• Quick note - the problem with Youtube videos not embedding on the forum appears to have been fixed, thanks to ZiprHead. If you do still see problems let me know.

Artificial Intelligence thinks mushroom is a pretzel

theprestige said:
You're describing a very different problem. This isn't a problem of bad AI.

This is a problem of you thinking you're the customer, but really you're the product.
Click to expand...

It's really both. Yes, it's a problem of misusing the customers not even as product, like for advertising, but as involuntary lab rats for training an AI. But at the same time it IS the problem of what said AI is capable of. And it turns out: not much.

And I'll further say that it is a more general problem of the silly notion that anything can learn to know what you'd like better than you do, while working with buggerall data and feedback.

I mean, think about it. Even if you had a human instead of the AI, did I leave without leaving a like or dislike because I'm not very interested in the genre, or the video is meh, or the guy said something stupid (but not enough to warrant a thumbs down,) or I had a bad hair day and didn't feel like it, or what? Why did I even get to that video in the first place? Is it a topic that actually interests me, or did some idiot coworker spam a link to everyone in the company, or did I mis-click, or was it mis-tagged, or what?

Or let's say I did leave a thumbs up. Am I interested in the topic, the genre, the author, or did I just think the guy was really funny (while being otherwise completely wrong on the topic, as humour kinda tends to be), or what?

Let's say I click on something I usually don't. Well, what happened? Did I suddenly get interested in something else entirely? Is it a one off curiosity? Did I just click on a link in an email or forum post? Did I mis-click? Or what?

Was it even me at the keyboard? Computers can have multiple users in a family.

Etc.


And the last one is actually a very important issue for a lot of products. If, to return to that example, you try to take control of my toast for me... what happens when mom visits and she wants it less crunchy? Hell, what happens when _I_ get a bad tooth and want it lighter toasted than usual?

Circumstances change in all sorts of ways, and the idea that anything can learn a one perfect point that fits everything for ever, is just flat out... not well thought through.
 
Last edited:
Roboramma said:
I am quite aware that it can't be done very well at present. That doesn't demonstrate that it's impossible.
Click to expand...


You wouldn't believe how many people in threads over the years believe that because something is not done NOW it cannot be done EVER.

It has barely been 100 years since we were cutting limbs off people for small nicks. We've only had reasonably decent technology for 70 years out of our 70 thousand year existence (give or take).

And yet in thread after thread we get people saying X can't be done EVER because we haven't perfected it NOW.

Also, as far as AI threads go in particular, it is stunning how many people literally don't understand a damn thing about what they are talking about, yet will expound on AI anyway. Every AI thread here is a Dunning-Kruger orgy.
 
On a (way, way, way) less serious note, we don't want a perfect A.I. mostly because it has been sufficiently proven that it would eventually impose hell upon people, even people from the past!

I'm going to put it in spoiler tags to play along, since the theory is that only people aware of this during their lifetimes will be included in its hell. :D

Don't read this if you have anxiety problems.

Roko used ideas in decision theory to argue that a sufficiently powerful AI agent would have an incentive to torture anyone who imagined the agent but didn't work to bring the agent into existence. The argument was called a "basilisk" because merely hearing the argument would supposedly put you at risk of torture from this hypothetical agent — a basilisk in this context is any information that harms or endangers the people who hear it.


https://rationalwiki.org/wiki/Roko's_basilisk


https://wiki.lesswrong.com/wiki/Roko's_basilisk


Where to donate to not spend eternity in hell:

https://en.wikipedia.org/wiki/Machine_Intelligence_Research_Institute
 
This is The End said:
You wouldn't believe how many people in threads over the years believe that because something is not done NOW it cannot be done EVER.

It has barely been 100 years since we were cutting limbs off people for small nicks. We've only had reasonably decent technology for 70 years out of our 70 thousand year existence (give or take).

And yet in thread after thread we get people saying X can't be done EVER because we haven't perfected it NOW.

Also, as far as AI threads go in particular, it is stunning how many people literally don't understand a damn thing about what they are talking about, yet will expound on AI anyway. Every AI thread here is a Dunning-Kruger orgy.
Click to expand...
On the other hand, there are things that we know are so contrary to the way physics and the universe works that we can say with a great deal of confidence that it will never be possible with any kind of technology. Travelling through space at a speed that is faster than light is impossible not because of a lack of technology, but because the laws of physics make it impossible.

There are some (I happen to not be one of them) who feel that true General Artificial Intelligence is similarly impossible not because of a lack of technology, but because the laws of the universe prevent such a thing from existing.
 
Oh, I for one most certainly think it's possible to make a working AI. If anyone got that idea from my previous messages, let me dispel that misunderstanding.

What I do say is that currently the problem isn't as much the "artificial" part, but rather the "commercial" part of the equation. Namely:

1. Everyone seems to just want a quick and dirty hack. E.g., in the case of recognizing shapes, nobody actually wants an AI that learns what a mushroom is, or tries to reconstruct a 3D scene to figure out if that could be a mushroom, they just want something that looks at a 2D blob and just says "yep, it looks like other 2D blobs tagged mushroom."

