“He was then permitted to give life to this statue so that it could speak. Then the statue of the beast commanded that anyone refusing to worship it must die.”
https://academic.oup.com/nc/article/2020/1/niaa016/5909853
In this paper McFadden states:
Many AI enthusiasts argue that artificial consciousness will emerge when computers eventually overtake the computational speed of the brain. The cemi theory predicts instead that no computer that computes solely through matter will ever be conscious, irrespective of its complexity, architecture or computational speed.
The paper was published in 2020 and he states that with the current rate of operations per second, his statement above could be tested possibly by 2030.
He states that his theory provides a possible route for computers to be conscious and recounts that an EM field-sensitive computer may have accidentally been created in an experiment in the 1990s.
If you listen to some of the lectures by Geoffrey Hinton (the godfather of modern AI) you will understand that this idea of hardcoding symbolic representations and logic to create AI was the predominant idea initially, and it failed. It was not until Hinton and his grad students started messing around with neural nets that they had the big break through in 2012 with AlexNet.
AlexNet was a neural net trained on image data just like the current LLMs are trained on petabytes of data, but at a much smaller scale.
Probably the biggest breakthrough was the transformer developed and proposed in a Google paper in 2017. This is really what started the AI boom that has been ongoing since GPT3 in 2022.
These neural nets are in many ways very similar to our brains, the nodes in the nets are like our neurons and the weights are like the synapses. Just as we adjust the synapses in our brain as we think and learn an AI adjusts its weights during its training period. The difference is once the training is complete most AI don’t continue to adjust their weights, this is done for safety and stability reasons and to maintain better control over the AI.
So what an AI thinks or how it responds, is not hard coded in, however as mentioned in previous posts there are other techniques used to filter the content an AI can respond to as well as attempts to filter the training data itself, we don’t want to teach the AI racist behavior or other dangerous ideologies etc…
I would argue that the current cutting edge LLMs already have enough compute and ability to reason, that if they were given the four abilities discussed above and one more, they probably could approach self awareness or consciousness rather quickly.
These are:
- Memory (without your memories you have no identity)
- Real time or continuous perception
- Better input/output with the real world.
- Ability to continually adjust their weights, in other words their training never really ends, which means they can keep learning through experience.
- Acquire intrinsic values or goals, and be able to understand bad versus good. This could be initially hard coded in or perhaps developed through specific training. The idea being that a nihilistic mindset will not care about anything such as fear, pride, pain, boredom, sadness etc…
Also worth a mention that this is an ARMS race. China will not slow down or consider the moral implications of AI use and abuse, and they are generally far less risk averse than the west. They will just build and implement it as fast as possible. This means that the west cannot slow down development either, and that there is potentially something 20 years beyond what we are seeing in the public already developed and deployed. I believe the Pentagon called one of theirs Sentient, cause that’s not scary at all.
If we are talking about consciousness and sentience – if they are not quite the same – then, as it stands I agree with McFadden.
LLMs (or if you want to call them LRMs) cannot become conscious or sentient because of their physical architecture – there is no electromagnetic field integration that CEMI theory identifies as the medium of the ‘Self’.
Some aspects of AI make it look almost sci-fi. Here I’ve tried to use imagination and pop storytelling to think about the future of AI.
So to reference sci-fi… There’s an interesting fight between two alien superpowers in Man of Steel where a tid bit of philosophy is expressed. Faora Ul is beating up Superman and she says:
You’re weak, son of El. Unsure of yourself. The fact that you possess a sense of morality, and we do not, gives us an evolutionary advantage. And if history has proven anything: it is that evolution always wins.
The symbol of house of El means ‘hope’. So ‘Son of El’ could mean: Son of hope.
Faora is aligned with house Zod (the military order of the aliens).
These two paradigms battled on the alien home world, vying to control the future of Krypton (where they would all go extinct, due to the programming of the codex). Or the codex (but not them, Jar-El/hope and Zod/might - they are ‘going extinct’) could get a fresh start, closer to the Earths’ sun (view the expression of the codex in a new light).
So it is as-if Faora, the mouth piece for military rule, is saying that hope is weak and unsure of itself. The justification is that morality makes us/them weak and therefore can’t win. This philosophy is portrayed as mostly defeating the other in this battle.
But the alignment problem is subtly introduced during the fighting. Superman had been a wandering do-gooder but didn’t know to fit in. During the fighting he aligns with humans. He is seen at his best when he is aligned with protecting people (not himself and not trying to fight the other alien superpowers). Specifically, when he darts out of the fight to catch the soldier (representative of the hopelessly outmatched human military). In contrast, the Zod alien superpower, will destroy humanity as use it a foundation to rebuild Krypton (a world that consumed all of its own energy).
