[Link] Tools for thought (Matuschak & Nielson)
https://numinous.productions/ttft (archive)
An excerpt:
We’re often asked: why don’t you work on AGI or [brain-computer interfaces (BCI)] instead of tools for thought? Aren’t those more important and more exciting? And for AGI, in particular, many of the skills required seem related.
They certainly are important and exciting subjects. What’s more, at present AGI and BCI are far more fashionable (and better funded). As a reader, you may be rolling your eyes, supposing our thinking here is pre-determined: we wouldn’t be writing this essay if we didn’t favor work on tools for thought. But these are questions we’ve wrestled hard with in deciding how to spend our own lives. One of us wrote a book about artificial intelligence before deciding to focus primarily on tools for thought; it was not a decision made lightly, and it’s one he revisits from time to time. Indeed, given the ongoing excitement about AGI and BCI, it would be surprising if people working on tools for thought didn’t regularly have a little voice inside their head saying “hey, shouldn’t you be over there instead?” Fashion is seductive.
One striking difference is that AGI and BCI are based on relatively specific, well-defined goals. By contrast, work on tools for thought is much less clearly defined. For the most part we can’t point to well-defined, long-range goals; rather, we have long-range visions and aspirations, almost evocations. The work is really about exploration of an open-ended question: how can we develop tools that change and expand the range of thoughts human beings can think?
Culturally, tech is dominated by an engineering, goal-driven mindset. It’s much easier to set KPIs, evaluate OKRs, and manage deliverables, when you have a very specific end-goal in mind. And so it’s perhaps not surprising that tech culture is much more sympathetic to AGI and BCI as overall programs of work.
But historically it’s not the case that humanity’s biggest breakthroughs have come about in this goal-driven way. The creation of language – the ur tool for thought – is perhaps the most important occurrence of humanity’s existence. And although the origin of language is hotly debated and uncertain, it seems extremely unlikely to have been the result of a goal-driven process. It’s amusing to try imagining some prehistoric quarterly OKRs leading to the development of language. What sort of goals could one possibly set? Perhaps a quota of new irregular verbs? It’s inconceivable!
Similarly, the invention of other tools for thought – writing, the printing press, and so on – are among our greatest ever breakthroughs. And, as far as we know, all emerged primarily out of open-ended exploration, not in a primarily goal-driven way.
Even the computer itself came out of an exploration that would be regarded as ridiculously speculative and poorly-defined in tech today. Someone didn’t sit down and think “I need to invent the computer”; that’s not a thought they had any frame of reference for. Rather, pioneers such as Alan Turing and Alonzo Church were exploring extremely basic and fundamental (and seemingly esoteric) questions about logic, mathematics, and the nature of what is provable. Out of those explorations the idea of a computer emerged, after many years; it was a discovered concept, not a goal. Fundamental, open-ended questions seem to be at least as good a source of breakthroughs as goals, no matter how ambitious. This is difficult to imagine or convince others of in Silicon Valley’s goal-driven culture. Indeed, we ourselves feel the attraction of a goal-driven culture. But empirically open-ended exploration can be just as, or more successful.
I like Andy’s idea that there is a whole world of mnemonic products which have yet to be explored. And I am glad to see the insight on what is wrong with the emotional story telling of standard MOOCs. There is definitely work in the area of learning tools to be done. He’s convinced me that we can create far better learning tools without needing big technological breakthroughs. The wonders are already here.
One issue I have is his idea that the medium of content should be mnemonically based. This bothers me because I presume that if your content is really good, professionals and experts will read it as well. And since the way that they read is different from the manner of a novice, they should be able to ignore and not be interrupted or slowed by tools designed for novices.
Last month, I wrote an essay which started as a rebuttal to Matuschak’s “Why Books Don’t Work,” on r/ssc. In the end, I didn’t directly address his article, but instead explained how books in their most developed form are excellent learning tools.
In both this and previous essays, Michael Nielsen has specifically addressed the point that tools of thought should scale up to research-level thinking, so he (and presumably Matuschak) are aware of the issue of only appealing to novices.
I read the post you linked and I agree with your points about well-written books’ table of contents acting as a memory aid and outline. But, Matuschak’s points about books still ring true to me. Just because books are great tools for learning doesn’t mean we can’t do better. To give one example of a limitation of books outside of memory not addressed by your post, books don’t provide any way for me to answer questions about the ideas being discussed beyond what I can visualize in my head (in particular in cases where the ideas are heavily quantitative). To give one example of what this could look like, Bret Victor has posed the thought experiment of what it would be like if you could explore models of climate change as you read about them.
Yes, digital books offer far greater potential for visualization! Books do not offer a way to play with the inputs and outputs of models, and maybe one day even online academic papers will allow us to play with the data more. I look forward to the day when modeling tools are simple enough to use that even humanities people will have no problem creating them to illustrate a point. Although, Excel really is quite good!
Maybe it’s part of their excitement for their own research that Andy claims books are a bad medium for learning.
How would this be different from a textbook with problems to work through? Or did you mean good visualization (of data, imbedded in the text) as the link demonstrates?
The latter. To be clear, exercises are great but I think they’re often not enough, in particular for topics where it’s harder to build intuition just by thinking. The visualizations in that post would be an example of a prototype of the sorts of visualizations I’d want for a data-heavy topic.
Regarding textbook problems,the subset of things for which textbook problems substitute for rather than complement interactive visualizations seems relatively small, especially outside of more theoretical domains. Even for something like math, imagine if an addition to exercises, your textbook let you play with 3Blue1Brown-style visualizations of the thing you’re learning about.
To give another example, say I’m learning about economics at the intro level. Typical textbooks will have questions about supply & demand curves, diminishing marginal utility, etc. My claim is that most people will build a deeper understanding of these concepts by having access to some sort of interactive models to probe in addition to the standard exercises at the end of the chapter.
Is that How to Read a Book for Understanding?
Yeah!