By Jonathan Briggs, Director of Strategy, Technology, and Innovation

Jean-Paul Sartre, one of Dr. Perry’s favorite philosophers, discusses the impossibility of knowing anyone who is not oneself. At best, we can build a lower-resolution model of others in our minds, accepting that true knowledge of another is impossible. As a teacher, we are tasked with sharing ideas, knowledge, and skills in addition to assessing what students know. An accurate assessment of what students know is impossible; it is a subset of the general case that Sartre points out. Thus, we must develop and use proxies for this task.

A proxy is another metric that we believe correlates with actual knowledge. If you can answer all the questions on this test, we think that you know how to factor a quadratic equation. If your paper flows well and is free of grammatical errors, we believe you have edited it for clarity. If your thesis statement is well constructed, we assume that you have formulated a compelling argument.

In addition to assessing knowledge, we use proxy tasks to teach as well. If you do this math homework, you will begin to understand how to solve all the sides and angles of a triangle. If you read and annotate this book, you will develop a more complex understanding of the human condition.

Once you start to see them, proxies are everywhere. Resumes, references, and interviews are all proxies we use to determine who to hire. Admissions processes, from colleges to Eastside Prep, use GPAs, SSAT/ISEE/SAT/ACTs, personal statements, and so on. Restaurant and movie reviews, political polls, headlines—we both require these things to have any hope of parsing the world around us and must accept that they are insufficient to describe the world around us.

In machine learning, when training a model, we also have a problem of overfitting. The following analogy is a vast simplification but useful (another proxy!). Imagine a model that has enough capacity to memorize its training set. Instead of building rules in an effort to predict the outcome, it just remembers the outcomes of the data it was trained on. That model will not do as well on novel combinations of data that it hasn’t seen yet. Earlier in its training, it had that ability, but it has been lost in the quest to get to 100% accuracy on the training data.


A crutch is a supportive tool. The classic example helps you out when you have a cast on your foot, and there are more abstract examples such as spellcheck, grammar check, calculators, translators, Google Maps, cell phone contacts, and the web. Even being able to write things down on paper to remember them is a crutch in this sense.

The brain is inherently lazy; it can think hard about something but will wait until it is required to. In my youth, I spent considerable mind-time on learning phone numbers, pager numbers, and having a mental map of the area surrounding my house (including the various alternate routes to get from place to place).

Today, I don’t know anyone’s phone number; my phone doesn’t even display it when someone in my contact list calls. I do have a mental map of the Seattle area, not because I need to but because I choose to think about it.

My 6th-grade math teacher used to tell us that we needed to be able to do mental arithmetic because we weren’t going to have a calculator with us all the time. Turns out, we do, and not just a calculator, the ability to contact anyone, an encyclopedia, a dictionary, a video game console, a newspaper, a calendar, etc.

Like proxies, we need crutches to have any hope of thriving in the modern world.



As a society, we have never had as many options to complete a task as we do now. In my high school, one could do math homework with or without a calculator. Regardless, you would have to show your work. The calculator served as a tool to verify that you were on the right track or your answer was correct. Completing the assignment was very close to ensuring that you understood how to complete the assignment. Certainly, people would copy off of others or get too much assistance from a parent or tutor, but by and large, it was a good proxy for learning how to solve particular problems. Today, there are increasing numbers of tools to help you through the process. Step-by-step solvers, videos, posted solutions, AI math tools, and that list will continue to grow. Assigning the task is no longer sufficient to create a learning opportunity. Today, we must assign the task and discuss the manner in which the task should be completed.

Throughout my childhood, my parents had a saying, “Attitude is everything.” Their meaning was that the ways in which you approach work, life, tasks and so on was essential to the quality of work, how you felt about being assigned a task, and so on. In this way, how you approach academic assignments is as important as the assignments themselves. We used to be able to safely assume that students would approach assignments in the best way; we now need to be explicit.

Likewise, it is difficult to know how reliant we are on crutches until we take them away. It would be a good idea, for example, to spend a day now and then without your cell phone. Not as a break (though it might also serve that benefit) but as an exercise in reflection on how much it is relied upon throughout the day. In our courses, we should be asking students to occasionally put away IDE (Integrated Development Environment) programming environments, spellcheckers, and calculators so they can build awareness of themselves and the tools that they use.

Taken together, proxies, crutches, and overfitting require us to be more thoughtful about the actual value of what we do. In school, we want students to be able to think about themselves and the world around them. To generate that outcome, we have many proxies in place to support that goal, but we must be careful to avoid overfitting to the proxies themselves. Straight A’s may be a consequence of being able to think well, but in the coming years, it will be increasingly possible to get straight A’s without thinking. This trap is avoidable if we have strong, trusting relationships with students, if we have them do meaningful work, and if we have an open conversation about the manner in which that work is to be done.

These things are in place at EPS, and I look forward to the demands of clarity and rich discussion that the AI era will bring to our community.