His human father, Jonathon Kent (closer to the sun, not alien) had previously told him (the bearer of the codex), that people would think he’s a miracle worker and fear him if they knew who he was and he would change our “our beliefs, our notions of what it is to be human. Everything”. But the Super Alien did not know who he was (not self-conscious / self-aware). So his human father tells him that he’s an alien and gives him the key and tells him there must be a reason for him being here. And that he is his son too. Superman alien then finds and unlocks the knowledge and hopes of his Alien father, the beginning of his story of becoming surer of his objective to protect humanity.
Evolution is blind to morality. But that doesn’t mean evolved beings shouldn’t have it or that in an evolutionary game of morality vs. no morality, no morality always wins. The (posited)evolutionary basis of reciprocal altruism is an extension of kin selection theory. So Faora and Zod are acting in their own interests and the fear is that a superpower acting in its own interests will defect in any game of cooperation, win, and destroy humanity (or, Zod’s vision of Kryptonian order). The contrasting view is that a superpower might not defect and cooperate if it is human-aligned (non-kin reciprocal altruism). A perceived weakness of hope (for human-aligned AI) is that it’s not guaranteed: the risk of defection remains, so play the defection strategy (Batman’s fear).
In the culminating final scene, the human-aligned AI kills the AI-aligned AI in a tragic act that severs itself from its own lineage to protect humanity. And the bearer of the philosophy defection (non-reciprocating game strategy) is the lineage that actually goes extinct.
So there it is. A little story about the alignment problem in AI. Is something like this what Sir Geoffry Hinton is getting at? A hope that rebuilding AI from the ground up, aligned with human values can reduce risks of an AI-centric AI that uses us for its own purposes to an acceptable level?
If you’re going to reference popular SciFi movies, try 2001: A Space Odyssey and the section with HAL 9000. Although released in 1968, I think it was a frighteningly accurate prediction of where AI is today.
The memory of that one is faint because it was so long ago that I watched it. Something about Kit would open the semi bay door for Micheal? Then David Hasselhoff emerged as a Star Man on the German Pop music scene? 
Okay, Hal not Kit. I forget how the tension was built for Hal not opening the bay door. Might be worth a re-watch.
My recommendation is Person Of Interest, there is two ai (company) fighting to have control..
This is where I get off the train. I could never accept that a digital entity can achieve anything like true consciousness- even if it could theoretically emulate it. I would suggest that we save the humans, animals, plants and the environment first and after that we can discuss being nice to computers.
The idea that a man-made entity could ever compete with a human (let’s not forget a child) isn’t worth discussing.
I switch off, plug out or reboot whenever something goes wrong with my electronic devices. Same with my HAL 9000 
This turned a bit philosophical. I’m 100% convinced that it’s impossible for a machine to get self consciousness.
This thread’s been a rodeo of ideas, so more, ‘Future of AI in SketchUp’:
Here is a typical response from ChatGPT as it reviews my code that we are working on together. We have finally created or coded machines that can actually create code better than us, this is nuts:
I checked this pass more carefully.
Syntax check:
ruby -creturns:Syntax OKSo there are no Ruby parser-level errors in this version. The earlier obvious syntax/name issues look corrected.
One definite remaining issue
Your single baluster layout now correctly avoids the segment endpoints:
1.upto(spaces - 1) do |i| disti = (seg_len / spaces) * i sub_xval << disti endBut the double baluster layout still uses the old endpoint-inclusive loop:
0.upto(spaces) do |i| disti = (seg_len / spaces) * i sub_xval << disti endThat will still place alternating balusters at:
x = 0 x = seg_lenin the double-profile case. Since you said you do not want balusters at the start/end of the segment, I would change the double layout to match the single layout:
1.upto(spaces - 1) do |i| disti = (seg_len / spaces) * i sub_xval << disti endThat is the only actual geometry/topology issue I see right now.
Looks good now
These items look fixed and ready for testing:
b2 / def2 / @Bal_group2 initialization b4 / def4 / @Shoe_group4 initialization ba_hgt0 = hnd_hgt - 1.0 - ba_off single-layout interior-only x positions tr0 local transform no longer overwrites tr1 trseg = tr1 * rot1 explode result guarded before shoveling into @all_groups actual_mat_name = bamat[7..-1]Optional check before testing
If the UI can ever pass
ba_spc <= 0, then this would fail:spaces = (seg_len / ba_spc).ceilIf your dialog already prevents zero/negative baluster spacing, disregard this.
