimension of digital zakat distribution that needs to be underlined is human dignity. Where the traditional distribution methods require queuing up by participants, physical appearance and waiting at the place of distribution, and similar conditions, receipt of zakat may involve compromising one’s self-respect and dignity. For a self-respecting poor, a digital receipt or online credit will be a far welcome proposition. Indeed, this intends to protect the dignity of the individual (nafs), one of the five objectives (Maqasid) of the Shari’a.
2. Rationale for Digital Identity
In principle, before the distribution of zakat, an inquiry should be launched to ascertain whether the beneficiaries belong to one of the eight asnaf or categories (e.g., poor, needy, indebted, in-bondage) Shari’a had outlined for eligibility to claim zakat. Additionally, if the purpose of the zakat program is poverty alleviation or economic transformation of the recipient of zakat from mustahiq to muzakki (zakat recipient to zakat payer), they need to be monitored. With the digital payment of zakat, such monitoring of individual beneficiaries can be performed efficiently and effectively. However, this procedure of zakat disbursement may not be relevant in the context of humanitarian crises when zakat is distributed to meet an emergency.
The beneficiaries in this context should only qualify as eligible from Shari’a point of view. It is essential for the beneficiary to have a distinct identity for the digital payment of zakat. This identity shall be provided to the digital payment provider in order to claim digital zakat. Digital zakat payment then essentially is about assigning a monetary value to a digital identity. While digitizing the notion of a hundred dollars into zeros and ones is easy; giving zakat beneficiaries the means to unfailingly prove their ownership over that money (i.e., their digital identity) can be far more challenging with diverse mechanisms in place. It may be as simple as a random number/ token assigned to an individual where the identity need not be linked to anything intrinsic about the individual. It is not a tokenized form of his/her “real” identity.
A different form of identity linked to “real” identity is the Know Your Customer (KYC) requirements imposed by regulators on the various service providers. The KYC requirements comprise a set of verified personal information that service providers must ask for and hold about their clients. The information may vary widely and include combinations of name, address, date of birth, profession, education level, marital status, so on, and so forth. Since zakat payments involve financial flows, the authorities insist on such information to control money laundering, engaging in illicit transactions, and tax evasion. At a higher level, digital identity may require the addition of biometric information relating to an individual.
a) Identity and the SDGs
It is interesting to note here that legal identity and identification are also parts of the Sustainable Development Goals (SDG) agenda. As one of the proposed SDG targets (#16), it is also seen as a critical enabler of the efficacy of many other SDG targets. The SDGs underline the need to ensure that all individuals have free or low-cost access to widely accepted, robust identity credentials. While legal identity is frequently associated with a specific national credential, such as a National Identity Card, the appropriate concept in the context of the SDGs is broader than this. Not all countries issue National Identity Cards, and national status is not an essential component of identity as relevant for all SDGs. Indeed, there is a view that such legal identity can be provided far more quickly and widely by not requiring a prior determination of nationality, which can at times be a complex process.
It may be noted that robust identification is instrumental to achieving many of the SDGs. It directly relates to at least ten clusters: social protection floors, including for the most vulnerable; assistance in dealing with shocks and disasters; equal rights to economic resources including property and finance; the specific empowerment of women in this area; improvements in maternal and child health; coverage by vaccines and similar treatments; improving energy efficiency and eliminating harmful energy subsidies; child protection including the ending of harmful child labor; reducing the costs of making remittances; reducing corruption, fighting crime and terrorism and strengthening and improving the equity of fiscal policy. Most of these goals and targets may be directly mapped to the goals (Maqasid) of the Shari’a for which zakat funds may be utilized.
Robust identification is instrumental to achieving many of the SDGs
While identity may not be the most important requirement for pushing forward a zakat-funded development agenda, the SDG-related targets are made far more difficult if there are no robust identification means. At the same time, sound zakat management requires a delicate balance between requiring identity credentials as a condition to receive zakat versus creating an additional exclusionary barrier. Overzealous or inflexible ID requirements may sometimes block individuals from accessing the benefits.
b) Digital Identity for IDPs
Legal identity as a precondition for receiving zakat assistance can be difficult for the poorest of the poor and elderly, and for those affected by natural calamities (e.g., flood inundation) and the internally displaced due to conflicts and war situations. Moreover, most of the target beneficiaries may not even have legal documents to establish their identities. It is therefore, important that digital identities in such cases are assigned without linking the same to anything intrinsic about them. There have been some recent experiments in this regard using blockchain technology.
