Geochemistry, Mineralogy, Petrology & Volcanology

I asked ChatGPT to write a blog about lava lakes: here’s what happened

The lava lake of Halemaʻumaʻu at Kilauea, Hawaii

If you’ve not been living under a rock you’ve probably heard of ChatGPT, the AI text generator wowing the internet and striking fear into the heart of exam co-ordinators worldwide.

I’ve been meaning to write a blog about lava lakes for a while now, ever since rewatching Werner Herzog and Clive Oppenheimer’s excellent documentary Into the Inferno. But with deadlines looming and lab work and admin crowding in on my time, I kept not getting around to writing. So, I thought, why not get someone, or something, else to write it for me. And that’s exactly what I did. The results, as we’ll see, are mostly impressive, often well written, and occasionally bemusing.

I started off by asking “Please” (because it’s important to be polite, especially to an AI that might be your digital overlord in a few years’ time) “write a blog about lava lakes”. The following is what I got. To make it seem like I’m not entirely lazy, I’ve followed along, adding my comments and marking ChatGPT’s work. The only change I’ve made is to format the section headings.

Lava lakes are a fascinating and terrifying natural wonder. They are bodies of molten lava that sit in the crater of a volcano, forming a fiery pool that can be hundreds of meters deep. These fiery pits are mesmerizing to watch, with their glowing orange and red colors creating an otherworldly atmosphere. In this blog post, we'll explore the formation, behavior, and hazards of lava lakes.

Honestly this opening paragraph is pretty damn good. It’s well written and engaging. It’s not even wrong about lava lakes hundreds of metres deep. I probably would’ve done a worse job writing this – 8/10.

The only extra things I’d like from an introduction here is a sense of how rare, and often ephemeral, lava lakes are. In recent decades only eight or so volcanoes worldwide have had persistent or long-lived lava lakes: Mount Erebus (Antarctica); Kīlauea (Hawaiʻi); Nyiragongo (Democratic Republic of the Congo); Yasur (Vanuatu); Erta Ale (Ethiopia); Ambrym (Vanuatu); Masaya (Nicaragua); and Mount Michael (South Sandwich Islands). These are truly quite rare phenomena and all the more exciting for it, both for the pure fiery spectacle and for the ability to observe fundamental magmatic processes up close.

Lava lake and volcanic gases at Nyiragongo Volcano, Democratic Republic of the Congo

Lava lake and volcanic gases at Nyiragongo Volcano, Democratic Republic of the Congo (by Cai Tjeenk Willink – own work, CC BY-SA 3.0)

Formation of Lava Lakes

Lava lakes form when magma, the molten rock that resides deep within the Earth, rises to the surface and collects in a crater. The magma can come from either a deep-seated source or a shallow reservoir, but in both cases, it must be in a state of continuous eruption to maintain a lava lake.

The crater must also have a steady supply of magma to keep the lava lake going. If the supply is cut off, the lava lake will slowly cool and solidify, eventually forming a plug that can block the vent and lead to a more explosive eruption.

This is broadly correct – changes in magma supply will indeed lead to changes in the level of the lava lake. Another important effect is gas exsolution and the formation and coalescence of bubbles, which can lead to periodic rising and falling of the lake level in a similar way to how Strombolian eruptions occur.

The AI is a bit wide of the mark on the potential for transition from lava lake to explosive activity. Lava lakes may indeed freeze up if the rate of magma recirculation slows, and there are several volcanoes worldwide which have had periods of lava lake and explosive activity including Masaya and Nyiragongo. However, direct transitions from lava lake to explosive activity don’t appear to be common events. A big part of this is that the low viscosity basaltic magmas which tend to form lava lakes are quite different in composition from the viscous silicic magmas which are most likely to erupt explosively. The phenomena of conduit blockage by frozen magma as a potential precursor to explosive eruption is a real one and ChatGPT seems to have picked that up from somewhere and applied it to lava lakes. Here we start to see the limitation of ChatGPT and a lack of nuance or detail. Overall, 6/10 for this section.

Lava lake in Marum crater, Ambrym, Vanuatu

Lava lake in Marum crater, Ambrym, Vanuatu (by Geophile71 – own work, CC0)

Behavior of Lava Lakes

Lava lakes are constantly in motion, with the molten lava roiling and churning as gases escape from the magma. The lava is always at an extremely high temperature, usually between 1,000 and 1,200 degrees Celsius, which keeps it in a liquid state. The lava is also highly viscous, meaning it resists flow, which leads to the formation of lava domes and flows.

