Optical computing

A few nights ago I was thinking about the optical fibre memories that we were designing in the late 1980s in BT. The idea was simple. You transmit data into an optical fibre, and if the data rate is high you can squeeze lots of data into a manageable length. Back then the speed of light in fibre was about 5 microseconds per km of fibre, so 1000km of fibre, at a data rate of 2Gb/s would hold 10Mbits of data, per wavelength, so if you can multiplex 2 million wavelengths, you’d store 20Tbits of data. You could maintain the data by using a repeater to repeat the data as it reaches one end into the other, or modify it at that point simply by changing what you re-transmit. That was all theory then, because the latest ‘hero’ experiments were only just starting to demonstrate the feasibility of such long lengths, such high density WDM and such data rates.

Nowadays, that’s ancient history of course, but we also have many new types of fibre, such as hollow fibre with various shaped pores and various dopings to allow a range of effects. And that’s where using it for computing comes in.

If optical fibre is designed for this purpose, with optimal variable refractive index designed to facilitate and maximise non-linear effects, then the photons in one data stream on one wavelength could have enough effects of photons in another stream to be used for computational interaction. Computers don’t have to be digital of course, so the effects don’t have to be huge. Analog computing has many uses, and analog interactions could certainly work, while digital ones might work, and hybrid digital/analog computing may also be feasible. Then it gets fun!

Some of the data streams could be programs. Around that time, I was designing protocols with smart packets that contained executable code, as well as other packets that could hold analog or digital data or any mix. We later called the smart packets ANTs – autonomous network telephers, a contrived term if ever there was one, but we wanted to call them ants badly. They would scurry around the network doing a wide range of jobs, using a range of biomimetic and basic physics techniques to work like ant colonies and achieve complex tasks using simple means.

If some of these smart packets or ANTs are running along a fibre, changing the properties as they go to interact with other data transmitting alongside, then ANTs can interact with one another and with any stored data. ANTs could also move forwards or backwards along the fibre by using ‘sidings’ or physical shortcuts, since they can route themselves or each other. Data produced or changed by the interactions could be digital or analog and still work fine, carried on the smart packet structure.

(If you’re interested my protocol was called UNICORN, Universal Carrier for an Optical Residential Network, and used the same architectural principles as my previous Addressed Time Slice invention, compressing analog data by a few percent to fit into a packet, with a digital address and header, or allowing any digital data rate or structure in a payload while keeping the same header specs for easy routing. That system was invented (in 1988) for the late 1990s when basic domestic broadband rate should have been 625Mbit/s or more, but we expected to be at 2Gbit/s or even 20Gbit/s soon after that in the early 2000s, and the benefit as that we wouldn’t have to change the network switching because the header overheads would still only be a few percent of total time. None of that happened because of government interference in the telecoms industry regulation that strongly disincentivised its development, and even today, 625Mbit/s ‘basic rate’ access is still a dream, let alone 20Gbit/s.)

Such a system would be feasible. Shortcuts and sidings are easy to arrange. The protocols would work fine. Non-linear effects are already well known and diverse. If it were only used for digital computing, it would have little advantage over conventional computers. With data stored on long fibre lengths, external interactions would be limited, with long latency. However, it does present a range of potentials for use with external sensors directly interacting with data streams and ANTs to accomplish some tasks associated with modern AI. It ought to be possible to use these techniques to build the adaptive analog neural networks that we’ve known are the best hope of achieving strong AI since Hans Moravek’s insight, coincidentally also around that time. The non-linear effects even enable ideal mechanisms for implementing emotions, biasing the computation in particular directions via intensity of certain wavelengths of light in much the same way as chemical hormones and neurotransmitters interact with our own neurons. Implementing up to 2 million different emotions at once is feasible.

So there’s a whole mineful of architectures, tools and techniques waiting to be explored and mined by smart young minds in the IT industry, using custom non-linear optical fibres for optical AI.

