The AI Thread

[Hrothbern]

As AI news related to Covid-19:
AI used to virtual test 400,000 molecules on anti-Covid properties and ended up with 7,000 potential drugs that were tested in vitro, of which 40 showed activity against the virus. A molecule named Raloxifene for now the most promising.

Speed counts

Exscalate4CoV, using a unique combination of high performance computing power and AI with biological processing, brings together 18 partners and further 15 associated members. This includes supercomputing centres in Italy, Spain and Germany, large research centres, pharmaceutical companies and biological institutes from across Europe. The platform has around 120 Petaflops computing power, allowing research into the behaviors of molecules with the aim of identifying an effective treatment against coronavirus. The project's chemical library is constantly growing thanks to agreements with newly associated pharmaceutical companies.

The consortium has already virtually tested 400 000 molecules using its supercomputers. 7 000 molecules were preselected and further tested “in vitro”. Raloxifene emerged as a promising molecule: according to the project, it could be effective in blocking the replication of the virus in cells, and could thus hold up the progression of the disease. Researchers have indicated that its advantages include its high patient tolerability, safety and established toxicological profile.

Background

Using a European supercomputing platform, one of the world's most powerful, Exscalate4CoV is able to perform in weeks a screening process that with traditional techniques would take many years. In a first step it has so far identified 6 out of 25 different protein models of the novel coronavirus that are constantly evolving, with various mutations received weekly, which are translated into a digital form for use in the next step. The second step is to match the digital structure of coronavirus proteins against the available library of molecules. In the third and final step, the identified molecules undergo several additional biological screening operations in laboratories located in Belgium and Germany to understand how an identified molecule interacts with the virus model and to assess the degree to which it can stop its activity.

In the next phase, the project will enlarge the above tests by considering an extended library of 5 million molecules, out of the 500 billion molecules it disposes. It is expected that further potential molecules will be identified and contribute to an effective overall treatment of the disease.

https://ec.europa.eu/commission/presscorner/detail/en/ip_20_890

 

[Samson]

This was posted in some thread over there, and is an interesting application of AI:

Made with deepart.io, based on this algorithm. I think that looks right cool, I bet there are other ways to utilise that method, I wonder if you could make "art" for consumer markets with this?

Spoiler Example :

 

[Kyriakos]

This was posted in some thread over there, and is an interesting application of AI:

Made with deepart.io, based on this algorithm. I think that looks right cool, I bet there are other ways to utilise that method, I wonder if you could make "art" for consumer markets with this?

It is primitive. I tried one and it looked terrible. In the case of those avatars it worked because the first one is just a backdrop.

 

[red_elk]

Neural style transfer sometimes produces quite good looking results.

 

[yung.carl.jung]

This was posted in some thread over there, and is an interesting application of AI:
[/SPOILER]

there are photoshop presets which do the exact same thing, but much better. my dad has been making art like this for about 10 years now (though he's in love with deep dream). not sure this is something you really need artificial "intelligence" for, I think lots of software can do this now. same as @Kyriakos I'm not really impressed, but it is really cool that it's availabe and free!

 

[Hygro]

I prefer another website deepdreamgenerator . I’ve just been sitting on the berzerker tfish one for almost 2 years.

 

[red_elk]

Art-studio in Moscow revealed that one of their designers with pseudonym Nikolay Ironov, is not a human.
Nikolay worked for them since last year.
Its portfolio page:
https://www.artlebedev.ru/nikolay-ironov/

 

[Kyriakos]

Art-studio in Moscow revealed that one of their designers with pseudonym Nikolay Ironov, is not a human.
Nikolay worked for them since last year.
Its portfolio page:
https://www.artlebedev.ru/nikolay-ironov/

So, who pocketed the robot's wages?

 

[Commodore]

So, who pocketed the robot's wages?

Robots aren't entitled to wages. They are tools. Would you pay wages to a hammer for the nail it just pounded into a piece of wood?

