Lawrence Krauss interviewed by Steve Omohundro and Adam A. Ford after the Singularity Summit Australia 2011 summit2011.singinst.org.au It looks impossible to avoid [singularity] – if you can cool things down quite a lot. It is amusing the idea that the universe gives a limit in to intelligence in our human timescale. Mathmatical complexity is not as much of a constraint to thinking about physics as it used to be (thanks to things like Mathematica). Will AIs in the future have different understandings of quantum processes? Eugene Wigner – “The unreasonable effectiveness of mathematics in describing the physical world”. We had no reason to expect that mathematics would The playing field of physics has been determined by the symmetries of nature and the symmetries of mathematics – they restrict what is possible and what isn’t. It is hard to imagine that that aspect of why mathematics will change after we get machine intelligence. How we frame it and how we explore and explain it in a sense may change dramatically I guess, as may also the questions of interest. So it is hard to know.. so if you ask “if you thought about the world differently what would you know?”, “well if I thought about the world differently I would know.”. So it is sort of possible to imagine that the kind of questions that are of interest would change – once you know the symmetries of nature, the physics is constrained and its hard for me to imagine that this will change even with [intelligent] machines …
Video Rating: 4 / 5
Brilliant thinker.
He uses the same T-shirt inside the one 2+2=5 Lol , there’s a video where he show his t shirt to the audience ,….
Dr. Krauss is a Future Einstein.
Lawrence Krauss “You can make the decision to have your 20th kid like the Duggars… or you can develop a star-ship to travel intergalactic space.
……………………….
Having the chips designed via an evolutionary algorithm , while interesting, would complicate the job. The reason for that is it is quite possible to model chips in software, so a regular computer would do just as well with the software itself being evolutionary.
One other thing to point out is that it’s still early days, and our computers are nowhere near powerful enough to pull this off ( yet ).
I’ll believe in it when I see it. Perhaps an evolutionary algorithm could create the chips and software.
Yes, and destructive too, but only to those who can’t handle the truth.
You are missing the point slightly, while it’s no doubt true you cannot reduce thought to simple binary on off, it is certainly possible to model neurones in software. This is exactly the same principle as modelling any other complex thing on a computer ( like an aircraft ), where the “action” is at is in the software, not the hardware it’s running on. In the same way our thoughts are not dependent on the atoms that make up the brain, it’s the processes that run on the structures made of atoms.
This should have 10000s hits more.
Not really. A particular nerve may have dozens of synapses, whereas a transistor can only form 1 connection.. Also, different areas of the brain can perform more than 1 function through chemistry. But honestly, my knowledge of the brain limits how far I can take this argument. But there is one thing I do know for sure, human intelligence can not be reduced to the p-n junciton (which is what all solid state electronics break down to).
Certainly there are aspects of the human brain that are binary, though, no? Action potentials are binary. The distribution of transmitters and receptors confers a probabilistic quality to junctions, circuits, and local networks that makes the system as a whole seem not binary, but at its root it is binary. I think one could model this on an extremely powerful computer, in theory, if we knew enough about how these networks interact.
In a way, yes. Human brains aren’t binary, they are electrochemical and are probably the most complex system in the universe (that we know of). Things like imagination, which can not be accomplished by a computer at this time, will never work on a transistor based computer.
Intelligence is arbitrarily defined based on the useful context, isn’t it? It seems that you’re talking about human-like intelligence, which is really different computational modules exchanging information, compressing it, and ultimately sending it to a decision-making apparatus that is the basis of our “consciousness” – the generalized intelligence that we consider to be human, no?
You’re the only person I have heard use that expression in many years; I say it all the time!
As I am a big fan of Krauss, I will definitely have to give this a watch.
Of course of course. I don’t want to diminish the work people have done to computer’s brute force power or the research that has been done, I just wanted to explain that there is a pretty fundamental difference in intellectual capability between a machine and higher animals. I’m confident this will be solved but I won’t put a date on it; could be 50, 150 or just 10 years out.
You’ve got to wait for computational neuroscience to catch up. Once there is a workable complete model of a simpler mammalian brain, then the CS researchers can utilize the information gleaned from this. I think biomimicry is the way to go here considering how much remains unknown about consciousness.
what a shitty camera work for such a good interview :/
I know this is a few weeks old, but… There’s the concept of an OODA loop. The second O is to Orient; to build a mental model which permits predictions. Software to do this is not yet invented; AIs use pre-existing mental models developed by their creators. You can define “intelligence” as a word that doesn’t include this, sure, but entities that can re-orient *will* beat entities that cannot at a variety of basic tasks. This is why many impressive AI demos are human assisted or worse.
Until it crashes into something or it’s prey runs up a tree. We should (and do) use computers to analyze complex mathematics.
At this point, I think we simply disagree on what intelligence is. Intelligence is very difficult to define in the first place. All the analogies are bad, like cheetahs because you are talking a helicopter which is the real advantage over the prey. We can program computers to do tasks very well, on that we can agree.
We already have AI, we just don’t have conscious sentience. Don’t confuse intelligence with consciousness and sentience.
But your claim was specifically that you could not have a cheetah robot, when in fact you could have one. Put a knife on the end of a car bumper and program the car drive around killing animals and other cars. You see there is no need for this sort of nonsense – animals are stupid, we should be using computers for intelligent things, which animals are not.