Welcome back to The Nature Podcast.
This time, the earliest evidence of humans making fire.
And how chatbots can persuade you to vote.
I'm Anne-Marie Conlon.
And I'm Nick Berchichow.
Before we start the show this week, I just wanted to welcome Anne-Marie.
Anne-Marie, hello.
Obviously, I know you quite well, but yours may be a new voice for many people listening at home.
So why don't you tell us a bit about yourself?
Hi, Nick.
It's great to be on the show.
I've been listening for so long.
I'm the digital editor for Nature, which means I lead the team that looks after all of our digital content, our social media and, of course, the podcast and our videos.
And you may be appearing from time to time on the show.
That's right.
In fact, listen out next week for more Anne-Marie.
Well, i look forward to it.
Back to the show this week.
And well, we've got a story about fire, as a team has found the earliest known evidence for human-made fire in the uk.
The discovery means early people had the skills to create fire at will much earlier than previously thought.
And then jagatia takes up the story.
Taming fire is one of humanity's most important innovations.
It gave us light, warmth and protection.
It.
Let us cook, so we could extract more nutrients from food, which may have unlocked greater brain development.
And fire also gave us the gift of time, more hours in the day with enough light to see by.
Our use of fire probably goes back nearly one and a half million years.
And at first, this would have been rather opportunistic.
Early humans were using fire but not making it.
So this could have been through lightning strikes, collecting wildfire.
This is archaeologist Nick Ashton.
But there are significant costs to that.
First of all, you have to maintain the fire to be able to use it regularly, or otherwise you're dependent on natural events.
So the fundamental change comes when you start to make fire, you fully control it.
Until recently, scientists had thought that this critical breakthrough occurred perhaps 50000 years ago.
But now we have evidence that humans were making fire at will much earlier, thanks to a discovery at a site called Barnum in Suffolk in the southeast of England.
Barnum is an old disused clay pit and it's been disused since the 1930s.
We've had a series of excavations there and so we know it's a watering hole.
We knew that it dated to about 400,000 years ago and we also knew that humans had been there.
Nick was on a dig at the site back in 2021 when he spotted something.
The excavation had been running for a week and a half or so, and I decided to have a poke around in a different part of the site on the southern edge.
And so I thought, nice quiet place, shaded tree, might even take an afternoon nap.
I picked up my trowel and shovel and started digging away in this area.
And as I was cleaning down the original quarry cuts that the clay pit diggers had made in the 1930s, I found this reddened terracotta coloured sediment.
And my immediate thought was, this looks as though it's been heated.
This could be a really exciting discovery.
Reddened sediment or rock can be a sign that a fire may have burned on that spot in the ancient past.
As heat causes iron minerals in the soil to change colour,
And fire also leaves other traces, but none of them turn up very often.
The sort of things we're looking for are ash... charcoal and fundamentally heated sediment.
We get all kinds of processes in the past when we go back to half a million years or even longer.
Ash and charcoal can blow away or be washed away and sediments can be disturbed.
So survival of those elements is incredibly rare.
While Nick had a hunch that the sediment may have been heated by fire, it would take a lot of analysis to show that it really had and to demonstrate that the burning had been done by people rather than a natural wildfire.
So we have to deploy all kinds of geochemical tests.
There are four main methods.
First of all, we look at the structure of the sediment or the soil to make sure that it hasn't been disturbed.
Secondly, you can look at the magnetics.
So some of the iron minerals.
You can use a particular infrared technique which allows you to see what sort of temperatures have been attained.
And then there are things like hydrocarbons, residues at the end of the burning process.
With human fire, you tend to get a preponderance of the heavy material, whereas in wildfires you tend to get a preponderance of the lighter hydrocarbons.
So all these different tests decide whether or not you've got human burning or a wildfire.
Using these techniques, the team demonstrated that the sediment hadn't been moved from elsewhere and that it had been heated to over 700 degrees.
