Trolls vs. Wizards

By Gully A. Burns

“Trolls were large monsters of limited intellect. They were strong and vicious, but they could not endure sunlight”
- tolkiengateway.net/wiki/Trolls

We live in interesting times. On the one hand, social media drives public discourse into polarized shouting matches. Astroturfing bots contort the information landscape by pushing false narratives on Twitter with incendiary, ill-informed talking points. The ease of web-publishing makes it relatively easy to drown out informative scientific work by spreading misinformation in coordinated online media campaigns.

But also, tech advances drive public engagement and citizen scientists contribute directly to research more than ever before. It’s easier to any number of technically-demanding subjects with Massive Open Online Courses (MOOCs). The information infrastructure of science itself is evolving to change and accelerate the path towards discovery.

These two aspects are sharply opposed. The first approach is based on misinformation, manipulation, provocation, and storytelling. Let’s call the instigators of such methods ‘Trolls’. The second approach is based on expertise, hard work, and a nuanced view of reality in order to find effective explanations for real phenomena. This work requires diligence, honesty, intelligence, and patience. Let’s call anyone pursuing this endeavor ‘Wizards’. Thus, I frame this argument as the perrenial struggle between wizards and trolls.

How to fight trolls - a wizard’s field guide.

Scientific discussions are typically based on abductive reasoning: argument from evidence to the best available explanation. In the blood sport where teams of trolls strive to defeat their opponents by any means necessary, the meagre weapons of abductive reasoning can do little to withstand the assault of scorn, derision, and ‘bullshit’.

But, knowledge is power. Science is, after all, very much like magic. Science is the basis for technology, and try as they might, trolls cannot accomplish anything remotely as powerful. They can sway public opinion, but they cannot cure disease or prevent natural disasters. They cannot send ships into space or understand the mysteries of fluid dynamics in the upper stratosphere.

We must enable the explanatory power of scientific knowledge to address these counter-arguments as explicitly and powerfully as possible. At present, the scientific community does not possess the necessary power to challenge the trolls. This is was best said by Ben Goldacre in the last chapter of his book ‘Bad Science: Quacks, Hacks, and Big Pharma Flacks’. He writes:

“To anyone who feels their ideas have been challenged by this book or who has been made angry by it - to the people who feature in it, I suppose - I would say this: you win. You really do. I would hope that there might be room for you to reconsider, to change your stancein the light of what might be new information (as I will happily do, if there is ever an opportunity to update this book). But you will not need to because, as we both know, you collectively have almost full spectrum dominance. Your ideas - bogus though they may be - have immense superficial plausibility, they can be expressed rapidly, they are endlessly repeated, and they are believed by enough people for you to make very comfortable livings and to have enormous cultural influence. You win.”

Personally, I am not comfortable with this easy capitulation. I prefer to take a leaf out of Gandalf’s playbook even when faced with a daunting powerful foe.

Gandalf vs. the Balrog

Countering the troll ‘full spectrum dominance’ will require the development of transformative approaches to remove barriers that laypeople experience in understanding complex scientific concepts. We will need to make the work more inclusive and more democratic. We may need to transform the way scientists work and the way science is taught. This will require impeccability, creativity, honesty, and courage.

Data Science Wizardry

So, here are some observations and strategies to consider going forward.

