I find this not only refreshing, but at some level astounding.
Nature.
Welcome back to The Nature Podcast.
This week, how hunger and hormones interact in mouse brains.
And what generative AI could mean for universities and for students' education.
I'm Sharmini Bandal.
And I'm Nick Perchichow.
Sensations like hunger are interpreted by our brains and ultimately result in a behavioural response.
Perhaps hunger will cause you to seek out a snack, for instance.
And while researchers understand a lot about how a single stimulus like this results in a behaviour, most often you actually encounter multiple different states at once.
Perhaps you're cold and hungry, or hot and thirsty.
Less is known about how these, sometimes competing, states interact in the brain.
A new study in Nature though, may shed a little bit of light on this, as a serendipitous observation led to researchers looking at how mouse hunger and hormones interact.
This could help explain how the brain condenses competing stimuli into an action.
I called up one of the team behind the new paper, Johnny Cool, and he explained how they got started.
This project came about quite serendipitously when one of my graduate students discovered that very mild food deprivation in virgin female mice leads to an aggression towards infants.
And that's something that you don't normally observe in these animals.
And when we're talking about aggression, what behaviours are you seeing in this case?
It's a form of aggression that's only directed at mouse pups and not at adult intruders, for instance.
And what we see here is that these virgin female mice they try to basically attack and bite these young animals, carry them around aggressively in their cage and so on.
And it's important to emphasize here that We never let this behavior escalate, you know so there's always an observer present and we interrupt as soon as this behavior is demonstrated.
And so it's not fully understood why rodents such as mice do sometimes have these aggressive behaviours, where they attack young.
But I know it's speculated to perhaps be due to resources being limited and there being lots of mouths to feed.
But in this case, you had this initial observation that a bit of hunger led to this aggression.
What were your first thoughts about what might be going on here?
Well, our first thoughts were that there must be a quite intricate interaction between those brain systems or brain circuits that mediate hunger-driven behaviors and those brain systems that mediate infant-directed behaviors.
So you had this idea that food deprivation was causing something to go on in the brain which was leading to these aggressive behaviors.
Where did you start to look into this?
So we started by looking at the function of these so-called AGRP-expressing neurons.
So that's a population of only about 8000 or so neurons that are really the mediators of hunger-driven behaviors in the brain.
So when the organism is calorically deficient, these neurons become active and then decompose, drive a very powerful food seeking behaviour.
And so how did you test what the involvement of these neurons might be?
So we first artificially stimulated these neurons using a technique called chemogenetics.
And when we did this in mice that were well fed, we could indeed stimulate additional feeding in those mice.
But also, we could now basically evoke pup-directed aggression in all of these mice.
And assated mice, that's something that you never observe.
They usually are not aggressive towards pups.
And conversely, when we artificially inhibited these neurons, this time in hungry mice, we could reduce food intake in these mice, but we could also abolish this food deprivation-induced pup-directed aggression.
So it seemed like these neurons were involved in this aggressive behaviour.
But it wasn't quite that straightforward, was it?
Because not all of these virgin female mice were aggressive.
Exactly.
And that was sort of the next conundrum.
Why do only about 50 to 60% of those food-deprived females show pup-directed aggression?
And I guess our first naive assumption was that maybe these aggressive mice are simply more hungry than those that don't show pop-directed regression.
So we assessed whether they differ in the amount of food that they consume after food deprivation or whether levels of the hunger hormone ghrelin differ between those two cohorts, but we didn't find any difference whatsoever.
And the next question we asked is whether these aggressive animals are somehow in a, let's say, hormonally permissive state that favors pop director aggression.
And that's where we got lucky.
So specifically, we hypothesized that the estrous cycle in these mice plays a role in shaping these pop-directed behaviors.
Now, the estrous cycle in mice occurs over the course of about four to five days.
It consists of four distinct stages and we found, when we look at the rate of aggression that we can evoke in these mice, that that fluctuates throughout the oestrus cycle.
And so what did you think was the mechanism behind this and how did you go about testing it?
So the first thing we tried to figure out whether there was a good correlation between the levels of individual hormones that fluctuate throughout the estrous cycle and the, let's call it, the aggression rate.
And when we plotted the levels of, for instance, estrogen or progesterone individually against the aggression rate, we didn't find a very good correlation between those two variables.
But interestingly, when Mingran the key person behind this project When she plotted the relative ratio between progesterone and estrogen, we found that this relative hormone ratio correlated really well with the rate of aggression in these animals.