In a pilot program implemented with a blockchain application, developed by AidTech along with the global innovation team of the International Federation of Red Cross and Red Crescent Societies, provided digital identity to Syrian refugees in the form of smart cards with unique QR codes. Each card entitled an individual to buy goods from partner merchant/vendors worth a certain value. It was also possible to replace smart cards with mobile devices. The technology enabled monitoring in real-time and with complete transparency about all the transactions online. In short, it ensured the delivery of digital entitlements via digital identity with complete transparency and accountability of the distribution of resources. This experiment and pilot study can be replicated by other zakat organizations in similar complex humanitarian settings. The use of unique digital identity may allow the zakat organizations to document information such as eligibility, entitlement packages, and automated periodic payments. It may enable them to transparently track resources distributed throughout the process with a high degree of accountability.
c) Ethical Concerns with Digital Identity
Identity brings many benefits to the process of zakat distribution. Lack of identity or anonymity on the part of the beneficiaries is certainly not desirable. It should be evident that a digital environment makes it possible for the zakat organization to collect many types of personal data. Zakat organizations may collect a range of data like personally identifiable information, e.g., names, phone numbers, and bank record details. Such data should be the bare minimum for the organization to target assistance and ensure that the correct participant receives such assistance.
Usually, the scope of such beneficiary data is enormous with digital zakat, where such data needs to be shared with third-party service providers, such as mobile phone companies and financial service providers, to execute the transfer of cash. This raises some concerns about the privacy of data. Furthermore, it is essential to realize that other actors may be interested in such data, e.g., persons/ groups. Some of these may indeed be hostile to the intended beneficiaries and target them for extortion and violence. Therefore, adequate care needs to be exercised about possible data breaches. In addition, the personally identifiable information (PII) itself should be minimized.
Zakat organizations cannot close their eyes to privacy issues of personally identifiable information relating to beneficiaries. Therefore, they should have some mechanism in place to determine which information could be shared with whom. They should also define regulations applied to the data and its compliance with the legally established privacy principles of the relevant jurisdictions (in addition to national regulations like KYC, regional or international agreements may also apply). In addition, the zakat organization may develop its privacy principles and ethical guidelines.
3. Sanctity of Waqf Records
The institution of waqf or awqaf, as discussed before, implies holding or setting aside certain assets by the donor (waqif) and preserving it so that benefits continuously flow to a specified group of beneficiaries or community. The nature and purpose of the waqf should be clearly stated in the waqf deed or document by the donor (waqif). The donor also specifies the trustee-manager(s) to ensure that the intended benefits materialize and flow to the community. The trustee-manager is variously described as mutawalli or nazir. The waqf deed provides the method of compensation of the trustee-manager, usually a part of the earnings or benefits from the assets under waqf. Thus, the building block of waqf is the deed that is a written record of the transactional process and the relationships. Since the stated intentions of the endower (waqif) are immutable and binding on all stakeholders, therefore, the sanctity of the waqf deed assumes great significance.
In recent times, with a growing concern regarding recovery of lost awqaf properties across the globe due to encroachment, often by the state or its agencies, powerful corporates, and individuals, there have been ambitious initiatives to create computerized central databases on waqf deeds/ assets. However, blockchain technology has immense potential for the awqaf sector as a new technology that can ensure far superior outcomes compared to centralized databases. While blockchain employs various modern technology and security steps absent in a written text, it is essentially a ledger or a record. Like a written record, it has chapters – or blocks – of information, each added sequentially over time. Experts note that the blockchain fundamentally differs from traditional databases or computerized/ manual records. It differs in two distinct ways.
First, the blockchain is a shared record. We are no longer talking about centrally controlled and updated records, whether written documents by individuals or digital files owned by database administrators. In these cases, a centralized authority would govern the records. By contrast, the blockchain is a distributed record. No single participant owns the blockchain or dictates additions to it. Instead, updates are a function of consensus amongst participants. In the context of awqaf, it is perhaps an extremely ideal scenario that the society as a whole or the waqifs (and designated individuals/ bodies, e.g., Islamic scholars/ jurists/ waqf-entities) collectively own the blockchain.
In some cases, the blockchain may be “public” with complete access to waqf records given to every member of the society. On the other hand, the blockchain may also be “private” with restricted access for specific parties with the ability to “modify” or “update” records by consensus. It would be a more desirable scenario where participants (society) by consensus decide to modify the intended use or develop the asset or replace the asset if considered desirable in societal interest, as compared to one where a waqf board/ waqf commissioner implements the changes.
