{"id":6088,"date":"2015-07-22T14:59:13","date_gmt":"2015-07-22T22:59:13","guid":{"rendered":"http:\/\/www.rifters.com\/crawl\/?p=6088"},"modified":"2015-07-27T11:49:21","modified_gmt":"2015-07-27T19:49:21","slug":"dr-fox-and-the-borg-collective","status":"publish","type":"post","link":"https:\/\/www.rifters.com\/crawl\/?p=6088","title":{"rendered":"Dr. Fox and the Borg Collective"},"content":{"rendered":"

Take someone’s EEG as they squint really hard and think Hello. <\/em>Email that brainwave off to a machine that’s been programmed to respond to it by tickling someone else’s brain with a flicker of blue light. Call the papers. Tell them you’ve invented telepathy.<\/p>\n

\"I<\/a>

I mean, seriously: aren’t you getting tired of these guys?<\/p><\/div>\n

Or: teach one rat to press a lever when she feels a certain itch. Outfit another with a sensor that pings when the visual cortex sparks a certain way. Wire them together so the sensor in one provokes the itch in the other: one rat sees the stimulus and the other presses the lever. Let Science Daily<\/em> tell everyone that you’ve built the Borg Collective.<\/p>\n

There’s been a lot of loose talk lately about hive minds. Most of it doesn’t live up to the hype. I got so irked by all that hyperbole\u2014 usually accompanied by a still from “The Matrix”, or a picture of Spock in the throes of a mind meld\u2014 that I spent a good chunk of my recent Aeon piece<\/a> bitching about it. Most of these “breakthroughs”, I grumbled, couldn’t be properly described as hive consciousness or even garden-variety telepathy. I described it as the difference between experiencing an orgasm and watching a signal light on a distant hill spell out oh-god-oh-god-yes<\/em> in Morse Code.<\/p>\n

I had to allow, though, that it might be only a matter of time before you could scrape the hype off one of those stories and find some actual substance beneath. In fact, the bulk of my Aeon essay dealt with the implications of the day when all those headlines came true for real.<\/p>\n

I think we might have just hit a milestone.<\/p>\n

*<\/p>\n

Here’s something else to try. Teach a bunch of thirsty rats to distinguish between two different sounds; motivate them with sips of water, which they don’t get unless they push the round lever when they hear “Sound 0” and the square one when they hear “Sound 1”.<\/p>\n

Once they’ve learned to tell those sounds apart, turn them into living logic gates. Put ’em in a daisy-chain, for example, and make them play “Broken Telephone”: each rat has to figure out whether the input is 0 or 1 and pass that answer on to the next in line. Or stick ’em in parallel, give them each a sound to parse, let the next layer of rats figure out a mean response. Simple operant conditioning, right? The kind of stuff that was old before most of us were born.<\/p>\n

Now move the stimulus inside<\/em>. Plant it directly into the somatosensory cortex via a microelectrode array (ICMS, for “IntraCortical MicroStimulation”). And instead of making the rats press levers, internalize that too: another array on the opposite side of the cortex, to transmit whatever neural activity it reads there.<\/p>\n

Call it “brainet”. Pais-Vieira et al<\/em> do.<\/p>\n

The paper is “Building an organic computing device with multiple interconnected brains<\/a>“, from the same folks who brought you Overhyped Rat Mind Meld and Monkey Videogame Hive.<\/a> In addition to glowing<\/a> reviews from<\/a> the usual suspects<\/a>, it has won over skeptics who’ve decried the hype<\/a> associated with this sort of research in the past. It’s a tale of four rat brains wired together, doing stuff, and doing it better than singleton brains faced with the same tasks. (“Split-brain patients outperform normal folks on visual-search and pattern-recognition tasks,” I reminded you all back at Aeon; “two minds are better than one, even when they\u2019re in the same head”). And the payoff is spelled out right there in the text: “A new type of computing device: an organic computer… could potentially exceed the performance of individual brains, due to a distributed and parallel computing architecture”.<\/p>\n

Bicameral Order, anyone? Moksha Mind? How could I not love such a paper?<\/p>\n

And yet I don’t. I like it well enough. It’s a solid contribution, a real advance, not nearly so guilty of perjury as some.<\/p>\n

