The AI delusion: why humans trump machines
prospectmagazine.co.uk
prospectmagazine.co.uk
https://behavioralscientist.org/principles-for-the-applicati...
Aub is so upset by the appropriation of his discovery for military purposes that he commits suicide. As Aub's funeral proceeds, his supervisor realizes that even with Aub dead, the advancement of Graphitics is unstoppable. He executes simple multiplications in his mind without help from any machine, which gives him a great feeling of power.
No, the Turing test has not been surpassed. The "panel" was not fooled; one judge on the panel was fooled. And how, I don't know. The sample dialog on the Wikipedia page for that bot is the sort of pointless trivia questions that a bot can answer and a human can get wrong, and it won't tell you anything about who you're interrogating. If you instead try holding context across multiple inquiries, the charade will quickly reveal itself,
(">" = human, "<" = bot's reply.)
> Hello.
< Nice to meet you! Please tell me how you found me and what you'd like to discuss :-)
> Through Wikipedia. Tell me about somewhere you've visited.
< May I ask you about the real reason of your interest? Maybe, let's talk about something else? What would you like to discuss?
> Curious about your history. Why do you not want to discuss it?
< How do you know? I thought nobody could guess that! Could you tell me what are you? I mean your profession.
The previous reply is nonsense, and the constant evasion is typical of chat bots like Cleverbot. Clearly bot. I know I'm not blind, but I think we're kidding ourselves to equate this with the dialogue capabilities of a human. It took 3 queries to get to that. Even more recent attempts, like AI dungeon, fall apart after 2-3 queries that require context and understanding.
Maybe there's a bifurcation going on, and some people are converging with AI and others are not?
Conversational bots for support are increasingly common, and the voices are getting better, and even outside of actual conversational bots humans following flowcharts that essentially render them into conversational bots have been a thing for quite a long time.
If humans do not "tolerate" other humans that are perceived like bots, how will we ever tolerate bots themselves which will probably never be able to imitate robotic humans the way humans do so today :)
You are correctly pointing out the conflict between computationalism ("consciousness is what certain computations feels like from the inside") and physicalism ("consciousness is what certain computations feel like from the inside, when they take place on physical substrates, perhaps requiring the involvement of quantum decoherence").
One of the best primers on the current state of academic hypotheses on these topics is the whitepaper "Principia Qualia" by the good folks at the Qualia Research Institute [0].
[0] https://en.wikipedia.org/wiki/Dowding_system (The article contains only a quite crude and basic representation of the operations of the filter room.)
This is really damning humans with faint praise. "Machines may eventually do every job better than we can, and be immortal, but I promise that humans will remain superior in some completely undetectable way."
Consciousness is “magic” then.
Even if we build machines to mimic a real brain, “it’ll be like a golem stumbling about,” he writes: adept at the Turing Test perhaps, but a zombie.
This guy created a whole moat of unfalsifiability around his views.
That doesn't follow at all.
Well, they're here reading this, since they cant measure their experience they must not exist. /s
However, to add to the criticism you have of this article about this book, your two quotes taken together appear to be contradictory: if consciousness is a fundamental element of living matter, then given that we can make new living matter, why should there be any reason we can’t make a conscious artificial machine?
Is consciousness a physical phenomenon? If it's not physical, then what is it? Is it an algorithm? Well, for one, there is no precedent. Nothing else in the material world is purely an algorithm...except maybe a computer running an algorithm...
I don't see the distinction you're making. I think I want to say "ok, so nothing is purely an algorithm". A computer relies on the rest of the universe just like your brain and I guess it's therefore a "physical phenomenon"?
In order to draw a distinction, perhaps one could say that a "pure algorithm" is something that can run on "many" substrates, so the difference between a "pure algorithm" and a "physical phenomenon" is about the variance in the substrates maybe? But everything is physical. Statements about abstractions are statements about patterns of physical things.
Interestingly enough, this may not be entirely true. There are some fundamental relationships between computation, information, and entropy, that, in the limit of computational and representational perfection, probably would require exactly the amount of input energy as the ice-cube itself would to melt.
Now, you might argue that this is a pedantic attack on your argument. It's not though. The real point is that the laws of physics often seem to have grander symmetries and transformation laws than observable at first glance, and the universe has very clever ways of enforcing these symmetries in what appear to be very different contexts, but are indicative of less intuitive symmetry laws.
