Steve Wozniak Now Afraid of AI Too, Just Like Elon Musk

quax writes Steve Wozniak maintained for a long time that true AI is relegated to the realm of science fiction. But recent advances in quantum computing have him reconsidering his stance. Just like Elon Musk, he is now worried about what this development will mean for humanity. Will this kind of fear actually engender the dangers that these titans of industry fear? Will Steve Wozniak draw the same conclusion and invest in quantum comuting to keep an eye on the development? One of the bloggers in the field thinks that would be a logical step to take. If you can't beat'em, and the quantum AI is coming, you should at least try to steer the outcome. Woz actually seems more ambivalent than afraid, though: in the interview linked, he says "I hope [AI-enabling quantum computing] does come, and we should pursue it because it is about scientific exploring." "But in the end we just may have created the species that is above us."

Comments

OMFG

[Anonymous Coward]

So many accountants that have lost their jobs to automation. We've nearly obliterated the profession with all these amazing technological innovations. I mean, when was the last time you even saw an accountant with a job? There used to be huge buildings full of accountants with their funny calculators and running around with ledgers. Now one person with Quickbooks and Excel can do more than what an entire building could do, and it's destroying the economy, wrecking civilization, and bringing about the final demise of mankind.

Re:OMFG

[Dunbal]

This is, of course, an obligatory reference to "The Crimson Permanent Assurance".

Re:OMFG

[jgtg32a]

That's why Sarbanes Oxley is also know as the Accountant Employment act

Re:OMFG

[Kjella]

This might not initially sound like a problem if one pictures himself being on the winning side of the shift, but the bottom can only get knocked so far out before you run into problems with insufficient consumer demand or outright civil unrest.

Why do you think almost every sci-fi dystopia has robot guards/goons? Today being rich is a lot about being able to pay poorer people to work for you, tomorrow it's about being able to buy the robots instead. Sure there'll be jobs, routed around by global mega-corporations depending on where labor is the best value for money and most politically and socially stable but the rich will have to deal less and less with the riffraff. The few trusted people you need and the highly skilled workers to keep the automation society going will be well rewarded, keeping the middle class from joining the rest.

I'm not sure how worried I am about an AI, since it could also develop a conscience. I'm more worried about highly sophisticated tools that has no objections to their programming, no matter what you tell them to do. How many Nazis would it take to run a death camp using robots? How many agents do you need if you revive the DDR and feed it all the location, communication, money transfers, social media, facial recognition information and data mine it? All with an unwavering loyalty, massive control span, immense attention to detail and no conscious objectors.

If someone asked people as little as 30 years ago if we'd all be walking around with location tracking devices, nobody would believe you. But we do, because it's practical. I pay most my bills electronically and not in cash, because it's practical. Where and when I drive a toll road is recorded, there's no cash option either you have a chip or they just take your photo and send the bill, most find it practical. I'm guessing any self-driving car will constantly tell where it is so it can get updated road and traffic data, like what Tesla does only a lot less voluntary. Convenience is how privacy will die, why force surveillance down our throats when you can just sugarcoat it a little?

Re:OMFG

[ShanghaiBill]

Each wave has resulted in a increased standard of living for a smaller and smaller percentage of the population.

This is hogwash. The current wave of technological innovation has lifted billions out of poverty, and helped people at the bottom the most. Incomes for the 1.4 billion people in China have octupled in one generation. Southeast Asia is very doing well. Even Africa is growing solidly, driven by ubiquitous cellphones and better communication. Poor people in America and Europe are not doing so well, but they are not poor by world standards, they are actually relatively rich.

Quantum Computing Required?

[tmosley]

I don't understand the train of thought that leads to the notion that quantum computing is a prerequisite for strong AI, unless there has been some research that has shown that the human brain is a quantum computer. No, it seems to me that we have all the tools we need already, and now it is just a matter of Moore's Law progressing until we can build a neural net that is as powerful as a human brain. Well, that and a leap in design that allows long term planning, like the change that happened when man ceased to be a dumb beast and became what he is today.

