Google scraps AI project with Pentagon after employees revolt

[jde3]

1 minute ago, Sauron said:

You're confusing "is AI a useful technology that works?" with "is Facebook a good service?".

I'm actually not talking about AI at all. There are two different topics going on here.

 

And you KNOW the algos are bad because Facebook won't reveal them even to congress. They know the algos are bad too.. but if the advisers catch on... the whole thing falls apart.

[Sauron]

3 minutes ago, jde3 said:

I'm actually not talking about AI at all. There are two different topics going on here.

 

And you KNOW the algos are bad because Facebook won't reveal them even to congress. They know the algos are bad too.. but if the advisers catch on... the whole thing falls apart.

Well sorry but I was talking about AI and you quoted my post about AI forgive me for thinking you were talking about AI there...

 

The reason facebook doesn't make their algorythms public is that they abuse the data they collect and they don't want congress to know it (or at least to know it for sure).

[jde3]

11 minutes ago, Sauron said:

Well sorry but I was talking about AI and you quoted my post about AI forgive me for thinking you were talking about AI there...

 

The reason facebook doesn't make their algorythms public is that they abuse the data they collect and they don't want congress to know it (or at least to know it for sure).

Linus is a great example here though because look at his success selling products like Tunnel Bear and PIA. Those ad's aren't through Google. I just searched google for VPN.. did I find PIA.. nope. Cyber Ghost. But I would buy PIA. The old school ads work better.

 

The algos are not giving you data you want and they are just not as technologically advanced as people think. We tend to as humans over estimate things we don't understand.. we call it magic.. but in reality it's just smoke and mirrors and a good act. That is Facebook.

 

Thats my take on it.. I'm gonna drop the thread though as I've said my piece.

[laminutederire]

30 minutes ago, Sauron said:

 That's not AI in the sense of a neural network though, it's just a normal algorythm (at least for now).

I think many of them are plain sparse SVD algorithms even today. That's a deterministic algorithm, but so are neural networks. The only difference is that we know what the result will actually be on a theoretical standpoint.

That's dumb AI, but AI still. So are k-nearest neighbours or regressions, support vector machines and random forests. If used right, they can extract a lot of information.

In a way, deep learning methods or any neural network based method are still quite dumb. They require a looot of computing power for relatively basic tasks. The only advantage over human they have is basically that they can train using a lot of examples. So they can be useful for overly complex tasks where humans cannot even read all the data, as there's too much data to read, but even then, they need a lot more data than they should if they were to be considered good.

[Fred Castellum]

The government will just find someone else now. *shrug* 

[laminutederire]

21 minutes ago, leadeater said:

Humans = global warming

Global warming = death of the planet and life on it

Recommendation = kill the humans

 

Well I don't know. That information is not present in the data set since kill all humans has never happened yet

That's how dumb those networks are: they can't even do what Greek philosophers told us not to do

 

[Cole5]

26 minutes ago, leadeater said:

Humans = global warming

Global warming = death of the planet and life on it

Recommendation = kill the humans

 

You joke but a lot fo anti AI people think this

Now Im not an infinitely learning machine sadly, but i feel like any godlike ASI would see the value in at least the majority of humans staying around, for things like maitnance and construction, and later for helping conservation efforts of earth and space craft, sure most Political figures, Poachers, and warmongers would be snuffed out, but would that be bad?

[Sauron]

1 minute ago, jde3 said:

Linus is a great example here though because look at his success selling products like Tunnel Bear and PIA. Those ad's aren't through Google. I just searched google for VPN.. did I find PIA.. nope. Cyber Ghost. But I would buy PIA. The old school ads work better.

Again... nothing to do with AI. Linus knows his own channel and can look for sponsors that are relevant for his own viewers - try doing that when you have a list of almost 100 subscriptions that may or may not be related to each other, for each user individually... youtube wants to show you ads that they get paid for, not only ads that are relevant to you, and their algorythm is an attempt to merge the two. There's nothing to be learned or improved there because they have no feedback - they can't know how many people will go and buy a coke because they saw an ad for it.

5 minutes ago, jde3 said:

The algos are not giving you data you want and they are just not as technologically advanced as people think.

What people? What do they think? Are you talking about a specific technology or just pointing a finger at "algos" because some websites don't behave exactly as you personally would want them to?

