Mills let us harness wind- and waterpower. Yokes let us harness animal power. The damper-flapper mechanism that Albert Butz invented saved us from having to get up and cross the room to open a furnace door, by harnessing electricity. These powered tools turn us from laborers to task-managers, steering and guiding the forces to do useful work.
Reducing Information Work
In addition to labor, technology can also help us with the information work involved in a task.
Of course, thermometers give a user some facts to work with in managing the heat of a space.
What’s the actual temperature here? Is it just me who’s feeling cold? What temperature is the thermostat set to? Is it currently putting out heat or not? Is it even on?
These metrical tools give us facts to help us make decisions while performing a task.
More sophisticated technologies can begin to understand the rules of how a task should be performed, and let the user know when good form is being violated or thresholds crossed. These corrective tools help them understand what’s going on and correct course if they’re off track. If a user sets a thermometer to a temperature higher than it could actually attain, for instance, it could immediately provide this feedback and suggest additional measures that could be taken. This would be corrective.
Putting Physical and Information Work Together to Become Agentive
It’s when someone takes these two things—information awareness and machines doing physical work—and connects the two that you begin to see some magic happen. That’s when the tools become agentive.
Drebbel’s incubator was the first tool to do this. It took in information about the temperature to open and close the damper. As brilliant as it was for its time, it was still something of a dumb temperature monitor. It only paid attention to a single variable, and only acted when that variable went above an amount. It couldn’t help if the eggs were getting close to freezing. It didn’t help the alchemist know when fuel was running low. You can consider this an agent, but just barely.
The Nest Thermostat is a much more complicated actor, able to track and manage many variables at once. It even learns over time, refining its model of what good behavior means in its particular household, on this particular day, and for the people it knows are currently present. It is a very powerful tool for managing temperature, and much more exemplary of what you can think of as an agent today.
The thing is, you can examine the history of technology solutions around a human need and find similar patterns. Tools will start out manual. Some evolve to reduce physical effort and become powered. Others evolve to help with the information work and become metrical for measuring or assistive for staying within known rules. And of late, you can see a few dozen examples of systems combining the information and the physical work to do work on behalf of its users, becoming agentive. That these patterns repeat across history is a big claim, but let’s use three examples to illustrate: writing, music, and search.
The Problem of Writing
The Paleolithic cave paintings at Lascaux illustrate an ancient human need of mark-making. This evolution splits in two directions. One is toward expression, like paint brushes, but let’s follow the other direction that veers toward writing. Around this human need, manual tools include burnt sticks, graphite pencils, and pens. There’s not much physical effort involved in writing, but typewriters, both manual and electric, are powered tools that let people output many more letters per minute with less muscle fatigue and much more precision to the letterforms.
In addition, the advent of background spelling and grammar checkers on computers provide both metrical and assistive tools to keep you within the many conventions for clear writing. They’ve now evolved from simple rule-checkers like Sector Software’s Spellbound to more sophisticated ones like Microsoft Word’s grammar checker and iOS AutoCorrect, which not only notes misspellings, but immediately corrects the ones in which it has a high degree of confidence. Recently, Google Inbox released its Smart Reply, which parses incoming emails and provides several short, likely responses from which the user can simply select.
It all becomes agentive with the introduction of x.ai. Subscribers to this meeting scheduler only need to CC “Amy Ingram” (we see what you did there, x.) in an email asking her to “find us a time to meet” and “she” handles the rest. If you prefer a dude, the agent is happy to go by Alex as well. X.ai finds good times in your calendar, suggests them to the other people, works through conflicts, and lets you know when a fitting time has been found and a calendar reminder has been added to your calendar.
The Problem of Music Playback
Musical notation permitted the “recording” of music onto vellum, parchment, and paper, which could be played back with the manual tool of an instrument like a guitar or piano and let the human do all the decrypting work of turning those dots and lines into sound. The invention of powered tools like gramophones and record players let anyone hear a recording of a particular performance—you just had to manage collecting the music and switching out the disk yourself. CD players were also powered tools for playing, but being electronic first added metrical data like track numbers and later included artists and song titles to the display. (There are very few rules to how the user plays music, so you wouldn’t expect to see any corrective technologies, though if you looked at equalizer controls there would be certain thresholds to be managed via a spectrum analyzer.) Radio stations have for a long time had disc jockeys act as a service for selecting and broadcasting music, but more recently Pandora and Spotify are popular services with agentive aspects that let individual music listeners provide the system with a song or two they do like, and thereafter just listen.
The Problem of Search
Search might seem like an odd example, since at first it seems just like information, but then you realize how much physical work used to be involved in finding information even in a well indexed system. Card catalogs were an early manual technology for providing search-like access to the information spread out in space in the stacks of a library. Microfiche is a powered system for reducing the amount of effort in looking through periodicals. Modern automated retrieval systems are powered tools that even bring a particular book to you on request. Metrical tools like tables of contents and indexes help you jump to particular parts of content.
However, once information exploded on the internet, Yahoo!, Google, Bing, and its ilk made the task of searching easier, and even helped you with corrective tools when you misspelled something, or when you were using poor search terms. Did you mean . . .? When Google introduced Google Alerts, it introduced low-level agents by which users could set up topics of interest and let the information come find them.
So that’s just three examples. I’ll cover many more in the next chapter and throughout the rest of the book. These three are, of course, cherry-picked from the vast history of technology, but should help to illustrate how these lenses are a useful way to understand the ways that various technologies have worked to reduce effort around particular