Second, the blockchain is immutable. It stores a history of itself back to the first entry, known as the genesis block. The identity of each new entry is created, in part, from the identity of the previous entry. Changing its content or identity is essentially impossible because every individual block is inextricably linked to all that precedes it. It is this feature of the blockchain that makes it most suitable for awqaf. A waqf, by definition, is inalienable, irreversible (for most Islamic scholars with some recent exceptions), and perpetual dedication of a privately owned asset for a public purpose by the owner. Unfortunately, throughout the history of Islamic societies, there have been cases of abuse, misuse, and usurpation of waqf assets, tampering, and destruction of waqf deeds by bad actors (which unfortunately included public or state agencies as well as private entities). A search for solutions to this intractable problem has eluded societies. Given the immutability of the blockchain, which brings in its unprecedented security in the form of tamper-proof records, impermeable to incursions by bad actors, we may just have hit upon a solution.
As an append-only database, blocks cannot be changed once committed to the blockchain. Instead, the blockchain only changes by the addition of new blocks. Thus, it appears most suitable for creating waqf-mushtarak or the use of new waqf resources for the development of existing waqf assets. It must be noted that awqaf have a business face too. The waqf resources must be invested in the best possible way to maximize returns, which may then be directed at the intended beneficiaries. Thus, the blockchain database has enormous potential for corporate applications.
Experts note that the blockchain eliminates the need for an intermediary third party, such as a state agency (often lacking credibility). Both transacting parties can trust that the information added to the blockchain cannot and will not be changed. Large waqf-based organizations could directly interact with each other, writing their contracts with no need to involve third parties or any other intermediary to assert correctness. Blockchain also allows participants to reach a consensus or settle a transaction quickly. Multi-day processes channelled through intermediaries are reduced to minutes, thereby enhancing the efficiency of waqf-based organizations.
Experts note that for shared records such as waqf deeds or contracts, the blockchain fundamentally transforms ownership, transparency, security, and consequently, the value of the records and the process they govern. In the context of a shared record or contract, blockchain reframes the concept of trust. The blockchain lets people (or companies) who have no particular confidence in each other collaborate without going through a neutral central authority. As a machine for creating trust, the blockchain could solve the problem of the massive trust deficit in the waqf sector. Indeed, the importance of trust can hardly be overemphasized in the context of the development of waqf assets, especially where it involves infusion of private capital. Often the trust deficit is linked to society’s concern for the preservation of waqf assets. For example, in the face of large-scale encroachment of waqf assets in India by state agencies, most proposals to develop a particular waqf property and transform the same into productive community assets have met with stiff resistance by community leaders. Often the contracts that go with waqf management evoke a trust deficit. For example, waqf assets are often leased (if not sold or offered as collateral) at grossly below-market rates or for a near-perpetual lease term to the lessee(s). While waqf laws seek to minimize such possibilities through provisions, e.g., the penalty for the mutawalli, these have generally been ineffective in most jurisdictions, aggravating the problem of trust deficit. The blockchain potentially offers a solution in the form of smart contracts.
A “smart ijara” or operating lease contract– that uses the blockchain and automates the periodic payment streams as well as the reversion of leased assets to the waqf at the end of the lease period – could be a self-paying and self-executing instrument
A smart contract is a computerized transaction protocol that executes the terms of a contract. It purports to, among other things, satisfy common contractual conditions (such as payment terms, liens, confidentiality, enforcement), minimize exceptions both malicious and accidental, and minimize the need for trusted intermediaries. Related economic goals include lowering fraud loss, arbitrations and enforcement costs, and other transaction costs. For example, the Islamic lease contracts can now take the form of self-executing digital or smart contracts with “electronically coded” terms. The contractual terms will execute only if the conditions are met. This will automate the entire contractual process for Islamic institutions. In addition, the Islamic contracts will now be easy to verify, immutable, and secure, mitigating operational risks arising from settlement and counterparty risks as well as administrative and legal complexities and redundancies. Thus, a “smart ijara” or operating lease contract – that uses the blockchain and automates the periodic payment streams as well as reversion of leased assets to the waqf at the end of the lease period – will now be a self-paying and self-executing instrument. Indeed, the most effective application of blockchain is to ensure the sanctity of the waqf deed and ensure that the conditions stipulated by the waqif are met most transparently. Thus, as rightly called a “trust machine,” it can effectively address the “trust-deficit” constraint that faces the global awqaf sector.
