Fearing AI: Possibility of Sentient Self-Autonomous Robots

[.Scott]

Well if I'm not mistaken then currently I think we have no evidence of whether the brain does any quantum effects at all and if it does then how much. So I think it is really hard to talk about it because this is one of those arguments that really needs actual repeatable evidence.

That's only because you haven't caught on to the more profound problem. If you eliminate QM, you need to presume new Physics. Which of those choices is most in need of "repeatable evidence".

You're following what I would call the "common argument" - that consciousness comes from complicated and/or fuzzy logic. I have mentioned IIT only because they have filled in this common argument with enough detail to bring its short-coming into easier focus.

But if you only go as far as that common argument, you only have an epiphenomenal effect. To make it "phenomenal", you have to describe exactly what constitutes that complicated and/or fuzzy logic. Then you need to postulate that when such conditions exist, something physically different happens. At that point, you are describing full-fledged physics. If it isn't QM, it's new Physics.

So the repeatable experiment that defeats the "common argument" in favor of QM (or new Physics) is simply asking people if they can truthfully report being conscious and feel a sense of reality or awareness.

 

[artis]

So the repeatable experiment that defeats the "common argument" in favor of QM (or new Physics) is simply asking people if they can truthfully report being conscious and feel a sense of reality or awareness.

I'm sorry but I'm not sure whether I follow your thought there, can you please elaborate , how would first person subjective experience told verbally prove anything beyond that which we already know?As for the new physics VS QM, I'd say I see an even bigger problem, I tried to put it forth but apparently I wasn't successful enough so far.
Let me ask this both to you as well as anyone else participating in this thread. Am I the only one who thinks there is a problem there.

All our computers so far , whether analog or digital and irrespective of their architecture , whether Von Neuman or other, follow certain known physics. In every computer you can actually trace out the path from the most abstract - a symbol within a user interface, to the symbol representing that within a program language to then the machine code that represents that, down further to the actual electrical signals that get created as the code is executed , then you can follow how those signals go back and forth as they are being processed by logic gates.
In fact we know that every single step, every single bite , every single half period of a square wave is deterministic within such system, because all of them have to be right and matching in order for the process to work.
Any single deviation from that operation is automatically an error, sure software are designed to tolerate certain amount of errors, but the bottom line is - these errors don't add anything of value to the process!
It's not like a CPU logic gate that is failing is creating new creativity within the computer. It adds no new information but only noise.
It's like corrupting a TV signal, you don't add information you can only add noise which at some point will completely destroy the original signal if left to increase.

Why am I saying this?
Because as far as we currently know and as far as I can understand it, our brains work differently.
There is no clear path one can trace out for every thought down to every neuron.
Sure, every thought is a spike of many neurons that progressed along the way , but there is no order as far as we know it. We can approximate the brain regions where the spike path begins, we can also see the regions that are most involved in certain activities, but apart from that every region has millions of neurons, and depending on the brain input , many neurons within a region can be "primed" or readied for spiking and yet not spike until some input is given, or maybe one spikes randomly.
In a way it is somewhat similar to nuclear decay, where each atom can decay randomly at any given time. Now unlike atoms , the neurons that are ready to fire can be , at least in theory , seen, by their increased potential that approaches but stays below the threshold, but there is no deterministic rule that determines which exact neuron will spike at any given time!

This is much more than just "fuzzy logic" or whatever you want to call it, this is essentially complete randomness with very little determination. It's like creating information from chaos, almost like the butterfly effect.
It also means that no two thoughts come from the same neuronal path, they may come from the same brain area but as probability would say, each has a different path, even if slightly.

Now add to that the property of brains that neuronal connections rewire with time - even more complexity that is completely unique for each individual. Almost like you would have 8 billion unique CPU architectures.

I see AI researchers making alot of brain to computer analogies but I believe they are premature and wrong, the brain is nothing like a computer, a computer is deterministic, and the determinism is built right into the circuit, the circuitry is fixed, it can;t change by it;s own accord nor can a error create new information.
Human brains on the other hand are working by what it seems, built in randomness.

