by Vic Grout
“Don’t you dare,” Jenny threatened.
But, of course, they did.
Chapter 12: Nodes, Edges and Faces
They worked frantically. Bob immediately connected Hattie to the network adaptor on his office laptop to record some noise readings to compare with later in the day. Jenny spent the rest of the afternoon trying to wrestle through the mathematics of combining self-similarity across both logical layers and physical regions with the patterns normally to be found in negative feedback control loops and the anticipated increase in the signal-to-noise ratio of the meaningful to nonsense data. Aisha made repeated calls to Professor Paulo Di Iorio to establish what aspects of these measurements were known from human brain studies. After a considerable amount of difficult clarification, he was eventually able to send through some relevant results from an H2020 project he was working on. Later, Bob carefully added extra code to Hattie to enable her to take the necessary readings from a network and crudely calculate the new ‘S Parameter’. In the evening, with Aisha’s input from the human brain data, Jenny was able to refine the maths and Bob gradually tweaked Hattie further until they had something that all three of them felt was a good as it was likely to be. During this time, Andy made several pots of tea.
None of this (apart from the tea) was easy. As soon as she looked up the details of the H Parameter (hampered, as was usual now, by a very slow Internet response), Jenny was reminded that it was essentially a theoretical figure: difficult to calculate exactly with real, finite data. The S value would almost certainly have to be approximated as best they could, with a complexity (and run time) that could be tolerated. (Jenny at first resisted use of the term ‘Smith Parameter’ but gave up when Bob shortened it to ‘S Parameter’ and finally ‘S value’.) She spent a couple of hours working out an appropriate neural network-based heuristic algorithm to estimate the S value for Bob to implement in Hattie, who was fairly powerful for a stand-alone system but no super-computer so some simplification was needed.
Also, the data Aisha received from Paulo Di Iorio on feedback loops in the brain, whilst useful and relevant, was not exactly in the form that was going to give her the calibration that Hattie would need to find the S value in real networks. Professor Di Iorio had never intended it for this purpose and was very curious as to what Aisha might want it for. Having given both her thanks and an unconvincing sequence of non-specific reasons why she needed it, she then spent three hours turning the information over in a spreadsheet to get the figures they required. Then finally, Hattie had to be set up in a mode that would allow her to automatically ignore the ‘proper’ user data and only register PDN noise so that this could be used to measure the increase in the signal-to-noise ratio, which also fed into the S value calculation.
*
At about eight o’clock they were ready to test the new-look Hattie. From a memory pen, they loaded in the complete, aggregated set of Professor Di Iorio’s brain sample figures. Bob set Hattie to run manually on the data; she had been configured to run Jenny’s heuristic AI approximation algorithm for about a minute. Longer would give a more accurate result; shorter, a less accurate one. No-one was entirely sure how much difference that would make.
Bob had made no attempt to refine Hattie’s user interface. The ‘Processing’ message scrolled across her screen as the 60 seconds passed. Then, entirely without ceremony, the result appeared.
S = 0.751
“Is that good?” Andy asked innocently.
“Yes,” said Bob, beaming a little. “That’s what we were trying to normalise the figures to.”
“Anything in the range 0.5 to 1 might represent some level of self-similarity, feedback and signal increase, which could be associated with what is being measured having achieved some control imperative,” Aisha explained. “The higher the better: 0.75 is what we get from a typical adult human brain.”
“So, it looks like we’ve got the calibration about right,” suggested Jenny, “and a minute seems to be enough for the heuristic to get a decent approximation.”
“So, anything above 0.5 means conscious?” suggested Andy.
“It might not be as simple as that,” Aisha said, wincing slightly, “but it may give an indication of that type of neural activity.”
“So, shall we try it for real?” asked Bob.
“Yes, let’s do it. Let’s get her onto a real network!” agreed Jenny.
But this proved to be less conclusive than they hoped. They first reconnected Hattie to Bob’s laptop. They then set her to run and waited for the minute to pass with even greater expectation than before. After what seemed like an eternity, the ‘Processing’ message was replaced by the result.
S = 0.523
“Ah,” said Andy.
“Hmm,” mumbled Aisha.
“Damn!” said Jenny.
“Bollocks!” spat Bob.
“So, where does that leave us?” asked Andy. “At least it’s above 0.5?”
“Not enough,” said Aisha, dejectedly. “Not conclusive.”
“I can’t go to Stephen with this,” complained Bob.
They sat in silence for a while. Suddenly, Jenny started forwards.
“Hang on a minute; you’re wifi here aren’t you?”
“Well yes, of course,” agreed Bob. “and yes, I know that PDN itself isn’t on wireless networks. But this machine’s still plugged into the mains so it’s still a part of It, isn’t it? The noise is still visible: look.” He pointed to one of Hattie’s oscilloscope displays in justification.
“Yes, but it’s only on the edge of It, as it were,” Jenny suggested. “It’s connected through the power network but nothing else. Perhaps we’re going to get lower figures on the periphery?”
“If I’ve understood Paulo’s figures properly, that is not the way it would work in a human brain,” Aisha interjected.
