Princeton University [image] has successfully simulated the neural network of a fruit fly's brain. The insect's brain capacity - comprising 166,000 neurons and over 125 million synapses—provides a model sufficient for interacting with the environment and engaging in autonomous reasoning.
The "toaster" metaphor for AI used by Prof. Pekka Abrahamsson as recently as six months ago—now falters more than ever, a fact he himself acknowledges: a toaster has tiny legs.
In a test environment, this tiny insect’s neural network—microscopic in scale and smaller than the head of a pin—learns to analyze and resolve various situations, such as steering a vehicle by performing the necessary reasoning.
About 15 years ago, I wrote one of the first overviews of the emerging world of AI for *Kanava* magazine. I used the "intelligence" of the dandelion as a coding example: the plant systematically shortens its stalk to protect itself from the blades of a lawnmower—an adversary that repeatedly attacks it. Eventually, the flower stalk becomes so short that the blade cannot harm it. Simulating a mosquito's flight was too complex
For a long time, the flight of a mosquito in a room remained such a complex multivariate problem that simulating it using a mechanical computer model was beyond reach. In a real mosquito's neural network - and the associated data fusion - far too many variables interacted simultaneously.
Multivariate modeling is advancing rapidly, yet there is still a long way to go.
A dog's brain contains an estimated 2.2 billion nerve cells, or neurons. Of these, approximately 430–623 million neurons in the cerebral cortex are responsible for complex thought and decision-making. A single cortical neuron forms, on average, between 2,000 and 10,000 synapses.
In the field of robotic automation, we are approaching a threshold where a machine's neural network—and its data fusion processes—might make unexpected, autonomous decisions, particularly if the boundary conditions governing the AI's operational scope or the "safety cut-offs" in its evolutionary code fail or leak.
Included here is an image from Princeton showing a fruit fly's brain. The fruit fly was already a classic subject for genetic modeling in high school biology classes. The Chronological Path to the Emergence of a Thinker
What has bothered me about the AI discourse since the beginning of the decade is that the loudest voices belong to parties with absolutely no background in computer science—not a single credit hour of study nor any coding experience involving even a single algorithm. This seems to result, among other things, in a failure to grasp the validity of binary logic. In my own teaching, I have used pseudocode to illustrate how intelligence functions in real-life scenarios, and I have witnessed the moment when the lights come on in people's eyes as they finally understand what is actually at stake.
Without this foundation, it is impossible to comprehend edge computing, language models, neural networks, or the workings of embedded circuits. While there is no shortage of popular, fictionalized threat scenarios, they lack any grounding in reality. I have observed that when an author truly understands and has mastered the subject matter, the reader is at least offered a chance to get a handle on the actual reality.
Princeton’s image of a fruit fly’s brain.
