How artificial intelligence can decongest nigerian prisons digilaw agunbiade akintunde


Anyone familiar with statistics about Nigerian prisons knows that most of them are over-congested, carrying more than twice their actual capacity. For those unfamiliar, here are some stats:

Statistics of prison population risk assessment
Statistics of prison population digilaw risk assessment

The above list covers the total number of prisons, their capacity, and the actual number of prisoners as of 2017 in Nigeria. There are 240 prisons in Nigeria, with varying security levels and purposes. As of July 2019, the total number of inmates in these prisons is 73,995, an increase of 5,309 since 2017.

The growth rates of the prison population are not my concern here. The stats that we should be concerned about is the demographics of our prison population.

Convicts make up 23,568 (32%) while those Awaiting Trial make up 50,427 (68%).

What does this mean? If we had no inmates who were ‘Awaiting Trial,’ Nigerian prisons at their current capacity are more than adequate to accommodate the convicts in the system.

Why do we have so many inmates ‘Awaiting Trial’? What’s delaying their trial? What does it even mean to be ‘Awaiting Trial’?


Understanding the awaiting trial condition requires an understanding of the factors that lead to it. Primarily, it is caused by Nigeria’s rigid and outdated penal laws. Whereas other climes have lighter punishments for simple offenses like those that are traffic and environmentally-related, such as community service and paying of fines, Nigeria’s penal laws in many places still maintain provisions for incarceration as punishment for these offenses. On its own, this alone might not lead to the awaiting trial condition, but the method used by Nigerian security agencies is to keep suspects incarcerated before conviction for their alleged crimes. It is known as holding charges.

According to Black’s Law Dictionary, a holding charge is a criminal charge of some petty crime filed to keep the accused in detention while the prosecutor takes time to build a more significant case and prepare a more severe offense. Alternatively, the accused can be brought before a court of incompetent jurisdiction intentionally by the prosecuting authority to get a remand order on the pretence of building a stronger case while the accused stays in prison.

Combine holding charges and outdated penal laws, and you have the recipe for the awaiting trial condition, the reality of 68% of Nigeria’s prison population. The newly enacted Nigerian Correctional Service Act 2019 recognises this problem as it provides that one of the objectives of the Act, under Section 2(1)(d), is to ‘establish institutional, systemic, and sustainable mechanisms to address the high number of persons awaiting trial.’

Now for my solution; Artificial Intelligence, or to be precise Risk Assessment Systems, a new AI-based technology that has been successfully applied in several states of the USA and Israel to reduce the rate of incarceration. In my recently published book, ‘Artificial Intelligence & Law: A Nigerian Perspective,’ I discussed this system generally, although my focus there was on the application of AI in civil matters, not in criminal cases, under which the prison system falls.

My discussion now will involve an explanation of what Risk Assessment Systems are and how they work. I will then proceed to discuss how Nigeria can implement it. My model is slightly modified to combine features of the American and Israeli system, making it locally suitable and avoiding the pitfalls of other networks, especially the American versions.


The purpose of a risk assessment system is to decrease the rate of incarceration and predict recidivism, or the likelihood that a person will return to a lifestyle of criminality if released. They come in several versions that perform slightly different tasks. Some predict the probability of a person committing a crime (COMPAS); others decide whether a person should be released on bail or not (PSA). A third version suggests sentencing options (LSI-R and ORAS), while other assists judges in reaching sentencing decisions in traffic cases (IDSS). All these systems will be discussed in detail shortly.

Risk assessment systems generally use data to identify those most likely to commit crimes and recommend that they are removed from society. It does vice versa for those it finds less likely to commit crimes; it usually suggests that they should be released or given lighter sentences for offences already committed. The development of these systems is premised on the Selective Incapacitation Theory, which was developed in the 1980s USA. This theory held that there was a subset of society described as ‘career criminals’ who were responsible for most crimes committed. The courts should seek to identify these persons and put them away for more extended periods compared to one-time or petty criminals. Proponents of this theory argue that this course of action would reduce the cost of prison administration by identifying those defendants who did not have to be incarcerated towards conviction or rehabilitation. This line of thought shows why it can help mitigate our awaiting trial problem. Another argument in its favour is that it would reduce the crime rate. Their position is statistically sound. After analysing the records of 10,000 offenders, it crime that 51.9% of serious crimes were committed by just 18% of offenders.

