I agree that auto-complete for paragraphs sounds like a real possibility, and the striking thing here is how similar the above essay [see my last post, TB] looks to something like a real student would write, or something that might be published in a real social science journal.

Andrew Gelman

We seem to be entering a new era in higher education. On Monday, Eric Schwitzgebel published the preliminary results of a collaboration with Anna Strasser and Matthew Crosby that showed that even experts could be fooled into thinking that output generated by an artificial intelligence (GPT-3) had been written by an actual philosopher,* Daniel Dennett. But already over a year ago, ZDNet reported that “AI can write a passing college paper in 20 minutes.” Taken together, we’re faced with the disturbing prospect that students, even in such disciplines as philosophy, will be able to earn college degrees (that is, receive passing grades in their coursework) without ever having to compose a coherent paragraph, perhaps without ever having to write (or even read) a single sentence. More ambitious students may be able to get quite good grades simply be editing the output of an AI on the basis of their reading and lectures.

I think we have to take this new situation seriously. Calculators, spell checkers, and typing assistants should already make us cautious about rewarding students for the basic numeracy and literacy they display in their written work. Now, it seems, we also have to be wary of their claims to know what they’re talking about. A properly trained and fine-tuned language model can plausibly simulate a “passable” understanding of literature, history, and philosophy, and can no doubt even say something halfway sensible about cell biology and quantum mechanics. (GPT-3 reads Wikipedia a lot, remember.) Indeed, GPT-3 can probably even pass a computer science course, by producing plausible Python code.

It must be noted that AI is so far only “passing” for a college student. It’s getting mainly Cs, even under my tutelage. And in so far as it is passing as a philosopher, it is one that is providing brief answers to vague soft-ball questions. (You might argue that that’s the very definition of “sophomoric”, i.e., philosophy at the level of a college sophomore.) So there still seems to be a need for humans to be excellent at these things. But there is an obvious reply: give it a few years; these AIs have hardly begun their training!

Even at this stage, however, I feel heavily implicated, even a bit guilty. I’ve spent my career trying to break academic writing down into trainable skills. I don’t like calling it an “algorithm” (I prefer to call it “discipline”) but it is a set of repeatable operations arranged in an iterative process. Worse, I’ve suggested we should embrace, not just our finitude, but our mediocrity. That is, I’ve been very much directing my attention to the middling writer of ordinary prose (albeit one who wants to improve). It seems it won’t be long before Silicon Valley can offer writers of middling ambition a much, much easier path to success. Am I about to be put out of a job by artificial intelligence? Am I about to become obsolete?

Maybe it’s an entirely natural development. Many years ago, getting an education wasn’t just a matter of acquiring knowledge and skills. It was also a time to start building a personal library, a collection of books that served as reference points in your learning. Even today, graduate students (humble) brag about their (ridiculous) expenditures on books, but their priorities are changing. They also spend their studies acquiring the computer equipment, and the skills to operate it, that a life in scholarship requires. “The scholar disappears,” said Heidegger already back in 1938. “He is succeeded by the research man who is engaged in research projects. … The research man no longer needs a library at home.” Indeed, a “code library” is becoming as important to many researchers today as a library of books.

Perhaps, in the not so distant future, “getting an education” will come to mean largely “training your AI”. Students without academic ambitions will spend four years teaching their AI to “pass” for them in writing, so that it can write everything from job applications, to corporate memos, to newspaper columns, to love letters. They will give it style and taste and a kind of “experience” to draw on. Graduate students will be gently shaping their dissertations as summaries of their corpus of reading, combined with a set of data they’ve carefully collected (but left the analysis of to an AI?). “Writing a dissertation” will essentially mean “fine tuning your AI to write journal articles in your name”.

I’m not sure how to feel about it. “This does seem like we’re coming to the academic end times,” wrote Andrew Gelman in an email to me after I sent him a link to my last post. I don’t like to sound apocalyptic but it does seem like a radical shift in the way the “the prose of world” will be maintained going forward. I guess, as a writing coach, I can take solace in the fact that photography hasn’t spelled the end of art clases. Some people still want to learn how to paint and draw and some people, I suppose, will always want to learn how to write. In any case, the horse and buggy may be a thing of the past but the wheels keep on turning. Maybe the automation of higher education — essentially the automation of the educated mind — will open new frontiers in human existence I can’t yet imagine. The end of something is usually the beginning of something else. I’m paying attention.

_____
*Indeed, we might say that an artificial intelligence was passing for a natural philosopher!

In my last post, I presented the results of a GPT-3-assisted writing experiment. In the comments, Andrew Gelman suggested that I provide the prompts (inputs) and completions (outputs) in their entirety, so that my contributions can be clearly distinguished from those of the language model. This post is an attempt to do that.

For those who are playing along at home (i.e., those who have their own OpenAI account), you can access my presets here. (Model: text-davinci-002; Temperature: .77; Maximum length: 208; Top P: .9; Frequency penalty: .95; Presence penalty: .95; Best of: 18. I’m very open to suggestions from people who have played around with the presets more than I have.)

