In one of my early experiments with ChatGPT, I asked it a question that I’m sure every architect has considered at some point: Why do lawyers make more than architects? I’ve been considering this haunting question since 2007, when I read Peter Eisenman’s lament that his son, who’d graduated from law school three years prior, made more than he did after 40 years of field-leading practice. 

I asked ChatGPT this question because I’ve never really found a satisfactory answer, either from architects, lawyers, professors, economists, or anyone else. It just seems to be one of those truths at which we collectively shrug and offer tepid rationalizations like Architects undersell the value of their work. Really? Why? Because they all have self-esteem problems? Or enjoy poverty? Do you know many architects who aren’t fans of their own designs?

I thought maybe ChatGPT would know, or at least have some ideas. It seems to know a lot. Its training model was purportedly the entire internet, and if I met a human who had read the entire internet, the first thing I would do would be to ask them all the stubborn, unanswered questions in my life.

Reasoning, Not Recall

Sam Altman, cofounder of Open AI, has repeatedly stressed that ChatGPT is a “reasoning engine,” not a knowledge database, which becomes clear when you use it regularly. ChatGPT doesn’t have the answers—it’s generating novel ones based on some insanely complicated prediction mathematics. It’s trying to figure things out, which is probably why it sometimes returns incorrect or even nonsensical responses. 

Case in point: it gave me a whole host of reasons that lawyers make more than architects, none of which made any sense (a bit disappointing), but largely in keeping with every human response I’ve ever heard, which, in itself, makes a certain kind of sense. If, in reading the entire internet, ChatGPT found 1,000 instances of the question about lawyers and architects and in 900 cases the response was Because lawyers require more rigorous training, it’s trained to anticipate that response as the most probable correct response, even if it isn’t true. I’m dramatically oversimplifying how ChatGPT works, but you get the idea.

In a chat that would eventually last almost four hours, I asked ChatGPT, Why do lawyers make more than architects? in a variety of ways and interrogated its responses. It initially gave me five reasons, paraphrased below:

1. Education and Licensing: Lawyers require more training and more rigorous licensure.
2. Supply and Demand: The demand for lawyers often exceeds supply. Demand for architects varies greatly, depending on the economy.
3. Billing Structure: Lawyers get paid by the hour, and on retainer. Architects charge by the project, resulting in lower compensation.
4. Risk and Liability: Lawyers take on higher risk and liability.
5. Perceived Value: Society perceives legal work as being of high importance, warranting higher compensation.

ChatGPT’s responses seemed plausible above all else, and roughly in line with basic macroeconomic principles. If any Profession A requires more training, takes on more risk, and is in more demand than any Profession B, we would expect Profession A to earn more. 

That Can’t Be Right  …

However, I knew immediately that some of these assumptions weren’t true: e.g., it takes far more training to become a licensed architect than a licensed lawyer. ChatGPT knows this too, and was separately able to accurately summarize all the training requirements for both lawyers and architects for me.

So how could it then reason that lawyers’ high compensation relative to architects was partially explained by the fact that:

Law is generally a postgraduate degree requiring a bachelor’s degree first, then typically three years of law school. In comparison, you can become an architect with a bachelor’s degree, though some may also pursue a master’s degree. Passing the Bar Exam to become a lawyer is also typically quite challenging and requires substantial preparation.

While simultaneously “knowing” that entry requirements for architecture are longer and more rigorous?

My intuition at the time was that it was just reasoning backward. Compensation generally correlates with higher levels of training and more rigorous exams, and lawyers make more than architects, so one might assume that the path to becoming a lawyer must be harder and more rigorous. 

It occurred to me that it wasn’t ChatGPT’s fault—this is also the way most humans think. It was merely imitating widely held human inferences. (So much for artificial “intelligence.”)

Rather than immediately confront ChatGPT with its own logical contradiction, I did what a good lawyer does: I gradually built a case by picking apart all five components of its initial reasoning. If it could admit that all of its component logics were incorrect, perhaps it would be willing to revisit its own answer to my initial question.

Reason 1: Education

Education was the first and easiest. I merely asked it to enumerate the education and licensing requirements for each profession and compare them side by side. Then I asked it to assess which profession had the greater, longer, or more rigorous training requirements, like so:

I then pressed the attack, pointing out that this wasn’t exactly an apples-to-apples comparison. While a bachelor’s-plus law degree would take 7 years, the first four years didn’t necessarily include the study of law. One could major in anything, really, and proceed on to law school. A better way to calculate the length of study would be through assessing the required credit hours necessary for each course of study. That becomes difficult, because there are several different acceptable degree courses in architecture. Moreover, each institution is different. But, as a ballpark figure, and excluding general electives, it takes about 115 dedicated credit hours of study to complete a B.Arch, and about 123 for an M.Arch (three-year program). 

