An unsuccessful plan to create a computer simulation of a human brain within 10 years. An exhaustive catalog of cell types comprising a specific class of inhibitory neurons within mouse visual cortex. What do these massive research programs have in common? Both efforts were conducted by large multidisciplinary teams at non-traditional research institutions: the Blue Brain Project based in Lausanne, Switzerland and the Allen Institute for Brain Science in Seattle, Washington.
BIG SCIENCE is the wave of the future, and the future is now. Actually, that future started 15-20 years ago. The question should be, is there a future for any other kind of neuroscience?
Despite a superficial “BIG SCIENCE” similarity, the differences between funding sources, business models, leadership, operation, and goals of Blue Brain and the Allen Institute are substantial. Henry Markram, the “charismatic but divisive” visionary behind Blue Brain (and the €1 billion Human Brain Project) has been criticized for his “autocratic” leadership, “crap” ideas, and “ill-conceived, ... idiosyncratic approach to brain simulation” in countless articles. His ambition is undeniable, however:
“I realized I could be doing this [eg., standard research on spike-timing-dependent plasticity] for the next 25, 30 years of my career, and it was still not going to help me understand how the brain works.”
I'm certainly not a brilliant neuroscientist in Markram's league, but I commented previously on how a quest to discover “how
the brain works” might be futile:
...the search for the Holy Grail of [spike trains, network generative models, manipulated neural circuit function, My Own Private Connectome, predictive coding, the free energy principle (PDF), or a computer simulation of the human brain promised by the Blue Brain Project] that will “explain” how “The Brain” works is a quixotic quest. It's a misguided effort when the goal is framed so simplistically (or monolithically).
“I hope that you are at least partly convinced that it is not impossible to build a brain. We can do it within 10 years, and if we do succeed, we will send to TED, in 10 years, a hologram to talk to you.”
This claim would come back to haunt him in 2019, because (of course) he was nowhere close to simulating a human brain. In his defense, Markram said that his critics misunderstood and misinterpreted his grandiose proclamations.1
Blue Brain is now aimed at “biologically detailed digital reconstructions and simulations of the mouse brain.”
In Silico
Documentary filmmaker Noah Hutton2 undertook his own 10 year project that followed Markram and colleagues as they worked towards the goals of Blue Brain. He was motivated by that TED talk and its enthralling prediction of a brain in a supercomputer (hence in silico). Originally entitled Bluebrain and focused on Markram, the documentary evolved over time to include more realistic viewpoints and interviews with skeptical scientists, including Anne Churchland, Terry Sejnowski, Stanislas Dehaene, and Cori Bargmann. Ironically, Sebastian Seung was one of the loudest critics (ironic because Seung has a grandiose TED talk of his own, I Am My Connectome).
In Silico was available for streaming during the DOC NYC Festival in November (in the US only), and I had the opportunity to watch it. I was impressed by the motivation and dedication required to complete such a lengthy project. Hutton had gathered so much footage that he could have made multiple movies from different perspectives.
Over the course of the film, Blue Brain/Human Brain blew up, with ample critiques and a signed petition from hundreds of neuroscientists (see archived Open Letter).
And Hutton grew up. He reflects on the process (and how he changed) at the end of film. He was only 22 at the start, and 10 years is a long time at any age.
Some of the Big Questions in In Silico:
- How do you make sure all this lovely simulated activity would be relevant for an animal's behavior?
- How do you build in biological imperfections (noise) or introduce chaos into your perfect pristine computational model? “Tiny mistakes” are critical for adaptable biological systems.
- “You cannot play the same soccer game again,” said one of the critics (Terry Sejnowski, I think)
- “What is a generic brain?”
- What is the vision?
The timeline kept drifting further and further into the future. It was 10 years in 2009, 10 years in 2012, 10 years in 2013, etc.
Geneva 2019, and it's Year 10 – only two Principals left, 150 papers published, and a model of 10 million neurons in mouse cortex. Stunning visuals, but still disconnected from behavior.
In the end, “What have we learned about the brain? Not much. The model is incomprehensible,” to paraphrase Sejnowski.
GABA Interneurons
Another brilliant and charismatic neuroscientist, Christof Koch, was interviewed by Hutton. “Henry has two personalities. One is a fantastic, sober scientist … the other is a PR-minded messiah.”
Koch is Chief Scientist of the MindScope Program at the Allen Institute for Brain Science, which focuses on how neural circuits produce vision. Another major unit is the Cell Types Program, which (as advertised) focuses on brain cell types and connectivity.3
The Allen Institute core principles are team science, Big Science, and open science. An impressive recent paper by Gouwens and 97 colleagues (2020) is a prime example of all three. Meticulous analyses of structural, physiological, and genetic properties identified 28 “met-types” of GABAergic interneurons that have congruent morphological, electrophysiological, and transcriptomic properties. This was winnowed down from more than 500 morphologies in 4,200 GABA-containing interneurons in mouse visual cortex. With this mind-boggling level of neuronal complexity in one specific class of cells in mouse cortex — along with the impossibility of “mind uploading” — my inclination is to say that we will never (never say never) be able to build a realistic computer simulation of the human brain.
Footnotes
1 Here's another gem: “There literally are only a handful of equations that you need to simulate the activity of the neocortex.”
2 Most of Hutton's work has been as writer and director of documentary films, but I was excited to see that his first narrative feature, Lapsis, will be available for streaming next month. To accompany his film, he's created an immersive online world of interlinked websites that advertise non-existent employment opportunities, entertainment ventures, diseases, and treatments. It very much reminds me of the realistic yet spoof websites associated with the films Eternal Sunshine of the Spotless Mind (LACUNA, Inc.) and Ex Machina (BlueBook). In fact, I'm so enamored with them that they've appeared in several of my own blog posts.
