by Vic Grout
Later in the morning, Bob called Jenny to bring her up to date with his experiences of PDN and the connection with RFS. She listened attentively, with just the occasional audible intake of breath, as he outlined the discoveries of the past few days in a broadly similar manner to his explanations for Andy the previous evening. The exception, given Jenny’s computer science experience, was the new focus he gave to the ‘frame on the wrong network’.
“I just can’t explain it, he said. I’ve double-checked and I don’t think Hattie’s made a mistake. OK, it’s daft enough there being bogus frame fragments on the line in the first place, whatever’s causing it, but this particular one was structured like an ATM cell whereas the HGMS network is essentially an ultra-high-speed Ethernet variant. I’ve checked their protocol diagrams and there’s nothing like it. That frame should not have been on that network!”
“To be fair,” observed Jenny, quietly, “none of those frames should have been on any of those networks if they weren’t legitimate control or data traffic!”
“Well yes, agreed,” conceded Bob, “but I just can’t help thinking this is significant. Even given that weird things are happening anyway, I’m sure that should tell us something.”
Jenny could not suggest anything that Bob had not already considered. She certainly did not manage the imaginative alternatives Andy had proposed the night before. She was able, however, to confirm RFS from personal experience. She had been entering the university library through a revolving door on Friday when the mechanism had suddenly locked. Her walking momentum was such that she was unable to stop in time and had slammed her face against the glass panel. Somehow, her nose had survived intact but instead transferred the force of the impact further back, resulting in a matching pair of black eyes.
At lunchtime, Bob received a variety of contacts from ‘Stephen’ (message IDs, numbers, emails, etc.) together with an unambiguous directive that anything Bob found out was to be relayed to him without delay. Bob complied immediately by contacting him, describing their ‘wrong frame’ observation. Stephen seemed less than convinced by the suggestion (almost implying he must be mistaken) but, explaining it again, sent the same question flashing across Bob’s mind: Why would there be a frame on the wrong network?
*
The European tour continued …
Bob spent a day in Madrid, reunited with Hattie for the first time since Darmstadt. The network problem he had been called in to look at was challenging – certainly compared to Paris – but, even so, only took a couple of hours to deal with. For the remainder of the day (with the network operators under the impression he was still working on their problem) he and Hattie were able to study both the local network and the wider connection to the outside world. The pattern was a familiar one.
Even once the local network had been put right (so far as the operators would be interested, anyway), PDN was clearly visible across each and every link (although only with Hattie tuned to a very fine setting). Once again, the noise was mostly low-level and meaningless but with occasional data fragments. Now that he had a clearer idea of what he was looking for, Bob was able to focus on the higher levels and bigger of these streams and find some more partial frames – a few of these being longer than the one Hattie had recorded in Darmstadt. Their redeeming feature, however – if such an idea applied in this mystery – was that they did at least match the network protocol. Bob performed some simple calculations based on the frequency of these structured streams relative to the unstructured ones. The change was small but statistically clear: more structured – potentially meaningful – data was on the Madrid network than the Darmstadt one. This, of course, could be coincidence or a local characteristic. Alternatively, it might be a sign that PDN was gaining in whatever structure it was tending towards: RFS would then probably increase as a result.
And the pattern was repeated in Cork. Bob had to stay until early afternoon on the network problem he had been called in for: some equipment had to be replaced and the new units were not immediately available. However, this meant he was able to spend four hours looking at the network traffic in detail. Once again, what Andy had described as the ‘random but not random’ pattern of spikes, bits, bytes and partial frames was there. Once again, the ratio of the various levels of structure to unstructured was slightly higher than Darmstadt.
But Cork also provided two further pieces of evidence. Midway through the afternoon, for the first time, Hattie recorded a complete frame. It was valid for the network so Bob had to check and recheck that it was not legitimate traffic, picked up by mistake – but it was not: it was a sensible frame for the network, just not generated (at least intentionally) by a network device on that segment. It had to be the ‘Shakespearean Monkeys’ or ‘random but not random’ effect of the wider network noise. Even then, such a frame, in isolation, would not have affected network behaviour – but it all pointed further towards the hypothesis of PDN causing RFS.
The second breakthrough occurred just before Bob finished for the day. Again, it took a considerable amount of cross-checking to eliminate possible error, but Hattie clearly recorded another wrong frame. This time, the discrepancy was harder to spot (and few other than Bob would have done so) but the protocol was wrong. More precisely, it was the wrong version of the protocol that was present. Subtle but wrong nonetheless. The frame had no place on the network. Why would there be a frame on the wrong network?
He reported both pieces of news to Stephen that evening. The first – the complete frame – pleased him, and appeared to confirm similar results being detected by his team across Europe. He remained sceptical, however, in respect to the second.
That evening, Bob received a somewhat panicky call from Andy.