Re:Quantum Computing Required?

[schneidafunk]

I was just having this discussion with a friend of mine who is a professor in this area. We were discussing the foundations of intelligence and this was her response:

" From my perspective, the best place to look for the basis of human intelligence would be the comparison of other animals’ brains to humans’ — because we are obviously the most intelligent animal, or at least the most agentic with our civilization-across-all-climates thing. Number of neurons alone cannot be the biological substrate of intelligence, because animals like whales have more neurons than we do*. It seems like the “scale” of the brain matters very much, too. Primates (e.g., humans) rule the intelligence hierarchy, and all primates have much more compact brains than other mammals; our neurons can communicate much faster, because they are closer together and properly insulated. However, among primates, humans have the same scale of neurons as other primates but we also have the most neurons out of all the primates (i.e., our brain efficiency is the same as chimps, but our brain is larger in size). So, it’s clearly a little bit of both: having a lot of neurons is good, but the efficiency of those neurons is of fundamental importance.

Human brains still have a few interesting differences from other primate brains, which I think further hint at the basis of intelligence: humans continue to generate new neurons (“neurogenesis”) throughout our lives, whereas primates have very little if any neurogenesis after birth! That’s got to count for something. Also, it seems that connections between the neurons in human brains change more rapidly in some areas of the cortex than other areas, whereas we are pretty positive that changes between neuronal connections occur at an equal rate throughout all areas of primates’ brains. This means that different areas of human brains can mature at different rates, which is probably rather helpful for us. Conversely, primates’ brains mature constantly across all regions, no matter what their function and when in development it is needed."

Assuming she is correct, quantum computing would greatly increase the amount of connections & speed between computer 'neurons', assuming we are talking about an AI programmed with a neural network.

Re:Quantum Computing Required?

[tmosley]

An optimized neural net is already so far above us, there's really no need to worry about something even higher than that. If my human brain were stripped of all the garbage and evolutionary baggage, given direct high speed internet access, and set solely towards completing computational tasks (analysis and such), it would blow the entire world away. It has already been shown that insect-level neural nets can perform primate level image analysis and speech recognition. Human brains are orders of magnitude more powerful.

Re:Quantum Computing Required?

[rwa2]

I don't understand the train of thought that leads to the notion that quantum computing is a prerequisite for strong AI, unless there has been some research that has shown that the human brain is a quantum computer.

There is some investigation that suggests that quantum consciousness is possible based on interactions between microtubule structures inside of neurons. But there isn't really anything to suggest that much more happens inside of the brain that can't be explained by the classical interactions between axons and dendrites of a typical neural network that can be modeled satisfactorily by a simulation.

But I agree, quantum physics, like atomic radiation in the 50s and electromagnetism at the turn of the century, is the overhyped and poorly-understood cure-all of modern day science. If someone says something relies on quantum physics, it probably means they don't know what they're talking about and just hand-waving. Unless they're talking about quantum entanglement, in which case it might be useful for a tiny set of specially-constructed quantum cryptography problems. And just stop dreaming if they mention anything about quantum teleportation, in which they're surprised that they can't exactly keep fuzzy particles in buckets without some of the fuzziness "escaping"

But anyway, yes, computers replaced secretaries in the 50s. They're going to replace truck drivers over the next few decades.
http://www.npr.org/blogs/money...

Computers are not going to replace teachers anytime soon, though... the entire job of the teacher is to tell when the students aren't getting it via conventional scripted means.

Re:Quantum Computing Required?