3 Investigation of cell types is big in the NIH BRAIN Initiative ® as well.
References
Abbott A. (2020). Documentary follows implosion of billion-euro brain project. Nature 588:215-6.
[Alison Abbott covered the Blue Brain/Human Brain sturm und drang for years]
Gouwens NW, Sorensen SA, Baftizadeh F, Budzillo A, Lee BR, Jarsky T, Alfiler L, Baker K, Barkan E, Berry K, Bertagnolli D ... Zeng H et al. (2020). Integrated morphoelectric and transcriptomic classification of cortical GABAergic cells. Cell 83(4):935-53.
Waldrop M. (2012). Computer modelling: Brain in a box. Nature News 482(7386):456.
Further Reading
The Blue Brain Project (01 February 2006), by Dr. Henry Markram
“Alan Turing (1912–1954) started off by wanting to 'build the brain' and ended up with a computer. ... As calculation speeds approach and go beyond the petaFLOPS range, it is becoming feasible to make the next series of quantum leaps to simulating networks of neurons, brain regions and, eventually, the whole brain.”
A brain in a supercomputer (July 2009), Henry Markram's TED talk
- from the transcript:
“Our mission is to build a detailed, realistic computer model of the human brain. And we've done, in the past four years, a proof of concept on a small part of the rodent brain, and with this proof of concept we are now scaling the project up to reach the human brain.”
Blue Brain Founder Responds to Critics, Clarifies His Goals (11 Feb 2011), Science news
Bluebrain: Noah Hutton's 10-Year Documentary about the Mission to Reverse Engineer the Human Brain (9 Nov 2012), an indispensable interview with Ferris Jabr in Scientific American
European neuroscientists revolt against the E.U.'s Human Brain Project (11 July 2014), Science news
Row hits flagship brain plan (7 July 2014), Nature news
Brain Fog (7 July 2014), Nature editorial
Human Brain Project votes for leadership change (4 March 2015), Nature news
'In Silico:' Director Noah Hutton reveals how one neuroscientist's pursuit of perfection went awry (10 Nov 2020), another indispensable interview, this time with Nadja Sayej in Inverse
“They still haven’t even simulated a whole mouse brain. I realized halfway through the 10-year point that the human brain probably wasn’t going to happen.” ...In the first few years, I followed only the team. Then, I started talking to critics.
"my inclination is to say that we will never (never say never) be able to build a realistic computer simulation of the human brain."
ReplyDeleteOf course. But that's because it's (a) the wrong question, and (b) premature. One of the speakers in the video you linked in your previous post argued that it's the cognitive end of cognitive neuroscience, not the neuro end, that's harder. She was, of course, correct. When listening to folks in the cognitive fields, I sometimes pull my hair out and scream at the screen, "You idiot, "cognitive" means "to think", and you've forgotten that we humans actually think*". All of which is to say, if you don't know what thinking is, even if you succeeded in simulating a human brain, you wouldn't learn anything, and your simulation wouldn't do anything, either. People take 3 years or so before they begin to have enough language to talk, and require appropriate inputs to make that happen.
And, of course, premature, because the AI/cog. sci. types who were trying to figure out how humans think (a) didn't get very far (when I'm grumpier I say "failed miserably", but that's subject to the "speak for yourself" criticism) and (b) got drowned out by the neural network hype. (Gary Marcus is trying to get the AI end of this area back on track, so at least someone's noticed the mess.) (Of course, this presupposes that "human thought" actually is "symbolic reasoning" and that can actually be represented/implemented in computational models. But I believe this to be true.)
*: I saw an interview with George Lakoff the other day, and was quite depressed. Even though he gets it that people think, and he's right that metaphor** is how (at least some of) language works, he still doesn't get it how amazing human cognitive abilities are (he thinks that metaphors are limited to "embodied" base concepts (front/back, up down, and the like), and that's just silly; people aren't that limited at all. Sigh. Still, it's be nice if the AI types could come up with computational models that get mileage from the metaphor idea.
**: I'm trying to push my Japanese vocab to the next level, so when a word trips me up, I make a flash card with the word and the English glosses from a Japanese->English dictionary. When I find a word that I can't remember for the life of me, I look it up in a Japanese->Japanese dictionary, translate the definitions of the different senses and copy the examples to my flash card. Then I can sometimes remember it. What's usually going on is that the main usage is a metaphorical extension from an historical/original usage, but the J->E dictionary glosses don't make it clear what's going on there. Thank you, George L., I find myself saying. A lot.
David in Tokyo, where even the footnotes have footnotes.
(I hope my ranting isn't getting repetitive: I'm not doing enough reading in this area.)
The funny thing is nobody seems to realize HBP is/was foremost a sneak infrastructure- and tool building project, with the "simulate a whole brain" shtick as PR bait for the funders and surrounding society, because non-scientists would never approve that huge amount of funding for infrastructure. The aim: to build infrastructure advanced enough to support future hypermodern neuroscience, attached to a community using modern tools. The now-in-production EBRAINS and the surrounding tool ecosystem is the real heritage of HBP.
ReplyDeleteAnonymous - Thanks very much for pointing that out. NIH BRAIN is more transparent about that, I suppose.
ReplyDeleteDavid in Tokyo - Thanks, I appreciate your insights. I've admired Lakoff's academic work on metaphors, but I stopped paying attention after his popular writing on framing had zero impact on the outcome of the US presidential election in 2015. This is not a unique failing on his part -- in general, mainstream social & cognitive psychology has not been able to influence (or even touch) the disintegration of "truth" and political discourse in America.