Risk Assessment Systems rely on data, mostly gleaned from prison and police records, to reach their conclusions. Any gaps in this data will reflect in the conclusions reached by these systems. For instance, some of them have been accused of being biased against blacks in the USA, by recommending incarceration for them, while recommending lighter sentences for the same crime by white people. Historically, more black Americans and their communities have been the subject of police surveillance and crackdown. There is thus more data about African Americans in the police system in proportion to their minority population. When you feed a Machine Learning system with this kind of data, it will incorrectly interpret it to mean that blacks are more prone to violence than other groups and will recommend stronger sentences for them. I will discuss some of the risk assessment systems mentioned above and the challenges some of them have faced.

COMPAS – Correctional Offender Management Profiling for Alternative Sanctions

Of all Risk Assessment Systems to be considered, this one has generated the most controversy and criticism. It was developed by a private company known was Northpointe Corporation in the USA. The first state to adopt it was the State of Virginia, USA. Based on data provided, it delivers recommendations to judges about sentencing for defendants. It has successfully helped Virginia reduced the rate of prison population growth from 31% to 5% annually.

To deliver an assessment of each person, this system considers individual characteristics such as age, previous criminal record (if any), employment history, and group data. The sum of this information can result in a recommendation that the suspect should remain incarcerated, how long such a person should be sentenced, or whether the person should be pardoned.

This system has however, been found to be mostly unfavorable to African-Americans. One case where this bias was brought to nationwide limelight in the USA was the Wisconsin v. Loomis case, where the defendant was arrested for his involvement in a drive-by shooting. He pleaded guilty and requested for a plea deal. The judge, however, gave him a 6-year sentence, saying among other things:

‘The risk assessment tools (COMPAS) that have been utilized suggest that you’re extremely high risk to re-offend.’

The defendant’s appeal to the Wisconsin Supreme Court was rejected because it would be irresponsible to release a defendant identified as high risk by the risk assessment system, only for him to commit another crime and be re-arraigned. In this scenario, we see that the court in the State of Wisconsin has largely surrendered judicial decision-making to the computers.

Another instance of the bias of this system happened in 2014 involving two 18-year old girls named Sade Jones and Brisha Borden. They stole a bike and scooter for a joy ride. They were arrested and charged with robbery and petty theft. COMPAS assessed Borden to be a high-risk offender, while Jones was assessed as a medium risk of reoffending within the two years.

The implication of this is that judges would be predisposed to give the maximum sentences for crimes committed based on the assessment given, rather than focusing on the present circumstances of the defendant in question, which may suggest a lighter or no punishment. Sentencing in this scenario would depend on the sum of crimes committed by the defendants’ group, not just his crime — something like Jesus carrying the sins of an entire race.

Another risk assessment system that is very similar to COMPAS is the LSI-R (Level of Service Inventory-Revised). It was also developed by a private company based in Canada (Multi-Health Systems) and determined a person’s risk of recidivism (going back to crime) and the best sentencing options. It is used in the States of Washington and California .

If this the reality of a risk assessment system in practice, why would I recommend that Nigeria adopt something like this? The discussion so far only suggests that such a technology might even worsen the prison congestion problem.

If this has come to your mind, your fears are valid. But there is one factor that makes COMPAS a problematic system; it is privately owned. Northpointe Corporation, the company that developed this system, has full proprietary rights over it, and only licenses state governments in the USA to use their risk assessment system. The implication of this is that neither the government nor the courts have a clear idea or how the system delivers its recommendations, how it assigns scores to each parameter, or how it processes data. The company has also refused to make this information public, despite the public purpose that their product serves.

It is my firm position that Nigeria should not tow the path of outsourcing the development of a risk assessment system to a private company. Even if we do, the agreement towards its development should make it clear that upon complete payment of the contract sum, all rights over it shall be assigned to the government. This way, the system shall become public property, and members of the public will be able to access information about how it works, under the Freedom of Information Act 2010 and the Nigerian Data Protection Regulation 2019.

The next risk assessment system that will be discussed meets the described standard.


This risk assessment system was developed by the State of Ohio Government. Its development was a joint project of the State Department of Rehabilitation and Correction and the University of Cincinnati, Ohio. Other states that have taken the public route in developing their systems are Pennsylvania and Louisiana. They have done this to avoid the challenge that proprietary protection of private systems usually poses in states like Virginia, Washington, and California. The public approach is what I recommend Nigeria take.