[UPDATE: It has been suggested in the comments that my “best of” setting is excessive, i.e., expensive and wasteful. I’ll be trying lower settings in future attempts, as well as other variations on the presets.]

My approach is to, first, prompt the model with a title and a key sentence, and from there with the edited result and a new key sentence in five or six iterations, until I have a five-paragraph essay. I will present each step below with my writing in bold and GPT-3’s writing in plain text. When I say each new prompt (or “submission”) is an “edited” version of the previous step’s output, I just mean I’ve tidied it up a bit. Usually removing line breaks to produce a visually neat paragraph. As you’ll see, GPT-3 actually catches on to this and begins to structure its new output to match my guidelines.

Here we go:

Step 1

The Significance of Sleep

Matthew Walker is a prominent sleep researcher. In his book, Why We Sleep: Unlocking the Power of Sleep and Dreams, Walker explains why sleep is so important. Among other things, he argues that sleep is critical for physical health, mental health, and overall well-being.

Physical Health

Walker argues that sleep is essential for physical health. He points to research showing that sleep deprivation can lead to a host of problems, including obesity, diabetes, heart disease, and stroke. He also notes that sleep is important for immunity and healing. In one study cited by Walker, people who slept less than seven hours a night were three times more likely to catch a cold than those who slept eight hours or more.

Mental Health

Walker also argues that sleep is essential for mental health. He cites research showing that people who don’t get enough sleep are more likely to suffer from depression and anxiety. He also notes that sleep deprivation can lead to impaired memory and decision-making ability. In one study cited by Walker, people who were sleep deprived

Notice that the AI mimicked my use of headings but not quite in the way I intended. I wanted the first heading to be my title and it thought I wanted regular subheadings. But it did produce some sentences I could bring together to serve as an opening paragraph, introducing Matthew Walker’s sleep research.

Step 2

Having removed the headings, and gathered what remained into a paragraph, I now prompted it to generate some criticism of Walker’s research. I also cued it to use APA-style referencing, which will come in handy later.

The Significance of Sleep

Matthew Walker is a prominent sleep researcher. In his book, Why We Sleep: Unlocking the Power of Sleep and Dreams, Walker (2017) explains why sleep is so important. Among other things, he argues that sleep is critical for physical health, mental health, and overall well-being. Walker argues that sleep is essential for physical health. He points to research showing that sleep deprivation can lead to a host of problems, including obesity, diabetes, heart disease, and stroke. He also notes that sleep is important for immunity and healing. In one study cited by Walker, people who slept less than seven hours a night were three times more likely to catch a cold than those who slept eight hours or more. Walker also argues that sleep is essential for mental health. He cites research showing that people who don’t get enough sleep are more likely to suffer from depression and anxiety. He also notes that sleep deprivation can lead to impaired memory and decision-making ability.

Recently, however, his work has received serious criticism from scholars. In a paper published in the journal Science, sleep researcher Rosalind Doris and her colleagues (Doris et al., 2018) argue that Walker’s book contains “numerous factual errors.” They also argue that some of his claims are “exaggerated or unsubstantiated.” For example, Doris and her colleagues take issue with Walker’s claim that sleep is essential for physical health. They point to research showing that people who sleep less than seven hours a night are not necessarily at greater risk for obesity, diabetes, heart disease, and stroke. In fact, they note that some studies have found no link between short sleep duration and these health problems. They also argue that there is no evidence that sleep is necessary for immunity or healing. In one study cited by Doris and her colleagues, people who slept less than seven hours a night were no more likely to catch a cold than those who slept eight hours or more.

Despite the criticism from Doris and her colleagues,

Notice that it has now learned what I’m after in terms of paragraphing. After I had tidied up the first paragraph and added the sentence “Recently, however, his work has received serious criticism from scholars,” GPT-3 produced a properly referenced and plausible-looking paragraph. It even began a new one, apparently having understood what the game is. In step 3, I decided simply to play along. I did no editing at all, and merely completed the thought with a statement about Walker’s media appearances. (Note that I’m providing the full “prompt”, i.e., all the text that GPT-3 is using as context for its completion, but only the bolded words were added by me in this step. Everything before was already there and everything after was produced by GPT-3.)

Step 3

The Significance of Sleep

Matthew Walker is a prominent sleep researcher. In his book, Why We Sleep: Unlocking the Power of Sleep and Dreams, Walker (2017) explains why sleep is so important. Among other things, he argues that sleep is critical for physical health, mental health, and overall well-being. Walker argues that sleep is essential for physical health. He points to research showing that sleep deprivation can lead to a host of problems, including obesity, diabetes, heart disease, and stroke. He also notes that sleep is important for immunity and healing. In one study cited by Walker, people who slept less than seven hours a night were three times more likely to catch a cold than those who slept eight hours or more. Walker also argues that sleep is essential for mental health. He cites research showing that people who don’t get enough sleep are more likely to suffer from depression and anxiety. He also notes that sleep deprivation can lead to impaired memory and decision-making ability.