It only takes 80 to 90 hours for a J.D. Somewhere between 30% and 50% more schooling is required to become an architect. It’s not even close. Case closed. We moved onto Reason 2.

Reason 2: Supply and Demand

GPT’s observation that the demand for legal services often exceeds the supply of lawyers might be true. Its observation that the demand and supply for architects can vary greatly based on the health of the construction and real estate industries is certainly true. But I couldn’t get it to explain why these two things being true would explain a wage discrepancy. One thing being consistently high and another thing varyingly greatly don’t mean that one would necessarily exceed the other, overall. Moreover, supply and demand issues tend to correct over time—as demand increases, wages increase, and more people flow into the profession. 

Between 2010 and 2020, lawyers and architects in the U.S. experienced nearly identical growth, relative to population, and fairly similar wage growth as well:

Which suggests that the number of architects and lawyers, as well as their relative compensation, is probably right where it should be, at least according to the market. ChatGPT vaguely agreed.

Reason 3: Billing Structure

According to ChatGPT, one of the reasons that architects make less is because they work off of a fixed-fee project basis, while lawyers bill by the hour. Therefore, when a project encounters unforeseen challenges, the architect ends up eating his own fee, thus earning less. While that’s not uncommon in our profession, I pointed out that architects have all sorts of fee mechanisms that they deploy—like contingencies, cost-plus, additional services, etc.—that can take some of the risk out of fixed-fee contracts. It agreed. 

I also pointed out that in most contractual arrangements, the party taking on the highest risk is accorded the largest fee, so even if an architect is on a strictly fixed-fee contract, they’re certainly absorbing a lot of the client’s risk in the progression of the project, and would therefore be entitled to higher fees, not lower ones. ChatGPT vaguely conceded the point, but then began to waffle.

It waffled through the whole experiment, actually. All four hours of it. When asked to combine several obvious truths like higher risk should mean more fees + fixed fee contracts means more risk for the architect = architects should earn higher fees, it seemed to get flummoxed. It seemed as if it could not square that conclusion with its prior one: architects earn less because they take on fixed-fee contracts and hack their own fees when the project goes sideways. As if asking itself Why don’t architects just price the risk and add that to their fee upfront? Why indeed, ChatGPT, why indeed.

Reason 4: Risk and Liability

ChatGPT had opined that one of the reasons that lawyers earn more is because they take on more risk. When asked for specifics, it enumerated five types of risk endured by lawyers: 

1. Financial risk (for lawyers who work on contingency)
2. Reputational risk (the risk of losing a case)
3. Ethical and legal risk (they can be sued for malpractice)
4. Career progression risk (taking on a high-profile case and losing)
5. “Emotional risk” (which it explained thusly: Law can be a highly stressful profession, with demanding clients, tight deadlines, high stakes outcomes, and other pressures. This emotional toll can affect lawyers’ mental health, and could be considered another form of risk they take on.) 

It was the first time I’d actually made a physical wanking motion at my computer during this conversation, but it would not be the last.

I pointed out that architects—and indeed all professionals—face these risks as well, and that architects face myriad external risks over which they have no control (unforeseen site conditions, mercurial clients, etc.). The lawyer’s risks that ChatGPT had pointed out were mostly related to a lawyer’s own performance. If a lawyer was terrible at his job and lost a case, it wasn’t so much a risk as much as it was that lawyer’s just being a bad lawyer. ChatGPT seemed to concede the point.

I argued that if an architect was taking on so much risk external to her own performance, she should be compensated more than the lawyer, who was only managing the risk of himself being a shitty lawyer. Again, it waffled: 

In essence, you are correct in observing that architects may assume more risks related to external, uncontrollable factors while lawyers contend more directly with risks associated with their personal performance. This difference in risk profiles is indeed one of the factors influencing the earning potential in the two professions. However, it’s also worth bearing in mind that these are just part of a wider array of factors influencing earnings, from market demand to years of training to societal perceptions and more.

It repeated this parry multiple times. Whenever I would point out the flaws in its component logic, it would return to a general, abstract there’s many factors that affect compensation argument. Interestingly, arguing like a human once again.