Some trees just don’t get barked up

Now and then, someone asks me for an old document and as I search for it, I stumble across others I’d forgotten about. I’ve been rather frustrated that AI progress hasn’t kept up with its development rate in the 90s, so this was fun to rediscover, highlighting some future computing directions that offered serious but uncertain potential exactly 20 years ago, well 20 years ago 3 weeks ago. Here is the text, and the Schrodinger’s Computer was only ever intended to be silly (since renamed the Yonck Processor):

Herrings, a large subset of which are probably red

Computers in the future will use a wide range of techniques, not just conventional microprocessors. Problems should be decomposed and the various components streamed to the appropriate processing engines. One of the important requirements is therefore some means of identifying automatically which parts of a problem could best be tackled by which techniques, though sometimes it might be best to use several in parallel with some interaction between them.

 Analogs

We have a wider variety of components available to be used in analog computing today than we had when it effectively died out in the 80s. With much higher quality analog and mixed components, and additionally micro-sensors, MEMs, simple neural network components, and some imminent molecular capability, how can we rekindle the successes of the analog domain. Nature handles the infinite body problem with ease! Things just happen according to the laws of physics. How can we harness them too? Can we build environments with synthetic physics to achieve more effects? The whole field of non-algorithmic computation seems ripe for exploitation.

 Neural networks

• Could we make neural microprocessor suspensions, using spherical chips suspended in gel in a reflective capsule and optical broadcasting. Couple this with growing wires across the electric field. This would give us both electrical and optical interconnection that could be ideal for neural networks with high connectivity. Could link this to gene chip technology to have chemical detection and synthesis on the chips too, so that we could have close high speed replicas of organic neural networks.
• If we can have quantum entanglement between particles, might this affect the way in which neurons in the brain work? Do we have neural entanglement and has this anything to do with how our brain works. Could we create neural entanglement or even virtual entanglement and would it have any use?
• Could we make molecular neurons (or similar) using ordinary chemistry? And then form them into networks. Might need nanomachines and bottom-up assembly.
• Could we use neurons as the first stage filters to narrow down the field to make problems tractable for other techniques
• Optical neurons
• Magnetic neurons

Electromechanical, MEMS etc

• Micromirror arrays as part of optical computers, perhaps either as data entry, or as part of the algorithm
• Carbon fullerene balls and tubes as MEM components
• External fullerene ‘décor’ as a form of information, cf antibodies in immune system
• Sensor suspensions and gels as analog computers for direct simulation

Interconnects

• Carbon fullerene tubes as on chip wires
• Could they act as electron pipes for ultra-high speed interconnect
• Optical or radio beacons on chip

Software

• Transforms – create a transform of every logic component, spreading the functionality across a wide domain, and construct programs using them instead. Small perturbation is no longer fatal but just reduces accuracy
• Filters – nature works often using simple physical effects where humans design complex software. We need to look at hard problems to see how we might make simple filters to narrow the field before computing final details and stages conventionally.
• Interference – is there some form of representation that allows us to compute operations by means of allowing the input data to interact directly, i.e. interference, instead of using tedious linear computation. Obviously only suited to a subset of problems.

And finally, the frivolous

• Schrodinger’s computer – design of computer and software, if any, not determined until box is opened. The one constant is that it destroys itself if it doesn’t finding the solution. All possible computers and all possible programs exist and if there is a solution, the computer will pop out alive and well with the answer. Set it the problem of answering all possible questions too, working out which ones have the most valuable answers and using up all the available storage to write the best answers.

Your phone is wasted on you

Between 1983 and 1985, the fastest computer on Earth was the Cray X-MP. Its two 105MHz processors and 16MB of memory provided a peak performance of 400MFLOPS. It cost around $15M + disks.

The Apple iPhone XS is 1500 times faster and 15000 times cheaper.

In 1985, our division of 50 people ran all of its word processing on a VAX 11- 780, that produced 0.5MIPS (32bit). On equivalent instructions per second basis, the iPhone XS is 2.5M times faster, so ought to be able to run word processing for a country of 125M people.

Think about that next time you’re typing a text.

 

Future AI: Turing multiplexing, air gels, hyper-neural nets

Just in time to make 2018 a bit less unproductive, I managed to wake in the middle of the night with another few inventions. I’m finishing the year on only a third as many as 2016 and 2017, but better than some years. And I quite like these new ones.

Gel computing is a very old idea of mine, and I’m surprised no company has started doing it yet. Air gel is different. My original used a suspension of processing particles in gel, and the idea was that the gel would hold the particles in fixed locations with good free line of sight to neighbor devices for inter-device optical comms, while acting also as a coolant.

