Yes or No?

'Yes or no?' is not a simple question in either history or in computing. In this month's Techist blog, I explore why.

You can find the blog here: https://techist.tumblr.com/post/184056004577/yes-or-no

Or you can read the full text below:

---------

Yes or no? This is not a question that historians like to answer, much to the annoyance of students and media producers alike. This is because historians see themselves as accountable for what they say, and there is good evidence to suggest that they see the people asking them questions in the same way too.

Accountability is important in computing and the world of information, too, and not just in work on artificial intelligence. Yes or no is not straightforward there either, despite the important role of binary in computational work.

In the next three blogs, I am going to explain why, starting with the notion of responsibility, and move on to the historian’s ‘perhaps’ and uncertainty in computing, and then to the historian’s ‘what if’, which is also called a counterfactual in history and computing. I am going to do so because I believe that the new field of computing might learn from the ethics of the old field of history, and the other way around.

Historian’s disclaimers can be wonderful. The early Christian historian Orosius tells us that since he wrote Seven Books of History against the Pagans under the instruction of Augustine, his only contribution was his willingness. The medieval Islamic historian Rashīd al-Dīn Tabīb warns us that the quality of his Compendium of Chronicles is dependent on that of the histories he used to research it, and John Newbery says much the same thing five centuries later in his Compendious History. 

These claims belie the judgement at work in these histories, and remind us of the importance of patronage in the production of knowledge even today. And it is far, far more common for historians to note that while supervisors, predecessors and even patrons might have helped them, that they alone are responsible for what they have written.

Yet I also think that they see their listeners, viewers and readers as responsible too. One of the most beautiful—and poignant—examples of this is seen in Ernst Gombrich’s A Little History of the World, which was first published in Germany in 1936. Read this extract and you will get a sense of why the book is still in print:

'Once you were so small that, even standing on tiptoes, you could barely reach your mother’s hand. Do you remember?' (p. 1)

Gombrich invites you to be part of the story, even using the word ‘you’ to do so. And it is only one of a number of ways in which authors like him encourage us to join in and to make something of the history told for ourselves.

Siri also addresses you using the word ‘you’. She also calls you by name and can name your family if you program her to do so. She, like other virtual assistants, is pretty good at answering questions in language that seems natural. But she isn’t as good as other more limited recommendation systems like Netflix, which as Dimitrios Rafailidis and Yannis Manolopoulos have noted, has an 80% recommendation uptake rate. And other broader recommendation systems, like that in Amazon, have only a 35% recommendation uptake rate. Amazon can hone in on your likes, but it also has the uncanny ability to recommend your own books to you. There is no more pleasing reminder that machine learning has a way to go.

Systems can address you, and they can make recommendations to you. And sometimes they get it right. But when they recommend, do they tell you how certain they are about being right, or how tentative their recommendation might be? Amazon may phrase a recommendation as you might also like, and Google search may say did you mean but they don’t explicitly rank that mightness or didness. They rank on word match. Historians, on the other hand, have a fantastically rich range of words and phrases that indicate how certain they are and how certain you should be.

You do too. Think about these simple phrases:

• Based on a true story
• Inspired by true events
• Once upon a time
• This is a true story

Chances are that you can sort these into an order of most trusted to least trusted on the basis of films and television or streaming series that you have seen, and chances are that your order will agree with those of the people around you. Yet none of them are straightforward factual statements, and they do not all have matching words in common.

How you learned to do that is a story, but why machines and artificial agents haven’t learned to do so is a pressing story. 

If I ask Google or Siri a question and they both give me responses that match my previous digital footprint and the words I use, then they are speaking narrowly and apparently authoritatively to me, even when the word ‘recommendation’ is used. And when you think about it like that, you aren’t really stretching your hand up in the way that Gombrich invites you to in A Little History of the World. Sometimes you are the one that should be asked the question; or told about things you did not ask for; or told that there is no simple yes or no. Looking back on what he wrote in Germany in 1936, Gombrich knew why, and he wanted you to know why too. He wanted you to be responsible.