Comodo Security A.I.


I made a few suggestion so far but didn’t get many responses. I hope someone from the Stuff is reading the suggestion. To honest I would wish that Mr. Abdulhayoğlu would read my suggestions and give me his opinion. I have the outmost respect for Mr. Abdulhayoğlu and what he has done and I absolutely like his way of his thinking.

So the AI, I know it’s farfetched but bear with me for a moment. I am just an IT guy (for over 20 years) not a developer.
Today we have AI, our best AI is at the “stage” of a 4 year old child, it’s good with numbers but that is logical it is a computer. In my opinion we are not going to surpass this stage for two reasons.

  1. Let’s say we are Intelligent enough to make an AI, the AI can’t work on a normal PC and his architecture, one processor, ram, harddisk, that is too little, AI can’t work on this environment, our brain is also not working that way, it is just too much.
  2. We don’t have the right programing language. The most powerful programing language we have is (arguably) C++, you can forget Java, it is to slow (in my opinion it has a bad compiler, Google made it better but still no cigar), most of the other programing languages are “purpose oriented”. Today, 2015, we have surpassed the doable with C++, it just can’t do more, even MS is mixing C++ with C# and other languages for Windows because it’s just not enough. Whatever you do with C++ has to follow one structure because that is everything the programing language is giving you. We need a completely new Programing language for the 21st century that is not based on anything existing.

In my opinion, we have rudiments to solve problem one, it is called Memcomputer (Meet The Memcomputer: The Brain-Like Alternative to Quantum Computing), it is just a Prototype but is very promising. We also have “QC” for years (hell even Google just bought one, it costs “only” 10 Million Dollars) but we still don’t know how to use it. We are trying to make a PC out of it with the same Architecture and it just doesn’t work. We don’t have a real OS, we don’t have real software, I would say because we don’t have the right programming language.
In my opinion there are only 0,1% of all mankind that could solve problem two but most of them are not interested, so that is going to take a while.

So why would I want an AI from Comodo if in my opinion it can’t work. Well there is another approach to AI, not to try to make it able to do everything, just make it able to do one task. There is a working example for that AI learns how to build Super Mario levels by watching YouTube - CNET
They made an AI for one task, learn from YouTube videos and then build it. It is working very well. I just hope they know what they are doing and don’t get greedy because building an AI that is able to program and reproduce itself could be dangerous, maybe they should run it in the Comodo Sandbox .

But this AI is the angle of my Idea. Build an AI for one task, security. You can leave the “white list” environment for enhancement (which in my opinion has one problem). The AI should not use white or black list, it also should not be behavior based. Do something completely different. Let the AI analyze the traffic that is going to the PC. Every virus, malware, spyware, Adware (whatever you call it) has similar patterns. It doesn’t matter in which language it is realized in, it still does almost the same from its behavior so there are similar patterns in the code. I know most of you would say, that is not possible because the AI had to decompile or reverse engineer the program to see its code. No, the traffic is just a bunch of ones and zeros but still since the behavior and the code are similar there are also similar patterns in the ones and zeros. The AI should analyze the traffic for that patterns and kill it before they even come to the PC. For a example, if you want the Password from a WAP secured WiFi you just need to analyze the traffic for a few minutes and you have the password, it’s also just ones and zeros. You are an AV developer so you have a huge Database of viruses and so on, let the AI learn from that database, let it learn the patterns that it should search for on its own.
As a second step you could let it also learn Firewall behavior from the user. A Firewall should be set to “custom ruleset” otherwise it’s useless (my opinion), but the AI could learn from the user, what to block and what not to block (Of course it would be useful if one Ai could learn from another, for an example a AI that learned from an experienced user could help an unexperienced user, but letting them connect and learn from each other could be a big problem, so please don’t do that).
As third step, you could let it learn what phishing is and how to stop it and also recognize Hacker Attacks and stop them.
I think for all of that the normal PC developing pattern would not be enough, one executable and a few dll’s can’t do the job. I think you need a “framework”. The purpose of the executable should be to start the framework and to know when to use what.
With that being said, a few concerns:
Don’t make it cloud based, that would be too dangerous in my opinion, it would be contacted from millions of computers, it would learn and get to big to fast because in the cloud it has enough power to do that. The more it learns the greater is the possibility for (let’s say) “mutation”, “ghosts in the machine”. There is no software without errors, why, because it’s man made, even a spelling error with an AI could have unforeseen effects. Let it run on every PC so it’s limited to the capabilities of one PC.
In fact you should restrain its Internet access only to one of your servers with a newly devolved encrypted protocol which is only used for the AI. On the Server it also should be restricted to the Server, not being able to learn on its own from the internet.
The AI install files and also after the install the executable and the framework should be encrypted and only decrypted “on the fly” when running, in the memory there should nothing that can show the functionality of the AI. With that no one could temper with it when running or reverse engineer it. In the wrong hands the AI could be very dangerous (but what couldn’t).
You should never allow the AI to reproduce itself, it could build a better version of itself and you could lose control.
You should built in a local and a remote kill switch.
(Joke on)
Last but not least you should built in the one line that so many websites have in they “robots.txt”.
“Don’t kill John Connor”
(Joke off, well joke for the Ai but in fact many websites have that line in the “robots.txt)
Thanks for reading

P.S. If you succeed building this, you won’t just be number one in Certificates, you will also be number one in AV and Firewall.

P.P.S. It could happen that I edit the text a few times as new Ideas come to me, I’ll make sure to mark the edited part so you won’t have to read it all over again….

we already use some sort of AI in our new Valkyrie that we developed for businesses…
AI will become integrated more into everything we do in future…we are constantly evaluating its use.

When you think that be ready

There was a Kaggle competition about this (Microsoft Malware Classification Challenge):

Still, it would give a lot of false positives (a human reviewer should check the output).


yes at the begging a human would have to check the output until the AI learns how to do it, just like with Facebook’s “M”