And what is funny is that so far you don't even need QM or new physics, we understand how the neurons can change their potential and activity and influence one another even under classical physics - the hard part is to understand how that seemingly random behavior creates meaningful and complex information.
And is also very robust against damage and external influence.
One good argument against QM is that we know so far that quantum phenomenon not only need low temperatures for low kinetic energies of the involved particles so that their states can be preserved but they are also very sensitive to external influences.
The brain on the other hand is I'd say extremely robust against damage and external influence of all kinds, from downright mechanical impact to chemical to radiation etc . For reference see my post #543 and how the man with most of his brain deformed was able to preserve almost normal conscious self awareness.

 

[Borg]

I'm not sure how it's being used above but Fuzzy Logic is an actual algorithm that can be useful for machine learning outputs.

 

[.Scott]

I'm sorry but I'm not sure whether I follow your thought there, can you please elaborate , how would first person subjective experience told verbally prove anything beyond that which we already know?

From post #509 in this thread:
One of the problems with epiphenomenalism is called "self-stultification", (as described in "The Stanford Encyclopedia of Philosophy"). I have quoted a portion of it here:

The most powerful reason for rejecting epiphenomenalism is the view that it is incompatible with knowledge of our own minds — and thus, incompatible with knowing that epiphenomenalism is true. (A variant has it that we cannot even succeed in referring to our own minds, if epiphenomenalism is true. See Bailey 2006 for this objection and Robinson 2012 for discussion.) If these destructive claims can be substantiated, then epiphenomenalists are, at the very least, caught in a practical contradiction, in which they must claim to know, or at least believe, a view which implies that they can have no reason to believe it.

The point is that it is not sufficient for a "consciousness model" to explain your own experience of qualia. It must also explain how you can claim that qualia exists - how information about the qualia can escape into the real physical world.

For this to happen, qualia must be full-fledged physics - responsive to exterior events and able to influence exterior events.

So when you describe precisely what causes qualia, you are also describing what causes qualia's outward physical influences. If you say that all that is required to create qualia is complicated logic, then you are also saying that "complicated logic" is something physically significant - something that has effects greater than what would be expected of an assembly of Boolean logic gates, voting circuits, etc.

As far as the technology issues you described (the brain uses much different information processing technology and strategies than Von Neumann machines), note the reasoning presented above is completely agnostic to technology and algorithms. Any information processing machine (biological or otherwise) which elicits qualia as we know it, will require the physics described above.

One good argument against QM is that we know so far that quantum phenomenon not only need low temperatures for low kinetic energies of the involved particles so that their states can be preserved but they are also very sensitive to external influences.

This is the "warm and wet" argument. But examples of quantum processes have been identified in nature. Photosynthesis comes to mind. Stuart Hameroff has his own specific ideas on this.

The brain on the other hand is I'd say extremely robust against damage and external influence of all kinds, from downright mechanical impact to chemical to radiation etc . For reference see my post #543 and how the man with most of his brain deformed was able to preserve almost normal conscious self awareness.

So, to respond, all I need to provide is an example QM brain model that would demonstrate the durability and severability you question. I'm going to put it in a box so that it is clear that I am only providing an example model that would "checks all the boxes":

Given the practicalities of QM data processing technology, it would seem unlikely to me that it occurs across cells. And yet, it is also not dependent on any single brain location. So it would seem that we must sport many consciousness engines. Presuming they are critical in the decision-making process, and given that two simultaneous decisions are worse than one, only one engine gets our main storyline at a time. Only one gets to write to our "storyline" memory at a time. And only that same one gets to act as the "first person" at a time - and claim our general attention to consider a proposed action.

I see AI researchers making alot of brain to computer analogies but I believe they are premature and wrong, the brain is nothing like a computer, a computer is deterministic, and the determinism is built right into the circuit, the circuitry is fixed, it can't change by it's own accord nor can a error create new information.
Human brains on the other hand are working by what it seems, built in randomness.

Two points on this: The brain/computer analogy predates AI by a lot. In fact, "computer" was originally a job title and the first computers were "analytical engines", "computing engines", "digital computers", or "automated computers". So, I suppose the question shouldn't be "Are computers human?", but "Are analytic engines computers?".
Those analogies are not wrong, but certainly they can be taken in the wrong way. And since people a very social animals, we are very ready (perhaps too ready) to befriend a machine as if it was another social animal.

