Artificial Intelligence
 

Help! I need assist!

So for my Information and Society course I need to make weekly discussion postings on various topics and I'm drawing an absolute blank for this one.

The point is to argue something good or bad about artificial intelligence. We had a few articles to read, but not much, and I have no idea where to go.

What's good about AI? What's bad about AI? There are potential economic issues if machines take over certain jobs, bumping out people. And, of course, everything mentions the 'machines get smarter than us and make danger" rebuttal, though the usual response to that of vague 'safeguards' never seems too persuasive to me. (Science fiction seems to find ways around those safeguards quite a bit. ;P)

So really though, anyone have thoughts? I think the basic idea is to stick with something realistic but I'm tempted fly out into the realm of super-intelligence and emotive robots.

What's good about AI?

Well, first off, since AI is a computer, everything that computers are good at are things that are good about AI: An AI doesn't get bored. An AI can do repetitive tasks very very quickly. An AI can work with HUGE amounts of information. Computers are cheap compared to manual labor.

Then specifically to real-world AI: AI can find relationships between things and use those to make useful predictions about properties of those things. AI can learn from its mistakes.


What's bad about AI:

I mentioned that AI can learn from its mistakes, but the hard part is TELLING the AI that it did in fact make a mistake. It's very easy to train an AI in a biased way -- I've heard it said that "machine learning is like money laundering for bias".

Also, machine learning AI is ridiculously hard to analyze. You can't look at an AI's code to understand how it makes its conclusions. This also means that the only way to fix an AI that isn't coming to the right conclusions is to either throw away its data and start over, or flood it with new data and hope it learns the right conclusions from it.

AI (whether machine learning or not) is limited by its input and its knowledge. It doesn't have creativity or ingenuity or inventiveness to figure out how to deal with an unexpected problem.

At the moment, humans are still better than AI at finding interesting things in massive amounts of unstructured data. Teaching a computer how to drive a car is a lot harder than it sounds because humans are aware of SO MUCH about the world around them that we take for granted. We are insanely good general-purpose pattern matching engines. Computers can outperform us on some specific pattern matching tasks (ironically, AIs are better than humans at solving distorted-text CAPTCHAs), but especially when it comes to identifying features of images... we're REALLY GOOD AT IT.


What's irrelevant about AI:

The threat of industrialization putting people out of jobs isn't actually a bad thing, nor is it a new thing. Every disruptive technology has shifted things around in the job market, but people are resourceful and will find new things to do. New technologies create new fields for jobs and enable people to do jobs that they couldn't have done before. New technologies also have a tendency to reduce the cost of living by making products cheaper, so people get more leisure time because they don't have to work as much to afford what they need. And when people don't have to work as much, that opens up room for more jobs for other people.

Obviously that sounds idealistic, and it is, but it isn't the disruptive technologies that cause problems in the job market -- it's population growth, with more people wanting jobs than the market (technology or no technology) can actually support.

That analysis seems to be hand-waving a lot of significant factors like training time for new technologies jobs / public interest in pursuing them, especially among middle-aged and older displaced workers / the economic impact of re-training itself in schooling costs / increasing machine efficiency making the sheer number of human overseers required lower at an ever-increasing rate / society as a whole and the government adapting to this transition amicably and with conscientious legislation to keep everything from falling apart / the work ethic society has ingrained into people where a life full of unregimented leisure time can drive people stir-crazy / the executive elite not just going full an-cap tyrannical and creating a dystopia with levels of wealth and standard-of-living inequality like nothing seen on earth... I could go on, but that's probably enough for now. :P

@Quiet, I know I probably already mentioned it, but another area you could explore is the meta-impact of AI culturally and the game-theoretical power it holds where just a leak stating that a rival power is close to developing a full AGI could ignite a world war.

Edit @Crystal: Certain transhumanists would call that progress toward the Singularity. It's hard to argue that the old values millennia of social evolution have imprinted onto us won't erode as we slide further up that slope. I'm partial to the school of thought that says we shouldn't worry about AI destroying us, because by the time we get to that point, we ourselves will already no longer be recognizably human in the traditional sense.

It's also hard to argue against the idea that our obsessive media consumption and constant digital stimulation don't already make us in some sense fundamentally different creatures from people who haven't lived with these things as the norm.

