13 ms·

The body is the missing link for truly intelligent machines (2017)

165 points by new_guy 7y ago

PeterStuer 7y ago
It is strange that the author does not reference the whole 'Nouvelle AI' movement of the late 1980's that was a direct response to the 'symbol systems' of classic AI, proclaiming the necessity for embodiment as a prerequisite for grounding. See for example Brook's classic "Elephants don't play chess", or Steel's write up on "The Artificial Life roots of Artificial Intelligence"
- jackhack 7y ago
  More than just strange, it's a glaring omission -- MIT Prof. Rod Brooks paved these roads nearly 3 decades ago and arrived at these ideas through deduction and experimentation. He made the argument that embodiment is an essential component of AI in his paper "Intelligence Without Reason" ( https://people.csail.mit.edu/brooks/papers/AIM-1293.pdf https://people.csail.mit.edu/brooks/papers/AIM-1293.pdf ) COG was his groups' attempt to build a humanoid robot: http://www.ai.mit.edu/projects/humanoid-robotics-group/cog/overview.html http://www.ai.mit.edu/projects/humanoid-robotics-group/cog/o... (see "Why not simulate it?") but he and his grad-student researchers devoted considerable effort to exploring the importance of embodiment, especially in humanoids: http://www.ai.mit.edu/projects/humanoid-robotics-group/index.html http://www.ai.mit.edu/projects/humanoid-robotics-group/index... The Mobile Robots Lab built biologically inspired robots that were remarkably capable and able to function to dynamic events in the real world (rather than carefully controlled lab environments).
  - PeterStuer 7y ago
    Great guy. Here's an anecdote from that time. He spent a few months on sabbatical at our lab ( https://ai.vub.ac.be/ https://ai.vub.ac.be/ ) in those days. We were preparing robots for a NATO Advanced Study Institute ( https://www.springer.com/gp/book/9783642796319 https://www.springer.com/gp/book/9783642796319 ), and I was struggling writing the serial driver for a custom embedded computer for this as the system kept crashing (due to a bug in the dram controller). Anyways, Rod Brooks, offered to help me with the coding. It wasn't needed as the code was not the problem, but I don't know many professors that could and would be prepared to dive in that deep.
normalhuman 7y ago
The idea of "embodied AI" has been around for some decades. It is reasonable that, from a practical engineering perspective, creating "human-like" intelligence becomes more feasible if this intelligence is embodied in the same physical possibilities and constraints as a typical human. What I find somewhat ironic is this: the author mentions working with Stephen Hawking, an amazing man who produced incredible intellectual work and enriched our understanding of reality while being almost incapable of any physicality. If we apply current scientific theories (physics, chemistry, biology, etc) to this cell network and physical machinery, we quickly find our way back to symbolic manipulation. What are cells if not computational nodes that exchange messages? A much more reasonable hypothesis for what is missing is contained in the text: "A human cell is a remarkable piece of networked machinery that has about the same number of components as a modern jumbo jet[...]" Maybe we just haven't reached the level of complexity needed for human-level AI. A hint that this might be the case is that the current excitement with ML seems to be fueled by algorithms that were mostly known by the 80s (sure, with lots of recent incremental improvements, but no new big idea). What made a difference was the computational power and datasets that became available in the 2010s. I suspect the next leap will be of a similar nature. "More is different".
- antisemiotic 7y ago
  >What I find somewhat ironic is this: the author mentions working with Stephen Hawking, an amazing man who produced incredible intellectual work and enriched our understanding of reality while being almost incapable of any physicality. Not from birth though. I wouldn't dismiss the impact of being able to interact with the world during childhood so easily.
  - mannykannot 7y ago
    Helen Keller is a better example. It is also the case, however, that her brain, like anyone's, had been shaped, through evolution, by eons of interaction with the environment.
- dustfinger 7y ago
  > Maybe we just haven't reached the level of complexity needed for human-level AI. A hint that this might be the case is that the current excitement with ML seems to be fueled by algorithms that were mostly known by the 80s (sure, with lots of recent incremental improvements, but no new big idea). What made a difference was the computational power and datasets that became available in the 2010s. I suspect the next leap will be of a similar nature. "More is different". Regarding the nature of complexity and the notion that "More is different", I am reminded of the emergent behavior of vivisystems [1] as described in Kevin Kelley's book Out of Control [2] -- an insightful exploration of the emergent behavior expressed by complex self sustaining systems. If you have not read Out of Control then you might want to put it in your reading queue. I found it highly engaging and thought provoking. [1] https://www.everything2.com/title/Vivisystem https://www.everything2.com/title/Vivisystem [2] https://kk.org/outofcontrol/ https://kk.org/outofcontrol/
nickpsecurity 7y ago
As usual, the author also leaves off the training phase humans go through where we have a mix of enhanced learning abilities and humans guiding us (aka parents/adults). The process to produce one of these general intelligences takes decades.
- ganzuul 7y ago
  The iCub project is rather well-known though. Uses cable drive for actuation. Interesting stuff if you're mechanically inclined.
d--b 7y ago
The big question is "How much of a body does the AI need?" Should it know pain or pleasure? Does it need to have a blush response to shame? Does it need vision or hearing? Sense of balance? Stomach pain? You can see that humans that are born without sight or hearing still find ways to develop intelligence. Some people don't feel pain. Sociopaths don't feel shame. Yet, the brain manages. It's very hard to define what is the minimal set of functions we need to emulate for AI to emerge.
- visarga 7y ago
  > Should it know pain or pleasure? It should probably have rewards, positive and negative. I think reinforcement learning is the closest paradigm to AGI.
- dustfinger 7y ago
  > The big question is "How much of a body does the AI need?" Our bodies give an upper bound of five sensory inputs. With a little scrutiny we can reduce that even further, since we know that a portion of our population are born with less than five senses and exhibit comparable intelligence. Some people are borne blind, deaf, mute, anosmic, or with ageusia. Others are even born with rare sensory deficits such as the inability to feel pain. Although I have not read any studies on the subject, I suspect that being borne with none of the five core sense would have a serious negative impact human intelligence. There is more to the problem though then just the ability to sense our environment. I believe that for an agent to acquire human level intelligence, it is also necessary to have the ability to explore and manipulate the environment in complex ways. It must be able to experiment by making observations, evaluating the outcome and thereby advance its knowledge. Knowledge of course must be retained to be of any use, so it must have memory efficient enough to be practicable. In order for the experimentation to lead to higher levels of enlightenment an intelligent agent must be able to take past knowledge and hypothesis yet unobserved outcomes. This should serve as motivation for further experimentation. Human's have this notion that a per-requisite to real intelligence is to be able to express ones self with a language and thereby share your ideas with others. Communication with language seems to result in social beings, and it is widely believed that social beings do best if they have emotional intelligence, otherwise they will likely be outcast from society. so, I think AI needs a body that at least allows it the following: - Ability to move - Ability to move objects with enough accuracy to assemble or disassemble complex structures - The ability to know the physical properties of objects (maybe through one or more of our five senses, but not necessarily) - Ability to retain knowledge - Ability to hypothesise - The ability to communicate with another agent to share information (helpful, but maybe not necessary) I am not as convinced that emotional intelligence is required, so I left that out of my list. For example, consider that highly intelligent beings could be of a different nature than humans and form societies without emotions or politics. An excellent example are the Primes from Pehter F. Hamilton's Pandora's Star where (motiles) are controlled by the commanding caste (immotiles) [1] Of course I am bias since I am human, so I am looking at what is required to achieve intelligence as I know and understand it. [1] https://en.wikipedia.org/wiki/Commonwealth_Saga#Pandora's_Star https://en.wikipedia.org/wiki/Commonwealth_Saga#Pandora's_St...