Lava lakes can also create spectacular fountain eruptions, where gas bubbles burst at the surface, spewing lava high into the air. These eruptions can be dangerous, as they can launch hot rocks and debris in all directions.

So much of this section is good: roiling and churning – tick (and bonus points for language); high temperatures – tick; gas and fountaining eruptions – also tick. But then it goes and states that the lava is highly viscous and forms lava domes. This couldn’t be more wrong; lava domes form from effusive eruptions of highly viscous magma, the complete opposite of the type needed to maintain a roiling and churning lava lake. This is one of those examples that show that ChatGPT doesn’t really understand the topic it’s talking about it’s just predicting the most likely words or phrases to follow. Perhaps the books or articles ChatGPT is pulling from go from talking about lava lakes to lava domes as contrasting styles of magmatism. Robbed of that context, we end up with something that’s just plain wrong. 3/10

Satellite picture showing the lava lake of Mount Erebus, Antarctica

Satellite picture showing the lava lake of Mount Erebus, Antarctica (by NASA/EO-1 – Public Domain)

Hazards of Lava Lakes

Lava lakes are incredibly hazardous environments. The intense heat of the lava can quickly kill anyone who gets too close, and the toxic gases that are released can cause serious health problems. Lava lakes can also be explosive, with the potential to generate ash clouds that can travel long distances and disrupt air travel.

Here we’re back on track for a bit. Lava is indeed dangerous, as are the gases they release. Another particular hazard of lava lakes is that they can breach their flanks and rapidly release large volumes of very fluid magma. A notable example of this was the 1977 eruption of Nyiragongo in the Democratic Republic of the Congo. Nyiragongo is a particularly steep sided stratovolcano and when the lava lake breached the flanks of the volcano it travelled rapidly at <60km/hr and killing upwards of 600 people. I’m not sure about ash clouds from lava lakes exploding and impacting air travel though. One again ChatGPT seems to be just pulling in common themes from volcanology and/or news about volcanoes and associating it with lava lakes. 5/10.

In addition, the lava itself can cause significant damage to nearby infrastructure and communities. When lava flows out of the crater and down the side of a volcano, it can destroy homes, roads,

And then ChatGPT just trailed and stopped writing. Maybe I’d used up my allotment of its valuable time for the day, or maybe it just got bored and its mind wandered off to other things.

So what have we learned? We seen that ChatGPT is really good at the basics, writing clearly and engagingly. However, away from the generalities, it seems to mix up various concepts from across volcanology. It is clear that there isn’t truly an intelligence underlying ChatGPT, or at least not one that’s at risk of putting volcanologists out of a job for another few years at least. It doesn’t really understand how concepts tie together (especially magma viscosity and eruptive behaviour) and so it doesn’t notice when it’s stating things that are directly contradictory. This isn’t a surprise when we understand the underlying model which, in simple terms, is based on reading a huge amount of written material and then predicting the most likely words and phrases to follow one another. It’s been well-documented that this can result in invention of non-existent academic articles and even completely invented lyrics to songs.

To round this all off, if you’re feeling lazy and need a vague and general introduction to something – or a contrived and desperate concept for a blog article – ChatGPT might be your bot. However, if you want it to write you a novel research article or an answer to a sufficiently esoteric exam question, I’m sorry to disappoint you but these are not the droids you are looking for.

Further reading

Oppenheimer, C., Lomakina, A.S., Kyle, P.R., Kingsbury, N.G. and Boichu, M., 2009. Pulsatory magma supply to a phonolite lava lake. Earth and Planetary Science Letters284(3-4), pp.392-398. link

Tazieff, H., 1994. Permanent lava lakes: observed facts and induced mechanisms. Journal of Volcanology and geothermal research, 63(1-2), pp.3-11. link

Helz, R.T., 1987. Differentiation behavior of Kilauea Iki lava lake, Kilauea Volcano, Hawaii: an overview of past and current work. Geochemical Society, Special Publication1, pp.241-258. link

Diller, K., Clarke, A.B., Voight, B. and Neri, A., 2006. Mechanisms of conduit plug formation: Implications for vulcanian explosions. Geophysical Research Letters33(20). link

Avatar photo
Elliot is a postdoc at Trinity College Dublin, working on the geochemistry of the North Atlantic Igneous Province. He recently spent 2 months aboard R/V JOIDES Resolution as part of IODP Expedition 390 and will tell anyone who will listen. You can find him on twitter @elliotjcarter

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>