And then the next question was where in the brain this hormone ratio is sensed to drive this behavioral change.
And here we tested whether these hunger neurons, which are hormone sensitive, whether the sensing occurs there.
So we used genetic techniques to make those neurons hormone insensitive and then checked whether the rate of animals becoming aggressive differed and didn't find that there was any difference.
So what we did next is we looked at where these hunger-mediating neurons actually project in the brain.
And we found that there was one projection, namely the projection to the so-called medial preoptic area, or MPOA, that really seem to drive this behaviour.
And the MPOA is an interesting area because that has been involved in driving parental behaviour.
So we have this section in the brains of these virgin female mice that is sensitive to oestrogen and progesterone, but only at a certain ratio.
So what do you think this all means?
So I guess it means that even those so-called instinctive behaviors, so very robust behaviors that animals can perform, they can actually be very flexibly tuned and are really subject to animals' internal states.
And what we've discovered here is basically a mechanism by which individual neurons in the brain can integrate several of those coexisting internal states at any given point in time.
Organisms, animals are subject to many, many different internal states.
So in this case, it's estrous cycle, it's hunger, it's stress.
There's many other that you can think of.
But all of these have to somehow converge on brain systems that make decisions on how to behave towards, let's say, a social target, because there are only so many ways that you can behave towards a mouse pup, for instance you can ignore it, you can attack it or you can take care of it.
So all of this sort of high dimensional state space has to somehow collapse into a few decisions.
And sometimes these decisions are essentially binary, you know, run away or stay and so on.
And what we think we found here is how individual cells in the brain can integrate those different internal states.
And so what would you say are the next questions for you?
What do you want to understand now you've completed this research?
One of the obvious next questions would be to see how scalable such an integration mechanism would be.
Rather than manipulating two internal states, what if you manipulated four or five or ten?
This will be challenging but I think it will be very insightful because this will get us closer to the truer situation that a neural system would encounter in the actual brain in a living organism.
And well, I mean mice and humans are obviously quite different, but do you think there are any lessons we can take from your research for humans?
So I guess it's tempting to call this hangry, but we don't know whether this directly translates into humans.
The only thing I can say is that all of these neural systems and the hormones that we're dealing with here are really highly conserved between mice and humans.
So my prediction is that these integrative mechanisms that we've discovered here exists in the human brain as well, even though behaviorally speaking, you know, our findings might not be directly translatable.
That was johnny cool from the francis crick institute here in the uk.
For more on that study, check out the show notes for some links Coming up.
What ChatGPT and other AIs mean for higher education.
Now, though, it's time for the research highlights with Dan Fox.
As the population of towns and cities grow.
Researchers have identified an overlooked source of greenhouse gas emissions – wastewater treatment.
A team has estimated methane, nitrous oxide and ammonia gas emissions for the United States and found that official inventories underestimate the contribution that wastewater treatment makes to emission of all three greenhouse gases.
Their updated calculation exceeded the estimates from the US Environmental Protection Agency by 24 times for methane, 19 times for nitrous oxide and more than two orders of magnitude for ammonia.
Given the growing global population.
The authors say that sustainability targets need to consider this underestimated source of emissions and efforts be made to mitigate it.
Sniff out that research in Nature Water.
Around 6000 years ago, people in the Andes developed ways to thrive in the challenging landscape, building a system of communal hunting that lasted well into the age of colonisation.
Using satellite data to study an arid valley in northern Chile, a researcher was able to identify chacus, V-shaped traps made from stone walls about 1.5 meters high and 150 meters long.
The traps funnelled animals into a central enclosure about 95 square metres in size and two metres deep.
The chacus were connected to hundreds of small settlements that probably served as outposts, helping people to move across the highlands in ways that matched the movement of wild animals.
Some chacus date back to around 4000 BC and lasted into the 18th century.
The author says that these findings suggest that Andean people kept hunting and gathering for thousands of years after farming and pastoralism began.
Hunt down that research in antiquity.
There is a sea change sweeping through university campuses right now, as institutions and students scramble to adapt to the use of generative AI tools such as ChatGPT.
This week, Nature is publishing a feature written by Helen Pearson looking at the rise in Gen AI in universities and the differing opinions surrounding the effects that this technology might have on higher education.
Helen joined Benjamin Thompson to talk about it, starting with what's known about how students are currently using Gen AI.
The surveys are showing, perhaps unsurprisingly, that a lot of students are using generative AI to produce text for them to edit their text.