AI Applications in Zakat and Awqaf
A digital and technology-enabled environment for zakat and waqf opens abundant new opportunities and possibilities for applying artificial intelligence (AI). As is explained below, AI can enhance the efficiency of the processes along the value chain in the IsSF sector.
Artificial Intelligence is the use of computers to mimic the cognitive functions of humans. Humans can see, listen, relate, analyze, compute, and make decisions. A computer system that does these things is called artificially intelligent. So, AI is about machines that can perform tasks typical of human intelligence. Computers need data that could be anything ranging from words, texts, images, and gestures – anything that conveys information. When one seeks to relate data, analyze data, compute, and make decisions, one uses algorithms.
1. Machine Learning (ML) Algorithms
A fundamental building block in AI is the algorithm, which means a set/list of rules to solve a problem. One needs a code to tell a computer what to do. However, before one writes a code, one needs an algorithm. An algorithm for a simple zakat liability estimation of an individual may appear like this:
- Get the values for Zakatable Assets (ZA) – Gold (G), Silver (S), Cash (C), Receivables (R), Investments in Zakatable Assets (I), Business Stock (B) – and Deductible Liabilities (DL)
- Get the current market price (P) of gold (say per 1kg)
- Calculate Nisab (NB) = P*0.085
- Find Sum of all Zakatable Assets [ZA= G+S+C+R+I+B]
- Find Net Zakatable Assets (NZA) = ZA-DL
- Compare NZA with NB;
- If NZA < NB, Zakat Payable = 0;
- If NZA >= NB, Zakat Payable = NZA X 0.025
There is little for the algorithm to do when all data are given other than to compute the result. However, there is no single answer once one seeks to predict the zakat liability of the individual into the future. Now all the inputs will be the predicted values of zakatable assets and liabilities.
The benchmark Nisab will also depend on the future value of gold. There is now the need for another algorithm to predict the future value of gold (e.g., exploring a pattern in the historical values of gold and extrapolating the same into the future). Alternatively, when some dynamic decision criteria are introduced (e.g., an individual’s intention to liquidate his/her investment or hold long-term), the zakatable investments will take different values. One would then be stepping into the domain of machine learning (ML).
Machine learning is a way of achieving AI without being explicitly programmed. Without machine learning, AI would require building millions of lines of codes with complex rules and decision trees. So instead of hard coding, machine learning is a way of “training” an algorithm to understand the logic and produce the results. Training involves feeding vast amounts of data to the algorithm and allowing it to adjust itself and improve. So these are self-reliant algorithms.
Machine learning outcomes can be of two kinds. First, it can be predictions about things that are not yet known but can shed some light on existing data. For example, what will be the price of gold next month? There is no single answer. Machine learning outcomes can also be about finding patterns in the data that are not entirely obvious because they are implicit or probabilistic. For instance, Twitter responses of visitors to the popular zakat crowd fund portal in Indonesia, which are just a bunch of texts can be analyzed to see a pattern hidden in such responses? Is the crowd fund providing for what it takes a visitor to turn into a donor? What are the chances that a visitor will pay zakat online through the portal or turn away? Now the machine needs to infer the idea of a “satisfied” visitor from a bunch of textual data.
However, AI is not just about predicting or classifying. AI is a much broader concept than machine learning. First, there are those particular categories of machine learning algorithms that carry the tag of deep learning. In addition to deep learning, other approaches to machine learning include decision tree learning, inductive logic programming, clustering, reinforcement learning, and Bayesian networks, among others.
Deep learning was inspired by the structure and function of the brain, namely the interconnecting of many neurons. Artificial Neural Networks (ANNs) are algorithms that mimic the biological structure of the brain. Deep learning models try to simulate more closely how our human brain works.
a) Prediction and Classification with AI/ML
In AI, the machines are expected to interact the way humans do. For example, when confronted with an individual, seeking a small sadaqa donation or a qard (loan), a potential donor or lender usually would like to ask a few questions about his/her financial conditions, make him speak, seek clues in his dress, countenance, and facial expression. Same data – pixels in an image, text in speech, numbers can all be used by a machine/computer. However, compared to humans – the ordinary individuals – a manager with a Qard Fund will perhaps seek more systematic data. Microfinance institutions (MFIs) insist that the client must perform salat (mandatory prayer) at the local mosque. With computer vision now fast becoming a part of good mosque management, data on mosque attendance should be available for the MFI and the machine. Additional data can be made available in utility bills, grocery purchases, type of mobile users so on and so forth. Indeed, there is data everywhere.