8 minutes ago, jde3 said:

We tend to as humans over estimate things we don't understand.. we call it magic.. but in reality it's just smoke and mirrors and a good act. That is Facebook.

Maybe you should make an effort to understand then... I've been trying to explain how these things work but it seems to me you're not listening...

4 minutes ago, laminutederire said:

I think many of them are plain sparse SVD algorithms even today. That's a deterministic algorithm, but so are neural networks. The only difference is that we know what the result will actually be on a theoretical standpoint.

I would classify that as a pretty major difference - "AI" is just an abstraction to help researchers think about what they're working on in a productive way, but that doesn't mean the technology isn't a field onto itself.

6 minutes ago, laminutederire said:

That's dumb AI, but AI still. So are k-nearest neighbours or regressions, support vector machines and random forests. If used right, they can extract a lot of information.

Information extraction is not the same as interpretation, and if you tell it exactly how to interpret something YOU are the one doing the interpretation, not it. IT is just helping you do it faster.

7 minutes ago, laminutederire said:

In a way, deep learning methods or any neural network based method are still quite dumb. They require a looot of computing power for relatively basic tasks. The only advantage over human they have is basically that they can train using a lot of examples. So they can be useful for overly complex tasks where humans cannot even read all the data, as there's too much data to read, but even then, they need a lot more data than they should if they were to be considered good.

Which is why it's an active field of research - make them faster, more accurate, more efficient. Even being on a remotely similar cognitive level with a human is a massive achievement compared to a "dumb" machine that does exactly what you explicitly tell it to and nothing more. If nothing else, it saves programmers a lot of headaches - suddenly, it's no longer your problem to figure out a great facial recognition algorythm; the machine will do it for you.

[leadeater]

13 minutes ago, laminutederire said:

In a way, deep learning methods or any neural network based method are still quite dumb. They require a looot of computing power for relatively basic tasks. The only advantage over human they have is basically that they can train using a lot of examples. So they can be useful for overly complex tasks where humans cannot even read all the data, as there's too much data to read, but even then, they need a lot more data than they should if they were to be considered good.

Well to put a bit of perspective on the human side the amount of years it takes for us to be 'functional' and partly self sustaining is actually rather long then we continue to learn from there. The amount of information we absorb and process to learn things is also extremely high and we still draw on that for even basic tasks, if we do it enough it becomes instinctive but ultimately when it comes to AI vs Humans and requiring too much data we both suck currently.

[jde3]

6 minutes ago, Sauron said:

Maybe you should make an effort to understand then... I've been trying to explain how these things work but it seems to me you're not listening...

Again.. (for the 3rd time) I'm not talking about AI.

 

About AI yes, skip logic.. or actually @laminutederire had it right. It's data mining.. thats a result of the "collect everything" practice.

[Sauron]

3 minutes ago, Cole5 said:

You joke but a lot fo anti AI people think this

Now Im not an infinitely learning machine sadly, but i feel like any godlike ASI would see the value in at least the majority of humans staying around, for things like maitnance and construction, and later for helping conservation efforts of earth and space craft, sure most Political figures, Poachers, and warmongers would be snuffed out, but would that be bad?

A machine couldn't care less about the environment and life on Earth unless you specifically instruct it to care. It wouldn't share human values, and getting it to is one of the hardest problems in AI research.

[leadeater]

5 minutes ago, jde3 said:

Again.. (for the 3rd time) I'm not talking about AI.

You'd be the only one not talking about AI then, the thread topic is about AI 

[jde3]

Just now, leadeater said:

You'd be the only one not talking about AI then, the thread topic is about AI 

Someone asked two pages back why I think the algos suck. I got roped in.

[laminutederire]

1 hour ago, Sauron said:

 

Except it's not that efficient. I'd even say, deep learning right now is just plain bad in most cases. What I'm saying is that every almost researcher and everyone else sees it as the pinacle of AI and as the right way to go, when it clearly just isn't. Because it's just not remotely similar in functions and efficiency of learning that the human brain is. There are orders of magnitude of difference between the two. Multiple orders of magnitude.

My point being that deep learning is not the future from a conceptual standpoint. At best it's a rock on the cathedral that will be true learning.

(And to be fair, the algorithm learns itself, but it's a headache for the programmer to make sure it converges and is accurate)

2 hours ago, leadeater said:

 

To be fair, a 2 years old need to see 2 or 3 cats to be able to recognize cats of all races in all contexts (almost). Deep learning algorithms need thousands and thousands of images to do that.

