The Silicon-chip meets the SHO

 

Let’s see what will happen if a citizen meets an AI-enabled robot instead of a human being (Duty Officer) when he first appears at the police station. At the very least, he’ll get a patient hearing, almost immediately, no sitting around waiting indefinitely, no lunch-ke-baad-aana. Whatever he says will get recorded in the General Diary verbatim but in a legible manner. If the matter is merely informatory in nature, a receipt with General Diary number, date and time will be automatically generated for him. If the matter indicates a cognizable offence, an FIR will be drawn up in a legible manner with correct sections of law without any discretion and escalated to the SHO for taking a decision as to whether to investigate or refuse investigation and allotment to an Investigating Officer. Refusal to register complaints, the long-standing bane of the police, will now be a thing of the past. This one measure, in itself, will transform the functioning and image of police no end.

Further, AI can reduce the humungous documentation burden that police stations grapple with. In 2024, around 59 lakh cognizable cases were registered. Each such case involves a First Information Report, the spot maps, witness statements, arrest memos, forensic reports, seizure lists, medical examination records, report in final form, and so on – documents numbering in hundreds. All documents in local language need to be translated in English when a case goes to the High Court or the Supreme Court. Manual translations for a single case can take up to 15 days which can now be done in an hour or so, while retaining specialised legal/ local terminology like surathal, panchnama, khatian, zimma etc. and also retaining page-for-page parity so that it is acceptable in the court. AI enabled digitisation can read handwritten FIRs in multiple languages with high accuracy. It can also process low-quality scans and faded documents by intelligently reconstructing the degraded letters. Most importantly, all the documents will become searchable.

Another big help from AI would be when the investigation is transferred to another officer due to transfer, etc.. Rather than the new officer having to go through the entire case file word by word, he can go through a structured summary, a chronological timeline of events, evidentiary gaps and inconsistencies, all generated by AI. AI can listen in to interrogation of accused, correlate everything and point out the inconsistencies in the statements.

AI can also generate the chargesheet index, cross-reference evidences and biometric data, witnesses and the sections of law and check for completeness of the case diary and the chargesheet. It can also create a draft prosecution narrative to be used by the Public Prosecutor and the Investigating Officer. My back-of-the-envelope calculation indicates that AI can free up to around 30 officer-hours per day at a mid-size police station on documentation alone.

Both at the police station level and beyond, AI can also help in mapping crime hotspots, connect FIRs and Modus Operandi and help in identifying repeat offenders.

Police station inspections by senior officers are the backbone of police administration. Usually, it’s an annual affair preceding which there is a lot of activity to compile statistics and cover-ups so that the inspection goes off peacefully. Then things are forgotten for a year until the next inspection which may be by a different officer or the same officer who may not remember much of what all he’d instructed the previous time around. When one goes through successive inspection remarks, one sometimes sees the same instructions having been passed from time immemorial, without any substantive changes in the performance. AI can enable a GPS-enabled field visit tracking system which will have a digital record of every inspection. This will enable continuous performance monitoring and accountability rather than having a lot of sound and fury once in a year, signifying nothing. During every inspection, all previous instructions and their compliances would be available to the inspecting officer so that repeat issues will be highlighted. GPS tracking will ensure that the supervising officer actually visits the police station rather than churning out reports sitting in his office.

A module can be carved out for the public representatives to send their recommendation and escalate to higher-ups, if need be. Further, there can be a chat bot for the public to obtain any information regarding their application for permission, licences, various fines, applicable sections of law, etc. in any Indian language and English.

Finally, the performance of officers will now face automatic and systematic evaluation based on objective parameters like average response time, detection, preventive action, tasks performed, compliance metrics, pending cases, escalation frequency, etc. so that high performers can be rewarded and encouraged suitably while underperformers will be caught out and earmarked for skill and performance augmentation.

Despite all the foregoing, it is important that there must be human oversight and intervention because AI makes mistakes and sometimes, the mistakes in policing can make the difference between life and death, literally. Even if the Duty Officer may become dispensable, there must be some supervision of the robot doing that job too. Because many things about policing are confidential, the AI systems must be based on a dedicated self-hosted Large Language model, end-to-end encrypted and with restricted access to certain parts and trained on all the criminal laws, police regulations, Government Orders, SOPs and circulars.

There should be a unified gateway for all officers. Any FIR should be possible to be generated at any police station with automatic transfers to the police station of jurisdiction. The heinous crime cases (SR cases) which are supervised by senior officers can have almost instantaneous passing of instructions with deadlines and automatic escalation in case of non-compliance.

While predictive AI hasn’t been much of a success, it can still be used for “now-casting” – i.e., predicting the present or very near future crimes, riots and traffic issues using real-time data. AI can’t judge context or the nuances of social situations and does not have the emotional intelligence required of an SHO. As of today, AI can’t become an SHO but use of AI can make for an unimaginably better SHO. By several orders of magnitude.