And yet I’m not sure I entirely trust<\/em> it.<\/p>\n

I can’t shake the sense it’s running some kind of con.<\/p>\n

*<\/p>\n

\"The<\/a>

The real thing. Sort of. (From Pais-Vieira et al<\/i> 2015.<\/p><\/div>\n

There’s much to praise. We’re talking about an actual network, multiple brains in real two-way communication, however rudimentary. That alone makes it a bigger deal than those candy-ass one-direction set-ups that usually get the kids in such a lather.<\/p>\n

In fact, I’m still kind of surprised that the damn thing even works. You wouldn’t think that pin-cushioning a live brain with a grid of needles would accomplish much. How precisely could such a crude interface ever interact with all those billions of synapses, configured just so<\/em> to work the way they do? We haven’t even figured out how brains balance their books in one<\/em> skull; how much greater the insight, how many more years of research before we learn how to meld multiple minds, a state for which there’s no precedent in the history of life itself?<\/p>\n

But it turns out to be way easier than it looks. Hook a blind rat up to a geomagnetic sensor with a simple pair of electrodes, and he’ll be able to navigate a maze<\/a>\u2014 using ambient magnetic fields\u2014 as well as any sighted sibling. Splice the code for the right kind of opsin into a mouse genome and the little rodent will be able to perceive colors<\/a> she never knew before. These are abilities unprecedented in the history of the clade\u2014 and yet somehow, brains figure out the user manuals on the fly. Borg Collectives may be simpler than we ever imagined: just plug one end of the wire into Brain A, the other into Brain B, and trust a hundred billion neurons to figure out the protocols on their own.<\/p>\n

Which makes it a bit of a letdown, perhaps, when every experiment Pais-Vieira et al<\/em> describe comes down, in the end, to the same simple choice between 0 and 1. Take the very climax of their paper, a combination of “discrete tactile stimulus classification, BtB interface, and tactile memory storage” bent to the real-world goal of weather prediction<\/em>. Don’t get too excited\u2014 it was, they admit up front, a very simple exercise. No cloud cover, no POP, just an educated guess at whether the chance of rain is going up or down at any given time.<\/p>\n

\"Hey,<\/a>

Hey, can’t be any worse than the weather person on CBC’s morning show…<\/p><\/div>\n

The front-end work was done by two pairs of rats wired into “dyads”; one dyad was told whether temperature was increasing (0) or decreasing (1), while the other was told the same about barometric pressure. If all went well, each simply spat out the same value that had been fed into it; they were then reintegrated into the full-scale 4-node brainet, which combined those previous outputs to decide whether the chance of precip was rising or falling. It was exactly the same kind of calculation, using exactly the same input, that showed up in other tasks from the same paper; the main difference was that this time around, the signals were labeled “temperature rising” or “temperature falling” instead of 0 and 1. No matter. It all still came down to another encore performance of Brainet’s big hit single, “Torn Between Two Signals”, although admittedly they played both acoustic and electric versions in the same set.<\/p>\n

I’m aware of the obvious paradox in my attitude, by the way. On the one hand I can’t believe that such simple technology could work at all when interfaced with living brains; on the other hand I’m disappointed that it doesn’t do more.<\/p>\n

I wonder how brainet would resolve those signals.<\/p>\n

*<\/p>\n

Of course, Pais-Vieira et al <\/em>did more than paint weather icons on old variables. They ran brainet through other paces\u2014 that “broken telephone” variant I mentioned, for example, in which each node in turn had to pass on the signal it had received until that signal ended up back at the first rat in the chain\u2014 who (if the run was successful) identified the serially-massaged data as the same one it had started out with. In practice, this worked 35% of the time, a significantly higher success rate than the 6.25%\u2014 four iterations, 50:50 odds at each step\u2014 you’d expect from random chance. (Of course, the odds of simply getting the correct final<\/em> answer were 50:50 regardless of how long the chain was; there were only two states to choose from. Pais-Vieira et al<\/em> must have tallied up correct answers at each intermediate step when deriving their stats, because it would be really dumb not to; but I had to take a couple of passes at those paragraphs, because at least one sentence\u2014<\/p>\n

