I think what Dijkstra means is: once we find a satisfactory solution to the problem of AI (or to put it simply - making a computer "think"), the way it will do this will be drastically different from the way we humans or other living things do it. At that point it will not matter to discuss whether computer "thinks" or not...
He uses submarine-fish-swimming but one can also use plane-bird-flying as analogy (car analogy left as exercise to the audience).
Disclaimer: Of course, this is my personal guess. He might have some other thing in his mind, too.
Exactly! Let the market decide. This is hardly a monopoly situation and there is no shortage of alternative hardware/software/middleware manufacturers and platforms in the (smart)phone market.
My personal disappointment is: I was planning to buy bunch of these for our research testbed to prototype stuff. After learning background processes and any interpreted shells (I need Python for various reasons) are not allowed, it is out of question now! Forcing developers to buy a Mac and insisting on ObjC as only programming language did not help either (Sorry, it is a resource allocation decision. In my case, there is no point in learning another language or getting a Mac just for this, no matter how shiny they are). Summary: Nokia gets my business with S60 and PyS60...
I am sorry to hear that corporate types have hijacked the term "risk management". I am as far from being a corporate type as one can imagine. All I am trying to say is (let me use game theoretic terms in order not to be confused with a corporate type again): knowing the action space of an attacker is only partially important. It is equivalently or more important to calculate with what probability attackers will choose specific actions (which in turn depends on the possible outcome of that branch of the game). Assuming some rationality on behalf of the attacker, a defender can easily decrease those probabilities by taking appropriate actions. Maybe the game even ends up at Nash equilibrium in the worst case. Then, the defender does not have to be paranoid about anything.
In practice this whole story might even degrade from glorious security warfare to boring accounting of how much the insurance company will pay back. I was just talking to a shop owner this afternoon who was telling about the last two thefts in his store only as incidence for evaluating his insurance company's quality.
By the way, these security games become quite interesting when the assumption of knowledge is removed and some learning dimension is added to the picture (e.g. repeated games, fictitious play, Q-learning etc.)
I have written a long long reply to his article at
my blog (no ads, etc.)
Short summary:
In my opinion, security in real life is not about "what can go wrong". It is about "how often and how much can it go wrong and am I prepared to handle those cases". In short it is more about how to calculate risks accurately and knowing when to take them.
Slashdot Mirror
I think what Dijkstra means is: once we find a satisfactory solution to the problem of AI (or to put it simply - making a computer "think"), the way it will do this will be drastically different from the way we humans or other living things do it. At that point it will not matter to discuss whether computer "thinks" or not...
He uses submarine-fish-swimming but one can also use plane-bird-flying as analogy (car analogy left as exercise to the audience).
Disclaimer: Of course, this is my personal guess. He might have some other thing in his mind, too.
Exactly! Let the market decide. This is hardly a monopoly situation and there is no shortage of alternative hardware/software/middleware manufacturers and platforms in the (smart)phone market.
My personal disappointment is: I was planning to buy bunch of these for our research testbed to prototype stuff. After learning background processes and any interpreted shells (I need Python for various reasons) are not allowed, it is out of question now! Forcing developers to buy a Mac and insisting on ObjC as only programming language did not help either (Sorry, it is a resource allocation decision. In my case, there is no point in learning another language or getting a Mac just for this, no matter how shiny they are). Summary: Nokia gets my business with S60 and PyS60...
I am sorry to hear that corporate types have hijacked the term "risk management". I am as far from being a corporate type as one can imagine. All I am trying to say is (let me use game theoretic terms in order not to be confused with a corporate type again): knowing the action space of an attacker is only partially important. It is equivalently or more important to calculate with what probability attackers will choose specific actions (which in turn depends on the possible outcome of that branch of the game). Assuming some rationality on behalf of the attacker, a defender can easily decrease those probabilities by taking appropriate actions. Maybe the game even ends up at Nash equilibrium in the worst case. Then, the defender does not have to be paranoid about anything. In practice this whole story might even degrade from glorious security warfare to boring accounting of how much the insurance company will pay back. I was just talking to a shop owner this afternoon who was telling about the last two thefts in his store only as incidence for evaluating his insurance company's quality.
By the way, these security games become quite interesting when the assumption of knowledge is removed and some learning dimension is added to the picture (e.g. repeated games, fictitious play, Q-learning etc.)
I have written a long long reply to his article at my blog (no ads, etc.)
Short summary:
In my opinion, security in real life is not about "what can go wrong". It is about "how often and how much can it go wrong and am I prepared to handle those cases". In short it is more about how to calculate risks accurately and knowing when to take them.