atroyn

Thanks a lot for your insight! I'm glad the US military takes Software licensing seriously - this means spending time on this may actually be valuable.

I don't want to go in to my specific objections because they're not material here, but I appreciate that militaries don't only make war. That doesn't change my objective.

No, I would prefer that the license not allow it.

I do want to prevent the military from using it, but I am also aware that militaries frequently do illegal things, and I don't have the resources to take them to task over it.

I am a signatory to the autonomous weapons non-proliferation petition from the future-of-life institute: https://futureoflife.org/open-letter-autonomous-weapons/

I am hoping that by including a non-military-use clause, if I ever do have the resources to fight military use of my software, I will be able to do so.

Thanks!

Thanks! This might be a good starting point.

Very interesting, and in more or less the vein I'm after.

> Even supposing such systems exist

They do, I gave examples.

> Conflating simulation with real time

I am not talking about real time, only about finite time for any Turing machine. No matter how fast you make your machine, I can ask for another digit of precision. As I mentioned earlier, you are also bounded on how fast your machine gets to be much earlier than I'm bounded on digits. You'll be running my double pendulum until the heat death of the universe, and using all the mass energy available to do it.

> The brain is a computer" is a very different claim to "the brain can be simulated by a computer"

These are equivalent by the universality of Turing machines. You are in fact making this claim, even if you don't realize it.

> There's frankly little reason to believe the brain is special in any particular way, simply complex and layered.

Which is precisely the point I'm making. I deliberately avoided any woo about non-determinism, biology, or consciousness by providing examples of the time evolution of purely deterministic dynamical processes, very simple ones, which produce behaviors that are provably not computable in any reasonable time and have infinite (informational, not thermodynamic) entropy.

The complexity classes of generating additional bits of precision are much worse than exponential in many cases, and these are just ordinary everyday systems - nothing so complex as a brain. You provide examples of flaws in human cognition, which I acknowledge, but pointing them out does nothing to advance the idea that the brain is a computer.

But what characterizes those other arrangements of matter?

What is their energy consumption? What are their switching speeds? Can cognition grow unbounded, and if so, why? Can any brain improve on its own process of cognition? If other arrangements of matter perform cognition equally well, why don't we see them in the independent evolution of life with wildly different structures for everything else?

There is far too much woo and not enough rigorous thinking around the notion of AI superintelligence, singularity, or whatever other incantation people care to use.

This says nothing about the time evolution of the state space of a system, only that the entropy as it relates to _the current state of the system_ is bounded - indeed the bound is on the entropy of a finite volume of fixed energy. Even if we expand the volume to encompass the entire universe, the bound informs only the _maximal_ entropy at any given time.

As this relates to the human brain, and even omitting any quantum weirdness (of which there is probably a lot), all we can say is the total state of the brain at time t requires at most N bits, where N is the bound over volume of the brain. It says nothing about the (information theoretic) entropy of the _dynamics_ of the brain, which is where the process of cognition actually occurs. In fact many (if not most) physical processes have infinite entropy.

For an example of a deterministic system that exhibits such behavior, Consider the logistic map at r~5.7. There exist some coarse partitions [0] of the state space such that the trajectory that can be expressed in a finite number of bits, but this is not in general the case, even though any _given_ state of the system can be expressed in a finite number of bits. Other examples include the double pendulum, three-body orbits, and in fact most real physical systems.

In fact it is not in general true that a system with N bits of state can be always simulated with <= N bits. We can obey the Bekenstein bound without necessarily being able to simulate the process, even a complete description of the state at time t.

For the time evolution of the state space of a system, the Lyapunov Time [1] and the Lyapunov Exponent [2] are much more informative. If the time required to reach another bit of precision grows faster than polynomial time, no Turing machine can hope to simulate that physical system in a practical amount of time. In fact we also have physical bounds on how much memory we have to work with, and how fast we can flip bits for a given volume of space - simply enumerating all possible state transitions of the brain using a Turing machine will require a volume of space much larger than the brain itself if the time evolution of the system produces large Lyapynov exponents (which it absolutely does). This should come as no surprise - known NP-hard problems have the same behavior.

I would also add, even if the brain turns out to have a representation in N bits, and the state transition is (somehow) computable in polynomial time, it may still be physically infeasible to do so without just building and 'running' an actual brain.