Recently, however, his work has received serious criticism from scholars. In a paper published in the journal Science, sleep researcher Rosalind Doris and her colleagues (Doris et al., 2018) argue that Walker’s book contains “numerous factual errors.” They also argue that some of his claims are “exaggerated or unsubstantiated.” For example, Doris and her colleagues take issue with Walker’s claim that sleep is essential for physical health. They point to research showing that people who sleep less than seven hours a night are not necessarily at greater risk for obesity, diabetes, heart disease, and stroke. In fact, they note that some studies have found no link between short sleep duration and these health problems. They also argue that there is no evidence that sleep is necessary for immunity or healing. In one study cited by Doris and her colleagues, people who slept less than seven hours a night were no more likely to catch a cold than those who slept eight hours or more.

Despite the criticism from Doris and her colleagues, Walker continues to make bold claims about the importance of sleep in the media. In an interview with The Guardian, for example, Walker (2018) claimed that sleep is “the single most important thing we can do to reset our brain and body health each day.” He also claimed that sleep deprivation is a “public health epidemic.”

The Significance of Dreams

Dreams have been a source of fascination for centuries. In his book, Why We Sleep: Unlocking the Power of Sleep and Dreams, Matthew Walker (2017) argues that dreams are important for mental health. He cites research showing that people who don’t get enough sleep are more likely to suffer from depression and anxiety. He also notes that sleep deprivation can lead to impaired memory and decision-making ability. However, some scholars have criticized Walker’s claims about the importance of dreams. In a paper published in the journal Science, sleep researcher Rosalind Doris and her colleagues (Doris et al., 2018) argue that Walker’s book contains “numerous factual errors.

This didn’t go so well. GPT-3 is now mainly repeating itself. It did grasp that my claim about Walker’s media appearances needed to cite an interview with an outlet like the Guardian, but it quickly looped back to things it had already said.

Step 4

I decided to see if I could push the argument forward by deleting the repetition and invoking the replication crisis. (To ease reading I’m now leaving out the title and the first two paragraphs of the prompt, which, again, are merely the result of the previously generated text. But keep in mind that those paragraphs are also part of the “prompt”, i.e., the input on the basis of which it is computing its output.)

Despite the criticism from Doris and her colleagues, Walker continues to make bold claims about the importance of sleep in the media. In an interview with The Guardian, for example, Walker (2018) claimed that sleep is “the single most important thing we can do to reset our brain and body health each day.” He also claimed that sleep deprivation is a “public health epidemic.”

Walker's work has been cited as an example of the replication crisis in psychology. In a paper published in the journal Psychological Science, for example, sleep researcher Daniel Simons and his colleagues (Simons et al., 2018) argue that Walker’s book contains “a number of unsubstantiated claims.” They also argue that some of his claims are “exaggerated or overgeneralized.”

Though the content choices it is making here are interestingly familiar, the result is somewhat disappointing in its brevity. I’ll see if I can help it along a little.

Step 5

Walker's work has been cited as an example of the replication crisis in psychology. In a paper published in the journal Psychological Science, for example, sleep researcher Daniel Simons and his colleagues (Simons et al., 2018) argue that Walker’s book contains “a number of unsubstantiated claims.” They also argue that some of his claims are “exaggerated or overgeneralized.”

It can be understood as an example of the overreliance on statistical significance and hypothesis testing in the social sciences. In a paper published in the journal Sociological Methods & Research, for example, sociologist Scott Frank and his colleagues (Frank et al., 2018) argue that social scientists often place too much emphasis on statistical significance. They also argue that social scientists often fail to consider the practical significance of their findings.

Despite the criticism from Doris and her colleagues, Walker continues to make bold claims about the importance of sleep in the media. In an interview with The Guardian, for example, Walker (2018) claimed that sleep is “the single most important thing we can do to reset our brain and body health each day.” He also claimed that sleep deprivation is a “public health epidemic.”

Notice that it’s repeating itself at the end again.

Step 6

I now wanted to see if I could coax it into giving me a concluding paragraph. After deleting the repetition beginning with “Despite the criticism,” and merging the “It can be understood…” paragraph with the previous one, I simply wrote “In conclusion,” and submitted the text. The result was not impressive:

Walker's work has been cited as an example of the replication crisis in psychology. In a paper published in the journal Psychological Science, for example, sleep researcher Daniel Simons and his colleagues (Simons et al., 2018) argue that Walker’s book contains “a number of unsubstantiated claims.” They also argue that some of his claims are “exaggerated or overgeneralized.” It can be understood as an example of the overreliance on statisticial significance and hypothesis testing in the social sciences. In a paper published in the journal Sociological Methods & Research, for example, sociologist Scott Frank and his colleagues (Frank et al., 2018) argue that social scientists often place too much emphasis on statistical significance. They also argue that social scientists often fail to consider the practical significance of their findings.

