AI tools are helpful and cool as long as you know their limitations. AI doesn’t exist. It’s machine learning and is as good as the data used in his training.
The general workflow of how AI works is:
-
Data Input: Collecting and preparing the necessary data for the AI system
-
Data Processing: The AI algorithms analyze the data, identify patterns, and learn from them
-
Decision Making: Based on the learned patterns, the AI system makes predictions, recommendations or decisions
Overall, the key to how AI works is its ability to learn from data and improve its performance over time.
But AI build on biased data sets will give biased results.
Some common issues that lead to AI mistakes include:
-
Brittleness: AI can only recognize previously encountered patterns and can be easily deceived by new situations
-
Embedded biases: If the training data for AI contains biases, the AI system will reflect those biases in its decision-making
-
Catastrophic forgetting: AI systems can abruptly forget previously learned information when trained on new data
So what?It is good or not? Answer: It’s a tool, use it as one and not as an oracle of truth.
More source
AI is NOT Artificial Intelligence, the real threat of AI is “Automated Stupidity”
Regarding biases, it’s worth noting that generative models (she talks mostly about labelling models) don’t simply reproduce biases, but also exarcebates them. My go-to example is an image generator that you “train” with four pics:
In this set, the typical tool shape is a wrench, and the typical colour is red. So every time that you ask the generator “pls gib tool”, it’ll give you a red wrench. If this doesn’t look like a big deal, now imagine that it’s outputting pictures of people, and couple it with what Angela says around 25:00.
I’m still watching the rest of the video. So far I think that she’s being spot on.