[0]http://tcode.tcs.auckland.ac.nz/~corpus/logistic.html [1]https://en.wikipedia.org/wiki/Lyapunov_exponent [2]https://en.wikipedia.org/wiki/Lyapunov_time

PS: To me all of the above suggests that whatever the universe is doing when it's doing physics, it certainly isn't computing in the Church-Turing sense.

I don't agree with your assertion that humans can't do that. Whether or not human cognition is a superset of computation is an unanswered question.

That aside, even if human cognition is a computable function, there are no guarantees that the physical process giving rise to human cognition is computable, nor that any process giving rise to cognition is computable.

I came to my conclusion independently by observing that computing the time evolution of most physical systems to arbitrary precision is impossible in finite time. More formally, the state space grows much much faster than polynomial time. Finding out if we can do better with quantum computing is an active area of research.

I haven't read Searle/Penrose

This is the wrong way around. It's possible for a physical system to perform computation. It's not necessarily true that all physical systems can be emulated by computation.

Dealing with poorly behaved surfaces (specular, luminous, reflective, deformity) in 3D reconstruction.

Figuring out human interaction intent from force measurements for cooperating with a robot in a collaborative task (moving a table together).

Both language agnostic.

This is a whole other set of problems, which I agree will only get worse. However they don't really represent the threat Musk discusses - superhuman / superintelligent paperclip maximizers and other Yudkowskian nonsense.

It's unclear that it's even possible to emulate general intelligence by computable functions, let alone that it's possible to improve it to superhuman capacity.

There are clear and present threats to civilization needing to be dealt with - superhuman A.I is, to quote Maciej/Pinboard, the 'Idea that eats smart people'.

What's your secret graffiti name?

I'd love for people to answer this honestly, and in that vein. I just can't ever picture myself playing that many video games.

Maybe the US cars. Our cars in Germany go fast enough that Aero certainly matters.

My team from TU Munich's self driving racecar (TUfast Driverless) are some of the best engineers I've worked with, including during my time in industry.

This is no surprise to me, and the sooner tech companies start tapping this well off talent, the better of they'll be. For now only the auto makers have caught on, but it's a matter of time.

Who would you give to, out of interest?

There's a point at which you have to wonder how wealthy the U.S actually is, if it can't even feed school children at no cost to their parents.

At some point it starts to resemble DPRK propaganda.

That's different to what the freelance visa is designed for.

Can't edit so I'd also like to add : Germany has a freelancer visa. If you can prove you have clients, contracts in the pipeline, and some savings, it's a matter of bringing some documents and filling out a form at the Ausländerbehörde.

It won't matter. You need the bachelor's degree in your field to get a work visa, but it need not be from Europe.

Sorry yes, 18 is correct.

Answering for Germany, Berlin/Munich in particular.

1. You'll need a bachelor's degree in your field, but you won't need a masters. That said, getting a masters may give you a leg up. Postgrad education in Germany is free even for international students, the standar is high, and it gives you the opportunity to get your German up to an acceptable level. Additionally, you will have 12 months following graduation to find a full-time job. If you're looking to work for a bigger Germany company like Siemens, BMW etc., I'd seriously consider the masters.

2. Research job openings before you go. Tailor your CV. Two places to look are AngelList, and also http://berlinstartupjobs.com/ Applying online has always worked for me, but you may also want to go to some meetups while you're there. Berlin Tech Meetup https://www.meetup.com/b-tech/ is one of the largest.

3. Develop your portfolio. I had plenty of Indian colleagues at several of the companies I worked at.

Try RemoteOK as well.

The difference between cold emailing / calling and spam is that cold calls should be individually targeted and well thought out messages, while spam is an identical mass communication to as many people as possible.

What is a good resource for getting started with this kind of development? I've done some write-an-OS tutorials but nothing with bare-metal or RTOS yet, and I'd like to.

It took a few tries, and I started off easy with the bed making, then building on that. It's not so hard to make yourself make your bed or drink some water. Then when I'd gotten a couple of those down as a regular thing, it was just a bit of reflection - 'well I didn't really feel like making the bed or walking to the kitchen all those days, but now that I do it it doesn't really matter if I want to. I guess eventually push-ups will work the same'

Turns out pretty much everything works the same. Maybe the trick is that I didn't really do anything that wasn't incremental, just a day at a time, but you just kind of keep going with the momentum.

Now I tend to just do that reflexively because it's just worked so many times in so many different areas.