Expanding the futures cone’s concept of the futures and the pasts ⊗ Reading as counter-practice ⊗ A bot that watched 70,000 hours of Minecraft videos

This week →{.caps} Expanding the futures cone’s concept of the futures and the pasts ⊗ Reading as counter-practice ⊗ A bot that watched 70,000 hours of Minecraft videos could unlock AI’s next big thing ⊗ Web3, the Metaverse, and the lack of useful innovation

A year ago →{.caps} A favourite in issue No.199 was It’s time to reimagine the future of Cyberpunk by Madeline Ashby.

◼{.acenter}

For the past few months I’ve been using the alpha and beta versions of Reader, a new ‘read later’ app (like Instapaper, Pocket, and Matter) from the folks at Readwise. It’s really really good, and recently they started testing some new functionality under the name Ghostreader. Basically, they leverage GPT3 to boost how you can interact with/understand/annotate what you read. It’s a bit stunning, and slightly scary, especially with the “Summarise” function, which takes a rather excellent crack at summarising any document you have in the app.

For someone writing a newsletter such as this one and making sense of things for clients, it’s definitely the “it composes music” or “it creates an original image” moment. Sure, AIs have been around to write text but it seems like we have a longer way to go before creative writing or laying out an argument in writing comes along. Just compressing an article to one paragraph though? It’s here.

So, I’m ‘pulling a Herndon’ (see the title/quote lower down) and obsolescing myself before someone else does it. I asked Ghostreader to summarise every article featured here and used those summaries instead of mine. In the first I changed one word, in others I cut a bit more but didn’t have to re-write anything. Those parts are highlighted in green with a 🤖 Summary{.ai} tag next to it. Sometimes it might be better to read the summary first, but since this is still my newsletter, I decided to put them after my commentary, just above the quotes.

I don’t know if this is going to be a one time thing, but probably not. Turns out having an assistant summarising articles forced me (in a good way) to actually add something to each. Nifty. So this is the first centaur version of Sentiers, enjoy.

Expanding the futures cone’s concept of the futures and the pasts

The link to Einstein’s Special Relativity theory and its geometry is a stretch, they are mostly using an associated visual; Minkowski’s space time. Regardless, I believe it’s a valuable upgrade to the cone. The main question is, who and what does the cone represent? Who is centered? It can’t be everyone, is it meant as a global view or specific to someone, some country, some organisation? By changing the visual to this new double cone fanning out from one point on the “hypersurface of the present,” the authors multiply the cones for each point, and integrate the influence of the past. Without replacing the cone, it contextualises it much more broadly.

In other words, it does something similar to the habit of using futures instead of future, it integrates an essential plurality. Voros represents a plurality of possibles, their model represents the plurality of perspectives that can be at the point of the cone.

🤖 Summary{.ai} This paper explores the concept of the Futures Cone, widely used in Futures Studies, in light of Albert Einstein’s Special Relativity Theory. It discusses the various versions of the Futures Cone, from Trevor Hancock, Clement Bezold and Joseph Voros, and compares them to the Minkowski space and its light cone. It suggests that the Futures Cone could be upgraded to include attributes from Special Relativity Theory, such as a representation of the absolute future, the plurality of the past, event causality, and the totality of the future. It also considers the possibility of events outside the Futures Cone, representing events that cannot be causally affected from the present moment.{.ailight}

in futures studies researchers study systems and utilize a plethora of tools for investigating the mid- or longer-term future. Like physics, futures studies are also intrinsically embracing uncertainty as it is emphatically underlined by Amara’s three laws of futures (Amara, 1981): (1) the future is not predetermined, (2) the future is not predictable, and (3) the future outcomes can be influenced by our choices in the present. […]

In addition, the current description of the futures cone does not explain the interconnection of the various “events” inside the cone: are these “events” inside the cone somehow connected and especially could one influence these events? ==Furthermore, the futures cone does not provide any info of the past (while the past is taken into in futures studies) and how the past is connected with the present moment. Finally, a question may appear: Is the past plural like the future?== […]

The past light cone contains all the past events that can casually affect P, while the future light cone contains all the events that can be causally affected from P. Thus, the Minkowski light cone manifests that the actions taking place at the present moment affect the future, and practically offers a direct poof for Amara’s 3rd law.