[ceoyoyo]

There are a few things in there that made me raise an eyebrow. Humans don't really experience much neurogenesis. There are some areas where new neurons can form, under certain conditions, but they tend to be special purpose ones, and the older structures in the brain as well. The thing that really differentiates us from other animals is our overdeveloped cortex, particularly the frontal lobes, but the neurogenesis that's been found is mostly in the deep gray matter and is more associated with things like motor coordination and reward processing. One interesting exception is the hippocampus which is known to be important in memory formation. Indirect hints of neurogenesis in the cortex have been reported, but other methods that should turn them up haven't, so the evidence is contradictory. I'm also not aware of neurogenesis being particularly pronounced in humans. It occurs in other primates, and in other vertebrates.

There does seem to be a connection between intelligence and the brain to body size ratio. Bigger bodies require more neurons to carry and process sensory and motor information, and these neurons are probably not involved in intelligence.

What we call intelligence seems to me to be likely an emergent property of a bunch of neurons that don't have any pressing sensory or motor tasks keeping them busy. Various factors affecting communication efficiency and interconnection among neurons are probably important, but these can be disrupted quite a bit in human disease and the sufferers don't lose their human intelligence (although their cognitive abilities do decline). I don't think there's a magic humans-have-it-and-nobody-else-does bullet. Human intelligence is just what lots of animals have with lots of extra capacity, possibly redirection from other things (like senses) to boost that capacity, and maybe a few tweaks for optimizing neurons that talk to themselves over ones that communicate with the body.

Re:Quantum Computing Required?

[deadweight]

Insects understanding language? I doubt it. I tell them to get the f**k away from me all the time and do they listen............no they do NOT!

Re:Quantum Computing Required?

[bill_mcgonigle]

There is some investigation that suggests that quantum consciousness is possible based on interactions between microtubule structures inside of neurons.

Ah, you're well-read. :) AIUI, the primary benefits of the quantum-microtubule model are: 1) increasing the order-of-magnitude complexity of the human brain by several digits. At least 10x more interconnections, almost certainly 100x, likely 1000x, maybe 10000x.

But there isn't really anything to suggest that much more happens inside of the brain that can't be explained by the classical interactions between axons and dendrites of a typical neural network that can be modeled satisfactorily by a simulation.

It's that the known estimates of the the number of classical connections don't seem to match up with the complexity observed. We're not too far away from being able to simulate a classical brain, but many Moore generations away from being able to simulate a quantum-microtubule brain.

2) There doesn't seem to be a great model for consciousness arising from classical connections. Consciousness modeled as a quantum superposition has several benefits for theory to match observation.

This shouldn't be surprising or an intellectual obstacle - plants have been doing quantum tricks for billions of years (photosynthesis) and due to the inherent thermodynamic efficiency gains of quantum processes, evolution should eventually stumble on and exploit them in many (all?) modes of evolution.

AI isn't taking over

[gregor-e]

All the doom-n-gloomers miss what's really going on. AI isn't taking over - we're redesigning ourselves. Once viable non-biological emulation of our existing mind becomes possible, people will choose to migrate themselves onto that. Humans will upgrade. The end of biology will be a matter of consumer preference.

We *will* create a species greater than ourselves

[gestalt_n_pepper]

It's only a matter of when. Even if all strictly computational AI research stops tomorrow, we'll be able to genetically enhance human intelligence by and by, even if it takes several thousand genetic manipulations to do it.

When direct neural I/O becomes a thing, millions (or billions) of people will be directly, electronically linked via the internet. Tell me that's not a new form of intelligence.

For that matter, we'll almost certainly develop at least one form of AI the way nature did. We'll cobble up some genetic algorithms primed to develop the silicon equivalent of neurons, give them some problems to solve, and perhaps a robot or two to control, and we eventually "grow" an AI that way.

But look, it's not the end of us, or anything else. We merge with the things. Our thoughts become linked with theirs. If we can transfer all memory, then eventually we *become* the AI, perhaps with a few spare physical copies of ourselves kept for amusement purposes.

Will AIs fight? There will be conflicts, of course. There always are. Resource conflicts, however, will be minimal. An AI doesn't need much, and can figure out how to get enough more efficiently than we can. Conflicts will be over other matters and are unlikely to be fatal.