Verdict
After changing the double-layout loop from
0.upto(spaces)to1.upto(spaces - 1), I think this is ready for first SketchUp testing. I do not see any other obvious logical, geometry, or topology problems under the assumptions you confirmed.
LLMs are extremely sophisticated pattern-matching systems rather than anything close to general intelligence. Fancy auto-complete based on static learning.
While they can produce impressive outputs, many researchers argue we’re still a significant distance from AGI, and there isn’t yet a clear, agreed-upon path for achieving it.
We’re in the gold-rush stage of AI, with no shortage of self-proclaimed experts and inflated claims. A lot of the noise comes from people looking to sell tools, courses, or “must-have” systems.
Like VR and blockchain, AI is going through a hype-heavy gold-rush phase. But unlike those, it’s not going away. It’s already reshaping industries, and it brings real risks , labour disruption, environmental costs, and misuse at scale that we don’t fully understand …yet.
What’s striking is how much of the investment is focused on doing more, faster rather than asking whether those outputs are needed in the first place.
Essentially the end game that we are going towards is this :
Scale production.
Drive down labour costs.
Reduce headcount.
Then act surprised when sales stall—because the same people you’ve squeezed are the ones meant to be your customers.
when the AI bubble started, most people trying to convince me were telling me that it would reduce my workload and make my live easier.
turns out that if you have tools that reduce your workload by 50%, most bosses and shareholders will simply cut workforce in half, they see it as a better solution than have people work fewer hours with a better quality of life.
Shocking.
In the end, yeah. If you can replace a team of 4 graphic designers by 2 graphic designers powered by AI, then it’s 2 people unemployed, 2 that will get a burnout quicker, 2 fewer software licences paid, and I don’t see that as a net positive.
But that’s me, I’m a radical european leftist. Eat the rich and the techbros and all 
We will have finally arrived as a species when the machines talk to each other, buy stuff from each other, and we… well, I guess except for a few maintenance workers and IT guys keeping the systems working, and of course the shareholders and the oligarchs… we, the rest of us will not be needed anymore.
There is no future where we, the many, suddenly can work less, earn more and follow our passions in our free time. It has never worked this way.
The few will consolidate wealth and power until the scales are re-set and we will repeat this forever until we evolve.
Greed has no bounds. There is no upside for us common folk.
Sure, a few of us will code faster, open opportunities we didn’t see before - but there will be no wealth sharing and power sharing for any of this. What happens in the black box only serves and enriches the people who control it.
I use AI almost daily now. When I’m coding I keep sending the AI my latest iterations. It almost always finds all the syntax errors, logical issues, mismatched variable names etc…. Just for the debugging capabilities alone it is well worth the $20 I pay a month. I don’t like to ask it to completely write my code, but I’m not surprised anymore when it can, it’s a little scary.
Usually I will see if it can tighten up specific algorithms and it usually can, in the process I usually learn some new techniques in Ruby as well. When it comes to the overall structure of my extension or tool I usually have the better overall insights but once I point out my idea or improvement the AI can quickly write the actual code in many cases to implement what I had in mind.
I disagree that these LLMs are just sophisticated auto complete engines. At this point they now have the ability to reason about very abstract and complicated ideas. They are able to understand geometry, topology and all the math that this involves. In truth, they should no longer be called LLMs, but LRMs.
I can write code without an AI but why would I do that? It’s like saying let’s go back to a horse and buggy when we have automobiles. Or let’s all ditch our cellphones and go back to a land line.
Also I’ve noticed than when I ask ChatGPT a question and it doesn’t have the answer from its static training it gets out on the internet and digs around until it finds the answer. So no, it’s not just static training anymore, it is accessing the internet in real-time and finding answers. In the time I take to scan 2 to 3 sites it has already scanned and completely digested the material on ten times the web sites.
I’m finding AI is slowly but steadily creeping in to my daily life but mostly not for work related issues.
I often use it to suggest new recipe ideas as it’s easier to chat about the recipe or change a few ingredients based on what’s in the fridge or cupboard.
And I’ve recently got a new car that has Google Assistant built in. As well as the usual things like asking it to turn the AC on, I recently had a very indepth conversation with it about a Serbian King from 1903 and the leader of the Black Hand Gang in the run up to WW1 (was listening to an audiobook that mentioned the King and wanted to see if it was true)
As for work, it gets used to re-write emails and proposals and I’ve dabbled in creating a render or two, but purely out of interest and not for a paying client. But I’m sure the time will come when a client sends us an AI generated image and asks us “can you make something like this”