“the memory of a tactile stimulus could only be recovered if the individual BtB communication links worked correctly in all four consecutive trials.”<\/p><\/blockquote>\n

\u2014 was simply wrong. Whatever the merits of this paper, let’s just say that “clarity” doesn’t make the top ten.)<\/p>\n

\"What<\/a>

What the rats saw. Ibid.<\/i><\/p><\/div>\n

\"More<\/a>

More nodes, better results. Ibid.<\/i><\/p><\/div>\n

The researchers also used brainet to transmit simple images\u2014 again, with significant-albeit-non-mind-blowing results\u2014 and, convincingly showed that general performance improved with a greater number of brains in the net. On the one hand I wonder if this differs in any important way from simply polling a group of people with a true-false question and going with the majority response; wouldn’t that also tend towards greater accuracy with larger groups, simply because you’re drawing on a greater pool of experience? Is every Gallup focus group a hive mind?<\/p>\n

On the other hand, maybe the answer is: yes, in a way<\/em>. Conventional neurological wisdom describes even a single brain as a parliament of interacting modules. Maybe group surveys is exactly<\/em> the way hive minds work.<\/p>\n

*<\/p>\n

So you cut them some slack. You look past the problematic statements because you can figure out what they were trying to say even if they didn’t say it very well. But the deeper you go, the harder it gets. We’re told, for example, that Rat 1 has successfully identified the signal she got from Rat 4\u2014 but how do we know that? Rat 4, after all, was only repeating a signal that originated with Rat 1 in the first place (albeit one relayed through two other rats). When R1’s brain says “0”, is it parsing the new input or remembering the old?<\/p>\n

Sometimes the input array is used as a simple starting gun, a kick in the sulcus to tell the rats Ready, set, Go: sync up!<\/em> Apparently the rat brains all light up the same way when that happens, which Pais-Vieira et al<\/em> interpret as synchronization of neural states via Brain-to-Brain interface. Maybe they’re right. Then again, maybe rat brains just happen to light up that way when spiked with an electric charge. Maybe they were no more “interfaced” than four flowers, kilometers apart, who simultaneously turn their faces toward the same sun.<\/p>\n

Ah, but synchronization improved over time<\/em>, we’re told. Yes, and the rats could see each other through the plexiglass, could watch their fellows indulge in the “whisking and licking” behaviors that resulted from the stimulus. (I’m assuming here that “whisking” behavior has to do with whiskers and not the making of omelets, which would<\/em> be a truly impressive demonstration of hive-mind capabilities.) Perhaps the interface, such as it was, was not through the brainet at all\u2014 but through the eyes.<\/p>\n

I’m willing to forgive a lot of this stuff, partly because further experimentation resolves some of the ambiguity. (In one case, for example, the rats were rewarded only if their neural activity de<\/em>synchronised, which is not something they’d be able to do without some sense of the thing they were supposed to be diverging from.) Still, the writing\u2014 and by extension, the logic behind it\u2014 seems a lot fuzzier than it should be. The authors apparently recognize this when they frankly admit<\/p>\n

“One could argue that the Brainet operations demonstrated here could result from local responses of S1 neurons to ICMS.”<\/p><\/blockquote>\n

They then list six reasons to believe otherwise, only one of which cuts much ice with me (untrained rats didn’t outperform random chance when decoding input). The others\u2014 that performance improved during training, that anesthetized or inattentive animals didn’t outperform chance, that performance degraded with reduced trial time or a lack of reward\u2014 suggest, to me, only that performance was conscious and deliberate, not that it was “nonlocal”.<\/p>\n

Perhaps I’m just not properly grasping the nuances of the work\u2014 but at least some of that blame has to be laid on the way the paper itself is written. It’s not that the writing is bad, necessarily; it’s actually worse than that. The writing is confusing<\/em>\u2014 and sometimes it seems deliberately so. Take, for example, the following figure:<\/p>\n

\"Alone<\/a>

Alone against the crowd. Ibid.<\/i><\/p><\/div>\n

Four rats, their brains wired together. The red line shows the neural activity of one of those rats; the blue shows mean neural activity of the other three in the network, pooled. Straightforward, right? A figure designed to illustrate how closely the mind of one node syncs up with the rest of the hive.<\/p>\n

