In 1986, the moral philosopher Harry Frankfurt published his famous essay On Bullshit in the Raritan Review.[1] Bullshitting, he says, is making a statement without regard for its truth value, in order to achieve some concealed goal. Politics and advertising, he said, are fields with a lot of bullshit.

Frankfurt is careful to distinguish bullshit from humbug (deceptive misrepresentation with a pretentious motive). It is closer, he says, to hot air and bluff. The analogy with hot air is striking, because hot air is an excrement from the other side of the body, that is just as necessary for emitter, and just as useless for others, as shit. The similarity with bluffing is close too. Bluffing your way out of a problem is a slightly politer variant of bullshitting your way out of a problem.

Already in 1986, before the advent of social networks and Twitter politics, Frankfurt though there was too much bullshit in the world. Today, thanks to the world-wide spread of information technology, we have reached volumes of bullshit unimaginable thirty years ago. This can be explained by the following properties of computing technology.

Physical symbol manipulation

Computers manipulate symbols physically. Data, at the lowest level, is a physical state of a physical object. It is a physical mark on paper, a pattern on a screen, a state of a disc or of a communication medium. Data comes with manipulation rules defined, ultimately, in terms of the physical form of the symbol. In other words, it is manipulated in terms of the physical properties of the symbol. Software programs are instructions, themselves encoded in physical symbols, to manipulate other physical symbols based on their physical properties.

This means that to program a machine you must formalize what you want the machine to do to the point that everything is encoded in physical properties of the symbols and in physical rules for manipulating the symbols. Formalization is the replacement of meaning with physical symbol manipulation rules.

This does not imply that physical symbol manipulation is meaningless. If the formalization is sound, and the data is meaningful, then the output of the computation will be meaningful too. Sound computations transform meaningful data into meaningful data. But with unsound computation rules or meaningless data the output will be meaningless. For the computer, there is no difference.

Physical symbol manipulation creates one condition for bullshitting: Computers manipulate data without a concern for truth. But computers have no purpose and therefore computers are not bullshitting us.


The second property of information technology that helps explain the volume of the heap of bullshit that the world is struggling with, is this: Over the past 30 years, computers were connected in a world-wide network. Now people anywhere in the world can view the results of computations done anywhere else in the world. The meteoric rise of mobile computing has not helped to reduce the problem. Anyone with a smartphone can browse the results of any computation available online and interpret the results according to their mood and context of the moment.

Computerized data is manipulated in a context-free manner, because all information needed to do the computation is encoded in the physical symbol manipulation rules. Users of the results must provide their own context to interpret the results. They can interpret the data with their own goals and knowledge level, regardless of where, when, how or why the computation was done.

If users don’t care about the meaning of the computation or the truth of the output, only care about the extent to which the output of a computer supports their goal, and keep their goal concealed, then they are bullshitting us in the old-fashioned sense defined by Frankfurt.

But there is a new sense in which they can be bullshitting us. One in which we can also be bullshitting ourselves.

Digital bullshit

People can believe that data produced by a computer is true just because it is produced by a computer and seems to support their opinion. They care about true and false but think that what is true is (1) what is online, (2) believed by many people, and (3) supports their goal. I call this digital bullshit.

World-wide computer and smartphone networks enable entirely new or greatly expanded ecosystems of conspiracy theorists, political manipulators, terrorists and pseudo-experts who use computing devices to collect whatever digital bullshit supports their goal.

A characteristic of digital bullshit is that believers do not point at facts nor use critical analysis to support their statements. Rather, they point at the fact that the data is online and believed by a large group of people. “Surely, 10 billion flies can’t be wrong. Eat shit.”

A relatively innocuous example are the relevance algorithms used by social networks and online shops. They do not measure relevance. They count numbers of physical clicks. Assuming that we click only on what is relevant, the number of clicks tells us something about relevance. But without this assumption, we are bullshitting ourselves about what is relevant. Even if 2.5 billion people click on a data item, this does not mean it is relevant.

Similar remarks apply to Friend links on social networks and to online news provision based on preferences measured in terms of numbers of clicks.

For big data analysis this does not need to be a problem, because statistical algorithms can still derive useful information from very large sets of inaccurate data. But it may pose problems for arguments based on a small number of individual data items. Like all data items, they have been physically manipulated by a machine without regard for their truth value. Adding human interpretation and purpose to the output, this may create information bubbles based on digital bullshit.

Now what?

There is nothing we can do about the prevalence of digital bullshit. And there is only one way to deal with it: Don’t contribute digital bullshit yourself.

Two questions suffice to put a brake on your own production of digital bullshit.

  • One: what is the meaning of the data in the real world?
  • Two: Does the computation transform true data in true data?

If you don’t know the answer to these questions, be careful and don’t claim more than you can know.

[1] It was reissued in 2005 by Princeton University Press as an 80-page booklet.