Reason 5: Perceived Value

Finally, I interrogated ChatGPT about why it thought that society perceived legal services were of “high importance” relative to design services. To be fair to ChatGPT, it made no claim that legal services were of more actual value than design services; it acknowledged that both lawyers and architects perform vital roles in society. But it opined that legal services were perceived as being more valuable, mostly because of two things: 

Visibility: The first was visibility. ChatGPT hypothesized (correctly, I think) that one of the reasons lawyers are perceived as having a greater value is because we see a lot more of them. Even if we don’t use them, they’re portrayed in all media forms as doing critical, dramatic, often life-saving work, even when it’s the case that most lawyers spend most of their days reading.

Direct vs. Indirect Impact: The second turned on direct vs indirect consequences. The successes or mistakes of an attorney are visceral and can be felt in real time. If you’re standing trial for murder and your lawyer does a bad job, you go straight to jail for the rest of your life. If you’re renovating your home and the architect does a bad job, those effects might not be felt for years or decades, depending on the error.

Taken together, architects make less than lawyers because, well, people generally think that they should. Where would they get such an idea? Unfortunately, I think it starts at home and finds its roots in something familiar: architects often value their work product more than their compensation for it. Once they do, the following nonsense starts to seem logical, even necessary:

• Working for free.
• Asking other architects to work for free.
• Doing a schematic design for a client, for free, that they can use in their marketing materials while they raise money.
• Bidding a project below cost in order to “get the job.”
• Balancing the books by underpaying junior staff.
• Accepting a bunch of risk on behalf of the client, without being compensated for it, in order to get the project moving.

Squaring the Circle

Once I had sufficiently broken down all of ChatGPT’s component rationalizations, I asked it to reassemble them in the form of a table, comparing each of its rationalizations on a line-by-line basis, and evaluate which profession took on the most risk, which had the hardest training requirements, etc. The table eventually looked like this:

By breaking down ChatGPT’s original answer to the Architects vs. Lawyers question, and getting it to analyze its own response at a more granular level, I was able to help it see that architects do indeed have longer and more rigorous training requirements, have analogous billing structures, take on more risk, and have similar levels of social value. Moreover, the categories where lawyers justify their inflated compensation seem exclusively perceptual. Across any measurable, objective criteria, architects and lawyers are either on par, or architects exceed their lawyer colleagues in the professional rigors that demand compensation.

To conclude the exercise, I asked it to evaluate the table and make a judgment about why lawyers tend to earn more than architects. It subsequently reversed its original answer, concluding that there was no good reason for the discrepancy, and that architects should make at least as much, if not more, than lawyers.

Just kidding. It immediately defaulted to its original conclusion. 

Much like a human, when ChatGPT is confronted with its own logical errors, it will sometimes self-correct. Also like a human, it will also stubbornly hold onto two contradictory truths and argue both of them at the same time. So ended my experiment.

The Middle Wisdom of Humanity

I came away feeling as if I’d had a glimpse into one of the shaky certainties of humanity. Our stubborn tendency to “know” things, even when the logics that underwrite them are exposed as incorrect—what Stephen Colbert called “truthiness,” back when he was funny.

If AI is going to be the “Google” of the future (the reference set that we all consult when we want to know the “real” answer to a question), then it’s troubling that this machine can’t ultimately explain why architects make less than lawyers, but seems to think that they should.  The ideas/biases/certainties that are being programmed into AI today may well become the default certainties of humanity tomorrow. And ChatGPT seems to have already assimilated the architects are underpaid trope, and is confident of its reasoning. It was tempting to come away from the exercise with a deep suspicion about ChatGPT’s ability to reason, but I don’t think that’s it.

I think maybe I was just asking it to make sense of something that doesn’t make sense. All that it is programmed to do is to fashion responses that an informed, competent human would offer. And why would any rational human go through more training, more risk, and more work in order to earn less? Why would any rational human be so passionate about their work product and its value, and then devalue the labor that created the product?

I had given it an impossible circle to square. Architecture’s approach to its own value proposition violates basic laws of risk and reward, supply and demand, etc., so much so that even a genius-level A.I. can’t make heads or tails of it. So maybe we’re the ones who are reasoning backward.

Featured image: Midjourney V5 responding to the prompt “a photograph of a full body shot of an average lawyer, white background” on the left, and the prompt “a photograph of a full body shot of an average architect, white background” on the right. There are many interpretations of the image, and much to be gleamed from what A.I. “thinks” a lawyer looks like, vs. what an architect looks like. (They’re all  white and male, for one thing!) But I think the relevant part for this article is that lawyers are depicted as white collar professionals, wearing suits, etc. Architects are portrayed as something less professional, and in two cases look remarkably like contractors. —EJC