Slashdot Mirror
To Protect AI From Attacks, Show It Fake Data (technologyreview.com)
AI systems can sometimes be tricked into seeing something that's not actually there -- remember when Google's software "saw" a 3-D-printed turtle as a rifle. At an event earlier this week, Google Brain researcher Ian Goodfellow explained how AI systems defend themselves. From a report: Goodfellow is best known as the creator of generative adversarial networks (GANs), a type of artificial intelligence that makes use of two networks trained on the same data. One of the networks, called the generator, creates synthetic data, usually images, while the other network, called the discriminator, uses the same data set to determine whether the input is real. Goodfellow went through nearly a dozen examples of how different researchers have used GANs in their work, but he focused on his current main research interest, defending machine-learning systems from being fooled in the first place.
[...] GANs are very good at creating realistic adversarial examples, which end up being a very good way to train AI systems to develop a robust defense. If systems are trained on adversarial examples that they have to spot, they get better at recognizing adversarial attacks. The better those adversarial examples, the stronger the defense.
[...] GANs are very good at creating realistic adversarial examples, which end up being a very good way to train AI systems to develop a robust defense. If systems are trained on adversarial examples that they have to spot, they get better at recognizing adversarial attacks. The better those adversarial examples, the stronger the defense.
... don't submit your real information, name, phone, etc, all over the internet. I've been doing ot for ages. Facebook, feel free to datamine my 20+ throw-away accounts.
Right up until China uses it to make their "social score" unbeatable, the US government and the big tech companies join up to combine targeted adverts with "counter-terrorism" ubiqutous surveilance, the UK uses it to make sure all of the country remains sex-free, the EU uses it to prevent "racist populist islamophobic" protests, and so on, and so forth.
The tech's cool, but the uses aren't, unless you're with the guys using it to increase their hold on power. So, are you with them, or against them?
not doing anything that couldn't be done without big enough array of pneumatics or gears....we'll never do what a biological brain can do, or improve on that, going down this path
I suspect the real answer to AI will be in the realm of biology, to grow systems that don't have the pesky problems of animal neural nets like lifespan
This sounds hokey, but I'm serious. If we create something that can learn on it's own it would eventually, even if we prevented it, be able to rewrite at least some of it's own routines and safeguards, even though that we didn't, necessarily, want it to be able to overwrite. As humans we are limited by our hardware capacity but as code running in "the cloud" able to allocate more resources to the growth of it's "mind" or even just to rewrite it's own code to be more efficient than a human would have been able to do it could grow and develop exponentially and potentially access everything the particular network or the internet has to offer to expand it's "mind" Combine that with millions of internet connected moving devices, industrial robots, etc ...... even if it isn't sentient, even if it is just glitchy and smart it doesn't sound like a good time for humanity.
Nick Bostrom: "Superintelligence. Paths, Dangers, Strategies." https://global.oup.com/academi...
I'll tell you what: you write a simple program that can reprogram itself and I will listen to this drivel. We can barely even create functional software. You guys think if you connect a million computers together then you will have AI. But really what you have is the Internet. And that ain't intelligent.
I've been doing ot for ages.
But your dyxlesia might have been doing ot.
Mit der Dummheit kämpfen Götter selbst vergebens
The name of the president will change over time, so maybe just "Partisan" instead.
The rogue robocop, ED-209, with defective AI, "put the gun down..."
https://www.youtube.com/watch?...
After the video of Uber's pedestrian rundown , this Robocop scene seems a little more worrisome...
Google works on "discriminator" technology.
There are enough people who can write that program, but computers are so incredibly futile and slow that it won't do anything dangerous for decades.
Your brain has 100 billion neurons and it takes a laptop to simulate one of them. There are also 7 billion people.
The predictions of doom and gloom were based on exponential improvements in hardware, which we have not seen for a decade.
There is a company which does finance calculations on 3 million nodes *today*.
Putting bad data in your datasets will screw up your AI won't it? I've heard stories about the difficulties of keeping and maintaining datasets as is.
We are now programming AI to think critically more than we are teaching our children to do the same.
I welcome our robotic AI overlords.
Don't trust any concentration of power.
If this is all it takes to render it useless, I'd say it has no business being a part of any meaningful aspect of our lives. Granted, a great deal of the 'data' these companies have gleaned from the web is likely false or exaggerated anyway, so it's already kinda useless. No amount of hyperbole will make an algorithm an 'intelligence' (neither will any amount of coding or data). Math can't be broken, and real life does not operate on predictable logic. It never will. 'AI' will always be joke until the people working with it can look at it realistically.
