I have completed my historical novel, using Gemini AI to help write the text (which is research-heavy since it’s set in the 1700s in Africa and the Mediterranean area). The book has over 25 AI-generated illustrations by Gemini, ChatGPT and Microsoft Image Generator. (Plus an original illustration by me, a 1710 map with markers I placed to identify the many locations in the story, and original family photos.)
I am amazed by the capability of free AI to generate beautiful pictures if they are given detailed prompts. The AIs will add period-appropriate details such as fabrics, chandeliers, etc. on their own.
AIs follow patterns in their incredibly huge memories. I was surprised to see that they do not understand spatial relationships the way a human would.
One item I wanted illustrated was a signet ring. Humans understand that the intaglio on a signet ring needs to be deep enough to leave an impression in wax and also needs to be the mirror image of the impression on the wax.
It was easy for all the AIs to understand the silver ring with the Hebrew lettering on lapis lazuli. It was easy for all the AIs to understand the wax with the impression of the Hebrew letters spelling the word “Azariah.”
It was difficult for the AIs to understand that the lettering on the ring needed to be the mirror image of the lettering on the wax. After several unsuccessful tries, I gave up on ChatGPT and Microsoft Image Generator. Gemini finally succeeded in generating the image I used in the book. But when I asked it later in a very long chat, it also failed. I kept the most successful image even though the impression in the wax is not the same size as the signet ring. Every time I tried to edit the images the AIs got more confused and the images deteriorated.
My takeaway is that AIs do not have the instinctive physical understanding of the world that humans take for granted and it isn’t easy for them to replicate it even when given specific orders. This could lead to unintended consequences. Doubly so if the AI is controlling robots which operate in 3-dimensional spatial reality.
A plug for my book: It’s a historical novel describing how a Beta Israel (Ethiopian Jew) born in 1732 makes his way through the challenging world of Africa and the eastern Mediterranean to eventually reside in newly-founded Odessa in 1795. My paternal great-grandfather was born in Odessa. Genetic testing shows that I have 0.1% East African DNA embedded in 99.9% Ashkenazi Jewish DNA. The book shows how that unlikely ancestor could have married into an Ashkenazi family 8 generations before I was born. It’s a fun book, loaded with caravans, sword fights, pirates, drug trading and gun running. Message me to get a link - it’s an e-book, $2.99 on Amazon.com.
Dad had 1% West African DNA and 3% Finnish. He was 96% Ashkenazi.
Ashkenazi in general terms is 50% Italian. The Jewish migration into Rome was not met by bans on intermarraige. Genetic testing pools people into groups that have other pools underneath.
Ashkenazi Jews are roughly 40% to 50% Southern European (often matching South Italian or Greek profiles) and 40% to 50% Levantine Middle Eastern, meaning they are close to half Italian by genetic ancestry. [1, 2]
Genetic and historical studies show that the Ashkenazi gene pool was formed during antiquity in the Roman Empire. Middle Eastern Jewish men migrated to or were brought to Italy, where they married local Southern European women. After this foundational mixing period, the community largely married within itself and later moved into the Rhineland and Eastern Europe. [2, 3, 4, 5, 6]
Typical Genetic Breakdown
Southern European / Italian: ~40% to 50%
Levantine / Middle Eastern: ~40% to 50%
Eastern European: ~5% to 10% (added later during the migration into Slavic regions)
Minor Western European/Trace: ~5% or less [1, 2, 7]
Would you like to know more about how commercial DNA tests (like 23andMe or AncestryDNA) report Ashkenazi ancestry compared to these academic models?
AI watermarks what it produces. In the contracts to use AI, free AI is not necessarily your property. I can end up in the CCO, public domain, or Creative Commons. Later claims, if AI productions become copyrightable, are possibly problematic.
You can demonstrate even worse understanding of the physical world by doing something as simple as asking AI to make a video of a person dropping a rubber ball on the ground. Or drawing a square on a white board, etc.
I would just add that, in my opinion, AI doesn’t “understand” anything (it’s just matching training data, even if that data itself has logical rules derived from human understanding).
I wouldn’t say AI has “instinct” (a vague term anyway).
AI is just a bunch of math that takes inputs in and calculates outputs.
It’s equations. Nothing more.
AI uses human language in a syntax like we do, and AI may or may not accurately render or navigate or otherwise represent 3-D physical space, but these actions are not like human understanding.
AI agents may cooperate or compete, but that’s, again, just math finding an optimal course of actions to achieve a goal/task.
All completely valid - but then we are only a few magnitudes more of little more than math.
Not suggesting AI is close to the same level, but all our decisions are little more than the sum of all our prior life experiences - input and output.
I foresee a day where someone wears an AI device that records all of their experiences (even mental) and that device would be able to effectively predict your future decisions.
At Home in the Universe: The Search for the Laws of Self-Organization and Complexity
A major scientific revolution has begun, a new paradigm that rivals Darwin’s theory in importance. At its heart is the discovery of the order that lies deep within the most complex of systems, from the origin of life, to the workings of giant corporations, to the rise and fall of great civilizations. And more than anyone else, this revolution is the work of one man, Stuart Kauffman, a MacArthur Fellow and visionary pioneer of the new science of complexity. Now, in At Home in the Universe , Kauffman brilliantly weaves together the excitement of intellectual discovery and a fertile mix of insights to give the general reader a fascinating look at this new science–and at the forces for order that lie at the edge of chaos.
“A very long chat” This is the problem. You can out of context window. Tell the chat it is not succeeding, have it write an instruction for a Project. Copy that instruction to a project instruction. Create a chat inside that project and tell it what you want.
You may have to refine the instructions a couple of times.