And then one of the key points of contention or concerns, I suppose, for universities is whether students are using generative AI in assessed work, as in the kind of coursework or their actual tests by which they'll be graded.
One survey I looked at, about 90 of students said that yes, they were using AI in some form in their assessments.
But most of those were using it to kind of explain concepts.
And it was about 25, who were using text generated from an AI in their assessment, but were editing it first.
And then about 8 admitted that they were actually using kind of raw text from an AI in their assessments.
Quite a broad spectrum then.
And, given the newness of this technology, the sense that I'm getting from your feature is that many universities it's almost a rearguard action right.
They're trying to work out what to do after the fact.
And the feelings, amongst academics in the first instance, are quite broad about whether this is good or not.
Yeah, I think that's fair to say.
So one person I spoke to who's an expert in AI and education described the situation, for example, on US campuses, as chaotic at the moment because, as you said, these tools have just arrived and now they're kind of scrambling to introduce policies which guide students and faculty into how to use AI.
Some surveys suggest, which is quite interesting, that faculty are not embracing AI as quickly as students, which itself presents challenges.
While some faculty are leaping on the bandwagon and they might even be using AI to do some of their own work, like to draft feedback on essays, for example.
Others are much more concerned, much more cautious.
So yeah, definitely a kind of real mix.
And there are various things to look at as a result then.
I suppose one of them is the impact that this has on the university system as we know it.
Because, of course, when I was a student, a lot of it was remembering facts and figures and being able to regurgitate those and link them together.
That's something that AI is, of course, very, very good at.
From the folk you spoke to.
Do you think there's a shift in how lectures and teaching may occur from now on?
Answer to that yet.
I think one of the key concerns is around what impact use of these tools is going to have on students and their critical thinking abilities.
So crudely?
Is it going to make students stupid because they're just outsourcing all their work to AIs, which I think is a little simplistic to think about it that way, because I've spoken to people, for example, who are interviewing students and students themselves.
And of course, they're concerned as well that if they use AI too much, that they too won't be learning the type of skills they need in the future.
And there are different studies coming out with different answers to that question at the moment.
I mean, the thing is, these tools are new.
Not many really rigorous studies have been done.
So there are some studies suggesting that, yeah, you know, if you use an AI, maybe...
It can sort of dull critical thinking.
And there are others that if you use it in a smart way, that it might be able to enhance learning.
So that's kind of one big area of research, I suppose, which is developing.
And there's definitely a need for really proper evaluation of these tools to show that they help.
But here we are, right?
These things are already being used.
And there are, say, concerns for students and for student welfare, of course.
But also there are potential benefits, as you laid out there.
Yeah, I mean.
One interesting approach which people pointed me to is around testing, because there's this really key question like how do you test what students know if in any unsupervised test, a student can go off and use an AI to potentially do some of the work for them?
And if you emerge with this kind of university stamped qualification, which doesn't really mean so much anymore because you don't know that actually the student does have those skills.
And that's like a really fundamental threat to universities.
And one approach which is being taken now at the University of Sydney is to rethink their approaches to assessments.
And what they've done from this year is to introduce two strands to assessment.
So they have a secure assessment which is like an old school in-person exam or test, where AI is not not allowed in the room.
And then you've got these unobserved assessments where students can freely use AI.
And the idea is that you do a unobserved assessment like, let's say, you learn in your unobserved one to write a lab report.
And then the skills that you learn through that will later be tested in the secure assessment you might be asked to analyse data from a lab practical, say.
So I think a lot of universities like that's a real key point that universities have got to work out.
And, as you say, the concern that gets the most headlines is that using AI is making students not use their brains right.
I mean, what evidence is there either way to show what really might be going on?
Because it seems like there are so many shades of grey as to what the answer might be.
So this question of like, what does it do to a student's brain if they're using AI, I think is fascinating.
And so, for example, there was one study that came out earlier this year and it was a preprint study called Your Brain on Chat GPT, which got like loads of headlines.
And in that study, which... you know, had plenty of limitations and was quite small.
They were looking using EEG at the electrical activity in students' brains when they wrote an essay.
And there were three groups One of them could use AI.
One of them could use search and the internet, but not AI.
And one of them had to just use their brains.
And what they found in this small group was that the students who used only their brains had the widest what they called connectivity, which is kind of communication between brain regions when they wrote books, their essay, and they also were good at remembering what they'd written, whereas the students that used AI tools had the kind of weakest connectivity and they often couldn't remember a word of what they'd written.
So, of course, this prompted widespread headlines.