“But we know It’s not exactly like the brain,” pursued Jenny. “It’s just based on the same rough model.” (The others smiled at her sudden, enthusiastic conversion to the cause. She noticed but continued undeterred.) “I don’t think we’re looking far enough into It here. We might be getting some local self-similarity and feedback but probably not anything from a wider area. In fact, if you think about it, we’re not part of any hierarchy here. We would be if the wifi was a part of It but it isn’t. There’s not going to be any self-similarity or feedback across regions or through levels. Perhaps we’ll get a better result – a higher figure – if we measure something a bit closer to the business end?”
So, they quickly tried another experiment. Bob took the cover off the broadband box and connected Hattie to the incoming ATM line. Those earlier weeks of programming her auto-configuration features might be worthwhile after all, he thought to himself. Hattie set off on her measurements and calculations and, after her allocated minute had elapsed, reported:
S = 0.551
“Strewth!” groaned Bob. “Does that tell us anything?”
“It’s higher,” noted Andy, somewhat obviously.
“Not high enough,” said Aisha.
“The problem is, I guess,” suggested Jenny, “that we’re still pretty much on the edge. Whatever It’s up to, we’re getting a sight of some of Its control data,” glancing again at Hattie’s oscilloscope, “but still not enough. We’re not seeing far up the ATM link and there’s effectively nothing below us because the wifi’s cutting off the rest of the house. We need to get further in to It to really test the concept.”
“So, in many ways, it would make our life easier if It had learnt to use wireless, then?” suggested Andy, as the most casual of throwaway remarks.
But both Aisha and Bob gaped at him in amazement: he had clearly said something utterly profound. Aisha was the first to speak.
“Yes, I had never thought of it like that. Learning to use different parts of Its structure is probably part of Its development. Although I suspect Its control imperative came into existence all at once, when Its combined structure reached some critica
l size, it is likely that It is currently having different degrees of success across different media …”
“Electrical, optical, etc.?” suggested Jenny.
“… Yes, different degrees of success across electrical, optical, and suchlike, media,” Aisha continued, “and with different protocol mechanisms: that does not seem unlikely. The improvement we think we are seeing in Its ability to produce meaningful signals, over time, is probably not homogeneous over Its entire structure. I expect we could trace that process if we had access to enough historical data?” She glanced questioningly at Bob, who appeared distracted by other thoughts. “Even now, It is probably still having varying degrees of success, in different places, as It learns. We could conceivably get different S measurements in different locations, even if the connectivity and hierarchical structures were the same.” She paused. “I suppose It will never learn to use wireless links, though, because there is no physical connection?”
Bob had been waiting his turn with considerable difficulty. “I’m really not so sure about that!” he said with some force. They all stared at him. He continued.
“Why shouldn’t it be able to use wireless links? Just think about it for a moment. They might seem to us like something completely different but they’re not. What’s a physical link? It’s just a transmitter and receiver with a carrier medium in between. OK, it’s pretty simple with a cable because, from Its perspective, there’s an obvious ‘in’ and ‘out’ but it’s essentially a ‘transmit-carry-receive’ system.
“Now, if you think about it, wireless isn’t any different. There’s still a transmitter and receiver. The only difference is that the medium’s the air. You still have the transmit-carry-receive system: just with a different carrier. OK, there are some issues with signals being multi-directional in a typical wifi environment but there are many wireless signals that are quite precisely focused.” As the thoughts in his head outpaced the words coming from his mouth, he suddenly banged the arm of his chair hard with his fist.
“I bet you It’s just not figured out how to use wireless YET!”
“You’re saying It will?” Aisha asked incredulously.
“I can’t see any reason why It wouldn’t,” Bob answered, slightly calmer. “There’s no essential difference. I can understand that it might take longer but I can’t see anything to stop it. It’s true that we find it easier to write network discovery algorithms for wired networks but we can manage for wireless too. Remember It has our data as a guide as to what can go where, and how. It can ‘see’ both ends of a wireless link because they’re both plugged into the mains: It just needs to realise what can happen in between. My guess would be that It first figures out how to use the uni-directional links because that’s not hugely different – not conceptually different, as far as the transmit-carry-receive model is concerned – from physical connections. From there, it would a simple step to getting to grips with the whole range of wireless communications. I can sort of understand why it hasn’t happened yet; but I’m sure it will!”
“And what would that give us?” asked Andy. “What would It become then?”
Jenny was first to reply. “I don’t think it would make It a lot bigger, would it? True, a lot of the current structure is over wireless but almost all of those networks and end hosts are connected to the main body through the power networks anyway. I doubt it would increase the overall neural count that much: not by a significant factor.”
“But it would increase the neural connectivity,” insisted Aisha. “As it stands, It is quite sparsely connected, as is the way of computer systems designed by humans. And this is dictated by the physical connections we have given It. Even if It was to develop Its own rules for internal communication, as it may be doing, It cannot use physical connections that are not there.
“But if It becomes able to control, or at least use, wireless connections, then It will be able to communicate internally through the air – probably through almost any medium. Rather than each network neuron being limited in its connection to those we have given it physical data links to, It will be able to establish a neuronal connection with anything in range.”