“AI can tell us whether or not theindividual belongs to one of the eight categories of mustahiq or person eligible to receive zakat in the eyes of Shari’a”
The AI-driven machine can then be trained on such data to enhance its intelligence and go on to classify the individual as conforming to the criteria of an individual in genuine need or not. For instance, AI can tell us whether or not the individual belongs to one of the eight categories of mustahiq or person eligible to receive zakat in the eyes of Shari’a. Moreover, AI can predict, should it be a case of qard, the probability that the borrower may default or delay repayment. AI can classify the default as a genuine or a wilful one. In the former case, the defaulter perhaps needs additional help. The latter case would call for a penalty.
There are innumerable ways in which AI can help humans analyze better, predict better and classify better. However, what needs to be underlined is that there is no known single correct answer in AI. One must accept the possibility of error. There are four possible outcomes. For instance, in the case of the above classification problem relating to mustahiq (zakat beneficiaries eligible according to Shari’a), the machine may end up classifying (i) a poor as a poor, (ii) a non-poor as a non-poor, (iii) a poor as non-poor and (iv) a non-poor as poor. In the first two cases, it would have done its job correctly. In the third case, it would deprive a genuine mustahiq of receiving zakat. In the fourth case, it would ensure a non-mustahiq to receive zakat. The error in the third case is perhaps more severe. Imagine a similar classification error in treating a Covid-19 patient – an infected patient who is not diagnosed as one and is not receiving the treatment! Fortunately, misclassification risks are far less harmful and fatal in the domain of Islamic finance than in healthcare. The computer is given a clear objective expressed as an optimization problem, such as minimizing the probability of error. Then data is provided to the computer, and it is required to optimize based on this data, which contain clues to solving the prediction/ classification problem.
- Types of ML
There are three broad types of ML – supervised, unsupervised, and reinforcement.
First, there is supervised learning, which is fast, accurate, and most commonly used ML. Machine learning takes data as input that is called “training data.” The training data includes both inputs and labels (targets). The data scientist trains the model with lots of training data (inputs & targets). Then with new data and the logic it gets, the output is predicted. Two types of problems are addressed through supervised learning – regression and classification. Classification separates the data. Regression fits the data.
Regression: This is a problem where one needs to predict the continuous-response value (a number that can vary from -infinity to +infinity). In the earlier zakat estimation example above, one needs to predict the prices of gold, silver, shares, value of inventories, agricultural produce, lease rental rates/ prices of houses and other assets, etc.
Classification: This problem predicts the absolute response value where the data can be separated into specific “classes” (one predicts one of the values in a set of values). Adding a few more to the earlier examples, one may note that the initial six classifications are binary (yes or no), while the last three are examples of multi-class classification.
- Is individual A in the poor (faqir) category of mustahiq or not?
- Is individual B in the traveler (ibn Sabeel) category of mustahiq or not?
- Are donors to Global Sadaqah (a crowdfunding platform) a satisfied lot or not?
- Will donors to Global Sadaqah (a crowdfunding platform) revert as repeat-donors or not?
- Is individual C going to default on his repayment of qard or not?
- Is this picture of Brother X or not?
Very high/ high/ moderate/ low/ very low
- What are the chances that a transaction by individual E will prove to be a fraudulent one?
- What are the chances that zakat beneficiary F receiving specific skills will be able to transform himself/ herself into a muzakki-entrepreneur?
- What are the chances that project D will utilize zakat funds in the Shari’a-stipulated way?
In unsupervised learning, the training data does not include targets. So, one does not tell the system where to go. The system has to understand itself from the data given. It has to understand patterns in the data itself. Here training data is not structured.
There are also different types for unsupervised learning, like clustering and anomaly detection. Clustering is a type of problem where one puts similar things together. It is similar to multi-class classification, but the system understands and clusters the data without providing the labels.
Some examples are:
- given questions or comments on a zakat portal, cluster them into payer-types
- given set of tweets on volunteering portal, cluster-based on the content of tweets
- given WhatsApp forwards in Islamic women groups, cluster senders into different types
Unsupervised learning is relatively more challenging to implement and not used as widely as supervised.