And it is my firm conviction that that's extremely limiting.

2. More generally, the problem has historically been that instead of trying to even talk to psychologists and neuroscientists, the domain has been left to idiot savants who just like maths and pulling postulates out of the ass, and idiots who don't understand what they're funding there when they decide which of those gets the funding.

E.g., we actually still have an AI prize for the algorithms with the highest compression ration, because an idiot once postulated that the best AI will be the one which compresses the most data in the tightest space, and ignored issues like also needing some kind of way to quickly access information X that is referenced by information Y. So virtually all winners for that most promising AI have been variants of arithmetic compression, which, sure, compress the whole Wikipedia in a remarkably tight space, but you need literally HOURS to access any particular page in the compressed blob. So it would be physically impossible to make any kind of real time inferences based on the compressed data.

Literally about 80% of the interval we've even had to research AI up to the present has been WASTED with such nonsense, before anyone even tried the only learning model that actually works at all, namely, the Bayesian kind. Now everyone's doing that, but see point 1, they're still trying to take some abstract shortcut with it, and still won't talk to a neuroscientist.

But the most important of them all is:

3. Not everything flippin' needs an AI in it. And more generally, not everything needs to hit every checkbox on a "latest and coolest nerdy buzzwords" list.

Even if you actually made an AI that is indistinguishable from a smart human, it's still faster and even more convenient to just let me set the timer on the microwave to whatever it says on the box of whatever I'm heating there, than have an AI learn over many months what is the perfect time for every kind of food.
 
Last edited:
arthwollipot said:
On the other hand, there are things that we know are so contrary to the way physics and the universe works that we can say with a great deal of confidence that it will never be possible with any kind of technology. Travelling through space at a speed that is faster than light is impossible not because of a lack of technology, but because the laws of physics make it impossible.
Click to expand...

I get your point, but that may not be the best example. Effectively FTL travel is still theoretically possible via wormholes, but they'd rely on the existence of exotic matter, which we don't know any examples of yet. So most likely impossible, yes :)

A closed system with decreasing entropy may have been a better example :D

HansMustermann said:
3. Not everything flippin' needs an AI in it. And more generally, not everything needs to hit every checkbox on a "latest and coolest nerdy buzzwords" list.
Click to expand...

Hey! You can't take away my deep and meaningful relationship with my electric toothbrush!
 
More images that AI cannot properly identify...

AI fails to recognize these nature images 98% of the time

The Next Web said:
...AI just isn’t very good at understanding what it sees, unlike humans who can use contextual clues...

...what AI doesn’t know can kill us. This happened when Tesla’s ill-named “Autopilot” confused the white trailer of an 18-wheeler for a cloud and crashed into it resulting in the death of its driver...
Click to expand...

https://thenextweb.com/insider/2019...up-the-price-for-hd-video-on-unlimited-plans/
 
Hellbound said:
I get your point, but that may not be the best example. Effectively FTL travel is still theoretically possible via wormholes, but they'd rely on the existence of exotic matter, which we don't know any examples of yet. So most likely impossible, yes :)
Click to expand...
That's why I specifically and deliberately said travelling through space.
 
The AIs don't have any context yet, they are too simple.
If you want it to identify pretzels you train it with many pics containing pretzels. Without any context the AI finds similarities between all the pretzel pics it was trained with.
In that situation it's both smarter than you and much more stupid. It is as likely to notice the pretzels in all the pics as it is to note other similarities elsewhere in the pics, similarities you would not even be aware of.
When it identifies a mushroom as a pretzel it's not really making a mistake, because the mushroom picture shares a large number of similar features to the pretzel pics it was trained with.
Humans on the other hand can't help but see things in context and therefore are unaware of how similar the mushroom and the pretzel pics actually are.
 
Cheetah said:
If you want it to identify pretzels you train it with many pics containing pretzels. Without any context the AI finds similarities between all the pretzel pics it was trained with.
Click to expand...
Why "train" it with photos of pretzels instead of real pretzels?
 
I could really go for a big soft pretzel with rock salt and mustard right now. Is there an AI powerful enough to get that for me?
 
Cheetah said:
A NN is made up of artificial neurons that functions like idealized biological neurons.
Click to expand...

They are so "idealized" that they have almost nothing in common with their biological counterpart.
That makes them unsuitable for simulating biological brain and if some future version of NN achieves actual intelligence, it will be intelligence completely different from human's.

So I still stand by my comparison. Birds and planes, biological brains and artificial neural networks.

Cheetah said:
They are feed, they are awashed in hormones and other stimulants, they can fail, they can degrade, they can change their state, break connections and create new ones. No neuron is exactly same.
Click to expand...
Which gives artificial neurons the edge.
Click to expand...
Not if you want to get something similar to biological neural networks. That was my point, they are nothing alike.

