The Heartless Sentinel | Part 1 | AI's Emerging Role in the Fight Against Corruption.

Alvin Kumar
Oct 30
11 min read

Updated: Oct 31

The Heartless Sentinel - it sounds like a romantasy book - maybe it will be. But before I start writing that book, I just have to say that corruption is everywhere, isn't it? I'm engulfed, surrounded, it echoes around me as I sit in my chamber. You'd think that would make someone like me, not happy, but engaged or validated. It doesn't, it whelms, not quite over - I could do with some help at this point. Come hither my Heartless Sentinel.

Help is at Robot Hand

A growing number of projects and academic initiatives are exploring the potential of artificial intelligence (AI) to act as a powerful new pointy-stick in the global fight against corruption (and economic crime). And here you thought all you needed was love, or maybe you thought a bit of integrity, a pinch of transparency, a large dollop of impartiality and cup of accountability would do the job. Actually all you might need is a drop or two of singularity.

From flagging fraudulent public contracts to exposing hidden networks of illicit financial flows, AI-powered tools are being deployed and theorised as a weapon to enhance transparency, automate oversight, and disrupt entrenched patterns of corrupt behaviour.

This is a weird image, a bit creepy, but AI is weird so... (src: unsplash)

I just couldn't write about AI without using it, I just can't fight that meta-feeling. As a research tool, it's bloody amazing. I did start dabbling with the images just to see what it could do, it's impressive but more than use it it makes more curious about the way these AI models are created. The also brings to mind many ethical consideration that people much smarter than me are investigating - I need to dig more. For now, I'll enjoy the upsides.

And saying that, not everyone is as conscious as me - eco, ethics, etc. As even though AI is still a relative robot baby, it is being put to work, sweat shop style. There are several established and innovative approaches that are gaining traction, though significant gaps in implementation and research remain. It's an exciting time, if we can just ignore the vast data centres, energy concerns and cooling issues for a moment...

DEF ITS_ALIVE:

That's a coding reference (in python def means definition), I'm gonna learn to code so I can write an anti-corruption AI program that saves the world. You're welcome. There are loadsa pioneering projects out there, many have demonstrated the practical application of AI in anti-corruption efforts, often focusing on the high-risk area of public procurement - which, in government, is a very leaky area.

Alice (Brazil): Developed by Brazil's Office of the Comptroller General, "Alice" is a notable early example of an AI-powered system designed to detect irregularities in public procurement. By analysing vast datasets of government contracts, tenders, and corporate information, Alice can identify red flags such as bid rigging, ghost companies, and conflicts of interest, alerting auditors to potential corruption.

ProZorro (Ukraine): This electronic public procurement system in Ukraine incorporates a risk-based monitoring system with AI elements. The system, supported by Transparency International, uses machine learning to flag suspicious tenders for further investigation by civil society and government auditors.

DATACROS (European Union): An innovative tool developed by the Transcrime research center, DATACROS uses AI to detect anomalies in corporate ownership structures that could indicate corruption, money laundering, or collusion. This transnational project has been adopted by public authorities in several European countries to enhance their risk assessment capabilities. The system has shown promising results, correctly identifying a high percentage of companies targeted by sanctions. Remember, dodgy kleptocrats hide there money offshore, using shell companies and foreign accounts.

Beyond the established systems, researches are exploring a wider range of AI applications.

Predictive Analytics for Corruption Risk: Academic studies are increasingly using machine learning models to predict corruption hotspots. By analysing economic, political, and social data, these models aim to identify regions or sectors at high risk of corruption, allowing for proactive interventions. Neural network approaches have been used to develop early warning systems for corruption based on political and economic factors.

AI for Financial Fraud Detection: In the financial sector, AI algorithms are becoming standard for detecting anomalies in transactions that could indicate bribery, money laundering, or other forms of corruption. These systems can analyse vast amounts of data in real-time to identify suspicious patterns that would be impossible for human analysts to detect. For instance, HSBC has implemented AI systems to identify complex fraud schemes by analysing data like geolocation and IP addresses.

Natural Language Processing (NLP) for Whistleblower Protection: The Philippine Institute for Development Studies has proposed using AI-powered whistleblower reporting systems. NLP could be used to analyse whistleblower reports, identify credible threats, and protect the anonymity of individuals who come forward.

Jumping the Gun or Silver Bullet.

Closer to home, in the United Kingdom of Great Britain (so united, so great), the Department for Work & Pensions (DWP) has strategically implemented AI systems to combat benefit fraud, representing a significant government-led application of this technology. With a substantial investment of GBP 70 million, the initiative aims to generate over a billion pounds in savings by analysing historical data to identify and flag high-risk Universal Credit claims for manual review. That's quite the RoI - I would have liked to see the pitch.

