Universal Credit – some lessons from using computers to calculate benefits in the 1980s

In the early days of home and office computers in the 1980s there was a voluntary and community sector national working group that looked at possible socially useful applications for these new machines, of which I was a member.

One surviving trace of our work was a published report on how to use these new word processors to print multi-lingual letters and leaflets in the non-Latin scripts used by people in Asian communities.

The other key piece of work we followed was the potential for these new computers to provide welfare rights advice, sadly not also reported at the time (unless colleagues have better records than mine).

On the face of it it looked so easy. There were quite a few welfare rights services which produced a poster every year with the current benefit rates, a bit like a large menu. You could look for your circumstances and fairly easily get a sense of how much money you could expect to receive, such a single parent with two children, one being disabled. It looked so easy to convert this ‘menu’ of benefit rates into a computer program. It should just be a simple logic tree of yes / no questions for someone sitting at the computer to answer until the program gets to the grand total.

However, the more we looked at it, the worse it got. Because behind this simple menu was a massively complicated rule book, one that welfare rights advisors kept partly in their heads and partly in large folders on a shelves. Citizens Advice Bureaux had a system of a nationally standardised filing cabinet if I remember correctly, one in each centre and updated by post.

We started with the simplest benefit at the time, Child Benefit, which was pretty much £x per child plus £y for the first child. But even this had something like 30 rules which had to be applied. Our working group soon concluded that our goal of a “welfare rights calculator” was an over-optimistic approach, and that more trained people were the best way forward.

So, roughly 25 years on, what are the lessons for the Universal Credit project to unify welfare benefits into one monthly payment, synchronised with any earnings received that month?

Well, one big change has been the growth of self-adjusting algorithms and of probability systems. Many large computer systems now use their own (secret) algorithms to decide everything from ranking Google web searches to Wonga credit worthiness. These systems have the advantage that they self-adapt, and the better systems also can spot trends early such that minority requests are not automatically disregarded. These probabilistic systems are highly sophisticated, often working with over 200 factors to arrive at a decision.

So, my conclusion here is that Universal Credit would work better as a probabilistic algorithm than as a deterministic rules machine. This would basically retain the underlying welfare benefits and their rules as applied, but get it roughly right very quickly and then smooth out the cash-flow. So, a single parent renting a flat at £a a month with two toddlers and an elderly parent would typically get £x a month based on thousands of people just like them. Safety limits would make no payment less than £b without a human authorisation. The rest is adjusted as you go along, and future claimants gain an improved service from the learning involved.

It would be introduced as a shadow scheme, sucking up data from the various current live benefits schemes to build up its knowledge base, and providing the opportunity for officials to “dummy run” the program to check it against real claims. I feel this is more ethical than trying the Universal Credit machine live on people, albeit in limited circumstances to start with.

Perhaps the political disadvantage is that the underlying benefits regimes remain as inputs, with a continuing resource cost. However, from the 1980s experience, the idea of a deterministic machine that can automate the entire benefits system without any catastrophic failure (such as people starving) was old-fashioned even then, as we found out, and as shown by the different path subsequently taken by leading commercial computing organisations.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s