the demon of the belfry

e
spring 2021 · volume 17 · issue 1

investigating the evolution of the penny

press and the emmanuel church murders

ALSO FEATURING
E L E M E N T S • The Digital Pandemic
the undergraduate research journal of boston college • The Force on the Front
 neuromorphic computing
Explaining how Projected SNN Training will Largely Impact
our Interactions with Technology
will riherd

in the face of increasingly large computational demands and the impending halt to

moore’s law, the semiconductor industry has been forced to re-evaluate the tradi-

tional computing paradigm. central to this re-evaluation has been the novel devel-

opment of neuromorphic computing–an approach that, at its core, seeks to replicate

the brain in silicon. despite challenges on the algorithmic front, neuromorphic

computing promises a massively parallel, efficient, and scalable computational solu-

tion with large implications on the daily lives of consumers. the future of the tech-

nology, however, is uncertain. with the rise of high performance and quantum com-

puting as promising alternatives, the semiconductor industry at large must consider

the extent to which neuromorphic computing can emerge as a viable and feasible

solution in the coming years. it is vital, therefore, to understand the theoretical

underpinnings of the neuromorphic approach and predict the likelihood of its im-

plementation within the next decade.

ed on the algorithmic front. In terms of research and re- learned parameters to an SNN (Tiernan, “Neuromorphic
sources, in October of 2020, Intel announced a three-year computing finds new life”). While parameter transferring
contract with Sandia National Laboratories (a government is a step in the right direction, Sejnowski admits that the
nuclear research lab) to advance neuromorphic computing next step—learning the SNN itself—is still “in the early
beyond edge applications to progressively complex compu- days,” but that there is “going to be another big shift, which
tational problems. The result will be the development of will probably occur within the next five to ten years.” Given
increasingly large neuromorphic systems with the intent both the difficulty and current progress of SNN training,
of integrating over one billion neurons. Creating a system there is moderate uncertainty regarding how quickly 35
with such neural capacity exhibits moderate uncertainty in development will occur on the algorithmic front, with a
the short-term. Given a) the progress Intel has made when general uncertainty factor of one and one-half to two times
it comes to scaling Loihi and b) the difficulty of integrating on the high-end. If an efficient learning rule is established,
an increasingly large number of neurons, the project has it is possible we will begin to see commercially viable
an uncertainty factor of around two on the low- and high- neuromorphic chips within the decade, allowing for
end (i.e., the system will be developed within one half to complex, energy efficient tasks on the “edge” (e.g.,
two times the expected implementation time frame). Neu- smartphones). Such commercial implementation,
romorphic computing will also experience strong research however, is highly dependent on the economic feasibility
in edge applications (particularly energy efficient robotics) of widespread production, an aspect of neuromorphic chip
as members of the INRC continue to deploy Loihi in real- development that is still in its infancy.
time audio and video processing tests.11 Long Term (10+ Years)
While Gartner predicts that traditional computing will hit Although the future of SNN training and the economic
a wall in 2025 (specifically mentioning neuromorphic practicality of neuromorphic hardware is uncertain in the
computing as the leading technology for a shift in the com- near-term, it is possible that beyond 2030 we begin to wit-
puting paradigm), it is unlikely that a robust commercial ness a shift in the current computing paradigm. Such a
product will be available within the next five years due to shift would be characterized by the wide-spread use of
software limitations. Unlike traditional ANNs, SNNs lack neuromorphic chips in energy-constrained edge devices
a learning algorithm when it comes to parameter optimi- (e.g., mobile devices, computers, robotics, etc.). Sejnowski
zation: it is impossible to use gradient descent given the predicts that once the software is developed and the hard-
non-differentiable nature of spiking neurons (Soni).12 ware is “sufficiently cheap,” the implementation of such
Yann LeCun, the head of AI at Facebook, is a leading skep- chips will be “ubiquitous...like sensors in phones.” Abu
tic when it comes to neuromorphic implementation, writ- Sebastian, Principal Research Staff Member at IBM Zu-
ing that it is “premature to build a chip” when SNNs face a rich, predicts that neuromorphic computing will play a vi-
fundamental training challenge.13 There is also skepticism tal role in the future feasibility of autonomous vehicles,
regarding the performance of the underlying SNNs; spe- making the point that you cannot “collect a frame, pass it
cifically, skeptics point to the fact that SNNs do not outper-
form ANNs (e.g. Convolutional Neural Networks) when it
comes to traditional tests of image recognition (Pfeiffer
and Pfeil).14