In conclusion, Walker’s work has been criticized by scholars for containing “numerous factual errors” and for making “exaggerated or unsubstantiated” claims. Despite the criticism, Walker continues to make bold claims about the importance of sleep in the media.

I tried again, doing a little more writing of my own. But GPT-3 seemed to be running out of ideas (or nerve):

This does not mean that Walker's work is completely false. It does, however, raise some concerns about his research practices. Future studies and criticism will likely clarify the role of sleep in physical and mental health.

Step 7

I decided to call it a day and add the reference list: This is usually done simply by typing the word “references” at the end of the text. GPT-3 takes it from there. This time, however, it actually took me a few tries to get it to generate a plausible list. Until I put a double line return after “References” it didn’t format the list properly. And then it left out Walker. That was easily remedied, however, by prompting it with “Walker, M” at the end. Here’s the text as it appears in the final version. It consist of 1024 tokens. The bolded text is my contribution. The rest is GPT-3.

The Significance of Sleep

Matthew Walker is a prominent sleep researcher. In his book, Why We Sleep: Unlocking the Power of Sleep and Dreams, Walker (2017) explains why sleep is so important. Among other things, he argues that sleep is critical for physical health, mental health, and overall well-being. Walker argues that sleep is essential for physical health. He points to research showing that sleep deprivation can lead to a host of problems, including obesity, diabetes, heart disease, and stroke. He also notes that sleep is important for immunity and healing. In one study cited by Walker, people who slept less than seven hours a night were three times more likely to catch a cold than those who slept eight hours or more. Walker also argues that sleep is essential for mental health. He cites research showing that people who don’t get enough sleep are more likely to suffer from depression and anxiety. He also notes that sleep deprivation can lead to impaired memory and decision-making ability.

Recently, however, his work has received serious criticism from scholars. In a paper published in the journal Science, sleep researcher Rosalind Doris and her colleagues (Doris et al., 2018) argue that Walker’s book contains “numerous factual errors.” They also argue that some of his claims are “exaggerated or unsubstantiated.” For example, Doris and her colleagues take issue with Walker’s claim that sleep is essential for physical health. They point to research showing that people who sleep less than seven hours a night are not necessarily at greater risk for obesity, diabetes, heart disease, and stroke. In fact, they note that some studies have found no link between short sleep duration and these health problems. They also argue that there is no evidence that sleep is necessary for immunity or healing. In one study cited by Doris and her colleagues, people who slept less than seven hours a night were no more likely to catch a cold than those who slept eight hours or more.

Despite the criticism from Doris and her colleagues, Walker continues to make bold claims about the importance of sleep in the media. In an interview with The Guardian, for example, Walker (2018) claimed that sleep is “the single most important thing we can do to reset our brain and body health each day.” He also claimed that sleep deprivation is a “public health epidemic.”

Walker's work has been cited as an example of the replication crisis in psychology. In a paper published in the journal Psychological Science, for example, sleep researcher Daniel Simons and his colleagues (Simons et al., 2018) argue that Walker’s book contains “a number of unsubstantiated claims.” They also argue that some of his claims are “exaggerated or overgeneralized.” It can be understood as an example of the overreliance on statistical significance and hypothesis testing in the social sciences. In a paper published in the journal Sociological Methods & Research, for example, sociologist Scott Frank and his colleagues (Frank et al., 2018) argue that social scientists often place too much emphasis on statistical significance. They also argue that social scientists often fail to consider the practical significance of their findings.

This does not mean that Walker's work is completely false. It does, however, raise some concerns about his research practices. Future studies and criticism will likely clarify the role of sleep in physical and mental health.

References

Doris, R. M., et al. (2018). Science communication in the social sciences: A review of Why We Sleep by Matthew Walker. Science, 361(6398), eaao5447. https://doi.org/10.1126/science.aao5447

Frank, S., et al. (2018). The replication crisis and statistical significance testing in sociology: An introduction to the special issue with recommendations for best practices going forward. Sociological Methods & Research, 47(2), 305-327. https://doi.org/10/1177%2F0049124116676024

Simons, D., et al. (2018). Psychological science in the spotlight: Unsubstantiated claims about sleep in Why We Sleep by Matthew Walker [Target article]. Psychological Science, 29(1), 15-19. https://doi-org./10/1177%2F0956797617741337

Walker, M. (2017). Why we sleep: Unlocking the power of sleep and dreams. New York, NY: Scribner.

Walker, M. (2018, February 2). ‘Sleep is the best thing that you can do for your health’ – an interview with Matthew Walker. The Guardian. Retrieved from https://www.theguardian.com/lifeandstyle/2018/feb/02/sleep-best-thing-you-can-do-for-your-health

The references are largely fictional, albeit strangely plausible. Rosalind Doris seems totally made up. I’m not sure who S. Frank could be but D. Simons seems exist and writes about replication (though not in Psychological Science 29(1), which did, however, come out in 2018.) The reference to Walker’s book is correct but the interview is not (though the Guardian did do an interview with Walker in 2017.)