Of course, a couple of lines weaving around a graph aren’t what you’d call a rigorous metric: at the very least you want a statistical measure of correlation between Hive and Individual, a hard number to hang your analysis on. That’s what R is, that little sub-graph inset upper right: a quantitative measure of how precisely synced those two lines are at any point on the time series.<\/p>\n

\"I<\/a>

What are you afraid of, Miguel? See how easy it is?<\/p><\/div>\n

So why is the upper graph barely more than half the width of the lower one?<\/p>\n

The whole point of the figure is to illustrate the strength of the correlation at any given time. Why wouldn’t you present everything at a consistent scale, plot R along the same ruler as FR so that anyone who wants to know how tight the correlation is at time T can just see<\/em> it? Why build a figure that obscures its own content until the reader surrenders, is forced to grab a ruler, and back-converts by hand?<\/p>\n

What are you guys trying to cover?<\/p>\n

*<\/p>\n

Some of you have probably heard of the Dr. Fox Hypothesis<\/a>. It postulates that “An unintelligible communication from a legitimate source in the recipient’s area of expertise will increase the recipient’s rating of the author’s competence.” More clearly, Bullshit Baffles Brains.<\/p>\n

But note the qualification: “in the recipient’s area of expertise”. We’re not talking about some Ph.D. bullshitting an antivaxxer; we’re talking about an audience of experts being snowed by a guy speaking gibberish in their own field of expertise<\/em>.<\/p>\n

In light of this hypothesis, it shouldn’t surprise you that controlled experiments have shown that wordy, opaque sentences rank more highly in people’s minds than simple, clear ones which convey the same information. Correlational studies report that the more prestigious a scientific journal tends to be, the worse the quality of the writing you’ll find therein. (I read one fist-hand account of someone who submitted his first-draft manuscript\u2014 which even he described as “turgid and opaque”\u2014 to the same journal that had rejected the much-clearer 6th draft of the same paper. It was accepted with minor revisions.)<\/p>\n

Pas-Vieira et al<\/em> appears in Nature’s “Scientific Reports”. You don’t get much more prestigious than that.<\/p>\n

So I come away from this paper with mixed feelings. I like what they’ve done\u2014 at least, I like what I think they’ve done. From what I can tell the data seem sound, even behind all the handwaving and obfuscation. And yet, this is a paper that acts as though it’s got something to hide, that draws your attention over here<\/em> so you won’t notice what’s happening over there<\/em>. It has issues, but none are fatal so far as I can tell. So why the smoke and mirrors? It’s like being told a wonderful secret by a used-car salesman.<\/p>\n

These guys really had something to say.<\/p>\n

Why didn’t they just fucking say<\/em> it?<\/p>\n

 <\/p>\n

 <\/p>\n

 <\/p>\n

(You better appreciate this post, by the way. Even if it is dry as hell. It took me 19 hours to research and write the damn thing. <\/span><\/p>\n

(I ought to put up a paywall.)<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

Take someone’s EEG as they squint really hard and think Hello. Email that brainwave off to a machine that’s been programmed to respond to it by tickling someone else’s brain with a flicker of blue light. Call the papers. Tell them you’ve invented telepathy. Or: teach one rat to press a lever when she feels […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[10,26],"tags":[],"class_list":["post-6088","post","type-post","status-publish","format-standard","hentry","category-neuro","category-relevant-tech"],"_links":{"self":[{"href":"https:\/\/www.rifters.com\/crawl\/index.php?rest_route=\/wp\/v2\/posts\/6088","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.rifters.com\/crawl\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.rifters.com\/crawl\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.rifters.com\/crawl\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.rifters.com\/crawl\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=6088"}],"version-history":[{"count":19,"href":"https:\/\/www.rifters.com\/crawl\/index.php?rest_route=\/wp\/v2\/posts\/6088\/revisions"}],"predecessor-version":[{"id":6115,"href":"https:\/\/www.rifters.com\/crawl\/index.php?rest_route=\/wp\/v2\/posts\/6088\/revisions\/6115"}],"wp:attachment":[{"href":"https:\/\/www.rifters.com\/crawl\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6088"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.rifters.com\/crawl\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6088"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.rifters.com\/crawl\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6088"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}