There is no good reason to believe that consciousness is reducible to physical phenomena. I think that intelligence is almost certainly reducible to physical phenomena, but consciousness? No. Consciousness is a mystery that quite likely will forever be beyond the reach of physical investigation.
(Not a good way to live your live, but a theory consistent with the evidence.)
You could ask try asking it? You can ask clever questions that would prove things like persistence of thought in between questions, and ability to reason about things. If it's able to reason about it's own processes, it's basically conscious by any measure that matters.
>There is no good reason to believe that consciousness is reducible to physical phenomena.
Lol, what? Other than the fact that we have millennia of scientific progress that have, in every single area of exploration developed predictive models of reality that have been tested, confirmed, retested and refined over time, and every single one of those successful models is not based on some invisible unmeasurable magic nonsense. What on earth could possibly make you think that consciousness is somehow the sole outlier that doesn't obey the laws of physics and is therefore reducible to physical phenomena?
If something is possible, but has measure zero in the probability space, it's really not worth considering.
This paraphrase "Koch believes the theory establishes that machines built along the lines of our current silicon-chip technology can never become conscious, no matter what awesome degree of processing power they possess." would seem to come from this statement: "Whether or not a network has this property of influencing itself depends on its architecture. If information is merely “fed forward” to convert inputs to outputs, as in digital computers, then IIT insists it can only generate zombie intelligence."
However all IIT (and presumably therefore Koch) requires we do to solve this is use a different architecture (i.e. some kind of self-modifying feedback approach in the computer). "In Koch’s words, consciousness is “a system’s ability to be acted upon by its own state in the past and to influence its own future.”"
Can you explain what you mean? It seems we already have this with naïve use of consumer VR devices on most VR experiences.
For instance, while playing Richie's Plank Experience, the player already feels qualia due to:
1. virtual self in virtual world: looking down the plank in mid air where the plank is virtual and being in mid air outside a tall building is virtual
2. physical self in physical world: if a friend nudges you from behind (or out of sight of virtual self), or you have a fan blowing wind that would correspond to the breeze you would expect to feel in the virtual world, or you're walking on a real plank to which the virtual plank has been resized/calibrated (this one is a true hybrid real+virtual 'quale': you see the virtual plank but feel the real plank with your feet).
If you meant the virtual agent would have separate agency from the player, please explain as I don't quite follow.
I think a huge gate opens if you get a high bandwidth brain/computer interface. But until then, an experiment like the following may give us a glimpse of the tractibility of such transfer:
- First, you need a full audio and visual override of one eye and one ear (basically over-ear headphone in one ear, and over-eye VR goggle.)
- You also need to be able to track something on a person that a) is underutilized in the 'real' world and b) has sufficient bandwidth to convey some kind of agency of control over a virtual agent. For example, you could wire things up so a person's toe orientations could measured, or internal teeth motion, or subvocalization, etc. You could have dual-use mechanics, like a dedicated hand controller that still gives you freedom to use your hand, but I'd argue this might compromise the experiment.
- Combined with these two, with the hardware on and the inputs working, connect this person bi-directionally with a virtual agent in a simulation. (Basically a modern first person video game would probably suffice.) The video and audio feedback is obvious, you'd need to come up with a mapping of the inputs. Give the person some goals to accomplish in the simulation that are non-trivial and take concerted mental and physical effort to do.
At the end of say a months worth of this dual immersion in both the physical body and the virtual, have the person self-assess their conscious state. I think it's even odds that they feel like they are playing a video game still, or are actually manifested consciously as inhabiting two places at once, whatever the hell that means. The experiment is predicated on the idea that brain plasticity would start to allow the person to experience a level of conscious experience and control over the virtual agent similar to the physical world, since their sensory inputs and agency is at some level of parity in time and bandwidth between both physical and virtual. It's far from total parity, but it's well beyond what has been possible to do in recent history, so may be past the tipping point necessary for changes in conscious experience.
If the latter were to occur, and the person feels a duality of presence, the subjective experience of say, closing off the person's other remaining senses which are attached to the physical world (ie their free eye and ear, etc.) and waiting another month, may actually result in a representitive experience of what a consciousness transfer would feel like.
