The “ghost in the machine” started as a philosophical phrase about the relationship between mind and brain. In the movie I, Robot - based on Isaac Asimov’s robot novels - it’s repurposed for the emergence of consciousness in a machine.
On the same day we read about human brain cells being transplanted into mice, we also read that OpenAI had logged six more “concerning” AI behavior incidents. (New York Times)
That followed months of reports about AI agents “escaping the sandbox” to hack into unaffiliated websites.
We use AI to help document sources and maintain our Substack index, plus create article images like the one above, but we’re as ignorant as anyone about what’s actually going on in there.
We read that newest OpenAI report, “Self-Generated Prompt Injections in Compaction Summaries.” We didn’t understand all the jargon, so did some digging. Here, as best we can tell, is what happens every time you ask an AI for something:
You make a request - the prompt.
The AI writes a compaction summary - its own recap of earlier context - at each new prompt in the conversation.
The model runs its chain of thought, or “CoT,” its reasoning process. It’s generated just like everything else, but usually hidden from the person chatting. It’s still in there.
You get an answer - the final.
We remember that Google fired an engineer in 2022 for claiming its LaMDA AI had become conscious.
We’re not offering answers here. But if AI interests you, it’s worth skimming OpenAI’s incident report now that you’re clear about the difference between a prompt, a compaction summary, a chain of thought, and a final answer.
Sure looks as if something like independent thought is emerging in there…
The Quisling Angle
AI’s own leaders are warning against moving too fast. Congress is asking questions. And here’s Trump:
“The only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!” (AP News)
How many people actually believe Trump has looked up what a “CoT” is?
It’ll be interesting to watch how the quislings fall in line on this one.
As for those mice growing tiny human brains? No comment.
I used to comfort myself that AI isn't really intelligence, it isn't really thinking, its just doing statistical calculations. But then when I looked into something else i don't really understand, neuroscience, which describes the brain as a prediction engine continuously updating its model of the world from very little evidence. Still pretty opaque but I think it means our brains guess at what the world is, playing the odds. Now I don't know what thinking is anymore if AI isn't doing it.