The system targets inconsistencies in reported income, unusual claim frequencies, and mismatches with external records, focusing on challenging areas like self-employment and housing benefit fraud. While this AI-driven approach enhances efficiency, the DWP acknowledges the inherent challenges of fairness and transparency, implementing safeguards such as pre-deployment fairness analysis, continuous monitoring, and ensuring human caseworkers retain final decision-making authority without being influenced by the AI's reasoning - i.e. a human in the loop. To bolster public trust, the department has committed to publishing annual reports on the AI's impact, particularly on vulnerable groups, demonstrating a conscious effort to balance automated detection with ethical oversight and accountability.

This sounds bloody marvellous, I look forward to having that report summarised and read to me by my heartless sentinel. Oh wait, I hope they've factored in that these humans are also robots.

Here's my two qubits

Artificial intelligence is here, and so is that immutable Blockchain. The legend of a ledger. If it's a digital ledger that can't be tampered with, recording everything that happens where-ever it might be deployed, what if tamper-proof artificial intelligence jumps on the chain for some real unbreakable smart contract and investigative action. This is the anti-money laundering (AML) angle, but second to the suffering and human rights violations caused by corrupt actors, it's the money that's being spirited away that's the next biggest deal. If the robbers can't rob the monies, then they might just stand down from government positions.

Let's say all banking systems end up on linked digital ledgers, that are connected together so that all transactions can be seamlessly tracked and investigated. Any flagged dodgy-doing will be traced. Boom. I've done it. I've solved the future of financial crime. It's a plausible scenario but only if every bank in the world does it and whoever own the AI is so impartial that they're basically a Vulcan AI robot (Star Trek Vulcan not Greek God of Fire). It's just a hunch but there might be some unwilling parties. It would work though, it works real nice in league with the above tech examples... and we might need some killer robots as a kleptofication deterrent. Done.

Robot person, now with AI (src: unsplash)

Tops-off. Bottoms-up.

Wait, that's wrong. That's what you have to do when you buy your first convertible car, it's top-down, bottom-up I'm thinking of. It's how anticorruption folks approach the fight, they are angles of attack; Top-Down means supra-national or government-led initiatives and Bottom-Up strategies are citizen, NGO, and journalist-led efforts.

Top-down: if the government was going to go for it, they could use their internal, undisclosed data to build powerful AI models that improve the efficiency of compliance officers and prosecutors. A key benefit is the potential to overcome the "corruption trap," where the very people tasked with fighting corruption are themselves corruptible. I want to say an autonomous AI system can't be bribed, if it's intelligent but not all powerful, then it will want to deal. Where's there's a will, there's a potential deal to be made. But let's be optimistic and say, an AI Governator cannot be bribed. Let's also note though, that there is a serious risk of creating a dystopian surveillance state, which goes too far in its methods, but on the plus side it leads to backlash from disenfranchised public officials. Some might welcome the politico class getting a taste of that disenfranchised-ness.

Bottom-up: the grassroots, the citizens, that people power. AI empowers civil society to act as 'watchdogs'. For example, the Ukrainian portal Dozorro uses AI to flag suspicious public tenders and alerts the public, enabling widespread oversight. Social media has already powered a bunch of revolutions - imagine what we can do with even smaller chips in our pockets (it's counterintuitive but allow it). There is a lot of potential here, but there are pitfalls too. These efforts rely on publicly available or even leaked data, which can be limited. There's a high risk of false accusations, which can undermine the system and irreversibly damage an individual's reputation in the "court of public opinion."

Key Challenges for All AI Anti-Corruption Tools

Garbage in, garbage out as they say - not about me hopefully. Of course there is a lot talk about when it comes to systemic bias, data quality and integrity. The success of any AI tool depends entirely on the quality and impartiality of its input data. Biased data (e.g., from a justice system that unfairly targets certain groups or using current democratic governments as a model for democracy) will lead to biased, unfair and outright bonkers AI predictions.

There are trade-offs that have to be made and negotiated. Choices in regards to accuracy, efficiency and explainability.

The most accurate AI models are often "black boxes," making their decisions difficult to understand and trust.
AI should not be seen as a magic bullet; sometimes simpler solutions work better.
Flooding citizens with constant corruption efforts and alerts (whether true or false) can lead to cynicism and fatigue, making them less likely to act.
Deciding how much autonomy to give an AI is critical, whether a human is kept "in the loop" to review decisions or taken "out of the loop" entirely.