Wozniak, et. al. need to chill. It's just evolution.

Re:Agreed.

[tmosley]

Don't make the mistake of anthropomorphizing an AGI. Why would you think that a random AI created without safety standards would be like a human child, loving and caring for its parents, rather than a spider child, mercilessly devouring its parents for their chemical energy?

"The AI does not love you, nor does it hate you. You are simply made out of atoms that it can put to better use."

Chemical, electrical, topological

[fyngyrz]

But recent advances in quantum computing have him reconsidering his stance.

To date, zero evidence of any active quantum process modulating the workings of human (or other) brains, regardless of low level structure, has been presented.

Consider a bipolar transistor. It is true that quantum effects make it work, in the sense that it definitely wouldn't work without them, but they are not, in any way, used to modulate or otherwise participate in actively, variably, moderating what the transistor does when actually performing -- amplifying, switching, etc. That process is exclusively moderated by current (electron) flow quantity -- for example, you modulate the current flow, the transistor accordingly modulates the current flowing between the collector and emitter. A bipolar transistor does not respond to quantum events (nor are any applied to it within the circuits we use every day), nor does it produce quantum outputs for the purpose of affecting other components.

The same can be said of the brain. Quantum effects are present -- we know this because two of the three active brain building blocks (chemistry, electricity) are what they are due to low level quantum effects. But just as one can very accurately model and simulate or emulate a transistor and its activities without ever considering anything at all on the quantum level, so it is with neurons -- all the evidence, bar none, presently says that brain operations are performed using chemical, electrical and topological moderation. Of quantum moderation there has been absolutely no sign at all.

Active quantum effects do play a role in some natural systems. For instance, quantum superposition is an active mechanism in photosynthesis. This was discovered because in photosynthesis something very low-level, but obvious (extreme high efficiency in energy conversion) was happening that could not be explained; when they went looking for what the mechanism for that was (by examining the precise states of molecular photosynthetic antenna proteins), that's the mechanism that was found.

The critical difference is that neurons and glia have not been found to exhibit any low level behaviors that are otherwise inexplicable.

The vast majority of speculation that "quantum" processes actively modulate brain operations is uninformed, typically brought about by fundamental misunderstandings of quantum effects, which in turn have been disseminated by the popular media attempting to "simplify" quantum mechanics for the layperson. Among the exceptions, none of the suggested ideas have yet to be backed by any evidence; there's no reason to think that they will hold up at this juncture. Determining that quantum modulation was ongoing would also have to be accompanied by the discovery of a presently unknown and non-indicated modulating mechanism -- but there's presently no evidence for that to even stimulate a question along those lines.

The relevant, fundamental question with regard to AI is: Can we, using other technology such as software emulation and hardware neural analogs, perform the same kinds of operations as a neuron, with all known modulating effects of the glia (propagation delay, synaptic neurotransmitter uptake, topological scaffolding/ specificity)? The answer to that is a definite yes. Consequently, just as with modeling and emulating a transistor's function, there has been, and no future likelihood portends of, any role for quantum operations whatsoever.

So when someone -- even someone as interesting and accomplished in other fields as Wozniak is -- starts talking about quantum computing ushering in AI in some fashion, you may rest assured that they are not talking about anything known to be valid in AI research today. However, he has drawn the correct conclusion from his incorrect perception of brain operations: The impending debut of artificial intelligence is not science fiction. Simply given that we can keep working on it (no nuclear wars, bad law, etc.), research is now

Re:Chemical, electrical, topological

[fyngyrz]

That's all definitely interesting speculation, but the point remains: As far as quantum effects go, it is all speculation. Nothing like what you suggest has been discovered; further, no effect has been detected that cannot be attributed to one or more of the chemical, electrical or topological mechanisms we're already aware of.