Medium Term (5-10 Years)

While SNNs currently face learning challenges, it is
expected that within the next five to ten years, researchers
will develop novel and efficient approaches to SNN
training. Current advancements have largely been focused
on implementing a form of transfer learning (i.e., a process
by which parameters are optimized in one “place” and
transferred to a new “place”). In March of 2019, Terry
Sejnowski of The Salk Institute of La Jolla and his team
were able to train a standard recurrent neural network artificial synapses based on ferro-electric tunnel
(RNN) via gradient descent methods and transfer the junctions (courtesy of wikimedia commons)

neuromorphic computing
 through to a deep neural net, and wait for the response treme integration of artificial intelligence into our daily
when you’re traveling down a freeway at 70 miles an hour” lives (Ferry). Suma describes a future in which a technol-
(Greengard). The long-term draw of neuromorphic com- ogy like Siri transcends basic voice commands (i.e., a Siri
puting, is energy-efficient, on-board processing that is ap- that listens and sees all of your conversations and interac-
plicable to a wide range of technologies, from autonomous tions). If you were to ask Siri what idea your friend Melissa
vehicles to independently operating interstellar spacecraft. came up with regarding your wife’s birthday gift, the effi-
Neuromorphic chips are also poised to play a large role in cient, continuous, on-board processing of data made pos-
36 future medical devices and artificial body parts that can sible by neuromorphic computing would enable Siri to re-
benefit from increasingly fast, on-site processing. call the conversation with your friend the week prior,
bringing up both the idea and a variety of similar, person-
While neuromorphic architecture could become ubiqui- alized ideas. The salient point in Suma’s eyes is that the
tous beyond 2030, it is important to note the false dichoto- efficiency of neuromorphic chips would enable the local
my between standard computer architecture and neuro- storage of information, resolving widespread privacy con-
morphic hardware.15 The promise of neuromorphic chips cerns regarding consumer data. From autonomous vehi-
lies in the creation of an adaptive system capable of the cles to smart homes and edge devices, Suma’s point is
efficient processing of highly noisy, increasingly complex, clear: the promise of neuromorphic technology is a future
spatio-temporal data, not the high-speed processing of pre- characterized by the extreme integration of artificial intel-
dictable, deterministic processes (Greengard). It is short- ligence into our daily lives.
sighted, therefore, to view neuromorphic computing as a Semiconductor Industry
fundamental alternative to traditional computing: the two
technologies are complementary to each other, excelling in Research scientist at Hewlett Packard Enterprise, Suhas
tasks suited to their specific design. It is also important to Kumar, summarizes the state of computing as a “huge
note the uncertainty surrounding the long-term ubiquity rush to find something” that can continue the improve-
of neuromorphic computing, namely, that there is a three- ment in computer science that we have witnessed the past
way race between neuromorphic computing, high perfor- half century. Semiconductor manufacturers, therefore, are
mance computing, and quantum computing (Vorhies).16 faced with a critical choice of whether to invest in risky
Given the advancements of competing fields, it is possible chip research to compete in a future characterized by ad-
the technology is outperformed before it is commercially vanced computing technologies. While IBM has seen suc-
viable (either by existing competitors or a novel approach cess with its TrueNorth system, Intel has exhibited the
yet to be developed). The future of computing, then, is strongest commitment to neuromorphic research, evi-
largely undecided, with the only certainty being that our denced by its continued progress in creating increasingly
current computing paradigm will shift in response to the scalable neuromorphic systems. Intel’s competitors are
impending halt to Gordon Moore’s 1965 prediction. also faced with the challenge of determining the type of
research they wish to undertake (i.e., whether they believe
high performance or quantum computing will ultimately
IMPLICATIONS
outperform the neuromorphic approach). Regardless, the
Consumers and AI Integration challenge is clear: the computing industry will witness a
Despite uncertainties regarding the path of computing, a shift in the coming decades, and the companies that lead
future characterized by the widespread adoption of neuro- this shift will capitalize on the widespread adoption of
morphic technology would have significant implications their novel technology.
on the way consumers interact with digital devices. Peter
Suma, co-CEO of Applied Brain Research, imagines a
world in which neuromorphic technology enables an ex-