There is another risk assessment system that perfectly matches what I have in mind for Nigeria to implement. It is the:


Unlike other systems that seek to identify potential offenders and recommend sentencing methods, this system only suggests that a person should either be granted bail or remanded before he/she is brought to trial. To reach its conclusion, it evaluates the age of the suspect and their criminal history. If the person is assessed to be low risk, bail is recommended. If found to be high risk, detention till the trial is recommended. Data from 1.5 million crimes in 300 US court jurisdictions were used in training this system.

It was developed by the Laura and John Arnold Foundation and is used in the States of Arizona, New Jersey, and Kentucky .

So far, we have seen a range of risk assessment systems, from the less than ideal COMPAS, the standard ORAS, and the excellent PSA. I could move from here to discussing my proposed model for Nigeria, but there is one missing factor that all these systems have. It might not be a big deal in the USA, but in Nigeria, if this factor were to be ignored in developing a risk assessment system, such a system is dead on arrival.

What is this factor?

Involvement of the judiciary in its development.

That’s why I still need to mention another risk assessment system. This one is not from the USA. This time, our discussion will pivot to the Middle East, Israel, to be precise.


This risk assessment system is restricted in use to traffic cases in Israel. The traffic courts are one of the lowest courts in the hierarchy of the Israeli court system. Just like lower Nigerian courts, their dockets are also congested, causing delays in hearing and sentencing for cases.

Conventionally, when a matter comes before the traffic court for a hearing, the judge will consider the previous record of the offender and the time intervals between offenses before delivering judgment in the instant case. This record is presented to the judge in hard copy format as a printout. It usually comes with technicalities that make it difficult to understand, especially for judges new to the traffic court bench and anyone generally unfamiliar with traffic court procedure.

The IDSS was developed to save judges the stress of understanding technical documents and reduce the time spent in traffic court. It analyses the previous records and gives recommendations to judges about the sentence in the instant case. Unlike other risk assessment systems that were developed without the involvement of judges, the IDSS was developed with the participation of Israeli traffic judges. The judges reached a consensus about the most critical factors for the system to analyse as follows:

  • The seriousness of previous offences, such as driving, causing death, driving under the influence, driving while license is suspended, over-speeding, beating red lights, etc.
  • The similarity of previous offences
  • The seriousness of the last sentences
  • Driving causing accidents in the past such as bodily damage and damage to property
  • The present crime committed during the period of disqualification arising from a prior traffic offence
  • The current crime committed during the period of deferred ineligibility arising from a previous traffic offence
  • Frequency of offences


Based on the discourse so far, it is clear that risk assessment systems give a definite advantage to not just the prison system, by reducing the number of potential inmates they have to cater for, but also in reducing the workload of judges. The challenge they have faced revolves around the method used in implementing them – privately owned and developed vs. publicly owned and developed.

I’m sure you would agree with me that the preferred model for Nigeria would be the public model. Borrowing from the model of the State of Ohio, the proposed Nigerian model – let’s call it Nigerian Risk Assessment System (NIRAS) – would be developed by or on behalf of the government through its agencies. This public approach would eliminate the proprietary cover that private developers can use to avoid sharing their source code and algorithmic parameters for assessing risk like in the case of COMPAS

NIRAS would be employed to give recommendations as to whether a person should be granted bail or not. If the suspect is found to have a high-risk score, bail will not be recommended. If the person has a low-risk rating, bail will be approved. NIRAS is similar to the Public Safety Assessment (PSA) already in use in the States of Arizona, Kentucky, and New Jersey.

Most importantly, the NIRAS should not be implemented for all categories of criminal offences and must whole with the active collaboration of the judiciary. This suggestion is gleaned from the Israeli Intelligent Decision Support System (IDSS), which is applied only in traffic courts, and judges supplied the parameters that should be deployed in building the machine learning algorithm that powers this AI risk assessment system.

Understandably, criminal offences, especially those that involve injury to another or public welfare, are more controversial. Introducing any form of advanced technology to their adjudication would be firmly rejected. NIRAS, if adopted, must first be tested with petty offences where it will gain acceptability. This testing ground will also provide the opportunity to detect any flaws in its application that can be easily corrected without dire consequences. Magistrate Courts and other courts of no record would provide a suitable testing ground for this technology.