The other point is about stochastic circuitry. There are stochastic algorithms (such as the Monte Carlo method) that depend on "randomness". I would be careful about describing any of these as "creating new information".

And what is funny is that so far you don't even need QM or new physics, we understand how the neurons can change their potential and activity and influence one another even under classical physics - the hard part is to understand how that seemingly random behavior creates meaningful and complex information.

As described above, the QM or new physics is only required of neurons that elicit our consciousness. So the project is not to explain how neurons can operate without QM/NP but to find the ones that cannot be explained that way.

 

[PeterDonis]

the QM or new physics is only required of neurons that elicit our consciousness

We don't know that this is the case. Some physicists believe it is, others believe it isn't. We have no real testable predictions either way.

 

[.Scott]

We don't know that this is the case. Some physicists believe it is, others believe it isn't. We have no real testable predictions either way.

If someone is claiming that a device that limits itself to known physics minus QM can truthfully report itself to be conscious, they need to explain where that report is coming from.

 

[PeterDonis]

a device that limits itself to known physics minus QM

The "new physics" part is the only limitation I was referring to in my previous post. I did not intend to include any "minus QM" limitation. Sorry if that wasn't clear.

If we take out the "minus QM" part, then as far as we know now, you are such a device. So am I, and so is every other human that says they are conscious. We don't know that any new physics is required for consciousness.

 

[DaveC426913]

Any new disruptive technology is always an opportunity to make a profitable business. But not just by use of the technology, but also by the development of counter-technology (and it has the added bonus of not risking being seen as evil).

Are there any efforts toward AI-busting technology? I suppose the obvious tech would be analysis software that produces a confidence level that some given piece of content (text, picture, video, etc.) was produced by AI.

That seems like it might be lucrative. There's my million dollar idea for any of you entrepreneurs out there. The wave of counter-tech will come; you could be on the leading edge.

 

[Borg]

This is the general idea for a class of neural networks called Generative Adversarial Networks (GANs). Two networks are built - one tries to create fake outputs (commonly images or audio) to fool the other and the second one tries to learn to detect the fakes. The problem is that they both get better so that it becomes harder for the detector network to detect the fakes.

That said, a common problem with GANs is that the generator network can often generate outputs that fool the detector but would be spotted by humans as obvious fakes.

 

[artis]

So when you describe precisely what causes qualia, you are also describing what causes qualia's outward physical influences. If you say that all that is required to create qualia is complicated logic, then you are also saying that "complicated logic" is something physically significant - something that has effects greater than what would be expected of an assembly of Boolean logic gates, voting circuits, etc.

Well actually that's the mainstream view so far. It's commonly known as "emergent properties", We do see similar properties for complex systems elsewhere so the researchers have a natural tendency to think that conscious information processing should be similar.
They also believe that because the mainstream view, still is, evolutionary biology that describes the human mind as just a result of long process of random mutations aided by natural selection. Now there are quite a few shortcomings to evolution , that are now even pointed out by evolutionists themselves, but in order not to diverge here let that be another topic.
All in all I do see this as the main belief, that consciousness emerges when sufficient complexity of specific circuits is reached.
I myself have doubts about this , mainly because , although I subscribe to the fact that many systems do show emergent phenomenon, it is by no means clear that it is even possible to achieve consciousness out of silicon logic irrespective of it's complexity, therefore it might be that silicon logic is incapable of this type of emergent property and some other systems needs to be used that can achieve it.

This is the "warm and wet" argument. But examples of quantum processes have been identified in nature. Photosynthesis comes to mind. Stuart Hameroff has his own specific ideas on this.

I think we need to put a distinction here. Quantum processes happen everywhere in nature if you zoom in to subatomic scales, that doesn't mean that the macro objects that hold within them these QM processes are conscious.
In other words, there is no proof so far that even if some quantum process does happen within the brain that it then directly influences consciousness or more so causes it to emerge.