- Model Evaluation
A business user of machine learning algorithms can, of course, leave the technical details out. One does not have to build one’s own machine-learning model. Instead, one can seek the services of dedicated data scientists. As Islamic finance or social finance professionals, one may encounter machine learning models that do one or more of the above tasks. Once one is done asking what it does and why one should need it, the next set of questions ought to be: how good is it, should one trust it? The process of model evaluation is about finding how good the model’s predictions are.
Interestingly, one can find this by comparing what the model predicted with what one already knows but has not shared with the machine. For example, a manager from Akhuwat (qard fund) is interested in a model that can correctly classify its new beneficiaries into good and bad borrowers (from a default point of view). It has data for the past two decades on its beneficiaries that includes the minuscule population of defaulters. It knows who defaulted and who did not. Assuming that it shares data for the first twelve years (60 percent) of the machine to train the prediction model, this set is called the “training dataset.” Then the machine uses the next four years of data to choose between alternative models. This data set is called a “validation dataset.” Finally, it uses the last four years of data (20 percent) to find the optimal parameters of the model that minimize errors.
Now, one may argue that there is no business case for Akhuwat type organizations to use such prediction models. An organization that enjoys minimal defaults (around one percent) may not find the idea of employing complex prediction algorithms a good business proposition. However, the issue of cost versus benefits of such an exercise is undoubtedly relevant. It perhaps serves well those entities experiencing high credit default risk and do not know how to tackle it. At the same time, such models may offer alternative value propositions to Islamic microfinance institutions that may like to predict the outcomes of their skill-enhancement initiatives – predict potential winners among micro-entrepreneurs based on data that go beyond their education and apparent competencies.
There is merit in the idea that machine learning models make predictions based on data beyond what is traditionally used to assess creditworthiness (such as proof of income or employment) or entrepreneurial traits in individuals. Poor people are financially excluded because they are data excluded. ML models use “alternative data” that has no apparent relationship with the financial and business capabilities of the client, such as the number of contacts on one’s phone, the make of the phone, one’s average mobile top-up value, one’s online access patterns, indeed, anything relating to their digital footprint.
Apparently of low value, there are hundreds of such data types available to our machine to explore and find relationships with loan repayment or entrepreneurial success. At the same time, the use of “alternative data” raises serious ethical issues, especially from an Islamic point of view. Islamic societies place massive value on “privacy,” and for good reasons. Can the machines come up as winners in the face of this ethical constraint?
- Natural Language Processing (NLP)
One can break up natural language processing into several activities. First, the machine needs to be able to recognize speech when it is spoken to. That is converting language from its speech form to its textual form. Then the machine needs to extract meaning from that text to understand. Second, the machine needs to be able to articulate response as a string word put together. Third, the machine may need to turn that into spoken words to synthesize speech. AI applications may include one or more of the above processes.
AI applications take care of the first step, extracting the text out of audio or video recordings. Some applications go further and analyze text contained in tweets or messages or emails or responses on a portal. Such an exercise may be helpful, for instance, for a waqf organization to know what the neighbourhood feels about the public goods (education, healthcare, orphan care, etc.) it provides. An association of zakat organizations in a country concerned about a high rate of “donor attrition” may feel the need for a “feel-good analysis” of donors. A microfinance institution serving micro and small businesses may be interested in its clients’ “business sentiment analysis”. All these would require analysis of texts in the form of stakeholders’ responses in various channels, e.g., tweets, emails, comments, and queries at the portal.
Web-based chatbots go one step further. They can formulate a response to individual customer messages and post that back as an instant message. Automatic response systems of the type one find with a few Islamic banks are similar, except that they need to understand spoken requests in the first place. So, they need to incorporate a speech recognition element. There are, of course, some basic ones that will not talk back beyond playing some standard pre-recorded messages. Instead, they note what the customer issue is. There are, however, the Kikus, the Siris, and the Alexas that are fully interactive voice assistance systems that cover the full range of activities.
There is no doubt that the chatbots of tomorrow will be much smarter than what we have today. A distinct possibility is a robot-scholar or a robot-Shari’a-auditor or a robot-regulator, well-versed in Islamic law & economics and national laws and regulations. However, one has to be content with Robo-advisors in place for Shari’a-compliant investments and wealth management. Since investments are an integral component of awqaf, the mutawallis and nazirs should find value in the recommendations of robo-advisors.
- NLP with Computer Vision
NLP with computer vision may help address “donor attrition” risk. This is a risk that all charity and non-profit organizations face. It is a significant risk. It can significantly affect their ability to keep their programs funded. A global report on Fund-Raising Effectiveness6 underscored this risk in simple words: for every 100 new donors in a year, the non-profits lost existing donors! One may add to this the estimate that it costs non-profits about ten times more to bring in a new donor than to keep an existing donor. Addressing this major problem may require a multi-pronged action plan – develop and use donor analytics, get feedback, and reach out to the lapsed donors.
