writing is dead (he writes)
IBM’s Watson computer, the one of “Jeopardy” fame, is being used today by insurance companies as they make decisions to approve or deny coverage of procedures. It can advise doctors on the best possible course of action. It can read and analyze every piece of medical literature ever published in less time than it takes a human doctor to drink a cup of coffee.
I see the next generation of computers that will come after Watson, contributing to the decline of writing as the primary technology of knowledge production.
But before getting to that point, we first need to go dress shopping and stop by a library.
Tailor made clothing doesn’t have “sizes.” A tailor takes measurements of each client and makes each dress to fit. Makers of readymade clothing, however, need to use the concept of “size” when sewing clothing ahead of time for unknown clients.
The clothing manufacturer cannot afford to make dresses for every human shape ahead of time. Instead they make dresses of different standard sizes. Later in a store, the clerk determines which size is correct for any given client—the client is classified as being of a certain dress size—and then the clerk presents a dress that was already made in that size.
The fit isn’t always perfect, and surely most people would prefer tailor made clothing, but that would be prohibitively expensive and time consuming.
Visiting a Library
When you ask that friend of yours who is a cookbook geek (she read and memorized them all) for advice about Moroccan Lamb Stew, she can point you just to the book you need.
In contrast, a reference librarian (they are good people!) cannot know all the books in his library. Instead, before they are put on a shelf, all the books are assigned to a subject category. When you ask the librarian’s help to find a book, he will classify your query as belonging to one of those subject categories and will present you with the books that were already placed in that subject category, say “Cooking, Moroccan.”
It’s a hit or miss kind of a system, just like trying on readymade clothing.
Nonetheless it has its advantages. During the 90s and even into the early 2000s some librarians and others advocated for organizing the web the same way they had organized books. That is, they wanted trained professionals to evaluate each website and place it in a subject hierarchy which users could later navigate to find the right resource.
They wanted something like Librarian’s Internet Index or Yahoo Directory. Do you remember it? Apparently it is still around, you should go there and play around. To find out about our Moroccan Lamb Stew, you need to go through the following steps: Directory > Society and Culture > Food and Drink > Countries and Cultures > Moroccan > Recipes. On the last page of your trek, there are three pages which may or may not have the information you are looking for.
I am not picking on Yahoo Directories here, rather I am pointing out the limitations of a system that relies on human expertise to make sense of huge amounts of relevant information. Such a system is likely to be outperformed by something like Google’s (or Yahoo’s own) search engine, which replaces human experts with computer algorithms and replaces the concept of general categories with analysis of links, keywords, and user behavior.
You prefer to use Google search over Yahoo Directory, am I right?
Going to a Doctor
What is a “disease” and why do we use this concept?
“Disease” can be thought of as a classification system that is used to pre-generate cures for individual patients just like dress sizes are used to make readymade clothing or classification categories are used to organize books in a library ahead of time.
Medical research does not come up with cures for individual patients. Instead it comes up with cures for given diseases. When an actual patient shows up at a doctor’s office, the doctor classifies the patient as having a given disease. Next the doctor prescribes a readymade cure for that particular disease to the individual patient.
Unfortunately, as with all pre-generated solutions, this one has the problem that actual patients don’t neatly fit into a few categories. What works for one patient doesn’t necessarily work for another.
But just as one can get a better fitting dress without using the concept of “dress size” and one can get a better search result without going through a classification system, in principle, one could get better healthcare without using the concept of “disease.”
Think about it. If the doctor somehow could apply the right treatment for the patient without knowing the disease, and if the treatment worked, every time, reliably, for sure, the patient would be healed and everyone would be happy ever after.
We need to accept that information relevant to our health is complex, maybe even too complex to be analyzed by humans. Our way of providing healthcare though human experts applying pre-generated, standard cures is not optimal because it cannot take advantage of all of this information. It is like browsing the internet using Yahoo Directories, rather than Google or Bing: you get results but they are not the best ones and you see only a fraction of what is out there.
Today, we are on the verge of being able to actually collect all of this relevant information which is increasingly stored in electronic form. Ideally, it could all be collectively analyzed. That includes every doctor’s note, every medical test, every result of every medical decision. Furthermore, other information that is relevant to our healthcare—what we eat, our blood pressure, with whom we hang out, with whom we have sex, what country we visit, how much we walk, or work—this kind of information is increasingly being collected by our phones and other devices. Imagine analyzing all of this data along with all the medical records and all the medical research information.
Imagine that such a system exists and imagine going to the doctor. The computer algorithm analyzes your tests, your medical history, your life history along with all of published medical knowledge and outcomes of all the patients that had an illness similar to yours. The computer system would give you a cure and it would work. It would never classify you as being an example of this or that disease.
This is similar to the notion of finding information via a search engine. During the 1990s it was difficult to imagine how an algorithm analyzing patterns in data could return relevant results without the use of categories, subjects, and human experts. Yet, as soon as we started using Google and were getting the results we wanted, we stopped worrying about classification systems. We were getting better results without them.
Watson is a first step in the direction of getting better healthcare out of data analysis. Another example is Google Flu which can more or less reliably predict the peak of the flu season months in advance.
In this imagined world I am suggesting, your personalized cure would be based strictly on data. There would be no need for the mediating, generalized concept of disease and we wouldn’t miss it. After all, are you bothered that your perfectly fitting, tailor made dress does not have a size?
In this world, human healthcare providers can concentrate on doing what machines cannot: giving attention and empathy.
When we are forced to create solutions before we face the problem to which they will be applied—by making readymade clothes, organizing books in a library, writing medical textbooks—then it is reasonable to proceed by first creating useful general categories, followed by creating solutions for problems of each of the categories, then classifying an individual instance as belonging to the general category, and finally applying the general solution to the particular instance.
However, if we have an option to create tailor made solutions for each individual instance, then going through the process of creating generalization is less useful because the custom-made solutions are often better than the pre-generated ones.
Creating general solutions and then applying them to individual cases relies heavily on writing—doctors and librarians spend years in school reading textbooks.
Creating solutions on-the-go relies heavily on computer algorithms acting on data. The role of writing as a technology of knowledge production is diminished.
w is dead (he writes)