And the dilemma is huge. Think about it. We know roughly how the brain works. We know there are neurons and synapses and neurons fire all the time. The problem is , how do you take that wet blob gray piece of matter called brain with all of it's billions of neurons and how do you then map out the neuron firing path on to the thought path. If you can't do this, how do you know which neurons made which thought , when and why?
Also, in order to prove a quantum process takes place one needs a highly prepared background. Now one can't just take the brain out of a living human and expect it will continue to function , so how do you prove a quantum process taking place by using the only working brain we have - that inside of a living human which is wet and warm.
This same problem applies to those that , I'd say naively, believe that there will come a time when we will be able to make an exact copy of one's consciousness at any given moment in time. But in order to do that you would have to know the state of each neuron within a brain at some specific time.
It is just an impossibility.

Now one could argue that we could take just few neurons, maybe even produce them artificially and then put them under the test to see whether QM is involved , but that wouldn't match the real situation, because you need them in their native environment to truly see how they work. That is because we don't know whether the brain is just a "sum of it's parts" or more and that is due to the emergent phenomenon law.

That said, a common problem with GANs is that the generator network can often generate outputs that fool the detector but would be spotted by humans as obvious fakes.

As you would expect because both are made from similar working principles. Almost like two thieves that work by the same methods would find it hard to rob one another.

 

[jack action]

Answering the thread title, I think we are still safe for a few years:

​

 

[jack action]

And again ...

​

 

[PeroK]

Human referee sends off wrong player in a case of mistaken identity!

 

[russ_watters]

And again ...

That one is bizarre. Yes it shows a limitation of current "AI" (quote unquote to indicate meaninglessness). But why? All you need is an rfid chip or something similar to make this work perfectly. It has nothing to do with "AI" it's just properly implementing '90s location/identification technology that's now mundane.

I suppose the real answer though is that the goal is "AI"-specific: they are trying to do it with "AI" because they are trying to advance "AI". Nevermind the pointlessness.

 

[jack action]

I suppose the real answer though is that the goal is "AI"-specific: they are trying to do it with "AI" because they are trying to advance "AI".

Yes, but this example is to help answering the OP's question:

Do you fear AI and what you do think truly sentient self-autonomous robots will think like when they arrive?

The answer seems to be that "sentient self-autonomous robots who think" are still closer to fiction than not. Therefore, it is impossible to determine what AI will or will not do once (if?) being sentient.

 

[pbuk]

All you need is an rfid chip or something similar to make this work perfectly.

It's a little more complicated than that.

It has nothing to do with "AI" it's just properly implementing '90s location/identification technology that's now mundane.

In fact in-ball technology for Association Football was only successfully implemented and approved in 2022 and is not yet in widespread use: I can't imagine it reaching Inverness's stadium for some time (the money would be better spent upgrading the toilets).

I suppose the real answer though is that the goal is "AI"-specific: they are trying to do it with "AI" because they are trying to advance "AI".

No, they are doing it with AI because this technology has been around for a number of years and is relatively easy to add to the existing infrastructure used to televise games. This clip is from a match in October 2020 and the software used has been updated many times since then at zero cost to the clubs.

 

[russ_watters]

It's a little more complicated than that.

In fact in-ball technology for Association Football was only successfully implemented and approved in 2022 and is not yet in widespread use: I can't imagine it reaching Inverness's stadium for some time (the money would be better spent upgrading the toilets).

No, they are doing it with AI because this technology has been around for a number of years and is relatively easy to add to the existing infrastructure used to televise games. This clip is from a match in October 2020 and the software used has been updated many times since then at zero cost to the clubs.

Ok, so I'm looking into the actual typical technology:
https://en.wikipedia.org/wiki/Hawk-Eye#Method_of_operation

That's been in use for 25 years, and is currently used in 20 different sports leagues. I don't think I've ever heard it described as "AI" before (that term only recently became fashionable). I don't see the it explicitly said how it identifies the ball, but what I'm reading implies it's visual/software, not hardware.

I think people trying to copy this technology today might just call it "AI" because it's fashionable.

[edit] When FoxSports did it in the mid-90s it was a hardware-based solution (an IR emitter in a hockey puck):
https://en.wikipedia.org/wiki/FoxTrax

 

[russ_watters]

Human referee sends off wrong player in a case of mistaken identity!

I think you posted that in the wrong thread. This thread is about AI.