Now, imagine an existing donor walking into the premises of an Islamic non-profit organization. A computer placed at the reception can instantaneously recognize him, cross-checks its database of his past contributions, and identifies him as an individual with a high propensity to donate. The next instant, the computer warmly welcomes him, offering to serve him with info on the latest campaigns matching his interest or providing impact feedback on past campaigns he contributed to. Won’t this gesture influence his decision to donate again to the same organization? If he is a zakat payer, the machine can even help him counsel and estimate his zakat liability. A machine that sees listens, and talks can undoubtedly help the non-profit retain him as a donor and a continuous supporter of its programs.
Here is another scenario. A poor micro-entrepreneur is seeking a qard loan or micro-Murabaha financing from an Islamic MFI. Usually, she should be ready to visit the nearest branch of the organization with a plethora of documents for KYC compliance opening an account. However, when she connects to the branch seeking an appointment, she is told not to put herself into the inconvenience of a personal visit. The process of visual authentication and KYC is now possible remotely with computer vision. What is now required is this. First, she needs to send a photo of her ID card. The computer will pick out her facial image, name, and other written text on the card. Next, she takes a selfie with her mobile phone against the photo on the ID card. The computer will compare the facial features on her selfie photo with the photo on the card. Her authentication is complete.
In the above two examples, enhanced customer satisfaction is made possible, most certainly, because of hearing, talking, and seeing machines.
Ethical Issues in AI2
As discussed above, a machine can hear humans talk, talk back to them, see and recognize them. Once one adds the internet-of-things, it can do things as well. A natural question that arises now is: Can such an “intelligent” machine enter a contract in a legal sense? Can an “intelligent” machine be held “accountable” for its actions? For example, a nadhir of cash or corporate waqf liquidates 20 percent of its equity portfolio on a signal from the Robo-investment advisor, only to see the markets bouncing back and experiencing a significant climb upward. Before one seeks an answer to such questions, it is perhaps helpful to differentiate between three different levels of AI. First, there is narrow artificial intelligence (ANI), which does one thing at a time. For instance, the AI algorithm enables one to convert speech to text. Second, artificial general intelligence (AGI) can do everything humans can at the same level of mental abilities.
Moreover, finally, the dreaded zone of artificial super intelligence (ASI) dominates and is far superior to human intellect. Humans would have no clue as to what the machine is thinking. Humans are currently at ANI, while experts disagree on whether AI will soon reach higher levels. One can comfort the forecast that AI will firmly stay within human control and conveniently trash the forecast of some that machines will become incomprehensibly smarter than humans with ASI. However, in the event ASI materializes, it will be beyond human control. One can only hope that, before the inevitable happens, humans would have inculcated in the machines Islamic ethical and moral norms and values.
- System Accountability
Experts in the ethics of AI consider three levels of ethical behavior by the machine. First, AI has ethical constraints programmed into it. Second, AI weighs inputs in a given ethical framework to choose an action. At the highest level, AI makes ethical judgments and defends reasoning. It is relatively easy to see the first level in action. The Islamic investments Robo-advisor will not touch a pork-producing company with a stick! It knows wine and pornography are harams and beyond its reach. This is because of Shari’a’s constraints programmed into it. It will never permit investments into any projects that violate the conditions imposed by the nazir. If it is into zakat advisory, it will never “clear” a list of beneficiaries that include the non-poor (unless there are other defendable reasons to pay zakat to them). If it is to assess the performance of a nazir or mutawalli, it will raise a red flag over benefits flowing to projects that are not in conformity with intentions of the waqif.
While we are still within the domain of ANI, the task seems to be more accessible. However, machines cannot be penalized for the consequences of their actions. Even though this may encourage people to exercise enough caution while creating AI, system accountability is suggested. However, system accountability may yield good results if ensured through government regulations and industry standards requiring developer companies to subject algorithms to rigorous scrutiny for ethical questions that may be lurking around the corner.