Reading as counter-practice

Something I’ve said and read often, is that creativity is recombination. An adjacent idea is to “Steal Like An Artist,” as Austin Kleon would say. Here L. M. Sacasas does just that, taking Maryanne Wolf’s conversation with Ezra Klein, a tweet by Molly White citing Vitalek Buterin and Sam Bankman-Fried, some Walter Ong, some Ivan Illich, and mixes the whole thing into a new one.

Instead of ‘just’ trying to re-master the skill of reading long form, something many of us have written about and struggled to do, he goes one step further and proposes reading as counter-practice, “a deliberately chosen discipline that can form us in ways that run counter to the default settings of our techno-social milieu.” In other words, the influence of feeds and speed is not just something to recognise and keep under control, but something that can be constructively counterbalanced by deep reading. I’d go further and make a parallel with meditation, which one can use as a foil to an anxiogenic world, here we’d have deep reading as a foil to predatory attention grabbers.

🤖 Summary{.ai} Examines the practice of reading, discussing its various forms and the differences between them. It looks at the history of reading and the physical form of the text, exploring how the affordances of the book might be uniquely suited to sustaining deep reading. It also looks at reading as a counter-practice and how reading for pleasure or wisdom can have its own rewards and consolations.{.ailight}

The problem is the further inference that how a text is materially instantiated is a matter of indifference. It is not. The form a text takes is not neutral, rather it changes our experience of the text. Or, to put it another way, the material form of the text mediates our relationship to the text. […]

At the very least, we should be mindful about pairing the kind of reading we set out to do with the most appropriate material form. […]

In these cases, reading is merely instrumental. It is good insofar as it is useful for the sake of some goal that is unrelated to the practice of reading itself. There are many reasons to read and many ways to read. Each may have its place. Sometimes we will read merely to absorb some bit of information. But we should occasionally resist the imperative to optimize all experiences for efficiency, which as a goal has a way of distorting every practice and vanishing all that cannot be quantified. […]

==I find it useful in this context to occasionally remind myself that I cannot read everything and that it would not be good for me to try. Better, perhaps, to read fewer things well. As is often the case, acknowledging and embracing our limitations can be freeing.==

A bot that watched 70,000 hours of Minecraft videos could unlock AI’s next big thing

I’m of two minds on this one. First, the humans doing the tagging were very likely paid a pittance, which is the usual modus operandi of creating an AI; stealing a dataset, using one with no specific permission, or badly paying someone to create one. Better practices ethically, with regards to biases, and with better remuneration are badly needed. Second, beyond the dollar distribution, it’s a kind of fascinating approach by stages. Humans document something they know how to do, one ‘bot’ multiplies it, and the other learns from it.

If, instead of fostering more inequality—not in this specific case but you now that’s what automation will do in the cases where it does “steal jobs”—, these tools served common goals for a fairer society, now that would be something.

🤖 Summary{.ai} OpenAI has developed the best Minecraft-playing bot to date by having it watch 70,000 hours of video of people playing the game. This showcases a powerful technique called imitation learning, in which neural networks are trained to perform tasks by watching humans do them. … [The team paid people to label 2,000 hours of YouTube video,] The researchers used Video Pre-Training (VPT) to generate action labels for 70,000 hours of unlabeled video, and then combined imitation learning with reinforcement learning to train the bot.{.ailight}

Large internet-sized data sets will certainly unlock new capabilities for AI, says [Natasha Jaques, who works on multi-agent reinforcement learning at Google and the University of California, Berkeley.]: “We've seen that over and over again, and it’s a great approach.” ==But OpenAI places a lot of faith in the power of large data sets alone, she says: “Personally, I’m a little more skeptical that data can solve any problem.”==

Web3, the Metaverse, and the lack of useful innovation

This is a bit of a weird one. In some parts the authors talk about current bubbles but not AI, then in others provide examples, including AI. If Web3, the Metaverse, and AI had been treated equally all along, that would have been more interesting. Then again, their thesis of current bubbles not being on pace to create any lasting value would have been wrong. Maybe.

If you enjoyed last week’s A Great Leap Forward in AI, which was specifically about AI showing “it has a story, and utility,” then this one can provide useful context and further details on the economics of those three trends/hypes/bubbles.