“Bob, I’m sorry: I may have said something out of place! I was speaking to Aisha this morning – I was telling her about us meeting in Luxembourg – and, without even thinking about it, I realised I was talking about your dirty networks. She was really interested and, before I knew what I was doing, I’d pretty much told her most of what you’d told me! We even had a wee laugh about some of my daft suggestions for what might be behind it. Then, suddenly, I thought, ‘bugger, I probably shouldn’t be saying any of this!’ I pretty quickly said I thought I might have spoken out of turn but it was all too late then. I’d trust Aisha with my life; you never know, she might even help. But I hope I haven’t got you into trouble?”
Bob laughed, immediately realising that he had needed to for some days. “No, it’s fine, Andy. I’d always assumed that Aisha would be told; I can’t see any reason not to: she is part of ‘The Desk’, after all,” he chuckled. “The guy I’m in contact with from Luxembourg said it wasn’t particularly to be kept secret – that I could talk to anyone that might be able to offer anything. I don’t know that Aisha’s going to able to help much, with her background, but it won’t do any harm. And I think it’s generally accepted that most of the detail will be publicly known pretty soon anyway.”
“That’s a relief!” admitted Andy. “So, how’s it going?”
“Slow,” Bob said after a brief pause to consider the question. “We’re starting to see the same patterns in PDN. It all seems to point in the same direction. We’re starting to see the longer frame sequences that might be causing the RFS. I’ve also found another frame on the wrong network. But that’s the bit I’m having trouble convincing the Euro guys about.”
“Do you think you can convince them?”
“I don’t know. I think they may have to see it for themselves.”
*
Fortunately, that was exactly what happened …
Bob’s plane had barely landed in Zaragoza when he received a call from Stephen. His tone was both excited and deferential at once.
“Bob, I have some news from the lab. And I have to apologise to you. Our technical people have found a number of frames of the form you describe. Frames with protocol structures that are alien to the network on which they are to be found. Until, now our equi
pment has not been sufficiently accurate to measure this with any certainty but we have modified it according to your suggestions and now we can do this. We have been able to identify at least fifteen ‘misplaced’ frames across our European measurements and three in the USA. There are likely to many more of course.”
“Millions, billions, probably,” agreed Bob, relieved. “We’re only picking up a tiny fraction of all the global traffic, remember. What we’re looking at only scratches the surface. We have to scale up the figures from our observations to the real world. The numbers will be huge. They have to be really because, buried in there somewhere, will be the frame sequences that actually cause the changes in behaviour: that’s our speculation, anyway.”
“Not speculation.” Stephen’s tone reverted to his normal calm and measured one.
“Pardon?”
“It is no longer speculation, Bob. We have found such a sequence.”
Bob swallowed hard. “You’ve found a frame sequence in the noise that affects the network?”
“Yes, at least, it could have had that effect. Whether it actually did or not, would depend on the device state; but, in this case, if there had been a Telnet session open, the sequence would have closed down the interface. It is a small, specific example but the point is made, we think?”
“Yes, that’s right!”
“So we think that we have established the principle of PDN causing RFS?”
“Agreed!”
*
After those colossal pieces of news, the remainder of the Zaragoza visit was something of an anti-climax. The original job proved to be trivial and Bob’s additional analysis of the network, whilst providing ample confirmation of much that was already known, did not advance their cause. However, on the way to his hotel that evening, Stephen called again.
“Bob, I just thought I would update you on two other points, in case you have not figured these out for yourself yet – you probably have, of course: but in case it helps your thinking at all.”
“OK.” Bob suspected there was still some element of amends in Stephen’s tone: making up for not believing him about the ‘wrong’ frames.
“You see, our guys in the labs have been widening their tests; experimenting with some of their ideas; trying some rather unusual configurations out.”
“Go on …”
“Well, we have identified a couple of places where we do not see PDN.”
“Ah!”
“Yes, we do not see it on completely isolated networks. If we build a completely separate network and power it independently (with its own power source), so there is no physical connection at all, we do not see PDN.”
Bob was uncertain of the significance of this. “What do you mean by an ‘isolated network’?”
“Well, we connect a few routers and suchlike together and run it off a stand-alone generator, so there is no network to the outside world and nothing is plugged into the mains; if we do this, there is no PDN.”
Bob whistled, thinking quickly. “I guess, on one level, that makes sense,” he muttered. “If the noise isn’t being generated simultaneously everywhere, it has to come from somewhere. If you make a completely disconnected system then I guess whatever it is can’t get through.” Another thought struck him immediately. “So, how about two independent networks that are only connected by wifi, but no cables?”
“No,” confirmed Stephen. “In fact, that is the second point: we do not see it on wifi at all!”
With something of a bolt, Bob realised suddenly that this was something he had not considered. All of his work over the previous few days had been on wired networks. He simply had not given any thought whatsoever to wifi, or wireless in general.
“What, never, nowhere?”
“No, Bob. There is no PDN on wireless. It appears to only be present on physical networks, where there is an actual, cabled connection.”
Bob continued to speak at about the speed of his thoughts (and that varied). “So, there has to be a physical electrical connection, then? No, wait; what about optical? Is it on optical networks?”
“Yes, Bob; it is on optical networks. It is on all physically connected wired networks; but not wireless. Does this help us?”