And when they do have those? That's one of the types we looked at is AI that is capable of problem solving to a degree that if you give it parts and a problem, it can figure out an answer in it's own fashion.



My problem with stuff like that Suze (forgot to quote) is that I only have a small 'fill in the space' sheet to stick it all in, and for the sake of ease I'm sticking to it. ;)

But I was just getting started D:

I'm not so much handwaving it as fast-forwarding past it. Disruptive technologies are disruptive, yeah. That's why they're called that. But they don't have a lasting impact.

Furthermore, they take time to spread through the market. People who are paying attention will start preparing for it early, and the last MAJOR disruptive technology (the Internet itself) didn't start having a significant impact on the job market for a couple decades.

And AI is going to be another slow-moving market, because the applications it's good at aren't the kinds of applications humans are good at anyway. They're not going to be taking away all that many jobs because the things they're best applied to are the things that people don't hire humans for anyway. Data entry is one role that would count as a place where AI competes with human workers, but frankly that damage is already done and we're not going to see it displace many more jobs than it already has in that field.

Regarding training: The entire point of technology is to make people able to work more efficiently. Applications of technology used to help the humans do THEIR jobs make those jobs more valuable. Cashiers became more valuable with computerized cash registers and laser barcode scanners, making it possible for supermarkets to hire dozens of cashiers because the demand for them was INCREASED by the technology. It's only been in the last couple years that self-checkout has started displacing that job role.

In my estimation, electric cars are going to be a bigger problem than AI when it comes to market effects -- they're popular and growing rapidly, but the support technologies for them are lagging way behind and the maintenance skills are pretty significantly different.

You're rather looking into the realm of science fiction now. No AI technology currently being researched (or even anticipated) is capable of doing that. It's possible for the programmers to add in more types of expected problems, but unexpected problems are always going to be beyond the ability of AI to correctly deal with until we get AI with much more human-like intelligence.

Because just because the AI comes up with an answer in its own fashion doesn't mean it's the right answer.

Sorry, I realized after I posted that "hand-waving" sounds kind of dismissive, didn't mean it that way.

As to your point, though, I can't quite tell if we're talking the wider-spread concern of automation exacerbated by burgeoning developments in AI or something more specific, but either way I guess where I come down on the issue is that I'm not convinced society is equipped to handle the transition -- ignoring the not-insignificant demographic that thinks concerns about automation displacing work are hysterical science fiction -- if there's news I've missed, I'd be happy to be better informed, but I have not seen almost any acknowledgement that this is even a thing that's going to happen on the broader political stage, where in my opinion it really should have been being debated for years and years now.

I might be letting my political leanings shine through a bit, but color me cynical as to the responsibility of the free market to see us through to the other side of this rabbit hole with the average person better off than they were before.

There is no limitation on time frame in the class chat we're having. If a AI can make an answer on it's own, how would we know it is wrong, unless we do the calculations ourselves, and then what's the point of having the AI doing it? This assumes we are using an AI to get to answers we haven't figured out yet.

Insight into the human brain and it's process is one goal of artificial intelligence (I forget who to refer to for that one). The self-problem -solving sort (so to speak) doesn't help in that regard because we can't see the path the AI takes.

On the efficiency, the fact that I'm sitting in front of a piece of technology that serves solely to amuse me would seem to argue against that, especially as I spent more money on a bigger and shinier one whose sole purpose is to amuse me with bigger and shinier version of the same things as the last one.

There are many, many, MANY problems in this world that are difficult to FIND an answer for, but much more feasible to VERIFY the answer once you have it. You set the computer up to figure out what mankind could never figure out on its own, and then you have humans check it.

To take AI out of the picture for an example: It requires an engineer to figure out how to make a car frame stronger, but any old schmoe can smash a car to see if it worked.

It would be exceedingly foolish to trust AI to solve problems that we CAN'T test the results of. At that point, your AI is God, and your science has become faith. And at that point again you'd be tasking an AI to do jobs that humans can't do in the first place, so it won't have a significant impact on the job market.

If we DO start moving into a world where automation is seriously displacing jobs and it looks like they're never coming back, then we as a society will need to adapt to that. We will need to give up on the idea that all able-bodied adults need to have a job. We will need to make sure that people who don't have a job are provided for without stigma or discrimination -- but that shouldn't be hard because you'll have automated machines producing the goods that they'll need.