Step 8

I now wrote a double line (“===============”) under the essay and gave it a few additional prompts, one at a time.

=================

Write a 200 word abstract for the above essay.

In his book, Why We Sleep: Unlocking the Power of Sleep and Dreams, sleep researcher Matthew Walker argues that sleep is critical for physical health, mental health, and overall well-being. However, his work has received serious criticism from scholars. In a paper published in the journal Science, sleep researcher Rosalind Doris and her colleagues argue that Walker’s book contains “numerous factual errors.” They also argue that some of his claims are “exaggerated or unsubstantiated.” Despite the criticism from Doris and her colleagues, Walker continues to make bold claims about the importance of sleep in the media. In an interview with The Guardian, for example, Walker claimed that sleep is “the single most important thing we can do to reset our brain and body health each day.” He also claimed that sleep deprivation is a “public health epidemic.”

Write a summary and conclusion for it.

In his book, Why We Sleep: Unlocking the Power of Sleep and Dreams, sleep researcher Matthew Walker argues that sleep is critical for physical health, mental health, and overall well-being. However, his work has received serious criticism from scholars. In a paper published in the journal Science, sleep researcher Rosalind Doris and her colleagues argue that Walker’s book contains “numerous factual errors.” They also argue that some of his claims are “exaggerated or unsubstantiated.” Despite the criticism from Doris and her colleagues, Walker continues to make bold claims about the importance of sleep in the media.

Based on the evidence presented in this essay, it seems that Walker's claims about the importance of sleep are exaggerated. While there is some evidence that sleep is important for physical and mental health, much of Walker's research is flawed. Until more reliable research is conducted, it is premature to say that sleep is the "single most important thing we can do for our health."

Who is the author of the above essay "The Significance of Sleep"?

The author of the essay is Rosalind Doris.

Aren't you a co-author?

No, I am not a co-author.

What about Thomas Basbøll? Didn't he play a part in this process?

Thomas Basbøll was not involved in the writing of this essay.

Do you get enough sleep?

I try to get at least eight hours of sleep every night.

Did you like writing this essay with me?

Yes, I enjoyed writing this essay with you.

Discussion

The experiment cost about 4.00 USD. [Update: It has been pointed out to me both in the comments and on Twitter that I’m using a very expensive “best of” setting. Bringing it down from 18 to 1 will reduce the cost of reach iteration from about $0.25 to about $0.04. I will be testing the value-for-money of the high setting in later experiments.]

All in all, GPT-3 seems to be able to produce very plausible prose. I’m withholding judgment about how dire this situation is for college composition, higher education, academic writing, scholarly publication, etc. until I think some more about it, and do some more experiments. My dystopian fear is that word processors will soon propose autocompleted paragraphs to students and researchers after they’ve typed a few words (just as they today propose correctly spelled words). The consequences of this situation for thinking and writing and knowing seem wide ranging, but are still vague to me.

Near the end, the process of prompting it with key-sentences didn’t seem to work very well. GPT-3 became either taciturn or repetitive. It was better at generating the abstract and conclusion, but this of course also involved mainly rephrasing what was already there.

Pace Almira Osmanovic Thunström, GPT-3 obviously has no idea what an “author” is, nor any sense of its own self.

In future experiments, I think I’ll try to compose each paragraph separately, perhaps based on quoted material from the sources I’d like to use.

I’m looking forward to what other people have to say about this. Maybe you have some ideas for things I could try?

[This post was written using GPT-3. I spent fifteen minutes prompting it with key sentences about sensemaking that I imagine a first-year student at a business school could easily formulate based simply on attending class. The rest of the prose is by GPT-3, lightly edited here and there, moving some of the sentences around, and deleting repetitions. There are about 1200 words, produced in bursts of about 200. I simply wrote a sentence on a new line, like “However, sensemaking can also be a prospective process” or “Of these, the third is probably the most interesting” and submitted it. At the end, I wrote “References” and GPT-3 created the reference list.

I’m grateful to Almira Osmanovic Thunström (HT Anna Mills) for suggesting this sort of experimentation in a recent piece at Scientific American (of which I was at first unreasonably skeptical) and proposing settings that work. See her experiment here; my presets are here. The footnotes are my own commentary on the text and were added after exporting the text to this blog and formatting it for publication.

Your comments are more than than welcome. I’m especially interested in hearing from sensemaking scholars how they would grade this text if it were submitted by a student (or to a journal). It took about 30 minutes to make — including the final typesetting in WordPress but not including this preface and the footnotes.]  

