TRENDS IN THE USE OF IMPRISONMENT public, punishment, giving the offender the opportunity to make amends, and expressing disapproval of offending behaviour.78 Elsewhere, countries have moved away from using imprisonment as a sentence for minor offences. In Cape Town, South Africa, a High Court judge ruled that all people given sentences of less than two years should instead perform community service, given that defendants are ‘generally poor and casualties of socio-economic conditions’.79 In Scotland, judges are now prevented from sentencing people to less than six months’ imprisonment in all but exceptional cases. In England and Wales, the government expressed a commitment to scrapping sentences of less than six months in order to reduce overcrowding, and also as recognition of evidence that short sentences are less effective at cutting reoffending than community sentences – they are ‘long enough to damage you and not long enough to heal you’.80 The language being used is of ‘smart’ sentencing – a choice between effective and ineffective justice rather than between ‘hard’ and ‘soft’ justice. (See Alternatives to imprisonment, page 40). Gender bias exists in sentencing decisions. A report on women on death row found that criminal justice actors frequently see women as victims and survivors, rather than as perpetrators of crime.81 While this assumption can benefit some women who receive reduced sentences, ‘[a]t the same time, women who are seen as violating entrenched norms of gender behaviour may be sentenced more harshly’.82 (See Women, page 20). Other biases related to an offender’s ethnic or racial minority background are also evident. (See Foreign national prisoners, minorities and Indigenous peoples, page 21). In 2017, the US Sentencing Commission found that despite the purpose of the federal sentencing guidelines to promote sentencing uniformity across geographic and socio-economic lines, black defendants received longer sentences than white defendants convicted of similar offences, even when factors such as criminal history, age, education and citizenship were controlled.83 In Australia, Canada and New Zealand, Indigenous sentencing courts have been established specifically to address the issue of overrepresentation of Indigenous 14 | peoples in prison and to facilitate the use of appropriate community-based and restorative sentences. Effective sentencing requires that the court has access to assessments of the risk an offender poses to public security, the factors that lie behind their offending, and the support they can draw on to move away from crime. The UN Standard Minimum Rules for Non-custodial Measures, also known as the Tokyo Rules, explicitly note the value of such assessments.84 Their importance was highlighted in research that found that the use of community sentences has fallen recently in England, which may be linked to an accompanying decline in the extent to which pre-sentence reports are provided to judges.85 In some jurisdictions there has been a move towards more evidence-based sentencing practices, which aim to improve sentencing decisions by giving courts access to an empirical assessment of the static and dynamic factors that make it more or less likely that an offender will reoffend. An offender is typically assigned a risk ‘score’ that is also used to inform rehabilitation interventions.86 A low-risk offender may receive a non-custodial sentence, while those deemed to be at a higher risk may be more likely to be imprisoned. The argument is that these risk assessment instruments can predict the risk of reoffending with a high degree of accuracy, impartiality and objectivity. Such risk assessments are often produced by probation or court welfare officers. However, a fast-moving trend in the US is the use of computer-based criminal justice algorithms that generate risk models for courts based on vast quantities of data. Many of these risk assessment algorithms take into account personal characteristics of the offender, such as age, sex, geography, family background and employment status. They are used to inform decisions not just about sentencing but also about bail. In practice, they mean that two people accused or convicted of the same crime may receive quite different bail or sentencing outcomes. The use of these tools has been heavily criticised on the basis that they are unreliable and can lead to racial profiling, and that there is little clarity or transparency about how the algorithms weigh different variables.87 They are often developed by private businesses, which keep their algorithms and underlying research a proprietary secret, even from the courts or government agencies that use them.88 The NGO ProPublica conducted an investigation of one such risk assessment algorithm, COMPAS, which is used as an aid to judicial decision-making, its conclusions being only one of the variables considered by the judge when deciding on the sentence. Defendants answer 137 questions, including whether they have a telephone at home, difficulty paying bills, family history and criminal history. COMPAS then rates the person on a scale from one (low risk) to 10 (high risk). ProPublica found that the formula used was particularly likely to flag black defendants as future offenders, labelling them at almost twice the rate as white defendants.89 In addition, the scores generated by the algorithm were unreliable in forecasting violent crime and only 20 per cent of the people predicted to commit violent crimes actually went on to do so.90 At present, this kind of artificial intelligence has not been widely adopted in other regions. In Europe, for example, only the UK has piloted predictive tools for judges to assess the risk of reoffending. However, in recognition of the associated risks, the European Commission for the Efficiency of Justice published an ethical charter on the use of artificial intelligence in judicial systems.91 This promotes a precautionary approach and spells out key standards that need to be in place, including ensuring that the design and implementation of artificial intelligence tools and services are compatible with fundamental rights; preventing the development or intensification of any discrimination between individuals or groups of individuals; making data processing methods accessible and understandable; and ensuring that users are informed actors and are in control of their choices.92 RECOMMENDATION 05 States should review the proportionality of sentencing policy and practice. Mandatory minimum sentences should be discontinued – particularly for non-violent crimes – and the use of alternatives to imprisonment should be promoted and strengthened. Penal Reform International and Thailand Institute of Justice | Global Prison Trends 2019