Of course, this is handwaving the fact that humans are apex predators.

I'm not sure that Homo sapiens sapiens can psychologically tolerate post-scarcity. As a whole, Man is not satisfied unless it is fighting for power. There are plenty of things that it accepts as stand-ins for power -- wealth, a large family, success, a full belly -- but in a world where AI is always going to be more powerful and wealth is meaningless, we're going to see hoarders, gluttons, tribal cliques, et cetera.

My sister and her boyfriend are graduate students of philosophy and her boyfriend just applied for a job to teach logic to ai. I told him it's going to be his fault when the robot uprising happens and that he should NOT teach robots logic, but him and my sister are saying, wouldn't it be better to have a moral person teach robots logic because this is happening and we can't stop it but if we get involved we can at least guide ai in the right direction.

I don't want robots passing turing tests, ai scares the hell out of me.

Teaching morality to an AI is rather more than complicated than, say, raising a human child. Except in outlandish sci-fi scenarios, you're not dealing with something that has human-relatable drives and desires. Violent opposition to us vis-a-vis a slave revolt seems unlikely unless major breakthroughs in replicating something resembling human consciousness in a synthetic substrate happen. We're not going to be looking at an entity that has such recognizable traits as identity, personality, self-interest, and motivations beyond its programmed purpose. Given how intrinsically emergent those phenomena are from things like hormones and genes, it might actually be impossible for a machine to develop consciousness as we typically acknowledge it to exist. (And I personally think research into machine intelligence is going to reveal some uncomfortable truths about our own sense of consciousness, but that's a different topic.)

Instead, you'll have something like the oft-invoked paper-clip maximizer that ends up enslaving or destroying humanity to turn the planet into a titanic paper-clip factory without any sense that it's done anything but carry out the purpose it was designed for. A lot of the scariest hypotheticals involving AI are where it can't be understood or reasoned with like a person, both because they're far more alien and far more likely than the alternative.

Then there's concerns with the fact that we more than likely won't be looking at a single AI but dozens or hundreds once the ball gets rolling, and then you consider the precarious balance of power between nation-states and suddenly you might not even want your AI to be moral as regards the "enemy". There's a lot of potential for disaster just in the reliable human fallibility surrounding the development of machine super-intelligence, let alone the impact that the intelligence itself ends up having.

Just wanted to add, anecdotally, that I personally am coping with that in my current life-situation, there's literally nothing in the world that I want that isn't already in my possession or available to me (aside from a cure to my disease that doesn't exist) and the level of disaffection and depression I feel despite that makes me feel (admittedly without any hard evidence) that post-scarcity is going to be one of the harder problems we'll have to face as a species. It might not be killbots that do us in, but rather ennui and loss of meaning. We didn't evolve to subsist in paradise. Our brains are like hamster wheels of desire-seeking but it's all illusory, it's the exercise that's the point, not the imaginary goal.

-dies laughing- Oh man, if you think this is talking heavily about one subject...

@Claire Bear: Computers have been passing the Turing test for several years now. It's rather a poor test. It's really not hard for a chatbot to pretend to be a human, at least at a passing glance. Meanwhile, the stricter Turing test doesn't actually work, because real live humans have been known to FAIL Turing tests.

Teaching logic to an AI is... really kinda strange. They're BUILT on logic. They understand nothing BUT logic. So if your sister's boyfriend is applying for a job that actually exists, I'd be curious to learn more about what it is he's actually doing.

Keep in mind that we're nowhere CLOSE to general intelligence or seed AI. We don't even really know where to START, so that's all entirely science fiction at the moment.

Like Suze said, you can't teach morality to a computer the way you teach it to a human child. It's a similar process, yes, in that you have to train the AI on repeated examples of good and bad behavior, with punishments and rewards, but humans are evolved with social behaviors and a certain set of base instincts that an AI simply won't have. So instead, what you have to do is figure out how to express morality in terms of logic. Asimov was attempting to do that when he formulated the Three Laws.

AI doesn't need to be as scary as science fiction makes it out to be. The fact that we HAVE those science-fiction stories as a warning is going to guide the people who are working on those projects to having suitable safeguards from the beginning.