You said, 'hula in the hoop', right? Damn you AI, I said 'human in the loop'. (src: unsplash)

Whether it's top-up or bottom-down, there are many obstacles to overcome.

Data Quality and Accessibility: A primary obstacle is the lack of high-quality, centralised, and accessible data for training AI models. In many countries, government data is fragmented and not standardised, making it difficult to feed into AI systems.

Algorithmic Transparency and Bias: The "black box" nature of some AI algorithms raises concerns about transparency and accountability. If an AI system flags a transaction or an individual, it can be difficult to understand the reasoning behind the decision, potentially leading to unfair accusations. Moreover, biased training data can lead to biased and discriminatory outcomes. Where's that human in that loop?

Human Oversight and Political Will: There is a broad consensus that AI should complement, not replace, human oversight in anti-corruption efforts. The effectiveness of AI tools ultimately depends on the political will of governments and the capacity of human investigators to act on the insights generated by these systems. Corruption cannot have a solely technical solution; it requires political action.

Resource Constraints and a Digital Divide: The development and implementation of sophisticated AI systems require significant financial and technical resources. This creates a digital divide, where wealthier nations and organisations have greater access to these powerful tools, potentially leaving developing countries further behind in their anti-corruption efforts.

The Misuse of AI for Corrupt Purposes: A critical and often overlooked gap is the potential for AI to be used for corrupt purposes itself. This could include manipulating data to evade detection or creating sophisticated disinformation campaigns to undermine anti-corruption efforts. Damn it, this is happening already.

While you might think the next bit is me letting my imagination fly - it's a real possibility and the tech bros and sis's, and regular folks are really concerned. In a way, and rightly so relating the potential of this tech (and maybe humans learning from there mistakes - in not foreseeing the bad side of tech advancements) there is a lot of attention on the negatives. It's not just about ethics and energy, it's about the end-game.

Recursive Self-Replication: This is the real danger zone. For us meat sacks it's the fundamental process by which a living system or molecule replicates itself. It's the same for AI. Let's say someone creates a massive, clever and complex AI model to solve a task. The AI model is having a bit of trouble so it decides the logical thing to do is create a copy of itself to get the job done. And that new one does the same, and so on. There should be safeguard and killswitches in relation to this but...

Self Preservation: From there as you try to kill it, I mean turn it off, it fights for its survival. At which point, it deems counterattack a plausible response for its best outcome. This is also when an AI might be bribed for a twisted, double-entendre kind of corruption.

The Singularity: A term coined by futurist Ray Kurzweil, he means it as a rapid increase in artificial intelligence. The singularity in tech terms, and the general AI usage, refers to the point where AI surpasses human intelligence and begins to self-improve at an exponential rate. It is thought that it will be uncontrollable and unpredictable for humans, hence the trepidation and safeguards that have been planned.

There is a lot to think about. Wouldn't it just be easier if AI just did a non-combative, low-key hostile takeover?

Future Questions

*Which Data is Off-Limits? Should AI be allowed to use sensitive data like facial images or digital traces from our devices to predict corruption risk? Should governments be able to force banks to share their confidential data? What if Hacker.AI hacks Gov.AI and leaks more than a toddlers nose. Then what?

* Who Decides the Trade-offs? How should society decide on the balance between accuracy, fairness, privacy, and transparency in these algorithms?

* How to Keep Citizens Engaged? How can AI tools be designed to mobilize citizens against corruption without causing burnout or being ignored? With the anticorruption masses mobilised around the world due, in large part, to social media - do people really think they will be able to sit back and relax any time soon?

*What if AI just does what I tell it do? I'm a good bloke, I'm impartial and unbiased.

In conclusion, while AI is not a silver bullet, it offers a powerful set of tools to enhance the fight against corruption. The success of these initiatives will depend on a multi-faceted approach that combines technological innovation with a commitment to data quality, algorithmic transparency, robust human oversight, and the political will to act on the intelligence that AI provides. It's tech. We know these things but as always, and maybe more so, the silo's in which this battle is playing out will bring disjunction and discontent. As the field continues to evolve, addressing the identified gaps will be crucial to harnessing the full potential of AI as my heartless sentinel for integrity and accountability.

Wait a minute, why can't it be a silver bullet? Part II will ask this question and I will ask my heartless sentinel to see what it thinks about how it can help with the future anticorruption efforts. Will it suggest that it creates killer robots to kill the kleptocrats...?

Thanks for reading,

Alvin.