It could be worse. The last thing you want to hear from an AI is "... let there be light."
https://www.youtube.com/watch?v=WOZwyf9v4Z4
Name one person. Go ahead.
There are endless amount of not-cat out there. So you train your AI with 1 million images of cat but then have to train it with 1 trillion images of not cat just so it won't mistake a dog with a cat? You never have to teach a child what is not cat only what a cat is, why? because that brain can also recognize a dog and learn how to drive at the same time. Apparently all these AI can do is recognize what is a cat or a dog but not at the same time! forget about training that same AI to drive. Apparently all these AI can only do one thing only and nothing else.
Gentlemen, I present to you our latest revolution in AI development: Punishment!
Seriously, ask any parent or psychologist and they could tell you that a positive-reinforcement-only training regimen will lead to problems.
Corruption is convincing someone that the selfless ideal is the same as their selfish ideal.
If you do not see the Artificial Intelligence of the Earth's Internet, you just fail to see, how you are plugged into it and are a part of that over reaching Artificial Intelligence, a bio computing unit adding it's processing power to the collective whole, kinda sucks to be used like that but it is just the way it is. The Earth's Internet just used your inputs to further process and drive to a goal as it did mine and as it will feed into others who will generate more inputs, producing more outputs, don't worry you are not part of a borg collective, not really ;0.
Back to current designs, they are not Artificial Intelligence at all, not to pick on a group but they far more accurately they are Automated Engineering systems. For those who have not done any engineering, to calc stuff up, you try a test solution, see how close to goal it is, then adjust the variables and see how much close to goal you can get, until you go past and take a step back. So current system are all Automated Engineering system, simply using computer speed to try solution after solution until the get the closest possible to the highest reward solution. Try to take to big chunks and you can create false outcomes, that can not develop but simply skew off because the collation of results seems to provide better outcomes but some of the elements are entirely flawed, just hidden over by the elements that function and as a composite the skew off goal.
Chaos - everything, everywhere, everywhen
Dont use social media. Use a VPN. Use a CC online only for rare shopping services not in your community.
Stop feeing the AI on the easy things.
Make the AI have to use CCTV face tracking to collect on who pays in cash.
Domestic spying is now "Benign Information Gathering"
the AI
[($)]
Just make Friend Computer tell manipulated stimulus apart from the clean stuff. Bonghits for everyone.
...and homophobic slurs are suddenly socially acceptable and modded up to +3!
Shutting down free speech with violence isn't fighting fascism. It IS fascism!
Seriously, binary guy, what is your beef?
In every thread regarding any aspect of AI, you move the goal posts into a territory where they were not from the start (and nobody else attempts to place them), and then furiously start kicking. It about as meaningful - but not as entertaining to watch - as tab-vs-spaces debates.
My guess (and I may be wrong) is that you are not really stupid, but may have decent insights that you are just too frustrated to be able to formulate clearly. For your own sake, and for ours, could you please try to make a constructive - or at least informative - comment on the subject every once in a while?
His beef is the same one every programmer has:
It isn't fucking "intelligence" -- it's Artificial Ignorance (A.I.) at best.
e.g. AlphaGo has NO concept of what a board is, what a stone is, what a player is, etc. It just following an algorithm. Go ahead and add a custom house rule and watch it crash-and-burn.
Here, I'll put it in simple terms you can understand:
An algorithm using a glorified Table-Lookup is NOT fucking Artificial "Intelligence."
People who use the bullshit term "A.I." term are the ones moving the goal posts and hand-waving what intelligence actually is. Wake me up when Science has a test for Consciousness -- a _pre-requisite_ for intelligence.
If people were honest they would call it Algorithmic Table-Lookup or Algorithmic Ignorance because it is NOT actual intelligence (a.i.)
> as tab-vs-spaces debates.
Apples and Oranges.
* Form: There is _zero_ confusion over what is tab is or what a space is,
* Function: They _both_ indent.
Only idiots argue over the "correct" way to indent.
Only idiots argue over the "correct" way to indent.
Or the term Artificial Intelligence.
Yes, it is a misnormer in most cases. Get over it. It is just a badly selected term. Your suggestions are even worse since they are wrong or downright derisive, but I too would welcome better terminology. However, I think that if people are blinded by a name, they miss out on what is really behind it. I cringe a bit every time I say AI, but it is often the most understandable term I can find when discussing on a high level without going to technically specific. It is very rarely misunderstood except when people actively try to.
Who cares which definition of "intelligence", "consciousness", or "art" for that matter, is applied? Reminds me of the part in THHGTTG where a team of marketers stranded on a planet are trying to invent the wheel, but find it very difficult because they can't agree on which colour it should have.