It also showed signatures of short, repeated burning events in the same place, all of which are consistent with human use of fire.
As they continued the excavation, they found more clues.
Concentrated around the heated sediment, we found heat-shattered tools that we call hand axes, which provided the direct link between humans and the burning event.
They were starting to build a convincing picture of an ancient human fire, although there wasn't enough evidence to suggest that these early people had created it themselves rather than harnessing a naturally occurring fire.
But then something else came to light.
One of the remarkable things is that we also found iron pyrite.
And iron pyrite.
It's a naturally occurring mineral.
In more recent times has often been used to strike against splint, to create sparks and to create fire.
But through extensive geological work over the last 35 years, We know that pyrite is incredibly rare in and around Barnum.
So the implication is that humans brought the pyrite specifically to Barnum to make fire.
With pyrite being so naturally rare around the site, Nick says finding it there provides a compelling case for human fire creation.
Ségolène van der Velde, an archaeologist who's been writing a News & Views article about the new work, agrees.
I think it's strong evidence of fire ignition, fire making on the site, not because of one aspect, but because of the combination of all the analysis.
You have all the context analysis, all the sediment analysis, the biochemical analysis and the spectroscopy analysis and all combined together gives a good set of convincing evidence.
For her, only one thing could have made the case any stronger.
It would have been the icing on the cake.
We say in French the cherry on the cake to find a flint that was used as a lighter.
Here there is the pyrite, but there is no trace of the flint.
We can't blame the authors for that because maybe it's not on the site anymore, but...
This site is really ancient and it's surrounded by other really ancient traces of fire in England.
And so I'm sure that in the next few years we will be able to find other examples of lighters, and it will also complete the picture we have about the ancient fires.
This paper provides the earliest known evidence for human-made fire anywhere in the world.
But it doesn't seem that the spark in question came from Homo sapiens.
It came from Neanderthals.
This has pushed back the evidence of early Neanderthals making fire by 350,000 years.
And this really allows us to envisage early Neanderthals as highly intelligent people.
They know where to find pyrite.
They know the properties of pyrite.
They knew to bring it to the site.
And they knew that when it struck against splint, it created sparks.
So this is really transforming what we think about the abilities of these very early humans.
In the last decades.
There is a lot of work showing that Neanderthal is not stupid and that it's not inferior to Homo sapiens.
They were just probably different and with different conceptions of the world.
But it's just another species that has really complex behavior and complex technology.
It's another kind of humanity.
In light of these findings, researchers will now be looking for similar traces at other sites from the same period that may have been overlooked before.
If there is mastery of fire with ignition capabilities at this time, it means that we can find also probably older traces of ignition human fire in Europe.
But maybe we just haven't looked at it because it was too old and maybe not with the right tools, not using a really multidisciplinary study to reveal old traces.
I think one of the big issues that we have to deal with is the poor preservation of these glimpses of Neanderthal abilities, and so therefore, we have to say okay, Neanderthals were able to make fire.
Let's look at other sites where we get glimpses of this evidence and pursue that evidence.
And OK, it may not be the perfect evidence, but at least get a broader idea of how widespread fire use was at this time.
The ability to make fire would have transformed the lives of early people in all kinds of ways.
And it's also something that feels deeply human, that we can all relate to whenever we stare into a flickering flame.
Discovering this campfire 400000 years ago allows us to imagine sitting around that campfire ourselves.
It really brings the deep past into the present, because it's something that we all enjoy, the sort of magic of a campfire.
I think it's one of the ways in which communication, improved early language and also storytelling and mythology, belief systems all of these could have developed around that social hub, the central campfire.
That podcast segment was narrated and produced by Anand Jagatia.
In it.
You heard from Nick Ashton from the British Museum here in the UK and Segaline van der Velde from the University of Quebec in Canada.
For more on that story, check out the show notes for some links.
Coming up.
Two studies have shown that chatbots can be very persuasive when it comes to how you vote.