“The promise of neuromorphic chips lies in the cration of

an adaptive system capable of the efficient processing of
highly noisy, increasingly complex, spatio-temporal data”

elements : : spring 2021

Large Technology Firms the brain’s neural wiring.
Beyond the semiconductor industry, large technology 3. The downside of traditional computer architecture is typically
companies (e.g., Google, Facebook, Amazon, etc.) are referred to as the von Neumann bottleneck—the notion that
despite increases in processing speed, processors are forced to
faced with the decision to embark upon internal research
remain idle while data is transferred to and from the memory of
regarding computing hardware. As deep learning process- a chip, resulting in increased latency.
es continue to surpass the capabilities of existing technol-
ogy, companies with large workloads and available capital 4. Neuromorphic chips employ analog circuitry to transfer elec-
are faced with the decision of whether to invest internally
trical signals between “neurons.” The idea is that the system will 37
be able to modulate the amount of electricity flowing between
in chip development. In February of 2019, Facebook’s nodes, mimicking the fact that brain signals naturally have vary-
Yann LeCun briefly described “internal activities” the firm ing degrees of strength.
was taking to address the growing computational divide,
5. A key emphasis of neuromorphic computing is the selective
referencing Google’s Tensor Processing Unit as another mapping of neural connections. In order to operate at maximum
example of the type of internal efforts large technology efficiency, the brain uses only the specific neurons and synapses
firms are capable of taking (Tiernan, “Facebook’s Yann Le- necessary to perform a given task. Neuromorphic systems,
Cun”). While the future is uncertain, it is clear there is a therefore, seek to emulate this efficiency by strategically and
demand for change, and the feasibility of neuromorphic selectively forming connections between neighboring neurons.
hardware should figure prominently in the minds of firm 6. Each “spike” in an SNN is a single-bit impulse that is analo-
leaders who are considering undertaking or continuing in- gous to an action potential in a naturally occurring neuron. A
ternal development activities. given node in the network is capable of spiking only if a state
variable exceeds a given threshold.
CONCLUSION 7. In fully-connected ANNs, all nodes compute and send their
output to the next layer at each time step, even if nothing has sig-
From computer science and biology to mathematics, phys- nificantly changed, making the overall network computationally
ics, and electrical engineering, neuromorphic computing expensive. Conversely, SNNs employ neurons that process input
is a cross-disciplinary challenge that requires a fundamen- separately and are only connected to local neighbors, implying
tal rethinking of the way computers operate. Neuromor- that the entire layer does not need to be calculated for informa-
phic research, however, is progressing: firms like Intel are tion to proceed to the next layer.
leading the charge in chip development, and the success of 8. In 1965 Intel’s co-founder Gordon Moore predicted that the
their efforts will have large-scale implications on consum- number of transistors that could be placed on an integrated
ers, the semiconductor industry, and technology firms circuit would double every two years while costs remain constant
alike. It is vital, therefore, that business leaders consider / decline.
the possibilities and threats of this novel technology, un- 9. Loihi comprises 128 neuromorphic cores, 131,000 “neurons”,
derstanding that the future computing paradigm will be and 130 million “synapses” (connections). The chip is named
decided by innovative firms who are rethinking current after an active submarine volcano off the coast of Hawaii that is
processes. set to emerge one day. The idea is that neuromorphic computing
is analogously “emerging” and will eventually break the “surface”
of the current computing environment.
10. In December 2018 Intel announced the creation of Kapoho
Bay, Intel’s smallest neuromorphic system consisting of two
endnotes Loihi chips. In July of 2019 Intel was able to scale Loihi into their
1. The term neuromorphic means “taking the form of the brain” 64-chip Pohoiki Beach, representing the neural capacity of 8 mil-
(Fulton III). Intel (a leader in neuromorphic research) summa- lion neurons. Most recently, Intel debuted its 768-chip Pohoiki
rizes neuromorphic technology as “chips that function less like Springs with a collective neural capacity of 100 million neurons.
traditional computers and more like the human brain” (“Intel The chips reside in a chassis the size of five standard servers
Newsroom–Neuromorphic Computing”). and is provided to members of the Intel Neuromorphic Research
Community via a cloud-based system. Intel hopes it will emerge
2. It can be argued that the original notion of neuromorphic as a “tool for researchers to develop and characterize new neuro-
computing was first proposed in Alan Turing’s 1948 paper, Intel- inspired algorithms for real-time processing, problem solving,
ligent Machinery. However, the concept is typically attributed to adaptation and learning” (“Intel Newsroom–Pohoiki Springs”).
Carver Mead’s 1989 Analog VLSI and Neural Systems, a paper
in which Mead argued that chips with an increasingly dense col- 11. In July of 2020 the National University of Singapore ran an
lection of transistors could best communicate via a replication of event-driven, visual-tactile perception test on Intel’s Loihi chip