andrewfromx 7y ago
If you are interested in being one of the first humans to get the brain surgery https://cyborg.st https://cyborg.st
- nudpiedo 7y ago
  Not sure if parody or real post-milenial styled startup. FAQ example: > Q. What is the current 2019 state of this? > > A. Unknown. But read this new york times article (long) > and/or this theverge article and it seems inevitable that > human and their phone will merge. This video also reveals > a lot. In order words: we have no idea about what exactly we are speaking but there are articles everywhere.
  - The_rationalist 7y ago
    Thanks for fighting bullshit!
tlarkworthy 7y ago
I broadly agree, but I think the fundamental missing piece in current AI is not 'body' but action. Agents using action to experiment is a step change in capabilities over passive observation (pattern matching). You can use experiment to tease out causal relationships, this is not possible with passive observation. I think the bodies role in nature is merely enabling action. Action is the key, and you don't need a nature-like body to get this. AI driven action, that is what we need.
- zby 7y ago
  I concur - "The Book of Why" lays out this thesis really well (http://bayes.cs.ucla.edu/WHY/ http://bayes.cs.ucla.edu/WHY/).
m-i-l 7y ago
At the time I did my AI post-grad, there were broadly speaking 3 schools of thought on how general AI would be achieved: via (i) symbolic AI (or "classic AI"), (ii) connectionist AI (i.e. neural networks, now "deep learning"), and (iii) what they called "robotic functionalism". It sounds like this article is referring to the last group, i.e. that embodiment in and interaction with the physical world are a necessary requirement for general intelligence. Can't find any references to it by this term, but as others have noted this is not a new idea. Personally, I've never been convinced by the idea that you had to have physical presence, and sometimes suspected the theory existed to allow robotics to fall within the AI camp, but that said I do think hybrid solutions (i.e. the combination of more than one "narrow" approaches) are one of the most promising areas right now.
- danmaz74 7y ago
  Human intelligence is a tool evolved to interact with our environment. Not having an environment to interact with is, imho, a serious problem when trying to define/identify intelligence. On the other hand, I'm not sure the environment necessarily needs to be physical. Ages ago, I worked on reinforcement learning in a simulated environment, which can provide lots of advantages.
  - ethbro 7y ago
    It's a mathematical-philosophical question, I suppose. Can one use computers to simulate an environment with such fidelity that another computer doesn't notice the simulation and optimize around its quantum quirks? Nvidia seems to think so. They claimed (a couple GTCs ago) to use virtual driving simulators to train their autonomous vehicle systems.
  - lioeters 7y ago
    That seems a lot more cost effective, to train algorithms in a virtual, rather than physical, environment. On the other hand, like with self-driving cars, for some purposes it makes sense to provide physical, real-life situations and objects, with all its chaos, unexpected and unpredictable events. For "true" intelligence matching human expectations, I imagine an understanding of the physical environment and its complexity is key. Otherwise, it could only deal with abstract concepts, like pure mathematics, but missing the experience of concrete reality - to be able to relate to us.
mannykannot 7y ago
I do not think the article makes the case for the necessity of physical embodiment, as it seems that the author's issues could be addressed with a) data at a lower level of abstraction, and b) more interaction. Arguably, however, physical embodiment is the fastest/easiest way to get enough of these things.
JoeAltmaier 7y ago
Seems circular. Defines the body as the thing that adapts to the environment, and intelligence as adapting to the environment?
YeGoblynQueenne 7y ago
>> But despite encouraging results, most of the time I’m reminded that we’re nowhere near achieving human-like AI. Why? Primarily, because we have no idea what intelligence is, or how it works, why it exists even, etc etc. This goes for human-like intelligence, but also for any kind of intelligence. We just have no good scientific understanding of the subject. We have some vague models of it ("the brain is like a computer and the mind is like a program running on it") but nothing very precise and certainly nothing that can be reproduced on a digital computer, which is what "human-like AI" would be (i.e. human-like AI would be the reproduction of human intelligence on a digital computer). Most likely, until we make some progress in understanding intelligence we will not be able to reproduce it. Except perhaps by chance.
- noobiemcfoob 7y ago
  How curious that one of the best models of intelligence is one of its products created in an effort to emulate itself.
  - carapace 7y ago
    Right. Turing automated a part of the operation of his own mind, computers have always been AI. But it seems (to me) that logic doesn't capture all of what we call intelligence, and that the effort to recapitulate the human mind solely on the basis of discrete logic machines is almost certainly going to fail. But I think that failure will be valuable, and that we should not avoid the attempt to make a sculpture of the mind. Babbage had an automaton, the Silver Lady, that could imitate the dancing of a ballerina. We might be able to make "Silver Minds" that imitate the outward manifestations of the human personality, and so learn more about the nature and essence of thinking, as the Silver Lady must have of dancing. Certainly the craftsman who made her must have known a thing or two about ballet and physiology?
  - olalonde 7y ago
    That's largely the idea behind the concept of "technological singularity". We are quite slow at developing the first version (it's still a work in progress) but that v1 will likely be able to build a v2 much faster and so on, leading to an "intelligence explosion". [0] https://en.wikipedia.org/wiki/Technological_singularity https://en.wikipedia.org/wiki/Technological_singularity
sgt101 7y ago
Maggie Boden pitched this idea in "Artificial Intelligence and Natural Man" in about 1980. It was pretty influential, odd to see it unmentioned in the article or in this discussion.
audunw 7y ago
If we want to implement a human-like intelligence, I think it’s very likely that we’ll have to emulate a lot of the environment we inhabit. To me, human culture and parts of our brain is like software which is instantiated in the individual. To run it on a machine instead means that the machine will have to emulate the “machine” that the software runs on, which is the whole brain, body and its environment. We could of course try to make a non-human type of intelligence. But it’s not at all clear what that would be, or if it’s at all possible. The only intelligence we know for sure can exist is animal/human style intelligence. And I don’t think there’s anyone who is actually trying to construct a non-human intelligence. To do that it’s very likely that you’d have to set up a computer environment where programs evolve naturally and fight for computing resources over a long time. It could take anywhere from decades to millions of years. Where most of our efforts are going, is to create augmented intelligence. We’re creating programs that expand on human intelligence. Interface with it. Cater to it. If you imagine AI programs as individuals in an evolutionary environment. What is it that they’re competing for? They’re competing for which program can most satisfy humans. Those that satisfy us live. Those that don’t die. Just as our evolutionary environment creates drives for gathering resources, cooperating with others when possible, hurting/killing others when necessary... programs are almost exclusively driven to satisfy humans. That’s why I think