- Big Data and Privacy Concerns
Any discussion about AI is incomplete without talking about big data. The narrative “data is new oil” is frequently used to underline that data has become the most critical resource of our times. As machines continue to gain intelligence by devouring more and more data – in the form of numbers, text, sound, images, and what-have-you – raise serious ethical questions that should concern an Islamic economist. Big data is about the massive volume of data. As our society gets more and more digitalized, data grows bigger and bigger. However, big data is not just about volume. It is also about the velocity of data. Unlike old days, social media apps now throw up massive amounts of data in real-time.
Furthermore, our data-devouring machines (e.g., robot traders) require fast feedback loops in order for the system to work. They need to sense what is coming to make real-time decisions constantly. One more distinct feature of big data is that it is mainly unstructured – text, images, and sounds – and the machines look for hidden patterns and signals in them.
Now, the first ethical question is: should big data be treated as a natural resource? Throughout human history, the privatization of natural resources – oil, coal, natural gas, forests and timber, minerals – have created large monopolies and contributed to massive wealth creation for a privileged few. A natural outcome of this is gross and ever-increasing economic inequalities. The idea of a monopoly is alien to Islamic economic ethics. However, a trend of cannibalizing small players and enhancing monopoly power should be discernible if we look around. A small number of tech giants have taken rapid strides in monopolizing this new resource. This demands legislative action. Further, unlike oil which gets depleted, the use of data creates new data. It is, therefore, never too late to curb the rise of tech- monopolies.
It is never too late to curb the rise of tech-monopolies
Second, the use of alternative data may raise serious privacy issues. In a fascinating 2019 publication, “On the Rise of FinTechs: Credit Scoring using Digital Footprints,” the authors find that easily accessible variables from the digital footprint – the device type, the operating system, and the email provider – a proxy for income, character, and reputation and are highly valuable for default prediction. It may be highly desirable to scrutinize whether such alternative data proxies for variables (e.g., race, religion) may lead to discriminatory behaviour on the lender. Regulators must also watch closely whether such alternative data violate individuals’ privacy rights. Both are of concern to an Islamic economist.
In conformity with the objective of Shari’a to protect the dignity of the individual (hifdh al-nafs), Islamic societies have always stood for the absolute privacy of an individual. The “purdah” system is a testimony to the same social and ethical norm that seeks to protect the dignity of women in Islam. On the possibility of sharing relevant information about the individual borrower (as in the credit-scoring model), one needs to be careful about a significant Islamic ethical Norman individual’s right to be protected against gheebah buhtan and nameema – that govern information-sharing.
Gheeba means information-sharing about a Muslim who is neither present nor approving, even when accurate. When data is inaccurate, sharing of information amounts to Bhutan. Nameema refers to the disclosure of data that may hurt the interests of the concerned party or lead to conflicts with a third party. Scholars assert that extreme caution should be exercised given the above norms while sharing data and information about others in general. Information-sharing may be undertaken only under specific conditions, e.g., when this is certain to bring some benefit to a Muslim or Ward off some harm. Shari’a, however, provides a window of permissibility for sharing personal data and information when it can potentially impact decisions relating to marriage, business, etc. At the same time, the data and information shared should be the minimum required to address the problem at hand.
Data privacy has been a subject of regulation that varies widely across countries. Perhaps instead of getting into the diverse range of regulations in practice, it will be helpful to underline some privacy-related guidelines and principles that do not violate Islamic societies’ fundamental beliefs and culture. They merit serious consideration by zakat and Islamic non-profit organizations, especially those that have gone digital, where the beneficiary is usually in a weak bargaining position.
- Organizations should respect the privacy of beneficiaries and recognize that obtaining and processing their data represents a potential threat to that privacy.
- Organizations should protect all personal data they obtain from beneficiaries either for their use or by third parties.
- Organizations should exercise extra care and sensitivity towards “purdah culture.”
- Among women in obtaining and processing their data.
- Organizations should ensure the accuracy of personal data and keep such information up to date.
- Organizations should obtain consent or inform beneficiaries as to the use of their data.
- Organizations should not hold beneficiary data for longer than what is required.
- Organizations should be accountable for holding the data and address any query/ complaint by a beneficiary regarding his/her data.
Finally, there is an ethical concern regarding possible bias in the data used for AI. If one uses partial data to train the machine learning AI, then one would get biased AI. Often the bias is unconscious or is based on deeply-rooted societal norms. For instance, it is common to notice a gender bias when all machine-voices talking to humans in lifts, cars, or virtual assistants on a portal are female voices by default. It is not hard to see how such biases in either direction can creep into an AI-based algorithm. Those concerned with and seeking to fight against highly destructive and pervasive societal biases will find the going increasingly difficult.