Cheetah said:
You would basically have to simulate entire human brain to atomic level, all with blood circulation, pressure, temperature, trillions of chemical reactions occurring in every cell at once etc.
Click to expand...
No you don't.
Click to expand...
Yes, you have to, if you want to actually simulate human brain. Nodes and weights won't be enough. For example, if your brain simulation cannot get drunk, it is not good simulation. NN cannot get drunk.

Well, I guess instead of atomic-level numeric simulation you could do atomic-level NN-based simulation. Taking aside practicality of it, it is absolutely not same thing as equating physical neurons with their artificial counterparts.

Cheetah said:
Theses are single function NNs with only a few neurons, of course they can be fooled. Even humans with these HUGE brains are fooled by stuff all the time.
Click to expand...
My point was that these NNs are failing in very different way than human visual system. Human never would see panda as gibbon just because some pixels in picture were very slightly brightened or darkened. Artificial neural networks "perceive" world very differently than humans, get over it.

This is The End said:
On a (way, way, way) less serious note, we don't want a perfect A.I. mostly because it has been sufficiently proven that it would eventually impose hell upon people, even people from the past!
Click to expand...

Concept of Roko's Basilisk exists because certain fringe transhumanist cult that created it has set of peculiar axioms (you could compare them to tenets of traditional religious cult). Basilisk is logical consequence of those axioms.

Critical one is that two copies of mind that are identical are exactly same mind. Like, philosophically. This is why you could be threatened (if you accept these axioms) in principle by act of creating and torturing simulation of you in not-yet-existent future. Because, well, it is actually you! Not separate, independent instance of you (as any sane human would think), but YOU, period.

There are so many practical problems with it, even when you accept their axioms, that most cultists distance themself from that idea.

As far I am concerned, this particular axiom was created because of wishful thinking - it makes possible to pretend that uploading mind to computer actually "moves" you into computer. Assuming creating mind in computer actually works, obviously you will be copied, not moved. While your digital copy frolicks in cyberspace, original YOU will be still stuck in meatspace in confines of your skull. And for certain people, that won't do.
 
Mader Levap said:
They are so "idealized" that they have almost nothing in common with their biological counterpart.
That makes them unsuitable for simulating biological brain and if some future version of NN achieves actual intelligence, it will be intelligence completely different from human's.
Click to expand...
Some evidence would be appreciated.
 
HansMustermann said:
You know you've described something that already exists and is crap, right?



Because you only need to look at Google's YouTube recommendations, to see how exactly what you describe goes horribly wrong. Because at some point Google quite conspicuously stopped giving a flying f-bomb about actually doing anything for the users, and just started using several of its services as testing grounds for its algorithms. Including stuff like giving everyone the same set of images of streetlights and street-side stairs in its captchas for a few weeks at one point, presumably because it was training its self-driving car AI, and who cares if it defeats the whole purpose of why a user might want to use a captcha.



Well, based on observing my behaviour and what it thought were clues from it, over the time it produced pretty much only the worst recommendations possible. E.g.,



- it kept recommending Turkish soap operas for years, until I switched my country from Germany to UK. Presumably because, yeah, there a lot more Turkish immigrants in Germany.



- it keeps recommending news -- granted, now about Brexit -- no matter how often I click the X in the corner to remove that. Presumably because, yeah, it looks at what other people are watching.



- it kept recommending wrestling videos for weeks, because ONCE I clicked on one video of Jeff Dunham doing a comedy show for the troops in an event that also featured a wrestling match. In fact, it's become an entertainment of its own to watch the recommendations go retarded if someone sends me a link to youtube.



- it can't seem to distinguish between playlists that are supposed to be watched in a sequence, like a TV series or a game Let's Play, and a random music list. Or at least I assume that's what happens when I watched episodes 0 to 9 of a Let's Play in order, and then it recommends that I watch one of episodes 19, 21, 13, 17, 8 again, or 12. Presumably because those got the most thumbs up. But conspicuously missing is the only one that actually makes sense to watch next, which is episode 10.



- it keeps recommending only the games I'm not interested in, but presumably the ones that kids these days play the most. Because, hey, they're all tagged as games. If you've watched strategy games before, surely you're interested in some twit bragging about using aim-bots to "pwn noobs" in games like Fortnite.



Etc.



Now all those could be solved MUCH easier by just letting me do the same thing as in Steam, namely let me set some options and tags/people I'm interested in, and ones which I don't. Especially the last one. Other than wanting to keep me as a lab rat in perfecting its learning algorithms, trying to guess what I'm interested instead of just letting me block certain twits, there is literally no advantage in doing the former instead of the latter.



And in fact, not only that, but its guessing game by now screws up even the search, where I can theoretically set some options. But for example, telling it to sort results by upload date is producing hilarious results that jump back and forth by months, while not showing at all more recent videos which I know exist.





And basically that's the problem with playing that kind of guessing game.
Click to expand...
Sounds like you are a faulty unit. Are you still under warranty, if so I'd see about getting yourself exchanged.
 
You must log in or register to reply here.

Back
Top Bottom