“Then yeh, you’ve got a big fucken problem.” Aussie bloke down the pub said, yesterday
It’s keeping tinfoil hat wearing folk awake at night. If Singuarity is the point at which the machines take over, it’s the bad thing. It’s Skynet, it’s Asimov, it’s Arthur C Clarke, it’s the end of the world
We’ve recently invented Large Language Models (LLM) which make things like ChatGPT work. Give it a text prompt and out comes human-like language. No sensible person has a clue how it really does that, but it must do because it’s doing it right in front of our eyes. That is wizardry
Now change the word Language to the words Human Behavior. Apply the above wizardry and you have an AI which teaches itself to predict and emulate human behaviour
Some experts believe that deploying AI like that to a drone with a gun might be a shit idea
Here comes the science bit
If challenged to explain this silly idea using actual brain knowledge, this is how AI works
Neural Networks
AIs use sophisticated types of artificial neural networks that are capable of processing sequential and non-sequential data, capturing the nuances of human behavior across different contexts
Unsupervised learning is key because it allows the model to learn from raw behavioral data without needing predefined labels or categories
New data
AIs are trained using extensive datasets of human behavior. This data comes from videos, CCTV, IoT sensor data, social media activity, and other records of human actions
Understanding Context: As the model processes this data, it learns to predict behaviors in certain situations based on previous observations. This process is driven by unsupervised learning, where the model identifies patterns and relationships in the data on its own, without explicit guidance on what to learn
Self Supervision
The model is given scenarios with partial data or hypothetical situations and asked to predict outcomes based on learned behaviors. It’s like observing a scene in a movie and guessing what a character will do next based on their past actions
Fine Tuning
After the initial broad training, the model could be fine-tuned with specific types of behavioral data or for particular applications, such as understanding customer behavior in business settings or aiding in psychological analysis
Additional Training
By incorporating new and diverse behavioral data over time, the model can adapt to changes in human behavior and understand new patterns that emerge
Feedback Loops
Feedback from users and behavioral scientists can help refine the model’s predictions and make it more attuned to subtle nuances in human behavior
Architectural & Algorithmic Updates
Ongoing research in neural network design and learning algorithms can enhance how AIs process and learn from behavioral data, improving their predictive accuracy and applicability in real-world scenarios
Drones
For a few thousand quid you could probably put together a drone with a 3D printed gun and a Raspberry Pi to run the AI needed to pilot it autonomously. It’ll be connected to the main AI Hive mind of course. It will use the mobile networks for that
Link: Speculative Fiction