neuromorphic computing
 as well as various GPU systems to compare power consump- NanoSystems Institute at the University of California at Los An-
tion (Russel). The Loihi test chip slightly outperformed the GPU geles, further describes how “conventional von Neumann-based
systems (in inferences per second) at significantly lower power computing systems” perform very well with “high-speed, pre-
consumption. dictable, deterministic processes,” but struggle with increasing
complexity. The promise of neuromorphic computing, therefore,
is the opening up of “an entirely new and unexplored area of
computing”—one that allows us to “do things with computers
that we couldn’t have imagined in the past” (See Greengard).
38 16. High-performance computing (HPC) can be generalized as
a process of optimizing chip architecture for existing deep learn-
ing algorithms (Vorhies). The majority of attention is currently
placed on HPC, with large technology firms beginning to enter
the chip manufacturing field by developing their own propri-
etary chips (e.g., Google’s Tensor Processing Unit) (Tiernan,
It is important to note that Loihi’s strong performance is largely “Facebook’s Yann LeCun”). Conversely, quantum computing is
the result of the entire system (in this case, robotic sensors, data analogous to neuromorphic computing in that it represents a
formats, algorithms, and the Loihi architecture) being “re-engi- rethinking of existing computer architecture. Rather than operat-
neered in an event-based paradigm” (Mike Davies, Director of ing in bits (0 or 1), quantum computing employs quantum bits
Neuromorphic Research at Intel. See Russel). Put simply, when (qubits) that can be set to 0, 1, or both simultaneously (Cho).
the system is consistent with Loihi’s event-based architecture,
neuromorphic hardware outperforms traditional GPU’s when it
comes to energy efficiency.
12. Existing deep learning algorithms depend on stochastic references
gradient descent and error backpropagation to efficiently “learn”
a given ANN. Since SNNs operate in discontinuous, non-differ-
Best, Jo. “Neuromorphic Computing Could Solve the Tech
entiable spikes, it is impossible to apply existing learning rules Industry’s Looming Crisis.” ZDNet, 5 Oct. 2020, `
directly to SNN training (Pfeiffer and Pfeil). www.zdnet.com/article/neuromorphic-computing-could-
solve-the-tech-industrys-looming-crisis/.
13. At the 2019 International Solid State Circuits Conference in
San Francisco, LeCun heavily criticized neuromorphic comput-
ing in his opening keynote address. While Intel’s Mike Davies Blain, Loz. “Intel’s Neuromorphic Loihi Chip Is Rapidly
agrees that there is a lot of progress to be made on the algorith- Learning to Discern Smells.” New Atlas, 17 Mar. 2020, ne-
mic front, which he claims is “holding back the field,” Yann’s watlas.com/science/intel-neuromorphic-loihi-smells/.
criticism resulted in a fireback from Davies citing the efficacy of
neuromorphic chips in a December report produced by Applied
Cho, Adrian. “The Biggest Flipping Challenge in Quan-
Brain Research of Waterloo, Ontario. Davies’s fireback was met
with a detailed Facebook post by LeCun outlining the issues he tum Computing.” Science, The American Association For
sees with neuromorphic hardware, specifically citing the lack of The Advancement of Science, 9 July 2020, www.sci-
an efficient training algorithm. encemag.org/news/2020/07/biggest-flipping-challenge-
14. It is important to point out that SNNs are not optimized for quantum-computing.
performance on existing AI Benchmarks (e.g., ImageNet). Just
how the brain is not optimized (but capable of) classifying an Davies, Mike, et al. “2018 CES: Neuromorphic Computing
image that is quickly flashed on the retina, SNNs struggle with Mimics the Human Brain.” Youtube, Intel Newsroom, 8
the typical frame-based test of software accuracy. The evaluation Jan. 2018, www.youtube.com/
of SNNs, therefore, requires benchmark testing that emphasizes watch?v=gXNCz26UhyY&feature=emb_title.
the functionality of spiking networks—i.e., “making decisions
based on continuous input streams while moving in the real
world” (Pfeiffer and Pfeil). Ferry, Jeff. “The Future of AI Is Neuromorphic.” Circuit
Insight, www.circuitinsight.com/programs/55035.html.
15. Dan Hutcheson, CEO of VLSI Research (an independent
market analysis and consulting firm that tracks the semiconduc-
tor industry), describes the false dichotomy as follows: “Today’s Fulton III, Scott. “What Neuromorphic Engineering Is,
computers are very good at what they do. They will continue to and Why It’s Triggered an Analog Revolution.” ZDNet, 8
outperform neuromorphic computing systems for conventional Feb. 2019, www.zdnet.com/article/what-neuromorphic-
processing tasks. The technologies are complementary and so engineering-is-and-why-its-triggered-an-analog-revolu-
they will coexist.” Adam Stieg, associate director of the California tion/.