What makes us take a long time to develop our cognitive functions is that we build ourselves. Most resources are dedicated to building ourselves. We learn a bit, then we exploit that learning a loooooooooooot, to self sustain and grow. Issue is that biological growth takes a lot of time.

That's a good point though, but I'd say that part of whybut takes that much time is that we're lazy, so we spend very few time to learn overall, and we learn a loooooooooot anyway from scratch and without having much informations fed. For instance when a child learns how to speak, it can't have some sorts of labels to know if he understood correctly like machine learning algorithms have. So there is a huuuge amount of information that a child doesn't have but still learns. Imagine that it takes non stop days for a datacenter to "learn" speech tasks, with a gigantic amount of labelled data. A child learns it in a year or so of non continued learning without labels at all.

[LAwLz]

2 hours ago, jde3 said:

Someone asked two pages back why I think the algos suck. I got roped in.

I think you're referring to me.

The reason why I asked that is because you said, and I quote:

5 hours ago, jde3 said:

AI is fraud

Then a bit later in the thread you said:

3 hours ago, jde3 said:

Sure man.. Machines don't learn at least not like we do. They basically follow instructions and the "algos" we all think are so great are pretty busted to be honest. It's a trend and a marketing term, nothing more. Skip logic.

A bit later in the thread you also have a fairly smug attitude where you talk about how other people don't understand things, yet you seem to be fairly ignorant about the topics you talk about yourself. Quote:

2 hours ago, jde3 said:

The algos are not giving you data you want and they are just not as technologically advanced as people think. We tend to as humans over estimate things we don't understand.. we call it magic.. but in reality it's just smoke and mirrors and a good act. That is Facebook.

 

Your examples makes with amazon and Facebook makes no sense, and now that people have started to question you you're trying to backpedal. "Oh, I wasn't talking about machine learning", (despite clearly specifying that you do not believe computers can learn, and that it's just skip logic).

You clearly have no idea how any of the things you talk about works. Your argument that the military hasn't driven technology forward because "it didn't invent philosophy" is laughable.

 

 

Machine learning is not skip logic. Machine learning operates very differently from traditional skip logic. Machine learning is not fraud or marketing hype.

 

 

 

By the way, for anyone interested in what neural networks are and how they work, I recommend this video. Actually, I think watching at the very least this video should be mandatory before commenting on neutral nets. There are way too many people who gets on soap boxes and preach about things they don't even understand the basics of.

 

[leadeater]

13 minutes ago, laminutederire said:

A child learns it in a year or so of non continued learning without labels at all.

With their parents constantly saying "Say dad/mum" lol, I'm glad we don't learn everything that way otherwise it would get real annoying fast.

[jde3]

6 minutes ago, LAwLz said:

I think you're referring to me.

The reason why I asked that is because you said, and I quote:

Then a bit later in the thread you said:

 

A bit later in the thread you also have a fairly smug attitude where you talk about how other people don't understand things, yet you seem to be fairly ignorant about the topics you talk about yourself. Quote:

 

 

Your examples makes with amazon and Facebook makes no sense, and now that people have started to question you you're trying to backpedal. "Oh, I wasn't talking about machine learning", (despite clearly specifying that you do not believe computers can learn, and that it's just skip logic).

You clearly have no idea how any of the things you talk about works. Your argument that the military hasn't driven technology forward because "it didn't invent philosophy" is laughable.

 

 

Machine learning is not skip logic. Machine learning operates very differently from traditional skip logic. Machine learning is not fraud or marketing hype.

 

 

 

By the way, for anyone interested in what neural networks are and how they work, I recommend this video. Actually, I think watching at the very least this video should be mandatory before commenting on neutral nets. There are way too many people who gets on soap boxes and preach about things they don't even understand the basics of.

 

Hahaha. ok

 

Your way too serious for me man. Skip logic is a joke I don't need you to define it for me. thanks tho.

[laminutederire]

15 minutes ago, leadeater said:

With their parents constantly saying "Say dad/mum" lol, I'm glad we don't learn everything that way otherwise it would get real annoying fast.

Sadly foreign languages teaching sometimes is a bit like that with bad teachers

[Serin]

Huh. How about that!
Maybe Faro Automated Solutions won't be a thing.