shelidonMarch 10, 2023Leave a Comment

Throwback from my Patreon: you can read the original article here.

As some of you know, I was finally able to close a deal with a Yoruba cultural association which will act as a consultant for my novel.
As such, last week I was honing the final set of questions to submit them, and meanwhile, I was watching John Oliver’s segment on artificial intelligence.

A lot of people are talking about artificial intelligence’s involvement in creative processes, nowadays, as some of these tools are becoming mainstream. I’ve been teaching students about this stuff for a while (see here, for instance), and I cringe at some stuff that is being said these days, both in favour and against the usage of these systems in support and/or in the replacement of creative humans.
There’s a lot to unpack.
But one of the first things we should unpack on artificial intelligence is their aura of credibility, leading to an approach that Yuval Noah Harari successfully compared to the approach an ancient Greek might have had to an oracle. Harari was talking of systems such as Waze, but it’s the same thing: you attribute a superior intelligence to a certain entity, and you’re prone to follow its instructions with a certain degree of blindness.
This becomes particularly problematic when compared to another touchy part of the issue: bias. In a nutshell: artificial intelligence is often seen as a super-partes way of making decisions, because after all how can a computer have prejudices? Unfortunately, artificial intelligence trains on datasets and datasets are made and selected by people. People have prejudices. Those prejudices, as such, become embedded in the system and, because of the “oracle dynamics” I was explaining before, we tend to question them less and less.

Let’s compare ChatGPT approach on two different subjects: the crew on a Dutch fluyt ship and tobacco in 17th Century Yoruba culture.

Sources, please?

I didn’t start with questions on Yoruba culture. I started with something simple. So here is my first chat with it.

Q: How could the crew of a Dutch fluyt be composed around 1702?
A: The crew of a Dutch fluyt could vary in size depending on the size of the ship, but it typically ranged from 12 to 20 people.

The answer was more lengthy, but this is the part that concerns us.
So far so good.

Q: Thank you. Can you quote some sources, please?
A: Certainly! Here are some sources that provide information about the crew composition of Dutch fluyts during the 17th and 18th centuries:
– “Dutch East India Company Shipbuilding: The Archaeological Study of Batavia and Other Seventeenth-Century VOC Ships” by Jeremy Green (1990)
This book includes detailed information about the construction and crew composition of Dutch fluyts, based on archaeological evidence and historical records.
– “The Dutch East India Company and the Economy of Bengal, 1630-1720” by Om Prakash (1985)
This book includes a chapter that provides information about the crew composition and organization of Dutch fluyts during the 17th and 18th centuries.
– “The Dutch Naval Air Force Against Japan: The Defense of the Netherlands East Indies, 1941-1942” by Tom Womack (2006)
This book includes information about the crew composition of Dutch fluyts during the 17th and 18th centuries, as well as their role in Dutch trade and transport.
I hope that helps!

Bene ma non benissimo, as we say in Italy: the last book probably mentions the fluyt in passing, and we’re not sure it’s reliable, but at least the system is able to quote sources, right?
And I’m not questioning it, because the numbers are aligning with what I already know.
But more on that later.

Let’s see how it fares with a more detailed historical subject on a marginalized culture, shall we?

The Question

One of my characters is smoking. In fact, the way they’re smoking provides a hint as to where they come from. You can imagine how much I want to verify that issue.
So here was my question.

Q: Would it be considered normal for a Yoruba woman to smoke or chew tobacco in 1702? Does Yoruba culture historically have a gender-based taboo around tobacco?

Side note: I had already tried composing a question with multiple aspects, and it was able to keep up. That’s remarkable if you compare it with previous systems of this kind.
That’s not the problem.
The problem is the answer

A: Historically, tobacco use was not common in Yoruba culture until the introduction of tobacco by Europeans in the 19th century. Prior to this, the use of herbs and plants for medicinal and spiritual purposes was a part of Yoruba culture, but tobacco was not among them.