elements : : spring 2021

Greengard, Samuel. “Neuromorphic Chips Take Shape.” Ray, Tiernan. “Facebook’s Yann LeCun Says ‘Internal Ac-
Communications of the ACM, 1 Aug. 2020, cacm.acm. tivity’ Proceeds on AI Chips.” ZDNet, 18 Feb. 2019, www.
org/magazines/2020/8/246356-neuromorphic-chips- zdnet.com/article/facebooks-yann-lecun-says-internal-ac-
take-shape/fulltext. tivity-proceeds-on-ai-chips/.

Hamblen, Matt. “Intel Shares Pohoiki Springs Chassis Ray, Tiernan. “Intel’s Neuro Guru Slams Deep Learning:
with AI Researchers.” FierceElectronics, 18 Mar. 2020, ‘It’s Not Actually Learning’.” ZDNet, 23 Feb. 2019, www.
www.fierceelectronics.com/electronics/intel-shares-pohoi- zdnet.com/article/intels-neuro-guru-slams-deep-learning- 39
ki-springs-chassis-ai-researchers. its-not-actually-learning/.

Irving, Michael. “Artificial Synapses Fill the Gaps for Ray, Tiernan. “Neuromorphic Computing Finds New Life
Brainier Computer Chips.” New Atlas, 23 Jan. 2018, newat- in Machine Learning.” ZDNet, 1 July 2019, www.zdnet.
las.com/neuromorphic-chip-mit/53081/?itm_ com/article/neuromorphic-computing-finds-new-life-in-
source=newatlas. machine-learning/.