“No, not really.”
*
In the morning, Bob headed for the airport, looking forward to going home for Christmas. But still the questions rang in his head: Who, where or what was PDN coming from? And, in particular, why would there be frames on the wrong networks?
Chapter 10: Awakenings
It was Boxing Day. Aisha Davies was reading from her tablet again. She had received a response from her Italian co-researcher, Professor Paulo Di Iorio. They were still discussing the cause of human creativity and how it might be measured or simulated.
“As to what advantage a comparator-based system might have over standard connectionist models in understanding intelligent processes: well, existing (artificial) connectionist models have difficulty with hierarchical processes, which are essential for control systems and for some types of creativity that are not just combinatorial. Here, perhaps, the goal state may represent the top-down element of a hierarchy and the actual state the lower level, and clearly one could have a cascade of such processes. Neuron–comparators at all levels of description will be situated within (mostly negative) feedback control loops, and the exercising of control is commonly hierarchical in nature. As to how comparisons are made within a comparator, nature is economical and it is possible that the neuronal primitive of regularity-spotting is involved, with the comparator seeking to identify regularities shared by goal and actual states. This sits well with regularity of firing as a desirable energetic state: that is, when goal and actual states share some regularity in common, the assessment of lack of control has been minimised, and there is therefore no output signal to indicate that there has been a prediction error. I don’t really understand why researchers seeking to simulate neuronal activity, and researchers in AI and robotics, are apparently not using, or at least exploring the use of, a framework based entirely on neuron-comparators, and assemblies thereof to build larger comparators. I call it a ‘comparators-all-the-way-down’ approach.”
Once again, Aisha found nothing particularly to disagree with in Professor Di Iorio’s thoughts. She was simply struggling to find the relevance to her own work – anyone’s work, in fact. She responded again in her typically brief style.
“Fine, I would not argue with that, but what use is it? Apart from anything else, how would we prove it?”
She scanned her BBC app for a time while she waited. There was little good news out there: several thousand deaths worldwide were now being attributed to RFS. Travel and communication were disrupted. People were beginning to find themselves stranded on holiday or business trips. A train had derailed at speed in Kenya – survivors unknown but probably few; a chemical plant had exploded in Canada with many casualties assumed.
Eventually, the message app alerted her to Paulo’s reply.
“Well, let’s start by thinking about what an architecture might look like then … A cartoon-like simplification might be this: At the bottom is the metabolic level of description, in which the brain burns glucose and oxygen to make energy available to do work. Above this, the ‘neuronal energetics’ level concerns the ways in which the flow of energy made available to do work drives and influences basic ‘control-processing’ by neuron-comparators and assemblies thereof, subject to the global imperative to minimise assessments of lack of control. This is where much hierarchical processing occurs. The level of description above this concerns the way in which memory is organised to handle different types of control-related information. Above this, presumably, is the level dealing with the more familiar divisions of memory, such as semantic, episodic, and declarative, and the integrated sensor-motor-affect simulations. At the top is domain-general intelligence that may interact with memory.
“So perhaps we could prove the theory by actually building such a system?
Maybe building an intelligent and creative device modelled at least loosely on living nervous systems? (I’m serious! Why not?) I suggest following the general trajectory that nature may have taken. Build from the ground up, and begin with general principles, especially neuronal energetics, and use comparators as the basic building blocks. As a cheap (relatively – compared to Jenny Smith’s ‘Big Brain Project’) project, one could consider having a ‘core’ or ‘hub’ of one or two research institutions, who begin by designing, say, half-a dozen different basic types of (virtual) comparators that can be fitted together like building bricks. These comparators are then made available, over a shared network, for the wider research community (including computer and brain scientists), as tools to experiment with as they wish. Nature’s scheme is conceptually mostly fairly simple, but there have been numerous fine-tunings in different species, and this is where the sophistication lies, I suggest: hence the proposed ‘open-source’ & ‘crowd-source’ empirical approach of ‘understanding by building’. Functionality can be improved as the project progresses, as in nature. Specific goals can be loaded onto specific comparators. It seems to me that comparators will fall into two main types; those where neuronal energetic protocols manifest fully minimised assessments of lack of control by not outputting any signal, as in the case of (usually) comparators regulating higher–level goals, and those manifesting fully minimised assessments of lack of control by outputting a regular signal, as is often the case with lower-level comparators such as those neurons in specific visual areas that only fire when a specific incoming signal is detected. The core group of researchers designs and retains control over the main architecture, but this is very simple and flexible, like a living nervous system. Comparators can be situated within negative feedback control loops, incorporating virtual sensors or actuators, in a virtual environment. If there is a ‘general explore and reinforcement-learning’ protocol, this should lead eventually to some form of virtual functionality. The key point here is that if one substitutes interfaces having ‘real world’ sensors and actuators for the virtual ones, one is on the way to a robot with primitive functionality. As a major safeguard, of course, an ethics committee/core group has to ensure that the system remains isolated: such a system needs to be ‘hermetically contained’, both to stop it getting out and so that it’s not influenced by the outside world!