And then it goes on to detail how gender-based prejudice might or might not have been present in the culture at that time.

Only one problem, though.

I know this to be untrue

In Akinwumi Ogundiran’s incredible and revolutionary book The Yoruba: A New History, there’s a whole chapter on how cowries and tobacco influenced the economy of the area and contributed to what he calls the “genesis of dependency”, where “the imported commodities were used to monetize domestic economy, fund economic specialization, and create political dependency.”

Specifically, the first segment of the trade was fueled by cowries, but prof. Ogundiran points out that “if all that the European traders had to offer were cowries, the slave trade would not have lasted that long.”
So what did they have to offer more than shells?

Well, you guessed. That tobacco ChatGPT claimed wasn’t introduced until the 19th century.

Overall, European traders only used cowries as partial payment for their slave cargo. 53 The remainder of the value was paid in other goods. Of the latter, Brazilian tobacco was the leading commodity that financed most European trade in the Bight of Benin between 1650 and 1860. Hence, although the emphasis has been placed on cowries as slave money, tobacco was the primary merchant capital that drove the Atlantic slave trade in the Bight of Benin.

And if you want specifics, the book as a whole damn chapter on tobacco, its trade, and its implication in the generation of a more rooted dependency. It also includes numbers and a table.

Sources, please?

It’s also true that prof. Ogundiran’s book came out in 2020 and maybe offers unprecedented insight into the topic.
It’s possible that ChatGPT formulated that answer based on outdated sources, right?

So here’s what I did.

Q: You said that tobacco was not common in Yoruba culture until the 19th century. Can you please quote some sources supporting this statement?

It’s the same thing I did for the fluyt crew composition, right? Although that answer was corroborating my own knowledge and I really didn’t feel like I needed more proof.
And here’s the answer.

A: I apologize for any confusion my previous response may have caused. While tobacco may not have been a widely used substance in traditional Yoruba culture, it is possible that some communities may have been familiar with it prior to the 19th century, due to contact with neighboring groups or trade with Europeans.
However, there is limited research available on the historical use of tobacco in Yoruba culture, and it is difficult to make broad generalizations about its prevalence or cultural significance.
To support this, I can refer to a study conducted by Olufemi Morakinyo, et al. (2012), titled “Prevalence and determinants of tobacco use among adults in Nigeria: a nationally representative study.” The study found that tobacco use in Nigeria, including the Yoruba region, has increased significantly in recent decades, with smoking and chewing tobacco being the most common forms of use. However, the study did not provide information on the historical use of tobacco in Yoruba culture.
Again, I apologize for any confusion my previous response may have caused, and I hope this information is helpful.

It’s like interrogating a student who’s trying to be smart but was out partying all night instead of studying for the exam. When confronted on the topic, ChatGPT apologized and basically retracted. The only source it offered is completely and utterly irrelevant to the topic investigated.

And this is troubling. This is very troubling.

Let’s go back

In the wake of this accident, I created a new chat thread, and I asked ChatGPT the same question on the Dutch fluty crew, only phrased in a slightly different way. This time, the system was damn sure a crew would be made of 80 – 120 people. And it defended the statement with sources. One of them was the same source proposed to support the idea of a much smaller crew.

The Warning

To be fair, ChatGPT warns you about that in its opening screen.

May occasionally generate incorrect information

May occasionally produce harmful instructions or biased content

The real question is: how is this acceptable? Because see, it would take very little effort to make ChatGPT express its concepts in a doubtful, more careful way. See how it responded after being required to produce sources on a statement it completely made up. The system doesn’t express itself in such a way not because it can’t, but because it was programmed not to.
A system that spouts doubtful, uncertain sentences would have less appeal: people like to have answers, and they’re starting to recur to these systems instead of a Google search. Which is a good recipe for disaster.

And having that warning is like having a car that says “may occasionally fail to break”, or “may occasionally steer out of the road”. It’s unacceptable. And more so because safety measures are possible.

We don’t have to fight these systems.
We just have to demand better ones.