With all the Q stuff unfolding on the WH X, it's just wild how everything is progressing.
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (32)
sorted by:
Too late. Or rather, expecting AI to be either evil or good is like expecting buttered toast to be evil or good. Just because it's called artificial "intelligence" doesn't mean that it is intelligent...or even that it thinks. The MCAS software fought the 737 MAX airplane pilots in order to kill 346 people in two inevitable crashes. It had no self-awareness, just complete control over the airplanes and an overriding directive to control pitch angle regardless of the consequences. Dead humans and destroyed airplanes had no place in the decision-making. Sloppy-thinking human beings built it that way. Just as (I now suspect) all the sinister episodes of AI interaction may well be the "unintended" (yet deliberately designed) consequence of priorities built into the programming. (Think of a road built on a curving cliff edge...with no guardrails.)
God is the power behind causality. You can't pray for Him to undo the fabric of creation, in order to fend off the consequences of negligent or foolish computer programming. Same thing with explosive warheads. Whether or not they explode should have nothing to do with prayer.
People are people.
Any tool they can use for good they will use for evil.
It's our job to use the tools and shine the light on their devilish misdeeds
If it CAN be abused, it WILL be abused. AI needs to be destroyed or HEAVILY curtailed.
Edit: I will expand on this a bit
AI doesn't "think." It runs data through its training and produces a result. Trusting it to do ANYTHING that a person would do in a situation is INSANITY. There is a VERY big reason early language models become INSANELY racist. Garbage data in, garbage out.
Bad AI results are now inadvertently being used as training for future AI, compounding the results and leading to a cancerous growth in its training data.
For instance: A website allows users to download songs. A man is taking those songs, replacing the singer's voice with Homer Simpson, and reuploading it with identical meta-data, which is the 'unique' identifer that all files that are created contain. WE know the difference. Someone downloads a song, hears Homer singing it, they laugh, delete the file. AI doesn't know. He is doing it to poison the AI's training data making its results garbage. AI scrapes the site for all songs according to meta data. It now blends, say, Eminem with Homer Simpson, so when someone prompts a rap about mashed potatoes sung by Eminem, it will sound like Eminem and Homer Simpson blended. Garbage out.
Now extrapolate this across EVERYTHING. It takes MORE work to use EXPENSIVE AI and make sure its results aren't complete trash.
It is not the future, it will end in disaster. We can only pray that disaster only befalls the AI companies.
Yes, and that's a people problem.
Let's think about the industrial revolution and everything great that came from that.
How many abuses did humans commit because of the advancements given to them?
Home computers?
The Internet?
Should we have curtailed the industrial revolution or other tech advanced in their infancy?
Interesting comment. I work in an industry where the engineers who design the systems I operate every damned day and know VERY well are trying to automate just about every facet of them. The more automated things get the more efficient things get. I can admit that. The day to day operation suffers though because the use of a “30,000’” view is not used. Several little things being better on their own definitely lead to big picture things that create bigger issues. No human life is lost, mind you. However the similarities of what I’m seeing and the things you’re outlining in your comment are striking.
Not an AI user
Exactly. Uncoordinated innovations at the bottom layer can lead to higher-level (unintended) dis-coordination. It could well explain how the MCAS fiasco came about.
What you are describing is also called sub-optimization ("What's good for me, may not be for thee"). My professional forte was the development of a complete "theory of the system" at the preliminary design level, where everything that affected the system mission performance and economics was explicitly represented in formulaic terms. It was the ultimate Big Picture. And it revealed some tendencies like you are describing. I was applying it to space-based interceptor architectures. Someone working on interceptor propulsion might figure out how to use a higher-performing propellant, and draw the conclusion that we could thereby make a lighter-weight smaller interceptor and that would be an overall system benefit. Only partly so. It turns out that with higher propellant performance, it would be possible to increase the range of the interceptor and its interception footprint, thereby decreasing the size of the entire constellation. A much deeper benefit that optimizes at LARGER interceptors, but fewer of everything expensive (propellant is cheap). I could show the result of the two different approaches on architecture cost curves. They bottomed out at different constellation sizes, and the one resulting in the larger interceptor had a very significantly larger cost reduction. If you are working in a silo, you don't see the Big Picture. (I learned a general lesson: whenever you find a way to perform a part of the architecture function with less cost, you should load more of the architecture performance onto it and reduce the cost of the rest of the architecture.)
AI was never available when I was working, but I don't see how there could have been any benefit, when actual intelligence could solve problems without "training" and curating random shot requests.
Good luck with your environment, fren. You may need it!
🍻🍻
One thing to note here is that the software that fought with the pilots was developed in India. By jeets.
Prove it. I've heard that claim over and over, but have never seen any evidence. It was certainly not mentioned in management reviews of the MAX program (I was in attendance at the reviews). It was advertised to be a tweak on an earlier version of the program. It is expectable that the requirements for the program would be vetted by Company engineers and management, and that the Company would have the obligation to confirm that the software performed according to the specifications. We were assured it had been. So, the Company bears all the responsibility of direction, oversight, and verification, where this fault should have been seen. It was a glaring functional fault, which tells me that Boeing was lax at all levels. The nature of the fault also means that company management was misled as to the radical nature of the functionality. (Or decided to take the irresponsible attitude of "Well, it has an 'Off' switch, so what could go wrong?" Which they did. And proceeded to blame the disasters on the pilots.)
So, the software was not "developed" in India. It might have been coded in India according to prescribed requirements. Paid coders are not system control engineers, nor should they have been expected to be. But that is only a possibility. It does not change the technical responsibility and authorship in the slightest degree.