Speaking of the Three Laws, got any thoughts on those?

EDIT: Sorry, Quiet, hijacking yer thread.

It's a good first start, especially considering that they were written very early on in the genre. However, they're shortsighted on their own, and if you bring the Zeroth Law into the picture they give a risk of scenarios like robots enslaving humans for their own good. There are certainly better laws that could be written.

Er... what's the zero-th law? I'm actually not really well-versed in, uh, "classical" sci-fi.

the job is to teach something my sister said is called 'ontology' to ai so robots can "learn to reason with common sense". maybe logic was the wrong word, i'm not as smart as my sister so i simplified.

i dont want no robots with common sense. dont trust it. ya'll should be more scared of scifi scenarios, this shit happens faster than you think. twenty years ago you wouldn't have belevied me if i told you everyone would carry a computer in their pocket so quickly. science moves fast. robots will end up smarter than us someday. some robots probably are smarter than us, beating us at chess and whatnot.

i do not trust the kinda people who can afford to advance science. we're gonna end up in a world like Snowcrash one of these days.

Oh, hijack away, this is pretty cool.

Are those three laws things what they referenced in I, Robot?

And Suze, I suspect the meaning search is part of the reason I put up with my school grievances. I remember first encountering the 'humans need hardship' think in Appleseed, some time ago.

The paperclip thing falls into the "what if" when it comes to self-improving AI. That was another detail from somewhere. (I should just find where the teacher got that clip and post it). They didn't go as far as taking over the planet with a factory, but the hypothesis is that if you had a self improving AI, and gave it a job with the purpose that it find the most efficient way to do it, it could come up with a response that puts it into competition with people for resources.

I tempted to skip the debate form and just write a mini essay on why debating a particular idea is hard! That way I can blather about a pile of things, and hopefully still get credit. ;)

To paraphrase Sam Harris (and he might be quoting someone else like Bostrom for this, but), "The only thing scarier than the potential threat of AI is the potential loss of not developing it." I know some people are okay with writing off things like Alzheimer's and cancer as facts of life, or the inevitability of a supervolcano or meteor impact annihilating civilization as we know it as something beyond our control, but that's pretty much exactly the point. Our best chance of making it to the next stage of civilization is with AI as a ladder. Check out the concept of the Great Filter as something related to this point, as well. (Though it's also possible that AI itself is a filter.)

@Claire, it's not that we're saying there's no reason to be afraid or worry -- far from it, it's just that the things to be concerned over aren't as basic as evil conquering robots with guns and red eyes taking over the world. They're a lot more subtle and insidious, working on the level of societal and economic shifts that could unbalance civilization, and the fact that it's not something most people worry about is actually part of what makes it so scary. It's a hurdle that we, as a species and global community, have to make it over, more difficult than arguably any that have come before, and the consequences could be catastrophic. But it's going to happen, to one degree or another depending on the limits of what's actually possible, and we need to figure out how to handle it.

(Not us specifically here on Trisphee, obviously :P)

...I don't think the rest of the world would want anything to do with any of our potential ideas on how to handle such a scenario.

Whacking metal with a crowbar isn't going to tell anyone why the new method works, just that it does.

Unless an AI can tell us how it did something, then us verifying an answer could take as long as finding one in the first place. We do do things without knowing exactly what we are doing because sometimes, the result is more valuable than how it works.

I'm never sure what my thoughts on AI and Cancer would be, unless it's information processing and pattern recognition and stuff. I get the image of nanobots or something that attack cancer cells, but then I imagine those bots going haywire for whatever reason and then kinda disintegrating from the inside out. Not a big fan of foreign objects but I have no real issue with new and curious organic things (would be rrreeaaaaally awkward if I did.)

I found the thing! It's just under 10 minutes. It's the definition that we've been using for class. I'm sure I can wrangle something now, but it's fun to keep going. ;)

https://www.youtube.com/watch?v=kWmX3pd1f10

I unfortunately don't have any statistics to argue about the hypothetical safety of a hypothetical cure for cancer created by a hypothetical super-intelligence. :P

I'd be wary of any reactions based on an aversion to "unnatural" methods, though. Natural methods have had a long time to find cures and failed thus far. That's a very close step to just admitting defeat and saying "best not to meddle in God's domain" or the like (an "argument" my father is annoyingly fond of). It might be that problems like incredibly deadly diseases and disorders are beyond human capacity to cure, sort of how a lot of the sciences have to use advanced math to express their ideas because human intuition isn't built to process information beyond a certain level of complexity.