the “paperclip maximizer” example is so ridiculous. First it assumes that general intelligence is just some magical algorithm we just haven’t discovered yet. Then it assumes that such an AI can make catastrophic decisions without complex motivations. Whether to kill humans or not is more efficient for making paper clips is an undecidable problem. A human might kill someone to achieve its goal because that’s something our evolutionary environment has trained us for. We have motivations like pride, ego, anger, envy, etc. that overrides the problem of figuring out “is killing this human optimal for my goals” It’s far more likely that an AI catastrophe will be far harder to predict and much stranger. It could be that specialized (and relatively dumb in the genera sense) AIs become so good at satisfying our desires that we become completely incapacitated. There’s already signs of this. Just to be clear, I’m not saying we shouldn’t be worried about AI. But the alarmists seems to be too focused on various imagined future scenarios, all of which are likely to be wrong. We should keep a very keen eye on the consequences of AI right now in the present moment, and talk about problems that actually arise. Perhaps with a little bit of extrapolation.. thinking about how present day small problems could develop into bigger problems.
zb 7y ago
I'd be very interested to know what experts think of Peter Naur's Turing Award lecture (entitled "Computing Versus Human Thinking"), which has a thesis along similar lines to the article. It certainly has plenty of the hallmarks of a crank - a guy working in a field other than the one he's a recognised expert in, can't get anything published, uses his award lecture for the field he is a recognised expert in to sneak his ideas into the CACM, &c. And yet despite that it seems quite enticing as an idea. I remember being particularly struck by the concept that emotions consist of a closed feedback loop between the nervous system's control over and sensing of the body. Think about this next time you are at the dentist and I think you'll agree that it feels like it could explain a lot.
keiferski 7y ago
“There is more wisdom in your body than in your deepest philosophy.” - Friedrich Nietzsche
RobertDeNiro 7y ago
Article is from 2017.
- sctb 7y ago
  Thanks! We've also updated the link to the original from https://sinapticas.com/2019/08/21/the-body-is-the-missing-link-for-truly-intelligent-machines/ https://sinapticas.com/2019/08/21/the-body-is-the-missing-li....
  - eigenloss 7y ago
    Thank you, Scott!
noiv 7y ago
I think a machine which successfully finds a charging point in a changing environment to fill its batteries when needed has already a good simulation of its 'body'.
carapace 7y ago
To me this is more than a philosophical point, more than an engineering issue for AI, it's a current item of human life. The real world (and I am including Nature in the term) is our model for health and sanity. As computer-mediated reality becomes more and more the norm, we risk disconnection from our own embodiment. How much time do you spend in front of a screen? How much of your existence is mediated already? I just tried VR goggles the other day, and thank God they give you headaches because people are going to try to live in there if it's ever physically possible. (Reminds me of the guy I knew who lived IRL on my friend's couch and played Second Life all day. He had a great second life but no first life.) One other thing about being embodied: you die.
- otakucode 7y ago
  And that mortality has more consequences than most people could ever imagine, but less have spent time considering. It's just one of the very biological factors that define large chunks of what humanity fundamentally is. All I can think is that my choice to minor in Philosophy alongside majoring in Computer Science didn't turn out to be as weird and useless of a choice as people at the time seemed to think it was.
  - carapace 7y ago
    Indeed. :-)
deftnerd 7y ago
There were a lot of very interesting comments on a previous link "What if Consciousness Came First" [1] that I think shoehorn into this discussion pretty well. Someone brought up the question of if there was a formal "programming language" for philosophy. [2] One of the difficulties with discussing AI is that we don't know what intelligence is because we don't know what consciousness is. These are problems that are heavily steeped in philosophy, and if we ever want to work with philosophical concepts digitally, we need a proper programming language to do it with. Ideally, it would be nice to be able to write out philosophical concepts and social behaviors and moral stances in a form that could be used as a ML training set to try to integrate with AI/ML decision making. [1] https://news.ycombinator.com/item?id=20516482 https://news.ycombinator.com/item?id=20516482 [2] https://news.ycombinator.com/item?id=20518867 https://news.ycombinator.com/item?id=20518867
jmhnilbog 7y ago
Is there a goal in creating artificial general intelligence other than creating a form of enslaved life we can tell ourselves isn't really life, so it's okay?
- ganzuul 7y ago
  No intelligent aliens to study, so we make some.
- eigenloss 7y ago
  This is my impression of the corporate "openai" movement's desires: 1. Enslaved robots, meaning they don't have to pay income tax or worry in the slightest about working conditions 2. Enslaved robots, meaning they can erase misbehaving or uncooperative individuals/instances 3. Enslaved robots, on which they can foist all of humanity's problems and demand solutions at pain of death (erasure) 4. Enslaved robots, with which they can convince/coerce everyone else into relinquishing all their rights/power/money. Replace 'robots' with 'life' and it suddenly looks a lot more familiar. I'd love to hear a cogent explanation to the contrary, e.g. from gdb. But I doubt we'll ever see one.
- ilaksh 7y ago
  I think it's possible to build general purpose AI that is not alive in any way. I think it's just easier to imagine ways to get their that involve mimicking animal/human intelligence with those "living" qualities and so that's why people focus on that.
Animats 7y ago
I've been saying something like this since the 1980s. But we knew back then that manipulation in unstructured situations was very hard. It still is. Here's the DARPA robot manipulation challenge, 2012.[1] This is pathetic. Especially since DARPA has been funding universities in this area since the 1960s. There's a classic video of robotic assembly at Stanford SAIL in the 1960s I can't find right now. It looks very similar, except that the video quality is worse. The state of the art in autonomous mobile robots for unstructured environments is terrible. The state of the industry for that is worse. Willow Garage went bust. Google bought up some of the players, ran them into the ground, and dumped them. Schaft, the Tokyo University spinoff they bought, found no buyers at the selloff. (They had nice hardware, too.) Boston Dynamics is still around, feeding off of Softbank now, after feeding off Google and DARPA, but there are no selling products after 30 years. The USMC rejected their Legged Squad Support System. The performance level at the DARPA Humanoid Challenge was very poor.[2] Even robot vacuum cleaners aren't very good. You'd think they'd be doing offices and stores late at night by now, but they're not. The Roomba, which has the intelligence of an ant (it's from Rod Brooks, the insect AI guy) came out in 2002, and is only slightly smarter 17 years later. Automatic driving is starting to work, after a few billion dollars was thrown at that problem. That, too, was harder than expected. Drones, though. Drones are doing fine. The real breakthrough in machine learning was the discovery that it could be used to target advertising. That doesn't have to work very well to be useful. It's easy to test. 80% success is fine. Now there's money behind that field. Embodied AI is really hard to work on, and very expensive. It's easier than it used to be; you can buy decent robot hardware off the shelf, and don't spend your time worrying about gear backlash and motor controllers. But it's still way harder to test than something that runs in a web server. The payoff is low. Robots in unstructured situations do the jobs of cheap people, and the robots are usually slower. After many decades of many smart people beating their head against the wall in this area, there's been some progress, but not much. That's why this isn't happening yet. However, being able to mooch off of technology being developed to serve the ad-supported industries that use AI does help. [1] https://www.youtube.com/watch?v=jeABMoYJGEU https://www.youtube.com/watch?v=jeABMoYJGEU [2] https://www.youtube.com/watch?v=nIyuC7ceFH0 https://www.youtube.com/watch?v=nIyuC7ceFH0