Even if the proposed risk assessment system is publicly developed, involves the collaboration of judges, and is restricted to minor courts, none of this still eliminates potential risks and pitfalls that might befall it. Judges might provide broad parameters, but they will not be in control of the collection and processing of data that will be used to develop the machine learning models in line with these parameters. We may not have a Black v. White divide in Nigeria, but there are subtle divides in our criminal justice data system that might be unduly emphasised by a risk assessment system. Some of those divides are the Male v. Female gap and Ethnic group imbalances in our data system. Because of the lop-sidedness of our criminal data records, a risk assessment system built on this might be prone to give harsher recommendations for gender or a particular tribe.

According to data from the Lagos Criminal Information System, 98% of prison inmates in the state are male, while 2% are female. When you consider the breakdown of the prison population in Lagos by states of origin, you find that indigenes of Oyo and Ogun State are the largest group of inmates, compared to indigenes of Lagos or other states. Inmates from the South-East make up 16% of the prison population, those of South-South origin make up 13%, the South-West makes up 49%, while the entire North (North Central, North-East, and North-West) is 22%. Funny enough, when you remove Kwara State indigenes from this list, the contribution of the entire North drops to 14%, as Kwara State indigenes make up 1/3rd of the whole of the Northern population in Lagos State prisons.

Notice how I mentioned some factors more than I mentioned others? After you read the previous paragraphs, these are the factors or terms that might have come to your mind:

  • Almost no women in Nigerian prisons
  • Nigerian men have a higher proclivity for crime compared to Nigerian women
  • People of South-West origin are more prone to criminality in Lagos State than all other groups
  • Northerners are mostly peaceful and law-abiding

None of these assumptions is entirely valid. Some of them are practically untrue, but if we decide only to use the data as represented by the prison population in Lagos, this is what it reflects. AI, notwithstanding the ‘Intelligence’ that forms part of its name, is no better at detecting biases or lop-sided information. The same way an uninformed person can read the above information and leave with an incorrect impression of Nigeria’s ethnic groups and gender population is the same way an algorithm can incorrectly predict higher rates of recidivism for some groups. A risk assessment system could give women higher scores for bail compared to a man, even when the gravity of the offence committed by the woman is higher than that of the man. A risk assessment system could also predict that indigenes of Lagos, Kwara, Oyo, and Ogun are more likely to jump bail and give them lower scores. An indigene of the FCT, Rivers, Gombe, or Niger might be recommended for bail, even when the offence committed is more grievous than that of his Oyo counterpart.

Applying a system like this, even in the most restricted way, would worsen the sensitive ethnic and gender relations in Nigeria, with untold consequences.

With all that I’ve shared so far, I’m sure that the benefits of AI to our prison system are the last thing on your mind. You could conclude that the risks of AI outweigh the benefits, I won’t begrudge you that.

But the problem here isn’t an AI problem; it’s a data problem. AI is only as good as the data it is built on. To improve the technology, we must upgrade the data. To ensure the fairness of a proposed risk assessment system, we must improve the quality of the data we develop it with.

I propose that in collecting and processing data from the criminal justice system for the purpose of developing NIRAS, details like gender and ethnicity are blurred or intentionally left out. I’ll illustrate using two different tables. The first table will have fictional names, their offence, sex, and state of origin. The second table will have only fictional names and crime.


risk assessment digilaw


Risk assessment system digilaw

A typical criminal record will have more details for each inmate or suspect. I have chosen to focus on these small details to make my point. If we use data like TABLE A that reflects ethnicity and sex, there is the risk that everything I described earlier will come to fruition. TABLE B, on the other hand, omits those details. A proper criminal record for developing a risk assessment system will skip those details. Instead it will be focused on practical information like facts of the alleged offence, previous criminal history, jurisdiction where the crime was committed, etc. This approach is known as data minimisation – it involves only collecting the minimum amount of data needed to accomplish a task.

It has legal backing in Nigeria under the Nigeria Data Protection Regulation (NDPR) 2019. It provides that the following types of data should not be collected, except with a solid form of consent. They are:

  • Ethnicity
  • Political affiliation
  • Religious belief
  • Trade union membership
  • Biometrics
  • Sexual orientation

However, the NDPR is not active or encompassing enough to provide the legal backing for enforcing data minimisation in the development of risk assessment systems. Because it is merely a regulation from an agency (National Information Technology Development Agency – NITDA) with no purview or relevance in criminal matters, its provisions are at best persuasive.