  1. Modern information technology is changing many aspects of scientific work, creating new opportunities and paradigms. Amongst these methods include:
    • The scalability and speed of ‘big data’ systems allows easy analysis of very large data sets
    • Deep learning and modern machine-learning methods provide groundbreaking AI performance in tasks like Data mining, Natural Language Processing, Image Classification and Document Processing.
    • In particular, deep reinforcement learning permits robots and automated systems to win games and learn complex behaviors
    • Forecasting methods provide powerful new methods of generating predictions based on data.
    • Social Media research permits us to track and understand some of the impact of misinformation in society.
    • Science information infrastructure is beginning to provide systems for reproducibility, scalability and increased rigor such as Workflows, Ontologies, Information Integration systems.
    • A key capabilty is tracking the provenance of knowledge: What is the evidence that supports our argument that a given claim is true? This is an area of continuing, active research, (see Wadden et al. 2020).
    • Creative methods for data visualization are becoming more common and more powerful.
  2. Attempts where scientists have adopted ‘trollish’ methods failed spectacularly.
    • When climate researchers in England attempted to convert their scientific perspective into a political strategy, leaks of their emails that revealed their attempts to construct a robust and compelling poitical argument involved ham-fisted attempts to silence critics and steer the conversation. ‘Climategate’ ensued, doing real and lasting harm both to the climate change debate and also to the credibility of academics as a whole.
    • More recently, attempts to create an expedient, powerful, simple talking point was the often-quoted ‘climate consensus’ figure. This was the product of a 2014 paper that boldly states: “Among self-rated papers expressing a position on AGW (anthropogenic global warming), 97.2% endorsed the consensus” (ref: Cook et al 2014). Unfortunately, even a simple review of this study’s own data, reveals vulnerabilities in the study that are easily revealed, see this blog post: ‘a climate falsehood you can check yourself’. Attempting to use the dark arts of spin when creating a scientifically-driven policy argument is a bad idea.
    • At the 2017 White House Correspondents Dinner, Hasan Minhaj said “We’re living in this strange time, when trust is more important than truth”. (video). We have to find ways to establish trustworthiness with people who currently don’t believe the research and choose instead to grasp, easy-to-understand, wrong answers.
  3. Currently, the most accessible repository of the world’s scientific knowledge is the scientific literature. This provides a valuable resource for knowledge engineering work and building models of what is reported in the literature can provide insight into the underlying subject and influence public opinion.
    • As a human endeavor, the scientific literature is far from ideal. People routinely game the system (see Greenberg 2009) to promote their own research, usually with honest intentions but sometimes not.
    • A systematic review of analyses of the accuracy of citations in the scientific literature shows that approximately
      14.5% (10.5% to 18.6% at a 95% confidence interval) of cited facts in biomedical papers are wrong, and 64.8% of those errors are serious (see Mogull 2017)
    • Systems such as CZI’s Meta, scite.ai, AI2’s Semantic Scholar and others use AI to improve and enable scientists’ interactions with the literature. These tools are generally open access, and collaborate to enable and support each other.
  4. Beyond scientist-facing work, there are also other related efforts attempting to render aspects of public life more fact-based and data-driven. This provides a working community of data-providers, developers and end-users as well as possible frameworks for increasing the scope and impact of technology in multiple areas.
    • The FORCE11 group (‘Future Of Research Communication and E-Scholarship) is a wide-ranging academic / industry community with a broad mandate to bring about the transformation of scientific communication.
    • ‘Solutions Journalism’ provides a powerful appraoch to communicating uncomfortable issues to the public. By surveying how problems are being solved and then framing discussions of difficult subjects, propopents of this approach have shown that members of the public are more engaged and receptive to reporting when framed in this way. Such an approach could work well for how we communicate the application of scientific methods to policy.
    • Steve Ballmer, the ex-CEO of Microsoft, has financed and driven ‘USA Facts’, a website that examines the financial ‘score card’ of the United States as if it were a business. A recent Freakonomics podcast (‘Hoopers! Hoopers! Hoopers!’) showcased this interesting project and dealt briefly with the subject of outcomes and how one might measure them. This is the purview of the social sciences and likely requires some expertise from within the field of education theory or psychology to really
    • A number of non-profits have similar missions: ‘Data 4 America’ is one such organization.
  5. The public are very receptive to scientific content when it is presented in a compelling and interesting way.
    • Organizations such as the Technology Entertainment Design (TED) conference are massively popular and provide an excellent template for packaging and presenting complex and compelling scientific ideas.
    • Blogs such as framework Radiolab, Science Friday and Freakanomics regularly describe new scientific developments in a public forum with great results.
    • MOOCs (such as Coursera) and online courses provide a wide range of course material to teach complex subjects. ‘Science Driven Public Policy’ could be a subject that we could develop and teach.
    • Citizen science projects (such as Galaxy Zoo and FoldIt) permit members of the public to directly contribute to the scientific endeavor. This is interesting, fun, educational and could be a vehicle for engagement for science-driven policy.
    • Academics typically present their ideas as powerpoint slides, but animation and storytelling methods could better illustrate their work to the public. An excellent example of this technique is this video by Pindex describing the Dunning Kruger effect (and narrated by Stephen Fry).
    • It is important to note that metaphor and analogy are crucial tools that help translate complex scientific ideas into commonsense language. Finding the right framework for this messaging is an important aspect of this communication.
  6. We must recognize that this is an adversarial situation where our opponents will use literally every rhetorical trick to counter a scientifically-defined viewpoint. We must counter the trolls directly by understanding, unpacking and attacking their arguments. We should do so explicitly, ruthlessly and with as much transparency and authority as possible.
    • A terrific example of how the trolls work is the PragerU, right wing ‘educational’ website. Consider this anti-climate change video which attempts to debunk the Cook et al 2014 consensus paper. Some of the arguments are nonsensical (including an absurd apparent attempt to appeal to an antivaccination argument), but some carry a little more weight. The production of the video and carries the listener through the logic of the argument well, making emphatic statements that boldy bullshit the listener to serve their underlying argument. To counter this, we should analyze, deconstruct and refute their argument, perhaps in the same format but with a great deal of underlying support from data and established existing research.
    • The work of Walton et al. 2013 on Argumentation Schemes provide a fascinating theoretical framework for formalizing how arguments are put together. This is an approach widely used in developing AI-support tools for legal argumentation, but could well be applied here.
  7. Finally, Computational Social Network Sciences is a powerful emerging field of AI research, that can provide insight in the emerging online world of politics. Colleages such Emilio Ferrara and Kristina Lerman study how people interact with policy through social media, social bots and each other. Understanding the dynamics of these interactions could be crucially valuable in developing effective technological strategies.