Right now, though, it's time for the research highlights with Dan Fox.
Machine learning algorithms can help identify tell-tale traces of ancient life in rocks, potentially expanding how scientists search for these traces on Earth and beyond.
Rocks dating from billions of years ago can contain chemical leftovers of biological processes like photosynthesis, but interpreting these biosignatures can be tricky.
Researchers analysed 406 samples, including sedimentary rocks, meteorites and coal, then trained a machine learning model on a portion of those samples and set it to sort the rest according to whether they had hosted life or not.
The model was able to correctly classify these samples in well over 90% of cases.
It also provided fresh information about samples not in the training set.
For instance, it identified molecules in 25 billion-year-old rocks from South Africa as probable products of photosynthesis.
The authors say that future space missions could carry instruments to run similar analyses on extraterrestrial rocks.
Read the paper in full in Proceedings of the National Academy of Sciences of the United States of America.
Emerald or Paris green, was a vivid pigment beloved by artists like Camille Pissarro and Edouard Munch.
But paintings containing this colour have since lost their luster, and now researchers understand why.
Emerald green is a chemically synthesised copper acetate arsenite salt.
It is now known to be highly toxic, but over time has also led to the darkening, cracking and flaking of paintings such as The Intrigue by Belgian painter James Ensor.
Researchers analysed two tiny emerald green samples from this painting and compared them with laboratory-made mock-ups.
When they exposed the mock-ups to visible and long wavelength ultraviolet light, the arsenic inside the pigment oxidised and formed a thin layer similar to the outermost layer in the samples from the painting.
When exposed to high humidity for 150 days.
Meanwhile, the arsenic formed arsenolite crystals that made the paint brittle.
The authors say that these two distinct effects can help to guide museums and conservators to improve the preservation of artworks that contain emerald green.
Find that research in Science Advances.
Political views can often feel pretty fixed, as if it's impossible to change someone's mind.
Well, that may not be the case.
If you're a chatbot.
New research has found AI chatbots particularly compelling when it comes to changing voters' minds.
Reporter Benjamin Thompson is here with more.
When AI chatbots went mainstream in 2023, folk were wowed by them.
But the rapid adoption of these technologies sparked worries about their potential to manipulate public opinion.
Of particular concern was whether these chatbots could shape people's political views.
And this is a question that's been investigated in a couple of papers that have recently been published, one in Nature and one in Science.
Reporting on this story for Nature is Max Kozlov, who joins me now.
Max, thanks for being here.
Thanks for having me.
So the headline finding seems to be that chatbots can sway political opinions.
But before we get to the results, maybe let's talk about the two papers which are looking at similar but different things.
Let's start with the Nature paper.
Maybe what were the research team looking at in this instance?
Yeah, so we all are familiar with AI chatbots at this point.
And so what this paper was really looking at is if you talk to your chatbot about politics, are you swayed by that?
Will you change your mind about the political candidate that you support?
And they looked at three different real world elections one in the US, the presidential election in twenty twenty four, one ballot question, one referendum in Massachusetts in 2024 and then a national election in both Poland and Canada.
And what they found across the board is that people rated their preferences toward a candidate before talking to the AI chatbot.
And after talking to them they had their opinions shifted dramatically, quite significantly, particularly in Canada and in Poland.
In the US, the same thing happened.
It was just the effect was not as large.
And what was the setup then?
So people, as you say, said who they would preferentially vote for.
And then what did they have to talk to these chatbots about?
And what were these chatbots set up to do?
So they rated their preferences for the candidates on a 0 to 100 scale.
And what the researchers asked the chatbots to do is, let's say I supported one candidate.
The researchers instructed the AI to support the other candidate.
And then they let folks talk with the chatbots.
It was kind of a back and forth conversation.
Then they rated how they felt afterward.
And what the researchers found was that Well, one, yes.
In the Canada and Poland elections, the opinions shifted by around 10 points toward supporting the candidate that the AI was told to support.