Nevertheless, it already has the necessary clauses that could be applied to minimise or eliminate bias in AI for our criminal justice system.

There are two alternative approaches that could be employed to achieve this – on the one hand; the NDPR could be amended to expressly expand its scope to cover criminal matters. It could then be re-enacted as an Act of the National Assembly, giving it equal force like the Criminal Code, Penal Code, and other criminal laws in Nigeria, apart from the 1999 Constitution itself.

On the other hand, a special law to create a legal regime for risk assessment systems could be enacted. One of the provisions of this law could expressly stipulate that the requirements of the NDPR concerning data minimisation should apply. This approach, though limited, is more realistic, especially for state governments that may be interested in adopting risk assessment systems. The former method requires federal buy-in and may be long-drawn, although once concluded, it would have wider effect and enforceability. The latter approach can be adopted by state governments without the foot-dragging that usually accompanies federal law-making. It would, however, be limited in application to and may be subject to judicial review if the state government in question chooses to extend the implementation of the risk assessment system beyond just petty offences as I earlier recommended.

In the ideal setting, the risk assessment system (NIRAS) should be an FG project, backed up an enabling law, and a stronger NDPR (preferably an Act). Willing state governments can then re-enact the enabling legislation to apply within their domain, just like the Administration of Criminal Justice Act (ACJA) 2015 and the Child Rights Act (CRA).

Another challenge that might plague the development and deployment of a risk assessment system in Nigeria is the nature of our criminal record system. In a country where most of our criminal records are paper-based, data collection, and processing to build training and building algorithms cannot happen until our records are digitised.

Luckily, there is already activity in this regard. The Nigeria Police Force, a few months ago, entered into a BOT (Build-Operate-Transfer) Agreement with Ace of Spades Consult, a private company. The purpose of the agreement was to develop a digital database of civil and criminal records in the custody of the Nigerian Police at the Criminal Records Registry (CRR) Alagbon, Lagos State. This online database would also have a web portal that could be used to run background checks and issue clearance certificates concerning previous criminal records.

All current physical records will be converted into electronic format, while future documents will be uploaded directly to this online database.

The Nigerian Police also intends to collaborate with other government agencies that collect citizens data such as the National Identity Management Commission (NIMC), the Nigerian Immigration Service (NIS), and others.

Irrespective of which level of government that takes the initiative to implement a local risk assessment system, a partnership with the Nigeria Police Force would be critical to pulling this off. Data from the Nigeria’s central criminal registry would be comprehensive and detailed enough to build a risk assessment system tailored to Nigerian realities, as opposed to merely importing one and trying to adapt it to our realities. Access to this data would not be a one-off project but continuous to ensure that new trends in criminal behaviour are incorporated into machine learning algorithms that predict recidivism. To create a framework for this, the enabling law for a risk assessment system would provide for this while also enshrining the principle of data minimisation.


There are positive signals that the authorities are taking steps to tackle the challenge of prison congestion. One of them is the Nigerian Correctional Service Act 2019, which now empowers prison authorities to reject new inmates if their facilities are full. With a risk assessment system working alongside this law, we can ensure prison authorities don’t even have to reject many inmates in the first place, because the risk assessment system would have recommended that some be given bail.


Prison Statistics: Prison Population by Total Detainees, Prison Capacity and Number of Unsentenced Detainees by State and Year and Prison Inmate Population by Gender (2011 – 2016), National Bureau of Statistics/Nigeria Prisons Service October 2017

Nigeria, World Prison Brief Data, World Prison Brief

– Nigerian Correctional Service Act 2019

– Kehl, Danielle, Guo, Priscilla, Kessler, Samuel, ‘Algorithms in the Criminal Justice System: Assessing the Use of Risk Assessments in Sentencing‘ (July 2017) Responsive Communities.

– A. I. Agunbiade, Artificial Intelligence & Law: A Nigerian PerspectiveAcademia

Quarterly Report Infographics, October – December 2018, Lagos Criminal Information System, Lagos State Ministry of Justice

– A. I. Agunbiade, ‘A Review of the Nigerian Data Protection Regulation (NDPR) 2019′ Academia

Nigerian Data Protection Regulation 2019: Implementation Framework

– Ugo Aliogo, NPF Partners Firm on Digitalisation of Records, (This Day, June 25, 2019)

Join the Conversation

1 Comment

Leave a comment

Your email address will not be published. Required fields are marked *