Opinion | A Finnish prison experiment and an ethical question

It appears that governments across the world are confused, even schizophrenic, about how they deal with advances in Artificial Intelligence (AI) and the human costs they may cause through loss of employment.

I have written before in this column about Finland’s experiment with Universal Basic Income (UBI). Under a UBI programme, all citizens get paid a basic wage whether or not they are employed. Many UBI proponents, including AI entrepreneurs such as Elon Musk and Mark Zuckerberg, have said that they see UBI as the only solution to the problem of mass unemployment caused by AI and other advances in information technology (IT).

The basic premise behind UBI is not unlike the premise behind universal healthcare coverage systems. Northern European welfare states have grappled with such issues for decades. They have outrageously high rates of taxation, but, relative to other nations at least, a semi-efficient way to plough these taxes back into welfare schemes for their citizens.

Finland first began experimenting with the concept of UBI in January 2017. The pilot programme allowed 2,000 unemployed Finns to receive a UBI dole, even when they tried out casual employment at odd jobs. These 2,000 were to be compared against a control group of 137,000 employed Finns to allow the government draw a conclusion on UBI’s efficacy.

There are two problems with every dole, however. One, it must be paid for by all citizens, which means higher taxes, and two, doles act as a disincentive for recipients who would otherwise be forced to go out and find paying work. The Organisation for Economic Co-operation and Development (OECD), an influential think tank, said Finland’s already high rates of income tax would have to further increase by nearly 30% for the country to be able to fund a UBI.

As many expected, Finland’s government finally decided after the two-year pilot scheme that UBI would end. It is now investigating other schemes for reforming the Finnish social security system. Meanwhile, apologists for the programme, including The New York Times, claimed that it was not UBI that failed Finland, but rather the reverse—that Finland failed UBI as the pilot programme was too limited in scope to produce meaningful results.

Surprisingly, there is now news from Finland that is the polar opposite of what the country was trying to do with UBI. Finnish prison labour is being used to feed AI. To throw light on this, I first need to explain one of the things that makes AI useful for applications such as driverless cars: inexpensive labour in countries such as India and Sri Lanka where employees spend hours on end categorizing and labelling data elements so that they make sense to an AI programme. In the case of driverless cars, these would be simple things such as a red light versus a green light at a traffic intersection, or more nuanced such as an older person who is crossing that intersection at a slower pace than other pedestrians. Finland has now decided to turn to a source of inexpensive labour, its prison inmates, to categorize and label these data elements.

Last week The Verge (a media website) reported that inmates at two prisons in Finland are performing a new type of labour: classifying data to train AI algorithms for a startup. While the startup in question, Vainu, sees the partnership as a kind of prison reform that teaches valuable skills, other experts say it plays into the exploitation of prisoners being required to work for very low wages. Vainu has been using Finnish prisoners for this work for the last three months.

Vainu is trying to deliver its AI operations in the Finnish language, which is why this work cannot be moved out of Finland. The company is building a comprehensive database of firms around the world that helps Finnish businesses find contractors to work with. For this to work, people need to read through hundreds of thousands of business articles scraped from the internet and label them accurately—in Finnish. In the English-speaking world, one could easily turn to outsourcing firms in India or Sri Lanka or to Amazon.com, Inc.’s Mechanical Turk or to a CrowdFlower, Clickworker or Toluna account that outsources the work.

Vainu found cheap labour by enlisting the Finnish Criminal Sanctions Agency (CSA), which had access to more Finnish-speaking labour than Mechanical Turk, according to The Verge. Vainu is now having its work done at two Finnish prisons, one in Helsinki and the other in Turku. In turn, Vainu paid the CSA roughly the equivalent of what it would have paid on Mechanical Turk. The Vainu team hopes to expand elsewhere in Finland and other countries where it can be hard to find people willing to do this type of work in local languages.

Mechanical Turk is itself no great paymaster. Research by Kotaro Hara, a professor at Singapore Management University, and his collaborators reveals that the median wage of a worker at Turk is around $2 an hour. Compare this to the US minimum wage of $7.25 an hour, and the savings become readily apparent.

The philosophical and ethical questions that this precedent of using prison labour brings to light are beyond the scope of this column. Suffice to say that if feeding AI algorithms is already exploitative, one shudders to think how much more exploitative the algorithms would be once finally in place.

Siddharth Pai is founder of Siana Capital, a venture fund management company focused on tech.