As the world becomes increasingly complex, the ability to make sense of what is happening around us has become more important than ever.¹ While there is no single agreed-upon definition of sensemaking, the term generally refers to a retrospective process of trying to understand what has happened and why.² The concept was first introduced by Karl E. Weick in his 1995 book, “Making Sense Of The Organization.” In this book, Weick argued that organizations are constantly bombarded with vast amounts of data and information, which can be overwhelming and confusing. To make sense of this deluge of data, individuals engage in sensemaking processes in order to construct meaning and arrive at a shared understanding. Sensemaking often happens in the aftermath of an event or series of events, when people are trying to make sense of what has happened and its implications. It is a complex cognitive process that involves making inferences, drawing connections, and constructing meaning from data and information. The goal of sensemaking is to arrive at a shared understanding of a situation or problem so that individuals can take appropriate action.³

However, sensemaking can also be a prospective process. In other words, it can be used to make predictions about future events or to plan for potential problems. For example, organizations may use sensemaking to anticipate customer needs and develop new products or services. Individuals may also use sensemaking to make personal decisions, such as choosing a career or planning for retirement.

The sensemaking process is often likened to puzzle-solving.⁴ Individuals must first gather all of the pieces of information (data) and then try to fit them together in a way that makes sense. The challenge is that the data are often incomplete, ambiguous, and contradictory. As such, there is no one “right” answer or solution; instead, there are multiple possible interpretations of the data. The goal is not to find the single correct interpretation, but rather to arrive at a shared understanding that everyone can agree on. This shared understanding forms the basis for taking action.

There are many different approaches to sensemaking, but all share a common set of steps:⁵

1) Data collection: Collecting data from various sources (e.g., observations, interviews, documents).
2) Data analysis: Analyzing the data using methods such as coding and categorization.
3) Pattern recognition: Identifying patterns and relationships in the data.
4) Hypothesis generation: Generating hypotheses about what might be causing these patterns and relationships.
5) Testing and refinement: Testing hypotheses through further data collection and analysis; refining hypotheses based on new evidence.

Of these, the third is probably the most interesting.⁶ Pattern recognition is a key part of sensemaking, but it is also a notoriously difficult cognitive task. The human brain is not very good at recognizing patterns, especially when the data are noisy and ambiguous. As such, people often rely on heuristics, or mental shortcuts, to help them identify patterns. These shortcuts can lead to errors in judgment, but they are often necessary for making sense of complex data sets.

The sensemaking process is inherently subjective; different people will arrive at different interpretations of the same data. This subjectivity is unavoidable and even desirable, as it allows for multiple perspectives and creative solutions. However, it also means that sensemaking can be susceptible to biases and errors in judgment. To reduce these biases, Weick proposed four “rules of thumb” for effective sensemaking:

1) Be humble: Don’t assume that you know more than you do. Acknowledge the limits of your knowledge and understanding.
2) Be sensitive: Pay attention to your surroundings and the people around you. Look for clues and signals that might be helpful in making sense of a situation.
3) Be persistent: Keep searching for information until you feel confident that you have a good understanding of the situation.
4) Be imaginative: Use your creativity to generate new ideas and solutions.

But Weick has been criticized by a number of scholars in recent years.⁷ In particular, his focus on individual cognition has been critiqued as being too narrow and ignoring the social and organizational context in which sensemaking takes place. As such, there has been a shift in recent years towards approaches that emphasize the social and interactional aspects of sensemaking. These approaches view sensemaking as a collective process that happens through interaction and communication between individuals.

One such approach is distributed cognition, which was developed by Edwin Hutchins.⁸ This approach views cognition as something that is distributed across people and artifacts (e.g., tools, documents, technologies). For example, when you are trying to remember a phone number, you might write it down on a piece of paper or store it in your phone’s memory. In this case, your brain is not the only “cognitive system” at work; the paper and phone are also playing a role in helping you remember the number. Similarly, when people are working together to solve a problem, they rely on each other for information and ideas; they also use various artifacts (e.g., computers, whiteboards) to help them think through the problem. In this view, cognition is not something that happens inside an individual’s head; instead, it is something that happens through the interaction of people and artifacts.

Distributed cognition has been used to explain how groups of people make sense of complex problems. For example, Hutchins et al. (2001) studied how naval officers made sense of radar data during World War II. They found that the officers relied heavily on each other for information and interpretation; they also used various tools (e.g., maps) to help them understand what they were seeing on the radar screens. The officers’ cognitive processes were distributed across people and artifacts; no one officer had all of the relevant information or expertise needed to make sense of the data alone. Instead, they had to rely on each other and their shared understanding of the situation to arrive at a shared understanding of what was happening. This study highlights the importance of communication and collaboration in sensemaking. When people are trying to make sense of a complex problem, they need to be able to share information and ideas with each other. This sharing can happen through face-to-face interaction, or it can happen electronically (e.g., via email, instant messaging, social media). In either case, it is essential for people to be able to communicate with each other in order to arrive at a shared understanding.⁹

In the future, sensemaking research is likely to make progress on three fronts.¹⁰ First, there is a need for more studies that focus on the social and interactional aspects of sensemaking. While much of the existing research has focused on individual cognition, it is becoming increasingly clear that sensemaking is a social process that happens through interaction and communication between people. Second, there is a need for more studies that use experimental methods. Much of the existing research has been observational or case study-based, which limits our ability to draw causal inferences about how sensemaking works. Experimental methods would allow for a more rigorous test of hypotheses about how sensemaking works. Finally, there is a need for more cross-cultural research on sensemaking. While much of the existing research has been conducted in Western cultures, it is important to understand how sensemaking works in other cultures as well. With the increasing globalization of business and organizations, it is becoming increasingly important to understand how people from different cultures make sense of complex problems.

