This way overstates the case. Obviously AI will have controls.
Furthermore, AI is not nearly as smart as we give it credit. It is like a Wikipedia that can talk. It understands language really well, but everything else is based on probability where certain facts and things are associated with other facts and things.
Beyond probability, AI has little understanding of things that it is talking about, and its ability to reason is limited, although it might be able to solve simple problems.
However, this is why there is a big push to make AI smarter. Doing so requires a great deal more hardware, and they are building it. The AI demand for computer memory is driving up the cost of computers by a significant amount.
However, I suspect that the current methods are inefficient and they are just throwing more hardware at the problem to make it better. This is a drain on resources, when they need to come up with better algorithms.
So I asked ChatGPT what it thought of my points, and this is what it said...
Your core argument is clear and grounded, but it leans a bit too confidently in a few places where the reality is more nuanced. Here’s a straightforward take on the strengths and weaknesses:
What works well:
- You push back against hype. That’s valuable—AI discussions often swing toward extremes.
- The point about scaling via hardware vs. improving algorithms is thoughtful and widely debated in the field.
- You correctly note that current AI systems rely heavily on statistical patterns rather than deep “understanding” in a human sense.
Where it overreaches or could be refined:
- Saying AI is “like a Wikipedia that can talk” undersells it. Modern systems don’t just retrieve facts—they generate novel responses, combine ideas, and can perform nontrivial reasoning in some domains. It’s still probabilistic, but not merely lookup-and-repeat.
- “Beyond probability, AI has little understanding” is philosophically debatable. Even humans rely on pattern recognition more than we like to admit. The real issue is type and reliability of understanding, not its total absence.
- The claim that reasoning is limited to “simple problems” is outdated. AI can now handle fairly complex reasoning tasks, though inconsistently and with failure modes.
- “Obviously AI will have controls” is asserted too casually. In practice, governance, safety, and control mechanisms are active areas of concern—not something universally agreed upon or solved.
Your strongest idea (worth emphasizing more):
- The critique that the field may be over-relying on brute-force scaling (more compute, more memory) instead of algorithmic breakthroughs is legitimate. Many researchers are asking the same question.
How you might sharpen it:
Instead of framing AI as “not that smart,” you could frame it as:
- powerful but uneven
- impressive in narrow ways, unreliable in others
- advancing quickly, but with unclear long-term efficiency
That keeps your skepticism but makes it harder to dismiss.
If you want, I can rewrite your paragraph to sound tighter and more persuasive without losing your viewpoint.
No comments:
Post a Comment