I'm not really talking natural, just organic. I trust flesh more than metal and plastic at this point. I have an kidney that is not mine already. I'd be perfectly okay if somewhere we managed to create some manner of flesh thread and organ recipe and Dr. Frankenstein a new kidney.

Of course, new ways to open and close skin holes would also be of benefit, too. Not sure if this requires and AI but it might help?

The Zeroth Law, in Asimov's writing, is an advanced form of the First Law that sufficiently advanced robots can formulate on their own. It states that a robot may not allow harm to come to humanity as a whole, or through inaction allow humanity to come to harm. This was an outgrowth of how AI had to reason about how a small harm to a human could prevent a larger harm (for example, punishing a child caused the child to be upset, which is harm, but it taught the child how to prevent bigger injuries).

Regarding AI and diseases: Any treatment that an AI comes up with would have to go through the same evaluation procedure that human-created treatments do. If a test population improves without side effects, you move on into further tests and then on into mass production. If a test population suffers horrendous side effects or gets worse, you abandon the drug and go back to the drawing board. So in that regard, AI is really just a tool for researchers to use in pursuit of what they're already doing.

Nanotechnology is a separate argument from AI. In particular, the problem with nanotechnology is the fact that they CAN'T have robust AI because they're tiny little things that act more like chemistry than like robots. There's no room for advanced computation. So while it's a worthwhile avenue of discussion, it's a DIFFERENT avenue of discussion.

Re: whacking metal with a crowbar -- You'd be surprised how much you can learn from whacking metal with a crowbar. It's a crude analogy, but the same techniques human engineers use to evaluate stresses in human-designed structures will work for evaluating stresses in AI-designed structures. The AI isn't going to invent brand-new laws of physics (unless they do! that would be super interesting!) -- the more likely surprise is going to come by developing a novel way of using things that can subsequently be analyzed to see what you can learn from it. That's how medicine has ALWAYS worked -- we discover molecules with medicinal properties first, then we start researching the pharmacology behind how it does its thing. Going the other way around was a flop -- pharmacologists spent a couple decades doing "rational drug design" and trying to engineer molecules specifically to pursue a hypothesized mechanism of action... it didn't really work any better than the typical method of "let's take this active molecule, swap out parts of it, and see what happens" and it cost a lot more. So an AI doing rational drug design would essentially be creating new base molecules for pharmacologists to study.

I'm just saying, if I had the choice of getting my eviscerated intestines replaced with a transplant or machine that does the job better, I know which one I'd take every time.

..but I'm filthy transhumanist. >_>

My point being that sometimes we start using things before we have that verification. A relative is taking meds right now that the doctors have no idea how it works, just that it does. I assume it's decided at this point, or case by case, that using it anyway is less harm than waiting to figure out what's actually going on.

Heh. Well, given those options...

But it depends on the situation. My body likes to super-cling to medical equipment so I'd rather avoid the extra bits and pieces if I can. (Images of having to remove a bionic kidney for some reason and the doctors are fighting with system going "NO! It's mine now!"

When it comes down to it, we don't know WHY anything works. In the last 50 years we've come a long way into getting more information about HOW things work at a fundamental level, but they've got major shortcomings -- even now we've only got some really vague theories that haven't even been fully ironed out yet to try to explain why gravity acts different from everything else.

AI doesn't change that. It shifts the landscape some, but in the end humans are still going to go on handling it all the same way we always have because that's all we can do. We try our best to understand, but in the meantime we verify that things do what we want them to do. That's how science has always worked.

Perhaps 'how' rather than 'why?'

So what you're saying is... no one can judge me for whispering fake Latin at my computer in attempt to appease the Machine Spirit next time it blue screens, right? :P

Does that work? My internalized rain dance to get Neir: Automata working is still somewhat hit or miss.

I mean, if you can empirically demonstrate that your rain dance has a greater probability of success than the control test where you don't do your rain dance, then sure, I'll accept that.