Air gel uses the same idea of suspending particles, but does so by using ultrasound, standing waves holding the particles aloft. They would form a semi-gel I suppose, much softer. The intention is that they will be more easily movable than in a gel, and maybe rotate. I imagine using rotating magnetic fields to rotate them, and use that mechanism to implement different configurations of inter-device nets. That would be the first pillar of running multiple neural nets in the same space at the same time, using spin-based TDM (time division multiplexing), or synchronized space multiplexing if you prefer. If a device uses on board processing that is fast compared to the signal transmission time to other devices (the speed of light may be fast but can still be severely limiting for processing and comms), then having the ability to deal with processing associated with several other networks while awaiting a response allows a processing network to be multiplied up several times. A neural net could become a hyper-neural net.

Given that this is intended for mid-century AI, I’m also making the assumption that true TDM can also be used on each net, my second pillar. Signals would carry a stream of slots holding bits for each processing instance. Since this allows a Turing machine to implement many different processes in parallel, I decided to call it Turing multiplexing. Again, it helps alleviate the potential gulf between processing and communication times. Combining Turing and spin multiplexing would allow a single neural net to be multiplied up potentially thousands or millions of times – hyper-neurons seems as good a term as any.

The third pillar of this system is that the processing particles (each could contain a large number of neurons or other IT objects) could be energized and clocked using very high speed alternating EM fields – radio, microwaves, light, even x-rays. I don’t have any suggestions for processing mechanisms that might operate at such frequencies, though Pauli switches might work at lower speeds, using Pauli exclusion principle to link electron spin states to make switches. I believe early versions of spin cubits use a similar principle. I’m agnostic whether conventional Turing machine or quantum processing would be used, or any combination. In any case, it isn’t my problem, I suspect that future AIs will figure out the physics and invent the appropriate IT.

Processing devices operating at high speed could use a lot of energy and generate a lot of heat, and encouraging the system to lase by design would be a good way to cool it as well as powering it.

A processor using such mechanisms need not be bulky. I always assumed a yogurt pot size for my gel computer before and an air gel processor could be the same, about 100ml. That is enough to suspend a trillion particles with good line of sight for optical interconnections, and each connection could utilise up to millions of alternative wavelengths. Each wavelength could support many TDM channels and spinning the particles multiplies that up again. A UV laser clock/power source driving processors at 10^16Hz would certainly need to use high density multiplexing to make use of such a volume, with transmission distances up to 10cm (but most sub-mm) otherwise being a strongly limiting performance factor, but 10 million-fold WDM/TDM is attainable.

A trillion of these hyper-neurons using that multiplexing would act very effectively as 10 million trillion neurons, each operating at 10^16Hz processing speed. That’s quite a lot of zeros, 35 of them, and yet each hyperneuron could have connections to thousands of others in each of many physical configurations. It would be an obvious platform for supporting a large population of electronically immortal people and AIs who each want a billion replicas, and if it only occupies 100ml of space, the environmental footprint isn’t an issue.

It’s hard to know how to talk to a computer that operates like a brain, but is 10^22 times faster, but I’d suggest ‘Yes Boss’.

 

Ultra-simple computing part 3

Just in time v Just in case

Although the problem isn’t as bad now as it has been, a lot of software runs on your computers just in case it might be needed. Often it isn’t, and sometimes the PC is shut down or rebooted without it ever having been used. This wastes our time, wastes a little energy, and potentially adds functionality or weaknesses that can be exploited by hackers.

If it only loaded the essential pieces of software, risks would be minimised and initial delays reduced. There would be a slightly bigger delay once the code is needed because it would have to load then but since a lot of code is rarely used, the overall result would still be a big win. This would improve security and reliability. If all I am doing today is typing and checking occasional emails, a lot of the software currently loaded in my PC memory is not needed. I don’t even need a firewall running all the time if network access is disabled in between my email checks. If networking and firewall is started when I want to check email or start browsing, and then all network access is disabled after I have checked, then security would be a bit better. I also don’t need all the fancy facilities in Office when all I am doing is typing. I definitely don’t want any part of Office to use any kind of networking in either direction for any reason (I use Thunderbird, not Outlook for email). So don’t load the code yet; I don’t want it running; it only adds risks, not benefits. If I want to do something fancy in a few weeks time, load the code then. If I want to look up a word in a dictionary or check a hyperlink, I could launch a browser and copy and paste it. Why do anything until asked? Forget doing stuff just in case it might occasionally generate a tiny time saving. Just in time is far safer and better than just in case.