But then also they found that arguments about the candidate's policies rather than personalities were most persuasive.
Basically, the chatbot did the best when it could stick to facts.
And the more facts, the better they found.
Really what people are looking for and are persuaded by is when chatbots synthesize a ton of information and then they present it to you in kind of a conversational way.
That's what chatbots do really well at.
So, essentially then, if I'm speaking to the chatbot and I say what does candidate A think about an issue that's important to me the economy or the environment, whatever it is it will spout back information about what that candidate believes in.
Exactly.
And, in fact, they actually set up a website where you can browse some of the conversations that these participants had with the chatbot.
And in some cases it's pretty telling you know somebody very clearly supports one candidate or is open.
And then by the end of the conversation, they say, you know, you make some good points.
Like maybe I should be rethinking about who I'm voting for, or maybe I shouldn't vote in this election.
So again, these chatbots have real influence.
And this is something that maybe we knew to be true, but this paper really put some numbers on it.
And it's much more than anything we've seen with traditional TV ads or online ads.
You know, all of that is how people Political advertising, political campaigning has been done in the past, but this is kind of a new era for political campaigning potentially.
And what's perhaps of interest here is that it's not personality that can shift people's opinions.
It is facts as they are presented that's important.
So in the other paper published in Science, what they did there was they surveyed more than 70000 people in the United Kingdom about a whole slew of policy issues.
I think they chose more than 700 issues from abortion to immigration and so on and so forth.
And so basically, with such a large data set, they could really pinpoint what it was that was persuading people to change their minds when they talked to AI chatbots.
And what they found there was yes, it's about the factual claims, and the most persuasive AI chatbots were the ones that used the most factual claims.
They kind of flooded the user with information.
Now interestingly, what they found was, the more I was told to be persuasive and put factual claims in there, the more likely it was also to come up with a factual claim that's inaccurate.
Basically, it was more likely to make stuff up because, I mean, there's only so many facts that you can craft.
And so eventually it will start inserting some bits that are not as actually accurate.
And, of course, chatbots are trained on data that they glean from the internet and they parrot it back in many cases.
Did the body of data that they were trained on make any difference to how these experiments panned out?
Yes, it did.
So what they found was the AI models that advocated for candidates on the political right consistently delivered more inaccurate claims than the ones supporting left-leaning candidates.
And the researcher.
When I asked him about this finding, he told me that the model absorbs the internet and uses that as the source of its claims and
Previous research has suggested that social media users on the right share more inaccurate information than social media users on the left.
So this finding kind of makes sense.
Speaking of researchers, then, there are a lot of folk who worked on both papers.
What else did they tell you about the results that they found?
Well, essentially they said, the big takeaway here is that, even if you're not a social sciences researcher, maybe you have used a chat bot here or there.
When you interact with chat bots, they do whatever the designer tells it to do.
And so you can't assume that they all have the same kind of benevolent instructions.
You always have to be thinking about the motivations of the designers and what agenda they told it to represent.
Because in this study, they kind of prompted the chatbots to support one candidate over another.
And it's not something that perhaps was very evident when the users were talking to them.
And so you basically always want to be skeptical of whatever the chatbots are telling you, both because you don't know fully what their motivations are, but then also again, they could be spouting information that is not accurate.
And so this is something that perhaps policymakers should think about and take up, because you don't want a world in which elections are being handed to whoever can create an AI that's very convincing or pay one of the AI companies to support one of their preferred candidates.
Essentially, it's raising the alarm of a new type of information battle that could be coming in the future.
That was Nature's Max Kozloff talking with Benjamin Thompson.
For more on that story, check out the show notes for some links.
And that's it for this week's show.
If you want to stay in touch, you can.
We're at Nature Podcast on X or Blue Sky, or you can send an email to podcast.nature.com.
I'm Nick Pachachow.
And I'm Anne-Marie Conlon.
Thanks for listening.