And interestingly, the researcher who ran that study said that since publishing it, she'd had about 4000 emails from around the world from teachers and university lecturers this huge outpouring of concern that this is what's happening to their students.
So that's kind of the one end.
And then, on the other end, I spoke to a researcher at Harvard who was doing this really interesting study, where he developed an AI tutor with his colleagues to teach physics to undergraduates there.
And they've taken what's known as an active learning approach.
So active learning is this approach in education, where it's all about engaging with the material, thinking about it, discussing it, problem solving and so forth.
And that's actually been shown in educational research to be quite an effective way to get students to learn.
So they took that approach and tried to build their AI tutor around that.
And in that case, in a randomized controlled trial, the students who worked with the AI tutor did appear to learn more actually than the ones who just had the human tutors.
And my sense, having spoken to people, is that's maybe where it's going.
Is that researchers are working out at the moment how to use AI in a way that supports learning and encourages these skills that we really want?
And maybe that active learning approach is going to be one way to go.
And so often when we talk about science, we say that more data is needed.
And it seems that in this case, that is, of course, absolutely true.
But in your feature you do touch on the fact that there will be a lot more data gained because a lot of universities are throwing their lot in with various AI companies.
And there are mixed feelings there too.
Yes, AI companies are going out and trying to integrate their tools into campuses.
We've seen firms produce versions of their AIs specifically tailored to higher education because, of course, for them It's an opportunity to integrate their tools into the lives of all these young people.
But equally that's eliciting some concerns from some academics because they feel like we shouldn't be saying the AI bandwagon is unstoppable.
Don't universities have a responsibility actually to be careful about inviting these AI tools onto campus when, as we've just discussed, we don't really know the cognitive impacts or the impacts on learning over time?
And they also have concerns about the ethics, the environmental costs of some of these tools.
So they feel like there should be a much more critical engagement, I think, with with ai, and is it possible to get a sense of the direction of travel on this helen?
As you say, we're three years into these systems existing.
What's the sense that you got from the folk that you've spoken to?
My sense is, although some academics push back at the idea of it being unstoppable, I would say there are more people who feel like it is unstoppable.
I mean, as we know in our own lives, right?
So I don't think at least from what I've heard, that there'll be a way to keep AI off campuses.
I think that researchers and teachers are going to be forced, if they're not already, to work out how best to incorporate those tools in a way which supports learning.
And one thing I haven't mentioned, which some people feel like there's a much kind of deeper conversation to be had about universities.
Because if you're going out into a world where some of the skills which universities have taught, like writing and analysing complex information, are no longer necessarily needed in the workforce because they're being done by AI, then what are the skills which young people need as they go into this sort of AI-powered future?
And so we'll.
Actually, we see universities pivot a bit more towards teaching young.
You know the human skills of communication discussion, judgment.
These human skills perhaps aren't going away and will maybe become more valuable.
Nature's Helen Pearson there.
To read her feature, look out for the links in the show notes.
Finally on the show, it's time for the Briefing Chat, where we discuss a couple of articles that have been highlighted in the Nature Briefing, Nature's daily round-up of science news.
Shamini, what have you been reading this time?
So this is some pretty cool news I read in Nature.
It's about blood tests for Alzheimer's disease, and this has been a topic that's been sort of developing for a while now, and we've now got the first blood test that can actually be used in a primary care setting.
So you could go to your doctor and potentially quite easily get a blood test for Alzheimer's.
And this is a new one that's just been approved by the FDA, the US Food and Drug Administration.
And so how might a blood test like this work?
What is it picking up to show signs of Alzheimer's?
Okay, so there are some main what we call biomarkers of Alzheimer's that you find in the brain, that are thought to be the things that are causing the damage and that's sort of causing the cognitive decline.
So there's amyloid beta proteins and then there's tau proteins.
And you can see in advanced Alzheimer's you get sort of plaques of amyloid and you get tangles of tau.
And these tau proteins get diluted phosphorylated modified, and that's thought to be what's sort of causing the problem.
So this new blood test is specifically looking at a particular phosphorylated form of tau p, tau 181.
And it looks at how much of the tau protein in the body has been modified as a result of Alzheimer's disease.
And then it will give you basically a negative predictive value.
So it will rule out Alzheimer's.
It will tell you if you don't have the disease.
And it's got a pretty good success rate.
So this clinical trial that's just been reported, 98 of the time it correctly ruled out Alzheimer's.
Some of them end up in the blood.
That can then be picked up by this test and that can rule it out for people.
