ANU Africa Network

By Bernard Baffour

This article was originally published by the ABC.

Immigration decisions are increasingly being outsourced in order to save costs and optimise resources. It is often efficient to rely on algorithms that analyse historical data and identify patterns to indicate potential risks or outcomes. In most government departments, resources such as time, staff, and finances are limited, and these algorithms can provide insights into where resources are most needed for efficiency planning.

In most cases, these algorithms can provide enhanced decision-making by distilling the complex and uncertain factors, and making data-driven recommendations. Human decision makers can then supplement their expertise and intuition based on these algorithmic recommendations.

The use of algorithms in this way and for these purposes raises some significant problems, however. Let me highlight them through the following scenarios.

How algorithms can reproduce bias

Officials are often faced with the difficult decision of screening for children at risk of child maltreatment. Most models usually include race (of child or parent) in their range of relevant factors. But although race is highly predictive, it can reflect inherent biases and prejudices. In this example, a predictive model which did not use race was found to be better, because the model which included race screened in low-risk black children and screened out high-risk white children.

In other words, this model made fewer false positive decisions (flagging children as “at-risk” in predominantly “black” families when the children were actually “non-risk”) and fewer false negative decisions (flagging “at-risk” children as “non-risk” in largely “white” families).

In the second example, it is well known that Australia’s Indigenous population has significantly lower levels of education, income, employment, and health and wellbeing, such that the proportion of Indigenous people in an area is highly correlated with socio-economic disadvantage. While this is a “statistically significant” variable, its use “problematises” Indigeneity.

Hence, in research I conducted with Sumonkanti Das, Mu Li and Alice Richardson, using geographic remoteness instead of the proportion of Indigenous people in an area was found to be most effective. This has the advantage of capturing the relative isolation of local communities both for Indigenous areas and rural neighbourhoods.

Applying a debiased predictive model to visa applications

The Department of Home Affairs deals with thousands of visa applications on a daily basis, and predictive analysis is extremely valuable in providing quick and efficient decisions. In most cases, “country of birth” or “nationality” is highly correlated with whether a person will violate their visa conditions. But this can be liable to prejudices and institutionalised racism. As a matter of fact, the country with most visa applicants who fail to comply with their visa conditions is the United Kingdom. This is not to say that British people per se are more likely to overstay their visas — it is, rather, back-packers and young people on working holiday visas who are the likeliest offenders.

In a fairer system, if two medical doctors of equal standing were both accepted to a conference in Sydney, then granting the visa should not depend on whether they possess a passport from Switzerland or Peru, from Germany or Ghana.

A few months ago, I wrote about the experience of my sister, who is an anaesthesiologist at the Korle-Bu Teaching Hospital in Accra, Ghana. She was repeatedly denied a visa to come to Australia for a brief visit to visit me and my family in Canberra, on the grounds that she “had not provided sufficient evidence that [she] genuinely intended to stay temporarily in Australia”. The entire decision-making process was conjectural, arbitrary, and highly prejudicial. Since that article was published, many people have reached out to me, expressing similar concerns, sharing similar experiences, and thanking me for articulating what had happened to them, but lamenting the fact that they felt they didn’t have the agency to do what I did.

This experience is not limited to applicants from sub-Saharan, South Asian, and South American countries, however. The same thing has happened to people from a number of European countries. While I believe that secure borders are important for ensuring our safety and managing immigration, there needs to be some transparency and objectivity in the way the visa application process works. The biggest complaint people have is the unaccountable and seemingly “God-like” powers visa officials possess.

Removing the country of birth (or nationality) from the algorithm, and using instead observable background characteristics — such as occupation/education or labour force participation of the applicant or sponsor — will arguably be less influenced by prejudices and biases.

Indeed, well-established sponsors who are in good employment and in good standing in the community are unlikely to want to break the law by encouraging or condoning their sponsored family members or friends to break the law and flout their visa obligations. Furthermore, applicants sponsored by well-established sponsors pose less of a risk to flouting their visa obligations since their sponsors can easily be contacted or traced in the unlikely event that their sponsored family members do not meet their visa obligations.

Well-established sponsors who are in good employment and in good standing in the community are also unlikely to let their sponsored family members break the law because it will adversely affect their chances of sponsoring any more family members or friends in the future.

Transparency and fairness matter — ethically and electorally

Prejudicial treatment of visa applications could loom as a significant issue in future federal elections — people from the Global South comprise around 18.5 per cent of the electorate. In fact, this is almost certainly an underestimate, because it does not account for migrants with Australian citizenship. With more than half of the Australian population born overseas, or with a parent born overseas, it is easy to imagine how the manifest unfairness in the treatment of such a large portion of the of the population could have serious electoral ramifications.

Nor should this be seen as a problem remote from “ordinary Australians”. After all, the very idea that some citizens are subjected to a different set of rules, and thus are denied the fair treatment extended to others, could hardly be more un-Australian.

Bernard Baffour is an Associate Professor in the School of Demography and the Associate Dean of Inclusivity, Diversity, Equity and Access (IDEA) at the College of Arts and Social Sciences at the Australian National University.