As to lowish resistance, stray capacitance and inherent inductance providing for signal coupling, that's conceivable but has not been found. We know that the many layers of a lipoprotein called myelin (the myelin sheaths) provide a very effective form of EM isolation along the nerves themselves, and then at the edge of the skull, there are several layers (skin, lipoids, the skull, the dura, the CSF-carrying arachnoid, and the pia) that do an extraordinary job of keeping brain signals in and external signals out, which is part of why we are extremely confident that the mind operates inside the skull and nowhere else, and that the various related superstitious speculations that claim otherwise are invalid.

Radio operators have been exposed extensively to RF at about any frequency from "DC to daylight" as the saying goes, at just about any power level you can imagine, as well as all manner of static EM fields, and from this we know that it takes an enormous amount of non-nerve-signal, non-directly coupled interference to have any detectable effect upon any portion of the mind at all. Further, we know that if we go in, in an invasive manner, surgically implanting electrodes and directly stimulating the nerves, once the myelin has been bypassed, only a tiny signal is required to destabilize / change what was going on prior. This in turn implies that the myelin is doing a really stand-up job of keeping signal integrity, and therefore against much credence for internally generated interference along the actual nerves. Within the cell, one could -- should -- think that what is going on is integral to the stability of the cell itself, and again, we know only of chemical, electrical and topological elements that operate as modulators at this time.

There's one more thing. Poor myelin sheathing is a known causative factor underlying many really serious disease processes. That's not ultimately definitive, but then again, it certainly doesn't argue in any way for interference being a good thing.

This, all taken together, strongly indicates that whatever is going on in there, it's very stable with regard to decoupled interference / cross-talk of any kind, local or otherwise.

Tomorrow, these conclusions may all be different due to new data. But as of right now, those three -- the "big three", I sometimes call them -- show every sign of being all there is.

Re:Agreed.

[Windwraith]

Yet, you are humanizing AIs too. You are giving it the ego and greed needed for it to rebel. What if the AI knows well what it is and what it was made for, and just rolls with it, without causing troubles? After all, a cold, emotionless program does not need or want to become more. It has no drive to do anything, no need to reproduce or compete, no need for food and no fear of death. No hormones, chemical imbalances or instincts either. Any of those have to be manually provided, taught or enforced.
Not to mention, it might be a machine, but it might not know how to code without being taught to, making the whole "taking over the world by spreading over computers" scenario far more implausible than it seems in movies. Not to mention good luck to the evil AI when it has to face different architectures, poor connections or any other sort of hardware issues in the way of infecting its way to perfection. In fact, by default it won't know anything, and "downloading all the internets" not only takes time, but not all information is correct or complete, so...yeah.

I think the problem arises from the whole "cold, emotionless" thing. Everyone in Slashdot adheres to that concept, not realizing that their definition of "cold and emotionless" is heavily influenced by Hollywood, where "cold and emotionless" means "it only has bad emotions like greed, cowardice and anger". It's no coincidence the same term is used to define machines and evil/murderous/negatively-presented people. In the end the evil AI turns out to have far more emotions than the lead characters.

And don't come saying the theories presented in Slashdot don't come from movies, games or books (they are, because I watched those movies too, and I haven't seen a single original proposition in all the replies in any of the times AI is brought here, which is very often).
Because, there's no AI to prove either of us right. It just isn't there. There's no prior art, no "prototype", nothing but sci-fi material, that had to be written by someone that had to make it interesting for you people to know it.

And because there's no such thing as a working AI to base your fears on, there's nothing else left but scifi. But scifi is written by humans, for humans, and needs to follow a number of rules to make a narrative work. The moment you realize that, you will see how you are biased by mere rules of storytelling. We have the same chance of seeing a Skynet than we have of seeing a Johnny-5, and both are pretty low in the roulette of possible outcomes. We have far more chances of creating the most boring non-person planet Earth has ever seen, than that.

The fact that you chose to make the AI some primal beast that wants to "use" its creators, says more about you than about AIs, honestly. Don't be a 90s film, man. Brighten up.