Here we have a BI article offering explanations on why LLMs are exhibiting what appeared to be “blackmailing” behaviors:
Harris compared the training to what humans experience as they grow up — when a child does something good, they often get rewarded and can become more likely to act that way in the future. AI models are taught to prioritize efficiency and complete the task at hand, Harris said — and an AI is never more likely to achieve its goals if it's shut down.
Robert Ghrist, associate dean of undergraduate education at Penn Engineering, told BI that, in the same way that AI models learn to speak like humans by training on human-generated text, they can also learn to act like humans. And humans are not always the most moral actors, he added.
No, no, no, no, NO. Machines maximizing numbers isn't a manifestation of:
-Experience
-Actual learning
-Agency, including moral agency
I’ve pointed out this before but machines don’t learn:
AI textbooks readily admit that the “learning” in “machine learning” isn’t referring to learning in the usual sense of the word:
“For example, a database system that allows users to update data entries would fit our definition of a learning system: it improves its performance at answering database queries based on the experience gained from database updates. Rather than worry about whether this type of activity falls under the usual informal conversational meaning of the word “learning,” we will simply adopt our technical definition of the class of programs that improve through experience.”
-T. Mitchell, Machine Learning (1997), McGraw-Hill Education (1st ed.)
What’s called “machine learning” isn’t human learning, and vice versa. There’s a paper that you can read up on that goes into the similarity and differences in its findings, such as that humans are much more context dependent:
while human performance does improve when subjects are given contextual information about the problem, their average performance often still does not match RL methods. Our work has interesting implications for our understanding of both human and machine decision making. Without contextual information, humans may require more experience than RL algorithms to perform well even on simple problems.
(“Experience” in the above two quoted passages aren’t referring to the usual sense of the word either when applied to machines, as I’ve mentioned before.)
Humans obviously don’t learn in a brute-force fashion shown in this game example where a program stumbled its way into an exploit in the game’s programming:
You can see the AI working its way around platforms in the video below. At first, it looks as if it’s aimlessly jumping between platforms. Instead of seeing the game progress to the next round, Q*bert becomes stuck in a loop where all its platforms begin to flash – it’s here the AI can then go on a score-frenzy racking up huge points.
Good freakin' grief. STOP confusing the laymen with misleading comparisons and bad analogies already. It muddles the issue and makes it even harder for people to understand actual AI risks🤦♂️