What can we do?

Our goal would be to enable the use of innovative, cutting edge, AI research within the context of policy development. Our strategy for doing this would be by developing methods to leverage and utilize scientific expertise and knowledge in politcally-relevant situations. If we could also better understand the rhetorical positions of identifiably-anti-scientific positions within public discourse. If we understand our adversaries, then we can defeat them more easily.

An initial pilot effort could be to re-examine the scientific literature described in the Cook et al. 2014 consensus study by developing detailed semantic models of the data being cited in those papers. Our job is to explore and explain the science to the public: exploring, explaining and educating through accurate reporting of the abudictive reasoning used to understand what is going on. We may explicitly contrast our approach to that of non-scientific arguments being made but always from the point of view of educating people to think scientifically, and never engaging in a polemic, fruitless discussion. If a climate-skeptic cites counterevidence, we will attempt to understand it rationally and scientifically. After all, as scientists we want people to poke holes in our models and find their flaws. Here, we would welcome such things with politeness and appreciation.

Multiple challenges threaten this plan:

  1. Scientific work is complex, difficult to understand and challenging to execute. In order to be able to execute their work well at all, scientists perform technical feats that are difficult to record accurately, let alone reproduce. Work in the field of semantic E-Science (including ontological modeling and workflow development) make it possible to reproduce even very complex data analytics, given the additional time and effort required to model them.

  2. There is a lot of knowledge to work through. curating information from the literature is a slow and laborious process (especially since scientific arguments do not tolerate errors well). Developing methods of automation of the curation task may speed up how representations can be populated. Needless to say, this is an area of active research and might form the basis of a focused study on the climate change literature.

  3. Communicating complex scientific ideas is difficult. The popular science media community provides a vehicle to do this through magazines, podcasts, books, television and other media. Relying on these traditional methods can only go so far. Data visualization methods can provide a far more compelling and exhaustive view of a complex subject. This astonishing representation of the casulties of WWII is an example of a series of data visualizations that tell a compelling story in a linear fashion. We may need to examine new, non-linear methods to explain and explore the complexities of the subject of climate change.

  4. Engaging people in this endeavor requires us to go beyond simple research. For this work to have an impact, we will require outreach and involvement far beyond the simple development of novel technology.

In ‘the Martian’, an incredible movie about a lone astronaut marooned on Mars, he says:

In face of overwhelming odds, I’m left with only one option. I’m going to have to science the s*** out of this.

Clearly, this is the situation we all find ourselves in now, and all unnecessary Tolkein references aside, solid science, honesty, and hard work are only things that can solve challenges like COVID-19, Rare Disease, Neurodegenerative Disease, and maybe even things like income inequality, homelessness, and poverty.

This is work for Wizards, let’s go.