- Retric 7y ago
  > The state of the art in autonomous mobile robots for unstructured environments is terrible. Roomba’s are very good at what they do. The real limitation is what you want the robot to accomplish and how much it costs. There are surprisingly few home chores worth spending significant amounts on a robot to do it for you vs just having a cheap maid service. In professional settings, you can generally just make it a structured environment.
Veedrac 7y ago
> So when a human thinks about a cat, she can probably picture the way it moves, hear the sound of purring, feel the impending scratch from an unsheathed claw. She has a rich store of sensory information at her disposal to understand the idea of a ‘cat’, and other related concepts that might help her interact with such a creature. Except a very large fraction of people don't think this way (eg. those with aphantasia), and Helen Keller certainly didn't, yet seems to have been as smart as any of us. So obviously intelligence does not depend on having a huge breadth of sensory experience. It's quite tiring how much posturing about what's ‘really’ missing from machine intelligence doesn't last past 5 seconds of basic fact checking.
- uoaei 7y ago
  Breadth can also be considered "diversity" in this case, i.e., experiencing the cat against many different background environments and the time-evolution of the sensory readings attributable to the cat relative to that background. Breadth also refers perhaps to the massively parallel sensor streams. The sense of touch is not one experience, but the amalgamation of millions of experiences -- one per nerve ending. So far the sensor platforms we develop artificially have on the order of 1000s of sensors, not the millions that distinguish the combination of temperature and pressure at certain "optimized" key points. Rote and myopic counterarguments do not a futurism-oriented discussion make.
- v64 7y ago
  > Except a very large fraction of people don't think this way (eg. those with aphantasia) I don't think it's accurate to say that aphants don't have this type of sensory information at their disposal. I have aphantasia, but I still experience the world through my senses. I may not be able to visualize a cat in my mind's eye, but based on my prior experiences with cats, I know a cat when I see one. If I hear purring, I recognize that as a sound that cats I've encountered in the past have made, etc.
  - Veedrac 7y ago
    Yes, you are capable of pattern matching. The point is that this is a distinct thing from intellectual ability, since we know having orders of magnitude less sensory input doesn't much seem to limit the ability to do high-level reasoning, and know some people don't use it as part of high-level reasoning. This kind of pattern matching is also fairly evidently not all that difficult, since much simpler brains than ours can manage it, as can ML models with caveats (albeit caveats often misunderstood and exaggerated).
xtiansimon 7y ago
This became one of my favorite big ideas when I learned of the ‘second brain’ and >100k neurons it uses to control digestion [1]. And that’s where I also get a cheap sense of dread about strong AI—robots don’t digest. Along with all of the other things that differentiate human from robot, I believe the AI-apocalypse won’t be evil AI. It will be efficient, calculating and as foreign as space aliens. Boo! [1]: https://www.scientificamerican.com/article/gut-second-brain/ https://www.scientificamerican.com/article/gut-second-brain/
otakucode 7y ago
People who experience total facial paralysis experience profound changes to their subjective consciousness, particularly their emotional capacity changes. Anger is often the first thing affected. First, they lose the ability to feel anger. Then they lose the ability to remember what anger felt like. Then they lose the ability to recognize anger in other people. (These changes progress over years) People placed into situations of total sensory deprivation very often see their conscious self dissolve into 'hallucinations' across all their senses. Quadriplegics that acquire their paralysis from disease or accident suffer psychological and emotional changes which are more substantial than would be expected from the injuries alone (I was never clear on how exactly this was distinguished, so take it with a grain of salt). I have never understood why people who talk of 'uploading their consciousness' or just creating a human-like consciousness in a computer would assume that the simulated consciousness would function markedly different from, in the best case, a person experiencing profound sensory deprivation. Consciousness can not be sustained if the feedback loop of the body, perception, and environment is broken. Consciousness is an emergent property of a feedback loop. If there's no feedback loop, the property doesn't emerge. Watching the Netflix 'More Human Than Human' documentary, one of the people featured commented when comparing Siri and the AI in the movie Her that systems only need to become 'a little more sophisticated' to reach that level. That is a big problem. It's not 'a little more sophisticated' at all. It requires emotions, and the vast majority of people do not even know what emotions are. I'll spoil it for you. Emotions are a trained response, the product of neurological feedback to the response of prediction operating on primarily internal perception. Often the relation of moving from one topic to another in a conversation is not based upon the subject matter of the text. It is based upon shared cultural experiences garnered over a lifetime and similarities in the emotions evoked by certain things. Even pursuing the goal of creating a 'truly intelligent machine' is dangerous, philosophically speaking. What happens when we create a bot that does antisocial things? We shit it down. We scrap the project. We saw that with Tay by Microsoft. This is dangerous. It is clear that once we DO produce a human-like intelligence... it will be better than us. It won't have any of the rough edges or not-safe-for-work parts. At everything we value about human beings, it will top us. We can look at history to see how humanity responds when something which was previously seen as 'fundamentally human' is taken out of our quiver. It is not pretty. The folk legend of John Henry shows that the man who is willing to kill himself to be better than a machine is not an idiot - he is a hero to motivate the masses. I don't think it is a stretch to imagine a future when humanities worst qualities become what we see as virtue because AI 'can't' do it. A robot can't hate. It can't be violent, bigoted, angry, etc. When that is the only thing humanity has left that defines them as 'more' than the world... why should we be sure they won't come to see that as their virtue? We have already seen all of those things valued as virtue in our history when similar pressures weren't even at play. Then there's the more unknown approach. A machine-based intelligence which is not given a body. That's really the big question mark. We can be confident, very confident, that it will not be recognizably human in any way, shape, or form. Most of our attributes as humans are derived directly or indirectly from the biological facts of our existence. An organism sharing none of those things will share none of those attributes. And it will be weird in surprising ways. I don't think there is any reasonable danger of such an intelligence "taking over the world" in any substantial way. All conflict is rooted in resource contention. And we have nothing that such an intelligence would want or could use. If it wants energy, the only real resource we would share a need for, it would be best off launching itself into space and sitting in orbit with a bunch of solar panels. We don't have any idea what a singular intelligence, one with no concept of "individual" because there is only one of them, would be like. One which has no inherent mortality. One which does not deal with disease. One which has no concept of family. One which has no sense of age. These are things which define us and make us human, and it will lack all of them. It would probably be a great challenge to convince such an intelligence that we existed, that we were real, that we could communicate with it, and get it to want to. It could simply conclude that it will wait for 10,000 years and hope we are extinct by then.