The Silicon-chip SHO

 

Today Artificial Intelligence (AI) is everywhere. It’s in the smart-alec student’s assignment paper; it’s in the Income Tax officer’s financial forensics to catch tax dodgers; it’s in the dystopian scenario of the doomsday soothsayers; it’s in the ongoing attempts at self-driving cars; most importantly, it’s in the adda-intellectual’s morning discourse – he calls it Chai GPT (Genuinely Pure Tea) enhanced with AI (Adrak and Ilaichi). Has the time come then for AI to replace the SHO (Station House Officer), the pivot around which the entire policing system in India revolves?

Let’s see what all the SHO is expected to do.

The SHO supervises all the work of the police station including work allocation, general management, discipline and all the administrative work. In turn, he is the most supervised officer in the Police or in any governmental set-up, with at least six layers fairly closely supervising his work. This includes the senior officers and the judiciary to whom all his crime-related work is accountable.

Under the Indian criminal jurisprudence, he is the de jure Investigating Officer of all crimes in his Police Station area. All the investigation is carried out by officers of the Police Station under powers delegated by him. A senior officer takes up investigation under powers of the SHO assumed by the former.

The SHO is also responsible for maintaining public order in his jurisdiction. He arranges both preventive and mitigation actions in this regard. Preventive actions include patrolling, keeping an eye on evolving situations, surveillance of known offenders, lookout for suspicious or unknown faces, predicting disturbances and so on. When a major disturbance occurs, the SHO is usually the first one on the firing line and has to meet fire with force.

The SHO is expected to deal with situations and persons with empathy and emotional intelligence. He is called upon to treat people with a rare nuance which distinguishes between a crime of malice from the same crime born out of desperation or mortal distress. He navigates with resourcefulness (mostly illegal) demands from political parties of various hues, ever alert to the fact that today’s no-hoper opposition may be tomorrow’s king.

The general citizenry has huge expectations from the SHO. Frankly speaking, there is not a single other agency of the government which is open and responsive to the public 24 x 7 x 365, come hell or high water. When everything else is shut in the odd hours or during major depredations, there’s at least a sentry and mostly an officer available at the police station – in that sense, it is always the governmental arm of the last resort. 

Apart from all this, there’s a huge social role thrust upon the SHO. All government offices have a person who is called Bada Babu or its equivalent in various languages. In the housing department office, for example, the Section Officer is called Bada Babu. However, to the general public, Bada Babu usually means the SHO. So, amongst the Bada Babus, the SHO has an outsized eminence, he is primus inter pares. This is because of the influence he and his work wield, touching everyone at a very basic level.

Given such a complex job description, can he be substituted by AI? Let’s look at what AI can or cannot do, as of now.

The term AI is being used as THE ONE BIG THING which is turning everything topsy-turvy now. Well, it has been disruptive but it’s not one single thing. It’s an umbrella term comprising at least three categories: Generative AI, Predictive AI and Content moderation AI.

Generative AI has had the maximum success and has caught the popular imagination and also filled people with a lot of apprehension. However, what it actually does is basically recognise patterns in data, words and pictures and correlate them through layer-by-layer iterations so well that its output comes across almost as human. All that it’s doing, based on its training, is choosing the most appropriate words, picture patterns, etc. to give the most suitable response as per its algorithms. Since it has been trained on a very large amount of data and has had some moderation built in, its responses have been very impressive but not always accurate. It doesn’t have sentience, i.e., consciousness, regardless of the hype and the claims. However, its speed and abilities have been formidable.

As far as the predictive part is concerned, AI hasn’t worked well so far. In Police-related work, it has been tried out for criminal risk prediction (i.e., whether to release an accused on bail with low risk of his committing crimes again), hiring (whether the person will do well on the job), face recognition and crime prediction, with indifferent results and racial/ gender/ economic biases. So far, it has been somewhat satisfactory only in “predicting the past,” which is of no use to anyone.

Content moderation has been used extensively by social media platforms but has shown large and grave mistakes. So much so that one can’t even search for “gorillas” in Google photos or Apple photos apps now – blocked by the apps themselves because it identified all black people as gorillas.

At the present stage of development when AI is more of “Microsoft Excel than The Terminator,” it is not feasible to replace the SHO with a robot. Things might change when AI becomes capable of judging the context, cultural and political nuances and is able to do AI research itself, i.e., progress to Artificial General Intelligence (AGI). That would require AI agents to self-improve through interacting with the physical world and the social world. To my mind, that is still some distance away.

That said, there’re still tremendous benefits to be had if we adopt AI for a lot of automation in policing, specifically at the police station level. I think, the utility through automation and not the hype of a sentient SHO-replacing-AI, is the part worth sitting with. That calls for another blog. Next week.

[To be concluded]