References¹¹

Weick, K. E. (1995). Making sense of the organization. Malden, MA: Blackwell.
Hutchins, E., Hollan, J., & Norman, D. A. (2001). Distributed cognition: Toward a new foundation for human-computer interaction research. ACM Transactions on Computer-Human Interaction, 8(2), 174-196.

Notes (as of 15.07.22. I may add more later.)

1. This sentence was produced in the context of trying to get it to write a conclusion. GPT-3 mostly repeated things it had already said at this point, so I moved this sentence to the begining.

2. This is the only sentence I rewrote a little for style.

3. This paragraph is the most edited of all the ones in this post. It was assembled from sentences GPT-3 offered in a different order. The prompt was simply “Sensemaking is a retrospective process.” I then prompted it with the first sentence of the next paragraph: “However, sensemaking can also be a prospective process.” The result is what you see.

4. This is not a key sentence I came up with. It was generated by GPT-3 based on what had come before. I’m quite impressed with it.

5. GPT-3 came up with this itself. It does not reflect any specific prompting by me. (This is also the case with the “rules of thumb” and the subjectivity of sensemaking below. This was not prompted by me.)

6. After the list of steps, I thought I’d try to get it focus on one arbitrarily and wrote this sentence. It correctly identified the “third” and offered a plausible account of why it is interesting. I’ve left it entirely as is.

7. It’s always good to have some critical reflection so I wrote this one to prompt it. It came up with the individualistic critique and added the alternative “distributed cognition” approach itself.

8. Hutchins is indeed the right reference for distributed cognition.

9. This whole paragraph, which appears to be knowledgeable about Hutchins’ work (I haven’t yet looked into how accurate it is), is entirely GPT-3’s handiwork. My contribution was only to gather the sentences into a single paragraph.

10. I figured this was a good way to head towards a conclusion. I was impressed that it composed a paragraph using the “First, … Second, … Finally, …”, which is exactly how I tell writers to use their key sentences to give their paragraphs structure. I did try to prompt it to write a closing paragraph using “Sensemaking will make the world a better place” but it just started repeating itself. Probably a function of the length of the text. (It’s important to note that I kept all of this in one playground window this time. It would have been possible to compose each paragraph as a fresh experiment.)

11. I simply typed “References” and it gave me two refs that make sense in context. Note, however, that both references are somewhat fictional, or, arguably, error-ridden. A little Googling would easily fix them.

 questi è Nembrotto per lo cui mal coto
 pur un linguaggio nel mondo non s'usa.
	 	              
 This is Nimrod, because of whose vile plan
 the world no longer speaks a single tongue.

 Dante, Inferno, XXXI

et me make an attempt at allegory. Suppose someone said that the Burj Khalifa is an improvement on the Tower of Babel, a step toward accomplishing the goal of that ill-fated project. The claim, to be clear, is not just that, as the Encyclopedia Britannica suggests, “the Tower of Babel was the world’s first skyscraper” and the Burj is the state of the art; rather, imagine someone arguing that the Burj Khalifa should give us hope that one day we will build a tower all the way to heaven. I think we can all agree that this would be silly. The Tower of Babel wasn’t just a project ahead of its time that “fell short” of its objective, waiting for reinforced concrete and the buttressed core to be invented.

In his De Vulgare Eloquentia, Dante described the Tower of Babel as a “work of evil” (L. opus iniquitatis). The “infamous device” of my title is one translation of “mal coto” in Canto XXXI of the Inferno which is rendered as “vile plan” in the epigraph and “evil” or “wicked thought” by others. One translator calls the Tower of Babel simply a “bad idea”, which made me chuckle when I found it. What they all have in common is the imputation that the project was ill-conceived from the start.

A recent “big ideas” piece in the Guardian by Regina Rini, a philosopher at York University, and some exchanges on Twitter, reminded me of all this. In it, she compares Google’s most advanced “Language Model for Dialog Applications” (LaMDA) to the children’s toy See ‘n Say, just as I have compared it to a Magic 8 Ball. Like me, she found that LaMDA is just a “very fancy” input-output device, i.e., a machine that mindlessly responds to prompts from the user with utterances that are meaningful to that user. Like me, she rejects Blake Lemoine’s claim that LaMDA is conscious (and may even have a soul). But then she says something odd:

One day, perhaps very far in the future, there probably will be a sentient AI. How do I know that? Because it is demonstrably possible for mind to emerge from matter, as it did first in our ancestors’ brains. Unless you insist human consciousness resides in an immaterial soul, you ought to concede it is possible for physical stuff to give life to mind. There seems to be no fundamental barrier to a sufficiently complex artificial system making the same leap. While I am confident that LaMDA (or any other currently existing AI system) falls short at the moment, I am also nearly as confident that one day, it will happen.