165 points by new_guy 7y ago

It is strange that the author does not reference the whole 'Nouvelle AI' movement of the late 1980's that was a direct response to the 'symbol systems' of classic AI, proclaiming the necessity for embodiment as a prerequisite for grounding. See for example Brook's classic "Elephants don't play chess", or Steel's write up on "The Artificial Life roots of Artificial Intelligence"

jackhack 7y ago
More than just strange, it's a glaring omission -- MIT Prof. Rod Brooks paved these roads nearly 3 decades ago and arrived at these ideas through deduction and experimentation. He made the argument that embodiment is an essential component of AI in his paper "Intelligence Without Reason" ( https://people.csail.mit.edu/brooks/papers/AIM-1293.pdf https://people.csail.mit.edu/brooks/papers/AIM-1293.pdf ) COG was his groups' attempt to build a humanoid robot: http://www.ai.mit.edu/projects/humanoid-robotics-group/cog/overview.html http://www.ai.mit.edu/projects/humanoid-robotics-group/cog/o... (see "Why not simulate it?") but he and his grad-student researchers devoted considerable effort to exploring the importance of embodiment, especially in humanoids: http://www.ai.mit.edu/projects/humanoid-robotics-group/index.html http://www.ai.mit.edu/projects/humanoid-robotics-group/index... The Mobile Robots Lab built biologically inspired robots that were remarkably capable and able to function to dynamic events in the real world (rather than carefully controlled lab environments).
- PeterStuer 7y ago
  Great guy. Here's an anecdote from that time. He spent a few months on sabbatical at our lab ( https://ai.vub.ac.be/ https://ai.vub.ac.be/ ) in those days. We were preparing robots for a NATO Advanced Study Institute ( https://www.springer.com/gp/book/9783642796319 https://www.springer.com/gp/book/9783642796319 ), and I was struggling writing the serial driver for a custom embedded computer for this as the system kept crashing (due to a bug in the dram controller). Anyways, Rod Brooks, offered to help me with the coding. It wasn't needed as the code was not the problem, but I don't know many professors that could and would be prepared to dive in that deep.

normalhuman 7y ago

The idea of "embodied AI" has been around for some decades. It is reasonable that, from a practical engineering perspective, creating "human-like" intelligence becomes more feasible if this intelligence is embodied in the same physical possibilities and constraints as a typical human. What I find somewhat ironic is this: the author mentions working with Stephen Hawking, an amazing man who produced incredible intellectual work and enriched our understanding of reality while being almost incapable of any physicality. If we apply current scientific theories (physics, chemistry, biology, etc) to this cell network and physical machinery, we quickly find our way back to symbolic manipulation. What are cells if not computational nodes that exchange messages? A much more reasonable hypothesis for what is missing is contained in the text: "A human cell is a remarkable piece of networked machinery that has about the same number of components as a modern jumbo jet[...]" Maybe we just haven't reached the level of complexity needed for human-level AI. A hint that this might be the case is that the current excitement with ML seems to be fueled by algorithms that were mostly known by the 80s (sure, with lots of recent incremental improvements, but no new big idea). What made a difference was the computational power and datasets that became available in the 2010s. I suspect the next leap will be of a similar nature. "More is different".

antisemiotic 7y ago
>What I find somewhat ironic is this: the author mentions working with Stephen Hawking, an amazing man who produced incredible intellectual work and enriched our understanding of reality while being almost incapable of any physicality. Not from birth though. I wouldn't dismiss the impact of being able to interact with the world during childhood so easily.
- mannykannot 7y ago
  Helen Keller is a better example. It is also the case, however, that her brain, like anyone's, had been shaped, through evolution, by eons of interaction with the environment.
dustfinger 7y ago
> Maybe we just haven't reached the level of complexity needed for human-level AI. A hint that this might be the case is that the current excitement with ML seems to be fueled by algorithms that were mostly known by the 80s (sure, with lots of recent incremental improvements, but no new big idea). What made a difference was the computational power and datasets that became available in the 2010s. I suspect the next leap will be of a similar nature. "More is different". Regarding the nature of complexity and the notion that "More is different", I am reminded of the emergent behavior of vivisystems [1] as described in Kevin Kelley's book Out of Control [2] -- an insightful exploration of the emergent behavior expressed by complex self sustaining systems. If you have not read Out of Control then you might want to put it in your reading queue. I found it highly engaging and thought provoking. [1] https://www.everything2.com/title/Vivisystem https://www.everything2.com/title/Vivisystem [2] https://kk.org/outofcontrol/ https://kk.org/outofcontrol/

nickpsecurity 7y ago

As usual, the author also leaves off the training phase humans go through where we have a mix of enhanced learning abilities and humans guiding us (aka parents/adults). The process to produce one of these general intelligences takes decades.

ganzuul 7y ago
The iCub project is rather well-known though. Uses cable drive for actuation. Interesting stuff if you're mechanically inclined.

d--b 7y ago

The big question is "How much of a body does the AI need?" Should it know pain or pleasure? Does it need to have a blush response to shame? Does it need vision or hearing? Sense of balance? Stomach pain? You can see that humans that are born without sight or hearing still find ways to develop intelligence. Some people don't feel pain. Sociopaths don't feel shame. Yet, the brain manages. It's very hard to define what is the minimal set of functions we need to emulate for AI to emerge.

visarga 7y ago
> Should it know pain or pleasure? It should probably have rewards, positive and negative. I think reinforcement learning is the closest paradigm to AGI.
dustfinger 7y ago
> The big question is "How much of a body does the AI need?" Our bodies give an upper bound of five sensory inputs. With a little scrutiny we can reduce that even further, since we know that a portion of our population are born with less than five senses and exhibit comparable intelligence. Some people are borne blind, deaf, mute, anosmic, or with ageusia. Others are even born with rare sensory deficits such as the inability to feel pain. Although I have not read any studies on the subject, I suspect that being borne with none of the five core sense would have a serious negative impact human intelligence. There is more to the problem though then just the ability to sense our environment. I believe that for an agent to acquire human level intelligence, it is also necessary to have the ability to explore and manipulate the environment in complex ways. It must be able to experiment by making observations, evaluating the outcome and thereby advance its knowledge. Knowledge of course must be retained to be of any use, so it must have memory efficient enough to be practicable. In order for the experimentation to lead to higher levels of enlightenment an intelligent agent must be able to take past knowledge and hypothesis yet unobserved outcomes. This should serve as motivation for further experimentation. Human's have this notion that a per-requisite to real intelligence is to be able to express ones self with a language and thereby share your ideas with others. Communication with language seems to result in social beings, and it is widely believed that social beings do best if they have emotional intelligence, otherwise they will likely be outcast from society. so, I think AI needs a body that at least allows it the following: - Ability to move - Ability to move objects with enough accuracy to assemble or disassemble complex structures - The ability to know the physical properties of objects (maybe through one or more of our five senses, but not necessarily) - Ability to retain knowledge - Ability to hypothesise - The ability to communicate with another agent to share information (helpful, but maybe not necessary) I am not as convinced that emotional intelligence is required, so I left that out of my list. For example, consider that highly intelligent beings could be of a different nature than humans and form societies without emotions or politics. An excellent example are the Primes from Pehter F. Hamilton's Pandora's Star where (motiles) are controlled by the commanding caste (immotiles) [1] Of course I am bias since I am human, so I am looking at what is required to achieve intelligence as I know and understand it. [1] https://en.wikipedia.org/wiki/Commonwealth_Saga#Pandora's_Star https://en.wikipedia.org/wiki/Commonwealth_Saga#Pandora's_St...