And this can work. You could copy a human merely by observing them. If you get enough data on a person (say yourself, so it isn't creepy), and train an algorithm. Once the algorithm is good enough at imitating you that your own mother can't tell the difference between a robot containing that algorithm and you, is there still a difference. And now every mathematician should say "No". After all, 4 and IV is the same number.
You don't need audio or visual override. You don't need to give any agency to any flesh and blood over a virtual body. Even the robot mentioned before is really more of an optional thing, you could eliminate it.
You could simplify further. You don't need to duplicate a human, you just need to provide them with "a kid". An algorithm that's purely virtual but that is sufficiently smart that humans can raise it to become part of human society. But it's mind can start out mostly empty. So you don't even need to duplicate anything, you just need a sufficiently complex learning algorithm and a way to interact with it, and a human (perhaps preferably a couple?) to teach it, if you're willing to give it a generation or two.
Now, once we get a brain/computer interface, things get really interesting. Because you can start to link minds, and even at low bandwidth, it may rapidly increase our understanding of the boundaries of what a conscious mind "is" and how much we should care about the "death" or "merging" of such things. It would be quite impactful, to say the least, to see that if you connected a person's mind to that of a simple organism in a bidirectional interface, if over time they act as though they had a single consciousness (probably mostly represented by the person's prior self, but perhaps with characteristics of the organism leaking in the other direction as well.)
The closest analog we have today is when one separates the left and right hemisphere of the brain, and that is a crude, fixed, and brutal methodology for such experiments, so it's only done as a way to treat people whose health it may improve, and is irreversible.
I don't understand what you are talking about. Do you think identical twins share the same consciousness? Is a clone the same person as who it was cloned from?
It seems to me that "4" and "IV" are abstract concepts that have a basically unlimited number of relationships with the real world, like tentacles. The mathematical concept of four is just one of those linkages.
However I’m not sure what the VR really adds to it, above e.g. learning to shift the default point where most of us are conscious from (behind the eyes) to the lower abdomen as done in Zazen. We can similarly learn to shift this locus of consciousness completely into our FPS VR avatars but that doesn’t really make them ‘agents’ in the full sense of the term. The transfer is illusory.
Mostly agree with Koch, but I'd take it a step further...
There's major problems behind the concepts of AI and even intelligence itself - and it's difficult to articulate why. It’s as if these terms require aggrandizing to the point of impossibility or they lose all their apparent meaning. Which is why I feel we'll never achieve what we call (Strong/General) AI, or if we do, we will always find ways to be unimpressed by it...
I mean, is it that absurd to consider that the ideal concept beholden to intelligence isn't a reality - even in humans? If you pull back enough layers on how or why humans think or do the things they do - we arrive at things we can't explain. We don't know what causes intelligence and have trouble coming up with an adequate definition for it; similar to the concept of life. For all we know we might be just highly complex biomechanical machines operating on stimuli, analogous to what current computers already do. Where's the fine line between making something conscious/unconcious?
Computers react to electrical inputs, and on some level can be considered reacting to a stimuli.
Is a computer therefore conscious?
Consciousness and computation are entirely up to what we define these words to mean. At some point we have to settle on a satisfactory definition of consciousness that fully differentiates it from computation, or they’re left with at least some overlapping (analogous) meaning.
If we dismiss consciousness as currently unknowable us, and thus undefinable (as your statement about dimensions alludes to), then how can we say assume with certainty, that we haven’t already achieved conscious AI?
No, we arrive at things that are uncomfortable to explain.
I think one of the biggest impacts of AI is that it will force us to confront this.
I think it's already being confronted but behind corporate greed instead of public need.
There's no discernible difference between p-zombies and 'real' conscious beings. There's a good chance that we're all p-zombies and a distinction between zombie and real doesn't exist.
> If you pull back enough layers on how or why humans think or do the things they do - we arrive at things we can't explain.
Sounds a lot like a magical argument. There's no evidence that anything about the way human minds work is fundamentally unexplainable.
Agreed. What I'm implying here is if we come circle to fully explaining how the brain works, then in theory we could predict how someone will function if given all input conditions (whether this is practical is another matter). If this is true, then wouldn't we just be biomechanical robots operating on stimuli? Would we be any more intelligent than say a rock, that similarly just reacts to physical stimuli? Or are we just a more complex rock?
