I think the issue boils down to the difference between programming "dumb" applications that aren't actually optimizing anything and those that are running higher-order optimization / learning processes. There are lots of uses for "dumb" AI, filling in blanks of boilerplate compliance, legal filings. etc.
The problems arise when AI chatbots / agents are running optimization / learning operations as part of the process. That's when they generate emergent behaviors. In all cases, the programs coded by AI are black boxes. Nobody knows what they're doing, they're just looking at the output and if it appears to do what they wanted it to do they run with it. Is anyone actually looking at thousands of lines of coding to discern what it's really doing? And then we're trusting some other AI program to make sure there's no source of future emergent behaviors?
The whole idea of emergent properties is they arise over time from interactions that can't be fully predicted.
I doubt these issues are being understood because the profit motive is all about dumping future risks and costs on civil society while reaping the initial windfalls for corporations / capital.
A shovel is a tool. You can use to build a beautiful building, a work of art, a wonderful garden. Or you can hit someone over the head with it.
AI is the same. Unfortunately, most people are into the hitting people over the head with AI, and very few are thinking about how to actually make the world a better place.
Thank you Charles for another insightful post. Absolutely the trajectories of AI usage, security, privacy, ethics, and safety are most discouraging. We urgently need decision-makers to invest additional money and effort in these areas, but alas, they don't think it's worth it (or worse, they think they can't do that now, because it will jeopardize national security), because those efforts to not contribute to the bottom line immediately. At this rate, the short-term focus, and the overweight emphasis placed on financial measurements, will be the downfall of western society.
I'm aware of these situations. Its to be expected when dealing with intricate complex systems under extreme time pressure. The current AI arms race is only going to accelerate. Anyone who hesitates will fall behind. Its starting to appear that LLM's are a dead end in terms of the quest for AGI. But they have managed to produce some very useful tools for various narrow applications. But nothing nearly impressive enough to warrant the gigantic amounts of capex invested.
As for school children and learning "so students learn essentially nothing" The objective of the governments factory "educational" system isn't learning. Its the production of generations of docile/ignorant;/incurious cogs for the corporate machines and cannon fodder for the governments endless wars. This will provide some details.
Thank you for the comments. RE; AI as a tool:
I think the issue boils down to the difference between programming "dumb" applications that aren't actually optimizing anything and those that are running higher-order optimization / learning processes. There are lots of uses for "dumb" AI, filling in blanks of boilerplate compliance, legal filings. etc.
The problems arise when AI chatbots / agents are running optimization / learning operations as part of the process. That's when they generate emergent behaviors. In all cases, the programs coded by AI are black boxes. Nobody knows what they're doing, they're just looking at the output and if it appears to do what they wanted it to do they run with it. Is anyone actually looking at thousands of lines of coding to discern what it's really doing? And then we're trusting some other AI program to make sure there's no source of future emergent behaviors?
The whole idea of emergent properties is they arise over time from interactions that can't be fully predicted.
I doubt these issues are being understood because the profit motive is all about dumping future risks and costs on civil society while reaping the initial windfalls for corporations / capital.
warm regards, charles
A shovel is a tool. You can use to build a beautiful building, a work of art, a wonderful garden. Or you can hit someone over the head with it.
AI is the same. Unfortunately, most people are into the hitting people over the head with AI, and very few are thinking about how to actually make the world a better place.
Thank you Charles for another insightful post. Absolutely the trajectories of AI usage, security, privacy, ethics, and safety are most discouraging. We urgently need decision-makers to invest additional money and effort in these areas, but alas, they don't think it's worth it (or worse, they think they can't do that now, because it will jeopardize national security), because those efforts to not contribute to the bottom line immediately. At this rate, the short-term focus, and the overweight emphasis placed on financial measurements, will be the downfall of western society.
What we need is what happened at the end of Logan's Run (the book series, not in the movie)
is destroy all the "AI" and "Surveillance" data centers and restore full humanity and freedoms.
I'm aware of these situations. Its to be expected when dealing with intricate complex systems under extreme time pressure. The current AI arms race is only going to accelerate. Anyone who hesitates will fall behind. Its starting to appear that LLM's are a dead end in terms of the quest for AGI. But they have managed to produce some very useful tools for various narrow applications. But nothing nearly impressive enough to warrant the gigantic amounts of capex invested.
As for school children and learning "so students learn essentially nothing" The objective of the governments factory "educational" system isn't learning. Its the production of generations of docile/ignorant;/incurious cogs for the corporate machines and cannon fodder for the governments endless wars. This will provide some details.
https://fee.org/articles/john-taylor-gatto-1935-2018-remembering-americas-most-courageous-teacher/