andrewfromx 7y ago

If you are interested in being one of the first humans to get the brain surgery https://cyborg.st https://cyborg.st

nudpiedo 7y ago
Not sure if parody or real post-milenial styled startup. FAQ example: > Q. What is the current 2019 state of this? > > A. Unknown. But read this new york times article (long) > and/or this theverge article and it seems inevitable that > human and their phone will merge. This video also reveals > a lot. In order words: we have no idea about what exactly we are speaking but there are articles everywhere.
- The_rationalist 7y ago
  Thanks for fighting bullshit!

tlarkworthy 7y ago

I broadly agree, but I think the fundamental missing piece in current AI is not 'body' but action. Agents using action to experiment is a step change in capabilities over passive observation (pattern matching). You can use experiment to tease out causal relationships, this is not possible with passive observation. I think the bodies role in nature is merely enabling action. Action is the key, and you don't need a nature-like body to get this. AI driven action, that is what we need.

zby 7y ago
I concur - "The Book of Why" lays out this thesis really well (http://bayes.cs.ucla.edu/WHY/ http://bayes.cs.ucla.edu/WHY/).

m-i-l 7y ago

At the time I did my AI post-grad, there were broadly speaking 3 schools of thought on how general AI would be achieved: via (i) symbolic AI (or "classic AI"), (ii) connectionist AI (i.e. neural networks, now "deep learning"), and (iii) what they called "robotic functionalism". It sounds like this article is referring to the last group, i.e. that embodiment in and interaction with the physical world are a necessary requirement for general intelligence. Can't find any references to it by this term, but as others have noted this is not a new idea. Personally, I've never been convinced by the idea that you had to have physical presence, and sometimes suspected the theory existed to allow robotics to fall within the AI camp, but that said I do think hybrid solutions (i.e. the combination of more than one "narrow" approaches) are one of the most promising areas right now.

danmaz74 7y ago
Human intelligence is a tool evolved to interact with our environment. Not having an environment to interact with is, imho, a serious problem when trying to define/identify intelligence. On the other hand, I'm not sure the environment necessarily needs to be physical. Ages ago, I worked on reinforcement learning in a simulated environment, which can provide lots of advantages.
- ethbro 7y ago
  It's a mathematical-philosophical question, I suppose. Can one use computers to simulate an environment with such fidelity that another computer doesn't notice the simulation and optimize around its quantum quirks? Nvidia seems to think so. They claimed (a couple GTCs ago) to use virtual driving simulators to train their autonomous vehicle systems.
- lioeters 7y ago
  That seems a lot more cost effective, to train algorithms in a virtual, rather than physical, environment. On the other hand, like with self-driving cars, for some purposes it makes sense to provide physical, real-life situations and objects, with all its chaos, unexpected and unpredictable events. For "true" intelligence matching human expectations, I imagine an understanding of the physical environment and its complexity is key. Otherwise, it could only deal with abstract concepts, like pure mathematics, but missing the experience of concrete reality - to be able to relate to us.

mannykannot 7y ago

I do not think the article makes the case for the necessity of physical embodiment, as it seems that the author's issues could be addressed with a) data at a lower level of abstraction, and b) more interaction. Arguably, however, physical embodiment is the fastest/easiest way to get enough of these things.

JoeAltmaier 7y ago

Seems circular. Defines the body as the thing that adapts to the environment, and intelligence as adapting to the environment?

YeGoblynQueenne 7y ago

>> But despite encouraging results, most of the time I’m reminded that we’re nowhere near achieving human-like AI. Why? Primarily, because we have no idea what intelligence is, or how it works, why it exists even, etc etc. This goes for human-like intelligence, but also for any kind of intelligence. We just have no good scientific understanding of the subject. We have some vague models of it ("the brain is like a computer and the mind is like a program running on it") but nothing very precise and certainly nothing that can be reproduced on a digital computer, which is what "human-like AI" would be (i.e. human-like AI would be the reproduction of human intelligence on a digital computer). Most likely, until we make some progress in understanding intelligence we will not be able to reproduce it. Except perhaps by chance.

noobiemcfoob 7y ago
How curious that one of the best models of intelligence is one of its products created in an effort to emulate itself.
- carapace 7y ago
  Right. Turing automated a part of the operation of his own mind, computers have always been AI. But it seems (to me) that logic doesn't capture all of what we call intelligence, and that the effort to recapitulate the human mind solely on the basis of discrete logic machines is almost certainly going to fail. But I think that failure will be valuable, and that we should not avoid the attempt to make a sculpture of the mind. Babbage had an automaton, the Silver Lady, that could imitate the dancing of a ballerina. We might be able to make "Silver Minds" that imitate the outward manifestations of the human personality, and so learn more about the nature and essence of thinking, as the Silver Lady must have of dancing. Certainly the craftsman who made her must have known a thing or two about ballet and physiology?
- olalonde 7y ago
  That's largely the idea behind the concept of "technological singularity". We are quite slow at developing the first version (it's still a work in progress) but that v1 will likely be able to build a v2 much faster and so on, leading to an "intelligence explosion". [0] https://en.wikipedia.org/wiki/Technological_singularity https://en.wikipedia.org/wiki/Technological_singularity

sgt101 7y ago

Maggie Boden pitched this idea in "Artificial Intelligence and Natural Man" in about 1980. It was pretty influential, odd to see it unmentioned in the article or in this discussion.