Jhonsa, Eric. “Intel Teams with Uncle Sam to Research Rouse, Margaret. “What Is Neuromorphic Computing?”
Chip Packages, ‘Neuromorphic’ Systems.” TheStreet, 2 SearchEnterpriseAI, TechTarget, 11 Feb. 2020, searchen-
Oct. 2020, www.thestreet.com/investing/intel-teams- terpriseai.techtarget.com/definition/neuromorphic-com-
with-uncle-sam-to-research-chip-packages-neuromorphic- puting.
systems-loihi-foveros-emib.
Russell, John. “Get a Grip: Intel Neuromorphic Chip Used
LeCun, Yann. ISSC Keynote Address. Facebook, 23 Feb. to Give Robotics Arm a Sense of Touch.” HPCwire, 17 July
2019, https://m.facebook.com/story.php?story_fbid=1015 2020, www.hpcwire.com/2020/07/15/get-a-grip-intel-
5808247852143&id=722677142. neuromorphic-chip-used-to-give-robotics-arm-a-sense-of-
touch/.
Molter, Tim. “A Brief History of Neuromorphic Comput-
ing.” Knowm.org, 4 Apr. 2015, knowm. Soni, Devin. “Spiking Neural Networks, the Next Genera-
org/a-brief-history-of-neuromorphic-computing/. tion of Machine Learning.” Medium, Towards Data Sci-
ence, 16 July 2019, https://towardsdatascience.com/spik-
Muir, Dylan. “Putting the Neural Back into Networks.” Me- ing-neural-networks-the-next-generation-of-machine-
dium, 15 Jan. 2020, medium.com/neuromorphic-intelli- learning-84e167f4eb2b
gence/putting-the-neural-back-into-networks-f0fd-
b19a5efd. Vorhies, William. “The Three Way Race to the Future of
AI. Quantum vs. Neuromorphic vs. High Performance
Norman, Jeremy M. “Carver Mead Writes the First Book Computing.” Data Science Central, Tech Target, 14 Nov.
on Neuromorphic Computing.” Carver Mead Writes the 2017, www.datasciencecentral.com/profiles/blogs/the-
First Book on Neuromorphic Computing : History of In- three-way-race-to-the-future-of-ai-quantum-vs-neuromor-
formation, 25 Oct. 2020, www.historyofinformation.com/ phic-vs.
detail.php?id=3900.
Wiggers, Kyle. “Intel Debuts Pohoiki Springs, a Powerful
Panetta, Kasey. “Gartner Top 10 Strategic Predictions for Neuromorphic Research System for AI Workloads.” Ven-
2021 and Beyond.” Smarter With Gartner, 21 Oct. 2020, tureBeat, 18 Mar. 2020, venturebeat.com/2020/03/18/in-
www.gartner.com/smarterwithgartner/gartner-top- tel-debuts-pohoiki-springs-a-powerful-neuromorphic-re-
10-strategic-predictions-for-2021-and-beyond/. search-system-for-ai-workloads/.

Pfeiffer, Michael, and Thomas Pfeil. “Deep Learning With “Neuromorphic Computing - Next Generation of AI.” In-
Spiking Neurons: Opportunities and Challenges.” Fron- tel–Neuromorphic Computing, Intel, 2020, www.intel.
tiers in neuroscience vol. 12 774. 25 Oct. 2018, doi:10.3389/ com/content/www/us/en/research/neuromorphic-com-
fnins.2018.00774. puting.html.

neuromorphic computing
 “Intel and Sandia National Labs Collaborate on Neuromor-
phic Computing.” Intel Newsroom, Intel, 2 Oct. 2020,
newsroom.intel.com/news/intel-sandia-national-labs-
neuromorphic-computing/.

“Intel Scales Neuromorphic Research System to 100 Mil-

lion Neurons.” Intel Newsroom, Intel, 18 Mar. 2020,
40 newsroom.intel.com/news/intel-scales-neuromorphic-re-
search-system-100-million-neurons/.

