Tools for thought — vision and philosophy
The idea of "tools for thought" has animated a strand of computing since the 1960s and 1970s, when pioneers including Douglas Engelbart, Alan Kay, Ivan Sutherland, Seymour Papert, and Vannevar Bush pursued a vision in which computers serve as genuine amplifiers of human intelligence. Kay summarized the aspiration: the very use of a personal computer would "actually change the thought patterns of an entire civilization." Retrospectively it is difficult not to feel disappointed — computers have not yet been nearly as transformative as far older tools for thought, such as language, writing, or mathematical notation. This essay by Andy Matuschak and Michael Nielsen, published as "How can we develop transformative tools for thought?" (2019), argues that now is a good time to work hard on this vision again.
The term "tool" understates the ambition. A medium such as Adobe Illustrator is essentially different from any individual tool it contains: it creates an immersive context in which users can have new kinds of thought, thoughts formerly impossible for them. The range of expressive thoughts possible in such a medium is an emergent property of its elementary objects and actions. When those are well chosen, the medium expands the possible range of human thought.
What makes a tool for thought genuinely transformative
The Hindu-Arabic numeral system serves as the paradigmatic example. Starting from Roman numerals, arriving at Hindu-Arabic numerals would require both extraordinary mathematical insight and extraordinary design insight — the two are inextricably entangled. The meaning of a digit changes based on its position; the same operation (2+3) applies coherently whether the digits represent ones or tens; commutativity, associativity, and distributivity are all embedded in the design. Any person who could invent this system from scratch would have to be simultaneously one of the great mathematical and design geniuses who ever lived.
This points to the general principle: the most powerful tools for thought express deep original insights into the underlying subject matter. They are not "applied cognitive science" — a collage of existing ideas pasted together using modern design practice — but cognitive science of the highest order, expressing insights no one else has had. Conventional tech industry product practice, which is excellent at creating businesses, is poorly equipped to produce this depth of insight.
The insight-through-making loop
What is required instead is a culture that combines the best of modern product practice with the best of research culture, operating an insight-through-making loop at full throttle: deep original insights about the subject feed back to change and improve the system, and changes to the system produce new original insights about the subject. This is a stronger claim than the common observation that making new tools can lead to new insights for the toolmaker. The claim is that making new tools can lead to new insights for humanity as a whole — significant original research — and vice versa, rapidly cycling.
This loop is extraordinarily rare to find operating well. Researchers who are brilliant in their domain often regard making as trivially implementational. Makers who don't understand research see it as a slow and dysfunctional version of making. The result is that the full loop rarely operates, and the most transformative tools for thought are underdeveloped.
Good tools arise from serious work, not toy problems
A related principle: good tools for thought arise mostly as a byproduct of doing original work on serious problems. They tend to be created by people doing that work, or by people working very closely to them. The failure mode is tools built for toy problems by people who have never done serious work in the subject they are supposedly building tools for. Seymour Papert's Logo is a cautionary example: designed to teach differential geometry, it was not used by actual differential geometers, raising the question of whether it left out the most important ideas. Peter Norvig's Jupyter notebooks on wealth inequality are a positive example: Norvig used the medium to explore questions he genuinely wanted to answer, and the educational value was a byproduct of authentic inquiry.
In serious mediums, there is a notion of canonical media — instances that expand the form and set a new standard. Citizen Kane did this for film; Grant Sanderson's 3Blue1Brown videos do it for mathematical explanation on YouTube; the Feynman Lectures do it for physics textbooks. Work on new tools for thought needs analogous canonical instances: ambitious enough to expand what the form can do, grounded enough in serious work to demonstrate genuine utility.
The public goods problem
Most fundamental tools for thought are public goods: expensive to develop initially but easy for others to duplicate and improve upon, free-riding on the original investment. This creates systematic underinvestment. Adobe invests in new interface ideas for Illustrator, but Sketch and Figma can copy the essential innovations at a fraction of the cost. Unlike chip design, where the difficulty of duplication creates incentives for deep research investment, software tools for thought are easily cloned.
The video game industry has partially solved this: game companies invest in novel interface ideas because a game earns most of its revenue in the first few months, before clones appear. That novelty-first-revenue model does not apply to tools for thought, where the goal is deep mastery developed over time. The most promising solutions are philanthropic funding for research (the model used in computer animation, which prepared the way for Pixar and Dreamworks) and the Adobe model, in which new tools for thought are central to operations but the competitive moat is built elsewhere — training, distribution, documentation.
The emotion dimension
Historically, work on tools for thought has focused on cognition and largely ignored emotion. But media forms like video games and film demonstrate that elaborate models of users' emotional responses — second-by-second emotional tracking, understanding of the overall emotional journey — are achievable and decisive. The mnemonic medium has the unusual property of delayed benefit: the greater the delay before seeing results, the more the benefit — making it almost the anti-product in terms of conventional engagement models. Mnemonic video is a direction worth exploring: combining the emotional range possible in high-affect video with the detailed retention of the mnemonic medium, producing both deep emotional connection and genuine conceptual mastery.
Why not AGI or brain-computer interfaces?
The question of whether to work on tools for thought versus artificial general intelligence or brain-computer interfaces is a genuine one. AGI and BCI have well-defined goals, which fits tech culture's engineering and goal-driven mindset. Tools for thought are more open-ended — the work is exploration of the question "how can we develop tools that change and expand the range of thoughts humans can think?" — which is historically unusual but not historically unsuccessful. Language, writing, the printing press, and even the concept of the computer itself emerged from open-ended exploration, not goal-driven processes. Alan Turing and Alonzo Church were exploring esoteric questions about logic and provability; the computer was a discovered concept, not a planned deliverable.
The three fields will likely merge over time. AI may be one of the most promising routes to discovering new tools for thought. BCIs may ultimately be used not just as memory chips or working-memory expanders but as interfaces that enable genuinely new mental operations and representations — which is the same territory as non-BCI tools for thought. Work done now on tools for thought may directly shape how BCIs are designed and used.