Yeah, and that's why it's sort of taken quite a while to get a blood test for it, because the concentrations in the blood and the sort of other things that are going on in the general blood, compared to the brain, it's a lot harder to pick up those biomarkers.
And actually the standard up until now.
So you can get a brain scan to look for Alzheimer's.
You can also get a lumbar puncture and they can look for the same biomarkers in your cerebrospinal fluid.
And that similarly has quite a sort of high level of accuracy, or it matches well with what the brain scans show.
And so these blood tests are now basically supposed to be a much easier, more accessible form of diagnosis to some extent, than having to do a lumbar puncture.
So in theory, you've got some signs that might be Alzheimer's, maybe some forgetfulness.
You go to your doctor, they do this blood test and then very quickly you get a sense of whether it is Alzheimer's or not.
Well, certainly with this one, it's whether it's not.
There are other blood tests, although they haven't been approved for primary care settings, but there was another one earlier this year that actually can do whether it is and whether it's not.
And yeah, the idea is that you would do these tests on people who already have some symptoms.
They'd be coming in with some cognitive decline and then you would do this test and this other test also had sort of pretty good numbers.
So this test that was previously published, again for the sort of Negative it was of 97 of individuals who did test negative.
Then when you did the lumbar puncture, that also sort of confirmed that negative diagnosis.
And 92% of those with a positive result then also had positive follow up tests.
And the reason this is important is because what you can do here hopefully, is catch Alzheimer's early.
So a lot of dementia in general goes just undiagnosed.
People don't end up knowing why it is.
And a lot of Alzheimer's disease ends up being diagnosed when it's very late, when it's really obvious or when people are able to do these more specialized tests and say, ah yes, this is Alzheimer's.
But early diagnosis could be really important because, you know, alongside developing these blood tests, researchers have been developing treatments that slow down the progression of the disease.
And it looks like or may well be that those kind of treatments are much more effective early on, before the brain damage occurs.
And so this certainly sounds promising and you said it's got FDA approval.
What happens now?
Well, neurologists think this is a pretty big deal, having these new blood tests.
And there's quite a few other blood tests, I think, in development.
So I think there's going to be a lot of competition in this area.
And all of this is hopefully really going to increase the access for patients to have this early testing.
They would then go on to have the more specialized testing.
You know, you might then combine it with a cerebrospinal fluid test or a brain scan.
But at the moment one of the researchers quoted in this piece was sort of keen to emphasise this is for situations in which you already have noticed some cognitive decline.
This isn't hey.
Will you just like check my blood every year and just tell me how I'm doing on the dementia front?
Although, you know, there are thoughts that maybe one day that would be possible.
Maybe one day you could have a blood test that you get when you go to your doctors, the same as sort of measuring your blood pressure.
Again, a researcher quoted in this piece says, well, we don't have access to the full trial data.
They've said it would be important to see the sensitivity, specificity and positive prediction values to assess the accuracy of this test.
And there are certain situations that we've seen previous blood tests, where quite a lot of samples fall into a sort of grey zone and they don't actually come back with a result one way or another.
They need follow up tests.
And we don't quite know the exact values for this new test that's just been released.
But it's certainly a really significant step in terms of the progress of monitoring and treating Alzheimer's.
No, it certainly sounds like it.
Well, for my story this week is related to actually what Ben was talking about with Helen earlier on the show.
It's all about AI language models.
These are generative AI models like ChatGBT and that sort of thing that do text-based generation.
And this is an article I was reading in Nature about, do we need a replacement for the Turing test?
Because these models have killed it.
The Turing test is dead.
It's over and done.
It was such an aim for so long, wasn't it?
Remind me what the Turing test says.
So the Turing test is basically a thought experiment where a judge has to determine whether...
Just through text-based communication, the thing they're communicating with is a human or is AI.
And as you have probably experienced, as many people have probably experienced, the most recent chatbots and Gen AI text tools are are really good at sounding human.
So the best of these models in March this year have sort of sailed through this Turing test.
And so the question has been raised at a meeting of the Royal Society.
In the past couple of weeks, do we need a replacement for the Turing test?
What's next to replace it?
It's really interesting because in a way these large language models are sort of almost designed to pass the Turing test right.
That is the point of them is to sound like people.
They take what people have said previously and try to sound like what they think a person would sound like.
So that is the Turing test, but that's not necessarily the be-all and end-all of what we might want from AI.
Yeah, and that's really where the next tests are, not sowing division.