So, an ultra-simple computer should only load what is needed, when it is needed. It would only open communications when needed, and then only to the specific destination required. That frees up processors and memory, reduces risks and improves speed.

Software distribution

Storing software on hard disks or in memory lets the files be changed, possibly by a virus. Suppose instead that software were to be distributed on ROM chips. They can be very cheap, so why not? No apps, no downloads. All the software on your machine would be in read only memory, essentially part of the hardware. This would change a few things in computer design. First, you’d have a board with lots of nice slots in it, into which you plug the memory chips you’ve bought with the programs you want on them. (I’ll get to tablets and phones later, obviously a slightly different approach is needed for portable devices). Manufacturers would have a huge interest in checking their  code first, because they can’t put fixes out later except on replacement chips. Updating the software to a new version would simply mean inserting a new chip. Secondly, since the chips are read only, the software on them cannot be corrupted. There is no mechanism by which a virus or other malware could get onto the chip.

Apps could be distributed in collections – lifestyle or business collections. You could buy subscriptions to app agencies that issued regular chips with their baskets of apps on them. Or you could access apps online via the cloud. Your machine would stay clean.

It could go further. As well as memory chips, modules could include processing, controller or sensory capabilities. Main processing may still be in the main part of the computer but specialist capabilities could be added in this way.

So, what about tablets and phones? Obviously you can’t plug lots of extra chips into slots in those because it would be too cumbersome to make them with lots of slots to do so. One approach would be to use your PC or laptop to store and keep up to date a single storage chip that goes into your tablet or phone. It could use a re-programmable ROM that can’t be tampered with by your tablet. All your apps would live on it, but it would be made clean and fresh every day. Tablets could have a simple slot to insert that single chip, just as a few already do for extra memory.

Multi-layered security

If your computer is based on algorithms encoded on read only memory chips or better still, directly as hardware circuits, then it could boot from cold very fast, and would be clean of any malware. To be useful, it would need a decent amount of working memory too, and of course that could provide a short term residence for malware, but a restart would clean it all away. That provides a computer that can easily be reset to a clean state and work properly again right away.

Another layer of defense is to disallow programs access to things they don’t need. You don’t open every door and window in your home every time you want to go in or out. Why open every possible entrance that your office automation package might ever want to use just because you want to type an article? Why open the ability to remotely install or run programs on your computer without your knowledge and consent just because you want to read a news article or look at a cute kitten video? Yet we have accepted such appallingly bad practice from the web browser developers because we have had no choice. It seems that the developers’ desires to provide open windows to anyone that wants to use them outweighs the users’ desires for basic security common sense. So the next layer of defense is really pretty obvious. We want a browser that doesn’t open doors and windows until we explicitly tell it to, and even then it checks everything that tries to get through.

It may still be that you occasionally want to run software from a website, maybe to play a game. Another layer of defense that could help then is to restrict remote executables to a limited range of commands with limited scope. It is also easy additionally to arrange a sandbox where code can run but can’t influence anything outside the sandbox. For example, there is no reason a game would need to inspect files on your computer apart from stored games or game-related files. Creating a sandbox that can run a large range of agreed functions to enable games or other remote applications but is sealed from anything else on the computer would enable remote benign executables without compromising security. Even if they were less safe, confining activity to the sandbox allows the machine to be sterilized by sweeping that area and doesn’t necessitate a full reset. Even without the sandbox, knowing the full capability of the range of permitted commands enables damage limitation and precision cleaning. The range of commands should be created with the end user as priority, letting them do what they want with the lowest danger. It should not be created with application writers as top priority since that is where the security risk arises. Not all potential application writers are benign and many want to exploit or harm the end user for their own purposes. Everyone in IT really ought to know that and should never forget it for a minute and it really shouldn’t need to be said.