“That might put scale-free network models back on the table,” said Jenny with some alarm. “So, it would begin to look a lot more like a human brain?”
“Yes; although Its neural size will increase only slightly, Its neural complexity will increase massively,” said Aisha. “I do not even know how much.”
“Several orders of magnitude, I’d think,” agreed Bob. “It would be a different beast entirely.”
“Not sure I like the ‘beast’ concept too much!” said Jenny.
“Let’s hope it doesn’t happen for a while yet then,” suggested Andy
*
“The fact remains, we’re still not getting far enough into It to get a convincing reading,” Jenny pointed out half an hour later, after a short tea-break. “For now, let’s ignore what happens if It manages to master the wireless capabilities It doesn’t realise It’s got yet and focus on what we have. Is it going to help if we set Hattie off to run for longer? Will that give us a better reading?”
“I doubt it,” said Bob, “but, while we’re not doing anything else, I guess there’s no harm in trying.” He leaned over, typed in a couple of parameter variations and set Hattie off again. This time, ‘Processing’ appeared on her screen and stayed there. “There: that should finish around midnight. But I don’t suppose it’ll give us much.”
“OK then,” said Aisha, “while that is running, how can we get further into It to measure Its control assessment capabilities?”
“You mean Its consciousness?” suggested Andy once more.
Aisha smiled. “I am still not quite comfortable with calling it that; but, if you like, yes, Its consciousness. But I am only saying that because it is shorter, you understand?”
Andy smiled in return. “So what is the best way to get deeper into the beast?” he asked. Jenny gritted her teeth but Bob had clearly been considering this already.
“I wish now I’d had a closer look at the ‘Network on Chip’ core they had at HGMS,” Bob said apologetically. “There would have been obvious network levels above and below that so there would have been more to see, I guess.”
“You mean the Darmstadt network?” Jenny suggested. “But even that was isolated, wasn’t it? And can Hattie get readings from an NoC? She’d have to somehow look inside it, wouldn’t she?”
“Well, not without some additional kit, I suppose,” admitted Bob. “Yes, I suppose you’re right. So, what else can we do? There are plenty of places where we would see big hierarchies and wide coverage but we can’t go and break into a BT box without getting locked up! And we can’t just walk into the ISPs and demand to set Hattie up, can we? They’ll want to know why. And what would we tell them? That we think the Internet’s come alive?”
“I’ve got a few useful contacts,” Jenny smiled. “In particular, there’s a research team at my university, who are looking at energy minimisation on NoC systems: they might have the means of taking measurements from the circuitry. Would Hattie be able to work on data supplied from an external source?”
Bob nodded. “No different to what we just did to test her with Aisha’s human brain figures.”
“Also, there’s a guy I’ve worked with at Imperial College,” Jenny said, “who may be able to get us into the Rutherford Labs at Didcot. That would give us access to the Janet Network.”
Bob nodded. Aisha and Andy both raised their eyebrows in query so Jenny continued.
“The ‘Joint Academic Network’,” she explained. “JANET for short. It’s the network used by all the UK universities and most of the further education colleges. It’s a pretty serious piece of work, which is what we want and, if my Imperial colleague can get us into the right bit at the Rutherford Appleton Labs, near Oxford, then we could be looking at one of the highest hierarchy nodes in the network. So we’d really be able to put our theory to the test. I can thi
nk of one or two other places as well,” she added.
The other three nodded agreement. The rest of the evening was spent planning an itinerary for the next day. Jenny made a number of phone calls and sent several messages. By the end of the evening, she had enough useful responses to put together a rough schedule for Bob and her to visit four places of possible value in London, Oxford and Cambridge. Hattie’s couriers were contacted to arrange for her to follow safely close behind. (Bob grumbled, under his breath, as to ‘someone paying for this at some point’.) Some of the visits would involve taxis and lifts from willing (but largely unapprised) collaborators so it was impractical, and generally considered unnecessary, for all four of them to make the tour. It being the period between Christmas and New Year, Andy and Jenny were on an enforced ‘operational efficiency break’ from their universities; Aisha had taken leave from the hospital; Bob, of course, was his own boss; so all were free. The plan then was for Jenny and Bob (and Hattie) to make the visits while Aisha and Andy took it easy for a day. They would meet back at Jill and Bob’s in the evening. All seemed fairly contented with the arrangements.
Before they went to bed, just after midnight, Hattie finished her extended run connected to the broadband box. A single light on the manual part of her dashboard alerted them. They moved over to look. Her simple text output showed:
S = 0.553
“So is that Hattie’s increased accuracy?” asked Andy. “Or just random variation? Or is It getting cleverer?”
“Who can say?” Jenny mumbled.
*
Jenny and Bob started early in the morning and were at the first London stop – her home university – by 8:30am. Dr. Jakob Sovel, leader of the NoC team, met them at the entrance, not in the best of humour.
“I trust this is important, Professor Smith? I have had to delay a family trip to the country for this. I appreciate you hold a senior position here but I hope you can explain why we are doing this?”
To avoid too many difficult questions, Jenny took a direct, and formal, line.