But there is a difference in opinion of what we should be doing with these next tests, because many researchers are saying okay, they passed the Turing test, but maybe we need tests that evaluate AI safety or maybe we need to test them for different capabilities.
Like, what are the capabilities we want?
What do we actually want from AI?
That's some of the questions that are being raised.
I'm going to read you a quote here because I think it sort of sums it up in quite a good way.
This is from Annel Seth, who's quoted in the article, and it says...
Let's figure out the kind of AI we want and test for those things instead.
Yeah, I mean, that's a huge question right now.
I don't suppose the Royal Society gave us a nice neat answer, did they, on what the next step should be.
No, I mean there was a lot of different discussion, and a lot of the discussion was about something you may have heard of, which is artificial general intelligence, or AGI.
This is AI technology. that could basically do any of the things that humans do.
It could be general purpose.
You could be like okay, do this task and it'll just do this task in the same way that a human might do.
And so this is a goal for many people, especially in the tech industry.
And it could be a thing that you devise a test for.
Like, how would you know if you've reached AGI?
And there are some tests... that have been made for this purpose.
There is a sort of abstract and reasoning puzzle which assesses AI's ability to adapt to new problems.
But the trouble is researchers don't really agree on what AGI is or what the benchmarks of it will be.
It's an open question as to what AGI is and, for some researchers, whether it should be a goal at all.
It feels like artificial general intelligence is also, as you say, it's a fuzzy concept.
We don't quite know what it is.
I guess people have vague ideas of like I want a little robot companion and I will tell them what to do and I'll be able to do anything for me.
That would be useful.
Is there any other sort of target that we might have that's not AGI?
Yeah, so for a lot of researchers they're interested in developing tests that robustly assess the safety of AI.
For example, it could evaluate whether you can make it go off the rails and produce harmful text or produce harmful information that people could then use.
And the reason that some of them argue for this as well is that this question of artificial general intelligence, this AGI, may be distracted from some of the potential harm to the technology.
Like there are risks of it de-skilling humans so it could do the things we do and therefore we don't use those skills anymore.
And also these AIs are known to produce delusions and they can amplify biases and those sort of things.
So some researchers are arguing that we'd be better placed making sure that the AIs don't do that, rather than trying to grasp towards some sort of fuzzy idea of artificial general intelligence.
Because this is such a rapidly changing technology.
It's sort of come so far so quickly.
It's so rapidly changing our society.
And so I suppose you've got the research goals of what is the next milestone that everyone needs to be working towards.
And then there's the question of um, maybe what we need to do right now in society, maybe that's to do with government regulations as well, but it sounds like the royal society are also saying hey, we as researchers need to think about its actual use, not just yeah, as you say, not just race on to the next marker.
No, exactly.
I mean, this event was organized for the 75th anniversary of the publication of the original Turing test.
And it was described as sort of a low hype approach to machine intelligence.
And It was actually really, really popular.
There were more than a thousand people watching the event online.
There was Peter Gabriel, the frontman of the rock band Genesis, that was there.
There was the star of The Matrix, Lawrence Fishburne, was there.
Oh, okay.
This sort of idea of like hey, let's step back and take a minute, seemed to prove really popular with people.
Additionally, the Turing test was never really supposed to be this sort of rigorous measure.
It was sort of like just a thought experiment, and it's been something researchers have been working towards for a while.
And yeah, these AIs beat it.
But also, if you know your way around them, they kind of don't as well.
If you ask one of them to label parts of an elephant or something, then they would fail the Turing test because they would do silly things like label the tail or trunk or that sort of thing that a human wouldn't do.
So is the Turing test really dead?
What are the best tests to use?
What is the best AI that we want and how will that benefit society and how do we get there?
These are all questions that we need to be asking as a society.
Yeah, does it matter that the AIs are passing a Turing test and we no longer know who is real and who is not?
Nick exactly, or are you to clarify the listeners?
I am real, we're on a call.
I can see him although again, generative video right now.
You could have scanned your face into one of these new apps.
I'm just saying i would have made myself pretty if i did.
Well, I think we should close that there.
And listeners, if you want more on these stories, you can check out The Nature Briefing.
It was not written by an AI, but some of our real wonderful colleagues.
They get the best science stories and send them directly to your inbox every weekday.
And that's all for this week.
As always, if you have any thoughts or comments or accusations that I'm not a real person, you can reach out to us on X or Blue Sky.
We're at Nature Podcast.
Or you can send us an email to podcast at nature.com.
I'm Nick Petrichow.
And I'm Sharmini Bundell.
Thanks for listening.
