The recent tentative listing of a new Immigration and Foreigners Bill, 2025 during the Budget Session of India’s Parliament has set the cat amongst the pigeons. As is the norm, there has been animated commentary and judgment on the proposed Bill by people who know no more about its contents than anyone else. This article will not do that; instead, this is an opportune time to discuss the larger issue of illegal immigration and its impact on communal relations – and how to deal with this using Artificial Intelligence, linguistics and regulation.
Before you roll your eyes about yet another article purporting to use the mythical powers of AI for something as complicated as Immigration, allow me to make the case first.
Prior to discussing possible solutions, it is important to examine two common claims that are made with regard to illegal migration.
The first claim is that there exists a moral duty for nations to treat illegal and legal migration equally, that we must help all as it is a human right. Now, at first glance, this idea seems extremely virtuous. After all, it sounds heartless to suggest that we should not help the needy. Yet, there are two issues with this argument, as principally ideal as it might appear.
For one, it is subject to the fallacy of composition, i.e. the wrong assumption that what is true for part of a group is true for the entire group. Consider the questions: How many are truly poor and needy? What are they employed as? Where do they live? How many are involved in criminal activities? How committed are they to Constitutional values? How many believe in harmonious multi-cultural and multi-religious living?
The answer is that we largely do not know – some may be, others may not. We are unlikely to find out either because an illegal migrant is not going to announce this to society but will seek to present themselves as a citizen or legal resident. So, the fallacy of composition undoubtedly exists, and it is not reasonable to assume bonafides in the complete absence of evidence/data, implying that there is a risk that must be managed.
This is not just a theoretical quibble: Yazidi women seeking refuge in Germany have come across their former captors from Iraq and Syria living there as well. One of the men involved in assassinating Sheikh Mujibur Rahman of Bangladesh was found to have been married to an unsuspecting Indian woman for almost a decade – and living in West Bengal for even longer.
How do we manage such risk in our lives?
This brings out the second issue with the moral argument: inconsistency across domains. Looking at it from first principles, it is clear that we don’t accept such a moral prerogative anywhere else in life. For example, how many individuals would be willing to fund an unseen stranger’s child’s education and living expenses the same way they would their own? Or let them reside in your house?
How many would like the financial system to extend loans without due diligence? How many would be happy to open an account with a bank that does this? Logically, we know that the lack of assessment doesn’t necessarily mean that the stranger’s child or unexamined loan recipient is undeserving of assistance – they might also need a livelihood and so on. Yet, without a (formal or informal) risk assessment, we are not willing to take on this risk at the personal and corporate level because it is not sensible. Why, then do we have different standards only at the national level?
And yes, knowing a person or donating to a charity is still a form of risk assessment. In the first case, you are still informally and heuristically assessing whether providing help is a worthwhile risk to take. Similarly, in the case of a charity, we are simply moving the burden of risk assessment onto an intermediary who we assume will put in the effort to identify deserving beneficiaries on our behalf.
So, we accept that in a world with constrained resources, we need to assess and manage risk appropriately. Unlike legal migration, where that risk is assessed, illegal migration comes with unknown and undefined costs and risks for the nation.
So, there are good logical reasons for someone to be in favour of legal migration but not illegal migration – and unlike what is claimed, religious bigotry or xenophobia do not have to be remotely involved.
Are risks overstated? – Effects on Communal Harmony
The second claim is that the effects of illegal migration are overstated as they are usually misidentified. We see this all the time when the issue of illegal immigration-associated conflict emerges: The media reports that one side feels aggrieved by the increasing presence of illegal migrants while the other convinces the reporter that they are indeed Indian citizens.
In fact, if we trust the media narrative and assume both sides are being truthful – that genuine citizens are being misidentified as illegal residents leading to conflict – it actually demonstrates the very harmful consequences of illegal migration on communal relations between genuine citizens.
It shows that the large migration of illegal residents (estimated at 12 million in 2004 during the UPA-1 government and at 20 million under the NDA in 2016), who may be claiming to be legitimate residents of (mostly) eastern Indian states, has created suspicion, tension and mistrust between communities because people no longer know what to believe or trust about each other’s nationality and origins. In turn, this becomes a vicious cycle as communal tensions are reinforced and patriotic citizens turn on each other in an atmosphere of distrust. For this reason, it must be a concern for everyone regardless of ideological belief – not just ‘conservatives’ or ‘nationalists’.
Modern Solutions for a Gordian Problem – Risk Management
If one wants to cut the Gordian knot, new solutions are needed that are high-throughput, reliable and minimise inconvenience to citizens.
Artificial Intelligence: AI has shown immense promise in pattern-recognition, and the launch of DeepSeek suggests that this may even be done much more cheaply than thought. India has also sought sovereign AI and has also created Bhashini, which aims for real-time translation between Indian languages.
Sovereign AI can, therefore, be trained on subtle speech, pronunciation and language patterns, based on both internal and publicly-available data, to identify which region a person comes from as they usually have characteristic linguistic differences. For example, Bangla spoken in Chittagong is not the same as that in Sylhet, or Dhaka, or Kolkata, or Cooch Behar. Similarly, in Bihar, Magahi, Maithili, Angika and Bhojpuri share similarities – but also have clear differences.
This AI should be technically validated through a rigorous process that compares it with the assessment of a human linguist. In fact, over time, it might be able to catch subtle inflections better than a human listener.
This would allow a simple mobile phone or portable electronic device to act as a powerful screening device, containing AI that will conduct a simple conversation OR ask for a read-back of a randomly-generated paragraph/text. It would analyse the person’s speech and make a probability-based assessment of which region someone originates from (and in turn, which country). It should be designed for high sensitivity (meaning a low rate of false negatives). This would be a high-throughput screening tool that would allow large-area assessment of legal status without personnel needing to have specific training while limiting process-related inconvenience and harassment for genuine citizens who have been mistakenly identified.
Those who are identified can then undergo further investigations, including linguistic assessments by humans and other details such as their travel information, whether their documents are genuine and so on.
Linguistics: Human linguists can be used for a second confirmation of origin, and this is a commonly used system of assessment overseas including in the United Kingdom, Netherlands, Switzerland and so on. For this process, a dedicated track of training linguists in subcontinental languages and dialects should be set up. Linguists emerging from this path should find guaranteed employment with the Government. As this process might take time, the development of screening AI has to be expedited to act as a force-multiplier that should eventually be able to directly process the vast majority of cases. In the long-term, the role of human linguists should be to assess cases that the AI classifies as indeterminate or in the case of disputes.
Regulatory Changes: The final change required is appropriate regulatory changes to ID documents. For example, the adult Aadhaar card has reached saturation in many parts of India, yet cases of people fraudulently acquiring the Aadhaar card continue to occur, often in border states. Registrations in border districts should be significantly tightened and require detailed Central Government assessments before being granted. Moreover, all entry/exit points should capture biometrics and identity data, and this should be cross-checked with all Aadhaar registrations by the system before being granted it. Finally, pattern-recognition systems should be implemented that would immediately go slow on ID registrations in a district – whether Aadhaar, passport, PAN, voter ID and so on – once an unexpected surge is noticed or if it does not match population data and trends triggering investigations.
Serious Discussions and Modern Solutions Needed
It is easy to caricature any discussion on this sensitive subject as being driven by bias or try to create a religious angle where there is none. Yet, a serious intellectual discussion needs to be had because rising tensions over this issue in India and the world (as the US and Europe have shown) cannot be good for national harmony and stability. We need modern solutions to minimise inconvenience for citizens while identifying and assessing those who are not. It’s nothing less than the need of the hour.
Comments