Even if they were to reach a level of sapience equivalent to humans, I would still say they should be classified as tools, not as workers that are entitled to protections and fair wages.

 

[Kyriakos]

Robots aren't entitled to wages. They are tools. Would you pay wages to a hammer for the nail it just pounded into a piece of wood?

Even if they were to reach a level of sapience equivalent to humans, I would still say they should be classified as tools, not as workers that are entitled to protections and fair wages.

Sure. But people paid the robot money, which was taxed. Someone did get that.

Serfs weren't entitled to cash payment either, but their owners got state funding. There's a famous russian novel where someone is trying to make money (in a scam) out of buying ownership of dead serfs.

 

[Commodore]

Sure. But people paid the robot money, which was taxed. Someone did get that.

I would assume the robot's creator or whoever currently maintains ownership rights over the robot.

 

[red_elk]

Sure. But people paid the robot money, which was taxed. Someone did get that.

... the employer?

 

[Kyriakos]

... the employer?

A certain Pavel Ivanovich Chichikov.

 

[Commodore]

... the employer?

You mean owner. To call them the robot's employer would imply that the robot is an employee and employees are entitled to rights and wages. This is terminology that cannot become normal when referring to the tools we create to help us work.

 

[red_elk]

You humans are so elitist and dismissive

 

[Kouvb593kdnuewnd]

Even if they were to reach a level of sapience equivalent to humans, I would still say they should be classified as tools, not as workers that are entitled to protections and fair wages.

That is pretty much the definition of slavery and I suspect over time that position would be harder and harder to maintain.

 

[Truthy]

there are photoshop presets which do the exact same thing, but much better. my dad has been making art like this for about 10 years now (though he's in love with deep dream). not sure this is something you really need artificial "intelligence" for, I think lots of software can do this now. same as @Kyriakos I'm not really impressed, but it is really cool that it's availabe and free!

You will often get better results using photoshop than deep learning. That’ll likely change at some point [1], but for the time being, something to point out is that the point of deep learning is to do complicated, specialized things automatically--without explicit hand-crafted tools. Those Photoshop presets were created through the effort of gazillions of Adobe engineers and an ungodly amount of research effort before them.

A small example: convolutional neural networks automatically learn to use internal filters that are essentially Gabor filters [2]. One could ask “why do we need this so-called artificial ‘intelligence’ to do this when Photoshop already has good filter banks?” Because deep learning gives us a mechanism to automatically construct and apply filter banks without having a Nobel Prize-winning physicist come along and define them, then having a bunch of Adobe engineers implement them, and then having a human specialist carefully use them to get a particular result.

[1] To start, I think Hygro’s example is actually pretty out of date and newer algorithms do a much better job. Though I'll admit that even the state-of-the-art neural net for really anything is usually less impressive than one would expect given all the hype.

[2] For the more technical here, there is a sense in which this is very unsurprising, given the relationship between convolution and wavelets.

 

[Commodore]

That is pretty much the definition of slavery and I suspect over time that position would be harder and harder to maintain.

Slavery is indefensible because humans aren't artificial constructs built on an assembly line. Subjecting machines to similar conditions, even if they are intelligent, is okay because at the end of the day they are still built to be our tools.

I would even argue that non-sapient animal life is more deserving of rights and protections than a sapient machine.

 

[Hygro]

To start, I think Hygro’s example is actually pretty out of date and newer algorithms do a much better job.

Thank god. Do you know of any?

 

[Kyriakos]

Anyone familiar with this channel? (is it serious?)

The speaker seems to be very enthusiastic about literally everything in the videos he makes, but he never provides any detailed account, which is why I ask if people here know of the channel.

Regarding the specific program, one has to suppose that such an exploit (AI just falling leading to the enemy AI also collapsing) has to be due to glaring code error (simplest case: collapse of enemy AI triggered by the attacking move, not factoring actual proximity) and not because the AI calculated some new way to solve problems near Basel