The noise generated by AI bosses on our extinction drove me back to “The Human Condition” by Hannah Arendt. Hannah Arendt published The Human Condition in 1958. It is astonishing how “prophetic” the book sometimes seems.
Here, she considered two possible criticisms of the political judgment of scientists who developed atomic weapons: (1) if they had character, they would have refused to develop them; and (2) they were naïve about how their invention would eventually be used.
The interesting thing is that these were not Arendt’s biggest concerns. Her deeper concern was that scientists inhabited a world where speech had lost its power.
The context of “speech losing its power” is that, in science, equations increasingly became a language for describing scientific discoveries. These equations can become so dense that it is very difficult to translate many non-trivial scientific statements fully into ordinary words.
The movement of waves and the movement of heat, for example, are expressed using differential equations. We can explain in English what these equations describe, but the English explanation does not contain everything that is expressed mathematically.
We had a version of this problem during the Great Financial Crisis. The mathematics and modelling required to understand something like a synthetic Collateralised Debt Obligation could become so complex that explaining and quantifying the risks in language that non-mathematicians could easily understand became a problem.
It was not simply that the mathematics was difficult. These instruments depended on assumptions about things such as defaults, correlations and how different risks interacted. You could express these things mathematically and build models around them, but translating what those models were actually saying about risk into ordinary language was much harder.
Meaning could be lost in translation.
And this mattered because the people making decisions about these instruments were not necessarily the people who built the mathematical models. A model could produce numbers, ratings and measures of risk that appeared perfectly understandable, while much of the complexity and the assumptions underneath those numbers remained hidden.
Move forward to AI and the problem becomes even more interesting. The mathematics involved in training AI assistants involves huge amounts of linear algebra, calculus, probability and optimisation operating across enormous numbers of parameters. The issue is not necessarily that every individual equation is more complicated than the equations used elsewhere in science, but that the scale and interaction of the mathematical operations become extraordinarily difficult to express meaningfully in ordinary language.
This seems to me to be part of what Arendt was getting at. Human beings develop increasingly powerful technologies, but our ability to explain, debate and judge what we are doing may not develop at the same pace.
There is another level to the problem in these social-media days.
Apart from scientific and mathematical language, we have invented another powerful way of making speech lose its meaning: misinformation and disinformation.
Humans are constantly and deliberately flooding social media with untruths.
This is not exactly the same problem Arendt was discussing. Scientific language can become so specialised that it becomes difficult to translate back into ordinary speech. With misinformation, the words themselves are perfectly understandable. The problem is that they become detached from factual reality.
People can therefore appear to be speaking the same language while inhabiting completely different factual worlds.
In that sense, misinformation and disinformation become another direct attack on our shared systems of meaning.
Arendt wrote:
“For the sciences today have been forced to adopt a ‘language’ of mathematical symbols which, though it was originally meant only as an abbreviation for spoken statements, now contains statements that in no way can be translated back into speech. The reason why it may be wise to distrust the political judgment of scientists qua scientists is not primarily their lack of ‘character’—that they did not refuse to develop atomic weapons—or their naïveté—that they did not understand that once these weapons were developed they would be the last to be consulted about their use—but precisely the fact that they move in a world where speech has lost its power.”
Perhaps there are now two ways in which speech can lose its power.
One is when knowledge becomes so specialised that ordinary language struggles to carry it.
The other is when ordinary language becomes so saturated with misinformation that words lose their connection with shared reality.
The first creates a gap between technical knowledge and public understanding.
The second undermines the common factual ground that makes meaningful public discussion possible