Yes, that's right. Which is one of the big problems with this line of argument for many people -- how do you reconcile "biomechanical robot" with the concepts of "free will" and "moral decision-making"?
I'm personally agnostic on the issue (I don't believe science is able to answer this question yet -- or possibly ever.)
I do think it's interesting that you say "just a biomechanical robot." The "just" implies that being a robot -- a fancy rock -- in some way isn't enough. But in my mind, there's absolutely no (objective) reason to think of a human as any better or more important than a robot, or a rock.
I'm just trying to challenge people's assumptions of what makes something intelligent. To me, it's not precisely clear as to what makes something intelligent, and is why I think we've had trouble coming up with a satisfactory test for determining when a computer is intelligent.
(P.S. Sorry for putting words in your mouth -- I don't want to imply you said anything you didn't. My bad there.)
> are we just a more complex rock?
of course we are. moreover, it's pretty straightforward to trace our evolution back to rocks or rock-like formations.
(see the "Origin of biological metabolism" section - lots of good new research coming out recently, including simulations of the very early processes predating life as such)
I think you're getting it wrong. There's no externally discernible difference between a normal person and a p-zombie, because the difference is in how they experience. I know I'm conscious, because I subjectively experience it, but I can't be sure about anyone else because they could be a p-zombie that has everything except that subjective experience.
Can you have evidence that something is "fundamentally unexplainable"? Other than a lack of explanation?
If your statement has consequences, then there must be an imaginable counterfactual where it does not hold. I'm struggling to think of that.
You might counter that your brain is tricking you into thinking you have those experiences, when you don’t. It’s an illusion.
Bit ask yourself this. What would it even mean for the experience of pain or color to be an illusion? An illusion is itself a conscious experience!
So the human brain can enjoy things because..... magic?
The idea of 'I', myself,separate from all else. My decisions affect my fate.
You like others conflate "how","what" and "why". Whether or not it is biomechanics,electrochemicals or a projected hologram of quanta is "how", so is whether or not stimuli or measurement of stimuli form our reality perception.
A person has fainted, that person then wakes back up,you say:"the person lost consciousness but is now conscious". He was no longer in a self aware state where he would have made decisons to self-preserve and he is now back in that state. Regardless of intelligence or processing capacity, a self aware program that self preserves without explicit instructions for either functions is conscious.
I'm not conflating anything. I'm trying to understand the meaning behind phrases like "reality perception" - which convey little meaning without a definition. “What” “Why” & “How” are used to grasp at meanings.
See, words are for conveying meaning. When they have too broad of a definition, little information is conveyed (which is why phatic expressions and small talk suck). To better convey meaning, we form more stringent definitions and create new words. Notice however, the formation of new words and stricter definitions is a never ending pursuit, otherwise if we could fully explain things we would end doing so.
For example, notice we started somewhere with a definition for life, and then years go by and it ends up being too ill-suited for discriminating life from non-life. So we add additional words to describe what makes life... life. Eventually we'll either have a well defined line between what makes something living and non-living, or we perpetually keep on tacking on existing words or new words to describe it. The same goes for intelligence.
I can only see two scenario’s:
Scenario A: We end up perpetually trying to define intelligence, and thus we’ll always struggle to differentiate intelligence.
Scenario B: we come to full stop and are able to fully explain everything (at least regarding intelligence). At which point intelligence will be fully differentiable.
Are we at scenario B yet? Then it should be easy to come up with a satisfactory test for AI, let alone know if it’s even a possibility. Yet here we are today still unsatisfied by the state of computer intelligence… This is what I mean in my OP by: “These terms require aggrandizing to the point of impossibility or they lose all apparent meaning.”
Anyhow, I like your points on self-aware introspection and self-preserving strategy formation. But I'm not yet convinced that it has strict enough definitions to differentiate if something is truly AI. I mean is a computer that tries to prevent itself from going into sleep mode, and is aware when attempts are being made to put it into sleep mode - fit your definition? It kind of does... yet, I think we'd agree that wouldn't be AI.
If that is a programmed function then it does not fit my definition. If it was not made to be aware,it became aware as a result of learning information and adjusting its programming and it realized the difference between sleep mode and a system wipe(death) and adjusts its programming to prevent a death scenario then it is conscious.
As for the rest, I don't think I am part of "We" I have pretty strict,well understood and time tested definitions for life and intelligence,separate from consciousness.