A phrase in the recent Economist obituary about Ebrahim Raisi, the president of Iran, immediately struck me.
According to the obituary, in the moments before his helicopter crashed, Raisi “stared sombrely” out of the window. And at the end of the obituary, “the president stared out of the window, unsmiling, as the fog closed in.”
I put it outside the ability of even US intelligence agencies to know that Raisi was staring at that moment and the way in which he stared.
Quotes like this are a symptom of the way publications like the Economist and others project beliefs onto historical occurrences.
But this symptom is not new. When I read the Economist piece, I was also reminded of a section from the book What Do You Care What Other People Think? by physicist Richard Feynman.
Feynman mentioned an experience he had as a child while going to what he called “Sunday School.”
The actual crisis came when I was eleven or twelve. The rabbi was telling us a story about the Spanish Inquisition, in which the Jews suffered terrible tortures. He told us about a particular individual whose name was Ruth, exactly what she was supposed to have done, what the arguments were in her favor and against her — the whole thing, as if it had all been documented by a court reporter. And I was just an innocent kid, listening to all this stuff and believing it was a true commentary, because the rabbi had never indicated otherwise.
At the end, the rabbi described how Ruth was dying in prison: “And she thought, while she was dying” — blah, blah.
That was a shock to me. After the lesson was over, I went up to him and said, “How did they know what she thought when she was dying?”
He says, “Well, of course, in order to explain more vividly how the Jews suffered, we made up the story of Ruth. It wasn’t a real individual.”
That was too much for me. I felt terribly deceived: I wanted the straight story — not fixed up by somebody else — so I could decide for myself what it meant. But it was difficult for me to argue with adults. All I could do was get tears in my eyes. I started to cry, I was so upset.
He said, “What’s the matter?”
I tried to explain. “I’ve been listening to all these stories, and now I don’t know, of all the things you told me, which were true, and which were not true! I don’t know what to do with everything that I’ve learned!” I was trying to explain that I was losing everything at the moment, because I was no longer sure of the data, so to speak. Here I had been struggling to understand all these miracles, and now — well, it solved a lot of miracles, all right! But I was unhappy.
In the two examples above, the reader can tell that the stories must have been enriched, whether in superficial ways (staring out the window somberly) or in more sophisticated ways (what we say Ruth thought).
It is deception. But rather than claim that readers are being maliciously deceived, these false examples could be seen as harmless embellishments that add meaning, or even enjoyment. But we should note them.
Actually Doing It
A counterpart to imagining what people think is seeing what people imagine. There are a few AI-enabled attempts at this today.
One is to recreate images seen by human subjects.
In the paper “High-resolution image reconstruction with latent diffusion models from human brain activity,” the authors describe how they used fMRI (functional magnetic resonance imaging) along with a latent diffusion model (Stable Diffusion) to learn what people see. From the sample images below, you can see that the results are pretty good. (Original images in top row, generated images in bottom row.)
As the technology progresses (and potentially does not require an fMRI machine), you can imagine both positive and negative applications: dream recorders, espionage, preference revelation, fear discovery, persuasion, and more.
Another version of this is to recreate text that a human subject reads. In the paper “Semantic reconstruction of continuous language from non-invasive brain recordings,” the authors describe how they used an fMRI machine (rather than a brain implant) to produce strings of text. Interestingly, the authors found that they could decode off multiple parts of the brain.
The challenge here is that it can be more difficult to identify individual words than it is to generalize an image. For that reason, the researchers’ system had to also guess at likely word sequences rather than consider random word strings.
From the sample above, meaning comes into play more than the exact words. You could imagine a scenario where the phrase: “take a bow” could be decoded differently depending on whether the subject read it as bowing after a performance or taking a bow made of ribbon.
Also, in this test, the subjects who were told to be uncooperative were able to reduce the accuracy of the output. That reminded me of the way people can beat lie detectors by modifying their stress rate.
Still, I’m struck by the accuracy of both examples.
Techno Richelieu Gets His Six Lines
Elsewhere in this blog I developed what I call the Techno Richelieu Effect, named after Cardinal Richelieu, who supposedly said “If one would give me six lines written by the hand of the most honest man, I would find something in them to have him hanged.”
While I earlier explored the way today we produce many more than the equivalent of six lines of text every day, the brain readings of above are different. Rather than reevaluations of what we write and say, we could uncover and interpret what we remember and think.
This blog is partly about imagining unintended future outcomes. Whether through brain scans or other means, what does the future of mental privacy hold? How might the brain reading technologies outlined above be used for goals only just imaginable today?
We could imagine a few including:
- Testing awareness of certain topics. Prompting subjects with images or text to check the responses.
- Lie detection. While people can learn to beat lie detectors and also brain imaging devices, there would be a learning curve. Would non-cooperation be more apparent in the brain scan devices?
- Morality confirmation. Does the subject have undesirable memories?
Another version of Techno Richelieu is “give me the man and I will find the crime.” The six lines are not even necessary. There are Eastern European and Soviet variations of this approach (and even American ones), but I thought that the Polish version was interesting.
In Censorship, Translation and English Language Fiction in People’s Poland by Robert Looby, we learn about the way the Joseph Heller’s novel Catch-22 was translated into Polish:
“Catch-22 might be considered an exception to the no-sabotage rule. The title was translated as Paragraf 22, which means ‘Paragraph 22’, as in a paragraph or section of the legal code. What in the original is left very vague is in the Polish translation concretised as a statute. There is a saying, often attributed by Poles to the Stalinist prosecutor Andrei Vyshinsky: ‘Give me the man; there’ll be a paragraph for him'”
How different would Catch-22 be if written today?