🤖 Summary{.ai} This document examines the lack of useful innovation in new technologies like the Metaverse, Web3, and blockchain, despite the high investment in them. It looks at the differences between the successes of the dot-com bubble and the lesser success of the 2010s, and attributes this to the decline of basic and applied research at companies and the focus on publishing papers at universities instead of developing new technologies. It argues that a new research system is needed to focus on practical improvements and useful technologies, rather than simple metrics pushed by bureaucrats.{.ailight}

If we are correct that the newest wave of hot technologies will do almost nothing to improve human welfare and productivity growth, then elected officials, policymakers, leaders in business and higher education, and ordinary citizens must begin to search for more fundamental solutions to our current economic and social ills. […]

==In the following year, an analysis by a pro-blockchain organization, the British Blockchain Association, showed that a vast majority of the blockchain projects they surveyed had no well-described rationale, no predetermined criteria for achievement, and no analysis of success or failure.== In other words, they were merely pie-in-the-sky ideas, based on hype rather than detailed analysis or justification. And none of the projects benchmarked themselves against existing, proven technologies. […]

To summarize, when the dot-com bubble deflated, we were left with lasting improvements such as e-commerce, digital media, and enterprise software, but our current bubble has involved investors running up the stock prices of firms working on technologies that have produced demonstrably less value. When the air goes out of this bubble, we very well may be left with hardly anything of value at all.

This singer deepfaked her own voice—and thinks you should too

* {.s}

This singer deepfaked her own voice—and thinks you should too →{.caps} “I do want people to understand how powerful these systems are and how having sovereignty over training data is really important. The thorny questions that are being asked right now—it’s really important that we get them right.”

Futures, foresights, forecasts & fabulations

Futures, foresights, forecasts & fabulations →{.caps} Digital Futures Gathering 2022. “[E]ssays written by participants at the Digital Futures Gathering, offer insights into the topics we discussed, the positions we explored and the conversations we had during the gathering.” ⊗ Disruptive seeds: a scenario approach to explore power shifts in sustainability transformations. “This paper responds to a need for methods that support the creation of imaginative transformation pathways while attending to the roles that power shifts play in transformations.” ⊗ The Spectacular Life of Octavia E. Butler

Asides

• 🌳 ⚡️ 🤩 Open Climate Fellowship. “… seeks seven (7) mid career professionals for the 2023 Open Climate Fellowship Program. We welcome applicants from backgrounds across the open and climate movements such as researchers, activists, developers, and educators”
• ⚡️ 🇨🇳 🇮🇩 😬 This one deserved better than just an aside, but here we are. Longscroll page with loads and loads of pictures + a longread. The dirty road to clean energy: how China’s electric vehicle boom is ravaging the environment. “In neighboring Indonesia, nickel extraction is causing environmental and social devastation.”
• 🎥 🕶 🤔 Debunking one of the Metaverse’s biggest myths. Good very short interview with Matthew Ball, I especially like the part where he argues that what’s being replaced (maybe, eventually) is not ‘real life’ but other media. The Metaverse wouldn’t replace taking a walk with your dog, it would replace re-runs of Seinfeld.
• 😍 😍 🚞 💰 Very klept, very Snowpiercer, but also very gorgeous. Step on Board the All-New Orient Express Train.
• 📚 📚 📚 100+ Must-Read Books in 2022 by DO lectures and Inga’s Gift Guide .
• 🌌 🪱 🕳 🤯 Longread and 17m documentary. Physicists Create a Holographic Wormhole Using a Quantum Computer. “The unprecedented experiment explores the possibility that space-time somehow emerges from quantum information, even as the work’s interpretation remains disputed.”
• 🌗 🛰 🇺🇸 How a tiny briefcase-sized spacecraft will prospect for water on the Moon. “Entering a highly elliptical orbit called a near rectilinear halo orbit, the satellite will make a close approach to the Moon’s south pole every six days, collecting data on each of 10 passes. It uses an infrared laser reflectometer instrument that operates at four different frequencies, bouncing lasers off the Moon’s surface to see whether the material being struck is ice or rock.”
• 🦠 🐺 🤔 That would explain Elongated Muskrat’s last few years. 🙃 Behavior-changing parasite moves wolves to the head of the pack. “_Toxoplasma gondii_ is a ubiquitous protozoan parasite that can infect any warm-blooded species. In lab studies, infection with _T. gondii_ has been shown to increase dopamine and testosterone levels along with risk-taking behaviors in hosts including rodents, chimps, and hyenas. Oh, and humans.”

{.miscellany}