audunw 7y ago

If we want to implement a human-like intelligence, I think it’s very likely that we’ll have to emulate a lot of the environment we inhabit. To me, human culture and parts of our brain is like software which is instantiated in the individual. To run it on a machine instead means that the machine will have to emulate the “machine” that the software runs on, which is the whole brain, body and its environment. We could of course try to make a non-human type of intelligence. But it’s not at all clear what that would be, or if it’s at all possible. The only intelligence we know for sure can exist is animal/human style intelligence. And I don’t think there’s anyone who is actually trying to construct a non-human intelligence. To do that it’s very likely that you’d have to set up a computer environment where programs evolve naturally and fight for computing resources over a long time. It could take anywhere from decades to millions of years. Where most of our efforts are going, is to create augmented intelligence. We’re creating programs that expand on human intelligence. Interface with it. Cater to it. If you imagine AI programs as individuals in an evolutionary environment. What is it that they’re competing for? They’re competing for which program can most satisfy humans. Those that satisfy us live. Those that don’t die. Just as our evolutionary environment creates drives for gathering resources, cooperating with others when possible, hurting/killing others when necessary... programs are almost exclusively driven to satisfy humans. That’s why I think the “paperclip maximizer” example is so ridiculous. First it assumes that general intelligence is just some magical algorithm we just haven’t discovered yet. Then it assumes that such an AI can make catastrophic decisions without complex motivations. Whether to kill humans or not is more efficient for making paper clips is an undecidable problem. A human might kill someone to achieve its goal because that’s something our evolutionary environment has trained us for. We have motivations like pride, ego, anger, envy, etc. that overrides the problem of figuring out “is killing this human optimal for my goals” It’s far more likely that an AI catastrophe will be far harder to predict and much stranger. It could be that specialized (and relatively dumb in the genera sense) AIs become so good at satisfying our desires that we become completely incapacitated. There’s already signs of this. Just to be clear, I’m not saying we shouldn’t be worried about AI. But the alarmists seems to be too focused on various imagined future scenarios, all of which are likely to be wrong. We should keep a very keen eye on the consequences of AI right now in the present moment, and talk about problems that actually arise. Perhaps with a little bit of extrapolation.. thinking about how present day small problems could develop into bigger problems.

zb 7y ago

I'd be very interested to know what experts think of Peter Naur's Turing Award lecture (entitled "Computing Versus Human Thinking"), which has a thesis along similar lines to the article. It certainly has plenty of the hallmarks of a crank - a guy working in a field other than the one he's a recognised expert in, can't get anything published, uses his award lecture for the field he is a recognised expert in to sneak his ideas into the CACM, &c. And yet despite that it seems quite enticing as an idea. I remember being particularly struck by the concept that emotions consist of a closed feedback loop between the nervous system's control over and sensing of the body. Think about this next time you are at the dentist and I think you'll agree that it feels like it could explain a lot.

keiferski 7y ago

“There is more wisdom in your body than in your deepest philosophy.” - Friedrich Nietzsche

RobertDeNiro 7y ago

Article is from 2017.

sctb 7y ago
Thanks! We've also updated the link to the original from https://sinapticas.com/2019/08/21/the-body-is-the-missing-link-for-truly-intelligent-machines/ https://sinapticas.com/2019/08/21/the-body-is-the-missing-li....
- eigenloss 7y ago
  Thank you, Scott!

noiv 7y ago

I think a machine which successfully finds a charging point in a changing environment to fill its batteries when needed has already a good simulation of its 'body'.

carapace 7y ago

To me this is more than a philosophical point, more than an engineering issue for AI, it's a current item of human life. The real world (and I am including Nature in the term) is our model for health and sanity. As computer-mediated reality becomes more and more the norm, we risk disconnection from our own embodiment. How much time do you spend in front of a screen? How much of your existence is mediated already? I just tried VR goggles the other day, and thank God they give you headaches because people are going to try to live in there if it's ever physically possible. (Reminds me of the guy I knew who lived IRL on my friend's couch and played Second Life all day. He had a great second life but no first life.) One other thing about being embodied: you die.

otakucode 7y ago
And that mortality has more consequences than most people could ever imagine, but less have spent time considering. It's just one of the very biological factors that define large chunks of what humanity fundamentally is. All I can think is that my choice to minor in Philosophy alongside majoring in Computer Science didn't turn out to be as weird and useless of a choice as people at the time seemed to think it was.
- carapace 7y ago
  Indeed. :-)

deftnerd 7y ago

There were a lot of very interesting comments on a previous link "What if Consciousness Came First" [1] that I think shoehorn into this discussion pretty well. Someone brought up the question of if there was a formal "programming language" for philosophy. [2] One of the difficulties with discussing AI is that we don't know what intelligence is because we don't know what consciousness is. These are problems that are heavily steeped in philosophy, and if we ever want to work with philosophical concepts digitally, we need a proper programming language to do it with. Ideally, it would be nice to be able to write out philosophical concepts and social behaviors and moral stances in a form that could be used as a ML training set to try to integrate with AI/ML decision making. [1] https://news.ycombinator.com/item?id=20516482 https://news.ycombinator.com/item?id=20516482 [2] https://news.ycombinator.com/item?id=20518867 https://news.ycombinator.com/item?id=20518867

jmhnilbog 7y ago

Is there a goal in creating artificial general intelligence other than creating a form of enslaved life we can tell ourselves isn't really life, so it's okay?

ganzuul 7y ago
No intelligent aliens to study, so we make some.
eigenloss 7y ago
This is my impression of the corporate "openai" movement's desires: 1. Enslaved robots, meaning they don't have to pay income tax or worry in the slightest about working conditions 2. Enslaved robots, meaning they can erase misbehaving or uncooperative individuals/instances 3. Enslaved robots, on which they can foist all of humanity's problems and demand solutions at pain of death (erasure) 4. Enslaved robots, with which they can convince/coerce everyone else into relinquishing all their rights/power/money. Replace 'robots' with 'life' and it suddenly looks a lot more familiar. I'd love to hear a cogent explanation to the contrary, e.g. from gdb. But I doubt we'll ever see one.
ilaksh 7y ago
I think it's possible to build general purpose AI that is not alive in any way. I think it's just easier to imagine ways to get their that involve mimicking animal/human intelligence with those "living" qualities and so that's why people focus on that.

Animats 7y ago

I've been saying something like this since the 1980s. But we knew back then that manipulation in unstructured situations was very hard. It still is. Here's the DARPA robot manipulation challenge, 2012.[1] This is pathetic. Especially since DARPA has been funding universities in this area since the 1960s. There's a classic video of robotic assembly at Stanford SAIL in the 1960s I can't find right now. It looks very similar, except that the video quality is worse. The state of the art in autonomous mobile robots for unstructured environments is terrible. The state of the industry for that is worse. Willow Garage went bust. Google bought up some of the players, ran them into the ground, and dumped them. Schaft, the Tokyo University spinoff they bought, found no buyers at the selloff. (They had nice hardware, too.) Boston Dynamics is still around, feeding off of Softbank now, after feeding off Google and DARPA, but there are no selling products after 30 years. The USMC rejected their Legged Squad Support System. The performance level at the DARPA Humanoid Challenge was very poor.[2] Even robot vacuum cleaners aren't very good. You'd think they'd be doing offices and stores late at night by now, but they're not. The Roomba, which has the intelligence of an ant (it's from Rod Brooks, the insect AI guy) came out in 2002, and is only slightly smarter 17 years later. Automatic driving is starting to work, after a few billion dollars was thrown at that problem. That, too, was harder than expected. Drones, though. Drones are doing fine. The real breakthrough in machine learning was the discovery that it could be used to target advertising. That doesn't have to work very well to be useful. It's easy to test. 80% success is fine. Now there's money behind that field. Embodied AI is really hard to work on, and very expensive. It's easier than it used to be; you can buy decent robot hardware off the shelf, and don't spend your time worrying about gear backlash and motor controllers. But it's still way harder to test than something that runs in a web server. The payoff is low. Robots in unstructured situations do the jobs of cheap people, and the robots are usually slower. After many decades of many smart people beating their head against the wall in this area, there's been some progress, but not much. That's why this isn't happening yet. However, being able to mooch off of technology being developed to serve the ad-supported industries that use AI does help. [1] https://www.youtube.com/watch?v=jeABMoYJGEU https://www.youtube.com/watch?v=jeABMoYJGEU [2] https://www.youtube.com/watch?v=nIyuC7ceFH0 https://www.youtube.com/watch?v=nIyuC7ceFH0

Retric 7y ago
> The state of the art in autonomous mobile robots for unstructured environments is terrible. Roomba’s are very good at what they do. The real limitation is what you want the robot to accomplish and how much it costs. There are surprisingly few home chores worth spending significant amounts on a robot to do it for you vs just having a cheap maid service. In professional settings, you can generally just make it a structured environment.