“Intel’s Pohoiki Beach, a 64-Chip Neuromorphic System,

Delivers Breakthrough Results in Research Tests.” Intel
Newsroom, Intel, 15 July 2019, newsroom.intel.com/
news/intels-pohoiki-beach-64-chip-neuromorphic-sys-
tem-delivers-breakthrough-results-research-tests/.

“Loihi - Intel.” WikiChip, 15 July 2019, en.wikichip.org/

wiki/intel/loihi.

elements : : spring 2021

list of artwork

13 the battle of the somme comic: the story of william mcfadzean

© Creative Centenaries. (n.d.). The Battle Of The Somme comic: The story of William McFadzean | CCEA. CEA.
Retrieved January 19, 2022, from https://ccea.org.uk/learning-resources/understanding-1916/creative-centena-
ries/battle-somme-comic-story-william

15 easter proclamation of 1916

© originally uploaded to the English Wikipedia by w:User:Jtdirl (https://commons.wikimedia.org/wiki/
File:Easter_Proclamation_of_1916.png), „Easter Proclamation of 1916“, marked as public domain, more details
on Wikimedia Commons: https://commons.wikimedia.org/wiki/Template:PD-old

28 figure 1
© originally tweeted by User:@realDonaldTrump (account suspended) on 26 May, 2020.

28 figure 2
© GDI report: Ad-funded COVID-19 Disinformation: Money, Brands, and Tech (https://disinformationindex.
org/wp-content/uploads/2020/07/GDI_Ad-funded-COVID-19-Disinformation-1.pdf), „Share of ad revenues
generated by COVID Disinfo Sites“.

28 figure 3, panel a
© originally published in Bursztyn, Rao, Roth, Yanagaziwa-Drott: Misinformation During a Pandemic (https://
www.nber.org/papers/w27417), „RF estimates of effect of differential viewership on cases and deaths (extend-
ed)“.

28 figure 3, panel b
© originally published in Bursztyn, Rao, Roth, Yanagaziwa-Drott: Misinformation During a Pandemic (https://
www.nber.org/papers/w27417), „2SLS estimates of effect of differential viewership on cases and deaths (ex-
tended)“.

29 figure 4, panel a
© originally published in Bursztyn, Rao, Roth, Yanagaziwa-Drott: Misinformation During a Pandemic (https://
www.nber.org/papers/w27417), „Implied COVID-19 Curves, Panel A: Estimates on Cases“.

29 figure 4, panel b
© originally published in Bursztyn, Rao, Roth, Yanagaziwa-Drott: Misinformation During a Pandemic (https://
www.nber.org/papers/w27417), „Implied COVID-19 Curves, Panel B: Estimates on Deaths“.
 35 artificial synapses based on ferro-electric tunnel junctions
© Mike Davies Intel Labs, Intel Corporation et. al: (https://commons.wikimedia.org/wiki/File:Artificial_syn-
apses_based_on_FTJs.png), https://creativecommons.org/licenses/by-sa/4.0/legalcode

38 table iii inference power and power utilization

© Tasbolat Taunyazov, Weicong Sng, Hian Hian See, Brian Lim, Jethro Kuan, Abdul Fatir Ansari, Benjamin
C.K. Tee, and Harold Soh: (http://www.roboticsproceedings.org/rss16/p020.pdf), „Table III: Inference Speed
and Power Utilization“

43 blache lamont at hecla, montana, with her students, 1893

© Unknown authorUnknown author (https://commons.wikimedia.org/wiki/File:Blanche_Lamont_with_stu-
dents.jpg), „Blanche Lamont with students“, marked as public domain, more details on Wikimedia Commons:
https://commons.wikimedia.org/wiki/Template:PD-US

45 theodore durrant’s mugshot

© originally published in Report of the trial of William Henry Theodore Durrant (https://archive.org/details/
reportoftrialofw00durr/page/10/mode/2up), marked as public domain.