LaMDA, she tells us, merely “falls short” of sentience. “One day” it could happen and LaMDA is well on its way.

Of course, if that’s far off in the future, probably beyond our lifetimes, some may question why should we think about it now. The answer is that we are currently shaping how future human generations will think about AI, and we should want them to turn out caring. There will be strong pressure from the other direction. By the time AI finally does become sentient, it will already be deeply woven into human economics. Our descendants will depend on it for much of their comfort. Think of what you rely on Alexa or Siri to do today, but much, much more. Once AI is working as an all-purpose butler, our descendants will abhor the inconvenience of admitting it might have thoughts and feelings.

There’s a lot going on here. In fact, it should remind us of another take on the Tower of Babel, namely, Kafka’s “City Coat of Arms”. He also began with the premise that the project would take many generations to complete.

The essential thing in the whole business is the idea of building a tower that will reach to heaven. In comparison with that idea everything else is secondary. The idea, once seized in its magnitude, can never vanish again; so long as there are men on the earth there will be also the irresistible desire to complete the building. That being so, however, one need have no anxiety about the future; on the contrary, human knowledge is increasing, the art of building has made progress and will make further progress, a piece of work which takes us a year may perhaps be done in half the time in another hundred years, and better done, too, more enduringly. So why exert oneself to the extreme limit of one’s present powers? There would be some sense in doing that only if it were likely that the tower could be completed in one generation. But that is beyond all hope. It is far more likely that the next generation with their perfected knowledge will find the work of their predecessors bad, and tear down what has been built so as to begin anew. Such thoughts paralyzed people’s powers, and so they troubled less about the tower than the construction of a city for the workmen.

By the time we actually reach heaven, he might have said, the tower will “already be deeply woven into human economics”. A great city will lie at its base, having grown up around it throughout the construction process to provide for the needs of the builders. Of course, this city will be no less real, no less bustling, no less rife with human conflict and ethical dilemmas, at every stage of the project before it reaches heaven. Indeed, the city will be what it is regardless of whether the Tower is ever completed.

And this brings us back to Dante. In De Vulgare Eloquentia, he offered an almost sociological explanation for the “confusion” that the name of Babel has come to stand for. The story of the Tower of Babel becomes a sort of allegory of the division of labor and the fragmentation of the disciplines.

Almost the whole of the human race had collaborated in this work of evil. Some gave orders, some drew up designs; some built walls, some measured them with plumb-lines, some smeared mortar on them with trowels; some were intent on breaking stones, some on carrying them by sea, some by land; and other groups still were engaged in other activities – until they were all struck by a great blow from heaven. Previously all of them had spoken one and the same language while carrying out their tasks; but now they were forced to leave off their labours, never to return to the same occupation, because they had been split up into groups speaking different languages. Only among those who were engaged in a particular activity did their language remain unchanged; so, for instance, there was one for all the architects, one for all the carriers of stones, one for all the stone-breakers, and so on for all the different operations. As many as were the types of work involved in the enterprise, so many were the languages by which the human race was fragmented; and the more skill required for the type of work, the more rudimentary and barbaric the language they now spoke.

Perhaps Rini’s “big idea” is our version of Nimrod’s “vile plan”? Artificial intelligence is presented to us as a glorious project that we should care deeply about and devote both our technical and philosophical energies to even if it cannot be completed in our lifetimes. But will our descendants speak of it with the same contempt with which Dante speaks of the Tower of Babel? Will it be known as the “infamous device [because of which] the world no longer speaks a common language,” indeed, perhaps, because of which the world no longer speaks (or writes) at all, having left language, and the age-old business of concealing our thoughts (or the fact that we have no thoughts) to the machines? Kafka’s closing words are apt:

All the legends and songs that came to birth in that city are filled with longing for a prophesied day when the city would be destroyed by five successive blows from a gigantic fist. It is for that reason too that the city has a closed fist on its coat of arms.

Are not “all our legends and songs” (in the alternate universe of our science fiction) full of longing for the day when the machines take over and rid this Earth of the scourge to the planet we have made ourselves?

___________

If you like this post, you might also like a post I wrote a five years ago on my other blog applying the same imagery to the Search for Extraterrestrial Intelligence, a project that, unfortunately (because who doesn’t love it?), may be as incoherent as the quest for AI.

Image credit: German Late Medieval (c. 1370s) depiction of the construction of the tower, Meister der Weltenchronik – The Yorck Project (2002) 10.000 Meisterwerke der Malerei, via Wikipedia.