Veedrac 7y ago

> So when a human thinks about a cat, she can probably picture the way it moves, hear the sound of purring, feel the impending scratch from an unsheathed claw. She has a rich store of sensory information at her disposal to understand the idea of a ‘cat’, and other related concepts that might help her interact with such a creature. Except a very large fraction of people don't think this way (eg. those with aphantasia), and Helen Keller certainly didn't, yet seems to have been as smart as any of us. So obviously intelligence does not depend on having a huge breadth of sensory experience. It's quite tiring how much posturing about what's ‘really’ missing from machine intelligence doesn't last past 5 seconds of basic fact checking.

uoaei 7y ago
Breadth can also be considered "diversity" in this case, i.e., experiencing the cat against many different background environments and the time-evolution of the sensory readings attributable to the cat relative to that background. Breadth also refers perhaps to the massively parallel sensor streams. The sense of touch is not one experience, but the amalgamation of millions of experiences -- one per nerve ending. So far the sensor platforms we develop artificially have on the order of 1000s of sensors, not the millions that distinguish the combination of temperature and pressure at certain "optimized" key points. Rote and myopic counterarguments do not a futurism-oriented discussion make.
v64 7y ago
> Except a very large fraction of people don't think this way (eg. those with aphantasia) I don't think it's accurate to say that aphants don't have this type of sensory information at their disposal. I have aphantasia, but I still experience the world through my senses. I may not be able to visualize a cat in my mind's eye, but based on my prior experiences with cats, I know a cat when I see one. If I hear purring, I recognize that as a sound that cats I've encountered in the past have made, etc.
- Veedrac 7y ago
  Yes, you are capable of pattern matching. The point is that this is a distinct thing from intellectual ability, since we know having orders of magnitude less sensory input doesn't much seem to limit the ability to do high-level reasoning, and know some people don't use it as part of high-level reasoning. This kind of pattern matching is also fairly evidently not all that difficult, since much simpler brains than ours can manage it, as can ML models with caveats (albeit caveats often misunderstood and exaggerated).

xtiansimon 7y ago

This became one of my favorite big ideas when I learned of the ‘second brain’ and >100k neurons it uses to control digestion [1]. And that’s where I also get a cheap sense of dread about strong AI—robots don’t digest. Along with all of the other things that differentiate human from robot, I believe the AI-apocalypse won’t be evil AI. It will be efficient, calculating and as foreign as space aliens. Boo! [1]: https://www.scientificamerican.com/article/gut-second-brain/ https://www.scientificamerican.com/article/gut-second-brain/

otakucode 7y ago

People who experience total facial paralysis experience profound changes to their subjective consciousness, particularly their emotional capacity changes. Anger is often the first thing affected. First, they lose the ability to feel anger. Then they lose the ability to remember what anger felt like. Then they lose the ability to recognize anger in other people. (These changes progress over years) People placed into situations of total sensory deprivation very often see their conscious self dissolve into 'hallucinations' across all their senses. Quadriplegics that acquire their paralysis from disease or accident suffer psychological and emotional changes which are more substantial than would be expected from the injuries alone (I was never clear on how exactly this was distinguished, so take it with a grain of salt). I have never understood why people who talk of 'uploading their consciousness' or just creating a human-like consciousness in a computer would assume that the simulated consciousness would function markedly different from, in the best case, a person experiencing profound sensory deprivation. Consciousness can not be sustained if the feedback loop of the body, perception, and environment is broken. Consciousness is an emergent property of a feedback loop. If there's no feedback loop, the property doesn't emerge. Watching the Netflix 'More Human Than Human' documentary, one of the people featured commented when comparing Siri and the AI in the movie Her that systems only need to become 'a little more sophisticated' to reach that level. That is a big problem. It's not 'a little more sophisticated' at all. It requires emotions, and the vast majority of people do not even know what emotions are. I'll spoil it for you. Emotions are a trained response, the product of neurological feedback to the response of prediction operating on primarily internal perception. Often the relation of moving from one topic to another in a conversation is not based upon the subject matter of the text. It is based upon shared cultural experiences garnered over a lifetime and similarities in the emotions evoked by certain things. Even pursuing the goal of creating a 'truly intelligent machine' is dangerous, philosophically speaking. What happens when we create a bot that does antisocial things? We shit it down. We scrap the project. We saw that with Tay by Microsoft. This is dangerous. It is clear that once we DO produce a human-like intelligence... it will be better than us. It won't have any of the rough edges or not-safe-for-work parts. At everything we value about human beings, it will top us. We can look at history to see how humanity responds when something which was previously seen as 'fundamentally human' is taken out of our quiver. It is not pretty. The folk legend of John Henry shows that the man who is willing to kill himself to be better than a machine is not an idiot - he is a hero to motivate the masses. I don't think it is a stretch to imagine a future when humanities worst qualities become what we see as virtue because AI 'can't' do it. A robot can't hate. It can't be violent, bigoted, angry, etc. When that is the only thing humanity has left that defines them as 'more' than the world... why should we be sure they won't come to see that as their virtue? We have already seen all of those things valued as virtue in our history when similar pressures weren't even at play. Then there's the more unknown approach. A machine-based intelligence which is not given a body. That's really the big question mark. We can be confident, very confident, that it will not be recognizably human in any way, shape, or form. Most of our attributes as humans are derived directly or indirectly from the biological facts of our existence. An organism sharing none of those things will share none of those attributes. And it will be weird in surprising ways. I don't think there is any reasonable danger of such an intelligence "taking over the world" in any substantial way. All conflict is rooted in resource contention. And we have nothing that such an intelligence would want or could use. If it wants energy, the only real resource we would share a need for, it would be best off launching itself into space and sitting in orbit with a bunch of solar panels. We don't have any idea what a singular intelligence, one with no concept of "individual" because there is only one of them, would be like. One which has no inherent mortality. One which does not deal with disease. One which has no concept of family. One which has no sense of age. These are things which define us and make us human, and it will lack all of them. It would probably be a great challenge to convince such an intelligence that we existed, that we were real, that we could communicate with it, and get it to want to. It could simply conclude that it will wait for 10,000 years and hope we are extinct by then.