60 graph illustrating the shifts in price and quantity supply

© originally published in 31.11 Efficiency and Deadweight Loss (https://saylordotorg.github.io/text_economics-
theory-through-applications/s35-11-efficiency-and-deadweight-loss.html), „Figure 31.8 Deadweight Loss“

61 graphs showing the number and value of mergers and acquisitions

© originally published in Statista: Mergers and acquisitions in the United States (https://www.statista.com/
study/65232/mergers-and-acquisitions-in-the-united-states).

61 graphs showing global query share by seelct search engines

© originally published in Statista: Global market share of search engines 2010-2022 (https://www.statista.com/
statistics/216573/worldwide-market-share-of-search-engines).

61 graphs showing domestic us search query share by select search engines

© originally published in Statista: Market share of search engines in the United States 2008-2022 (https://
www.statista.com/statistics/267161/market-share-of-search-engines-in-the-united-states).
list of artwork

62 corporate technology information services

© unpublished tabulation from Corptech Database

62 the breakdown of opinion regarding breaking up ‘big tech’

© Statista : Majority of Americans in Favor of Breaking up Big Tech (https://www.statista.com/chart/19440/
survey-responses-breaking-up-big-tech)

71 figure 1
© Samadi, Moradi, Moradinazar, Mostafai, Pasdar: Dietary pattern in relation to the risk of Alzheimer’s disease:
a systematic review (https://link.springer.com/article/10.1007/s10072-019-03976-3), „Figure 2: Healthy and
unhealthy diet in relation to AD.“

75 potential signatories, signatories, and parties to the aiib

© L.tak (https://commons.wikimedia.org/wiki/File:Asian_Infrastructure_Investment_Bank_map.svg),
https://creativecommons.org/licenses/by-sa/4.0/legalcode

78 china’s ongoing territorial disputes]

© Office of the Secretary of Defense (https://commons.wikimedia.org/wiki/File:China’s_Disputed_Territories.
PNG), „China’s Disputed Territories“, marked as public domain, more details on Wikimedia Commons: https://
commons.wikimedia.org/wiki/Template:PD-US

80 us department of justice among others announced 23 criminal charges

© U.S. Department of Justice (https://commons.wikimedia.org/wiki/File:U.S._Commerce_Secretary_Wilbur_
Ross,_Acting_Attorney_General_Matthew_Whitaker,_Homeland_Security_Secretary_Kirstjen_Nielsen,_FBI_
Director_Christopher_Wray_Announces_23_Criminal_Charges_Against_China’s_Huawei_&_Wanzhou_
Meng.jpg), „U.S. Commerce Secretary Wilbur Ross, Acting Attorney General Matthew Whitaker, Homeland
Security Secretary Kirstjen Nielsen, FBI Director Christopher Wray Announces 23 Criminal Charges Against
China’s Huawei & Wanzhou Meng“, marked as public domain, more details on Wikimedia Commons: https://
commons.wikimedia.org/wiki/Template:PD-US

82 america global trade deficit

© Statista: U.S. Trade Deficit in Goods Reaches Record High, (https://www.statista.com/chart/17281/us-trade-
balance).
 98 charles coughlin, leader of the christian front
© Craine, Detroit (https://commons.wikimedia.org/wiki/File:CharlesCouglinCraineDetroitPortrait.jpg),
„CharlesCouglinCraineDetroitPortrait“, marked as public domain, more details on Wikimedia Commons:
https://commons.wikimedia.org/wiki/Template:PD-US

100 german american bund rally, new york, madison square garen, feb 1939
© Department of Defense. Department of the Army. Office of the Chief Signal Officer. (09/18/1947 -
02/28/1964); ARC Identifier 36068 / Local Identifier 111-OF-2 1943 (https://commons.wikimedia.org/wiki/
File:Nazi_Strike_2.jpg), „Nazi Strike 2“, marked as public domain, more details on Wikimedia Commons:
https://commons.wikimedia.org/wiki/Template:PD-US