179 – Challenging the Dead Internet Theory

David: [00:00:00] Hi, and welcome to the IDEMS Podcast. I’m David Stern, a founding director of IDEMS, and it’s my absolute pleasure to be here with Santiago Borio, again, an Impact Activation Fellow. You have come back after heading into a school to do a maternity leave cover in Argentina. And now you are back with IDEMS. Welcome back.

Santiago: Thank you David. I’m very glad to be back. I’m not quite back yet as Impact Activation Fellow but I would call myself a collaborator for the moment. 

David: Well, for another month or so, by the time anyone listens to this will probably be correct.

Santiago: Okay, fair enough. Yes, I had a term and a half a bit more in a school here in Argentina, which was in many ways wonderful to be back in the classroom, [00:01:00] but I was craving a bit more social impact, and I thought it was the right time once that maternity cover was over to come back.

And I would like to thank IDEMS, you in particular, David, for all the support you’ve given me. The move to the school was not just for the maternity cover, it was because I needed a bit more structure in my life at a difficult time.

David: Yeah, and I’m, delighted, you were IDEMS employee number one. You have now had two periods where you have for personal reasons taken on jobs, particularly in schools because you have a passion for teaching, and come back.

And this is something which in our conception of IDEMS we hoped would be possible, actually supporting people to get work [00:02:00] life in ways that sort of suit them even though as an organisation we are not necessarily good at providing those structures ’cause everybody’s remote and there’s a lot of work to be done, more than we can do with our small team at the moment.

Santiago: Well, one of the things that I learned, I think from this experience is that I need an office. I need to go somewhere and interact with people and see other faces on a daily basis. And I believe there’s other members of the team who do this, I’m gonna find a place to go and work at. It’s part of the flexibility that IDEMS offers, that flexibility is in some ways the support. 

David: But it’s also a double-edged sword. That flexibility means that people can find themselves in situations, as you have done in the past, where you feel you are increasingly isolated. These are decisions which are made, if you [00:03:00] want, at an individual level, but it is also related to institutional structures.

Santiago: Of course. But anyway, I’m thankful to IDEMS, and I’m glad to be back, I’m not full time yet I will be at some point soon. We said one month.

David: Whenever you are ready. There’s interesting projects piling up for you.

Santiago: Yeah. But I thought that for my first podcast back, I love doing podcasts, I didn’t think I would, but I do love challenging you on stuff. And there was one particular area or concept or idea that I wanted to get your views on. It’s quite pessimistic an idea, and you’re a very optimistic person. So I want to hear your views about the idea itself, and then I want to bring in some positives that I heard around this theory.

The theory is called the [00:04:00] dead internet.

David: Yep.

Santiago: Have you heard of it at all?

David: Nope.

Santiago: It started coming up in 2016, 2017 apparently, I did small research, and it suggests that the increasing number of bots that were coming up were starting to generate content online to the point that information online, on the internet, it’s completely unreliable and will eventually get to be almost no human content created. Especially with the rise of large language models and generative AI, in 2021 this theory became more prominent or it gained a bit of traction with the popularisation of generative AI.

And it suggests that in a few [00:05:00] years time, the internet will be dead in the sense that humans will no longer be creating content in the internet, it will be mostly generated by bots, AI, large language models, et cetera, et cetera.

David: Let me just frame this, the way that is phrased is obviously incorrect. I think what is the correct way of phrasing it is that the proportion of content generated by humans will diminished to the extent where you normally find generated content rather than human generated content.

That I can believe is a problem and a statement which could be corrected and it could be a worry. But I don’t believe that humans will stop generating content. I believe what we’re actually saying is that that will just be proportionately small, potentially compared to the amount of automatically generated content. Have I interpreted that correctly?

Santiago: I think that is a [00:06:00] sensible interpretation. I think that the theory itself is quite extreme and I don’t quite agree with it. But it is a rising movement, it is a rising theory, and one of the challenges that they present is that because of the training of the bots on the internet and the lack of reliability that some of the outcomes from generative AI provide, the internet will no longer be trustable.

David: Okay, so the internet was never something you should trust blindly. So that’s not really changed in a sort of fundamental black and white state. There’s always been questions of what you should trust online and what you shouldn’t trust online. The question of whether it becomes less trustable is an interesting one.

These extremes are not useful because they’re just not true. It isn’t that it suddenly it was wonderful and now it is terrible. No, it is maybe it is slightly worse or slightly better. And [00:07:00] I agree that if you interpret this as a statement that there is a possibility that it gets worse, actually I have a lot of other evidence that is probably true. The internet is getting worse. It is becoming less accessible to more people because if you are on older devices, there’s more and more of the internet, which is not accessible to you.

These are things which have been tracked for a number of years in ways which are much more serious in my mind than the problem that you are stating of AI generated content. I’m not saying that the problem of AI generated content isn’t important, but I am saying that there are other more important issues related to the internet, which are worrying. So I think there are genuine worries related to the internet going forward. However, I don’t believe things are as black and white as this seems to be portraying it.

Santiago: No, of course. And Wikipedia describes it as a [00:08:00] conspiracy theory. So it is not an evidence-based theory, as far as I’m aware. But I thought it was an interesting concept, particularly because of the two positive spins that I found.

David: Okay.

Santiago: And you didn’t give me either of them. One of them, and I suppose with your challenges of the theory, the positives lose strength in some shape or form, but let’s see. One of the positives is that because there will be, according to the theory, less and less reliability on what’s online and what’s generated and so on, people will tend or progress or evolve back into community groups and start…

David: No. This is totally ridiculous. This is as ridiculous as the theory itself, because the idea that we can predict exactly how [00:09:00] people will respond to these sorts of things is so far from any reality I know, humans are extremely unpredictable, they don’t do something all in the same way. That some people might find community, this is probably more likely correlation than causation. There are people, there are lots of things happening in society that might lead to us finding community better and that is a possible thing which will happen and maybe some individual’s experience in some way might lead them to finding community.

Great. But these are complex systems which do not have simple causal links. This is why these are conspiracy theories rather than actual scientific processes that could be tested, that could be checked. And I want to be clear, I am not saying that there isn’t an element of truth to these, that there isn’t an element where there may be something correct, but formulating it in a precise way with [00:10:00] an extreme sort of framing and then drawing conclusions based on that, this isn’t an approach to understanding complex societies.

Santiago: No, and it is not very scientific in any shape or form.

David: I do like things which are scientific, I have no problem stating that, I think that the scientific method is one of the most powerful tools that we have to understand the world. And so I do believe in scientific method, but it’s not just about scientific method. It’s about understanding how the world works and how we build our mental models for how the world works. It’s important I think in many cases that when we build our mental models, we don’t get drawn to extreme exaggerations and then take them as a reality.

Santiago: Yeah, and as always you live in complexity.

David: I live in complexity. Exactly. I’m not looking for a simple [00:11:00] solution to understand the word as something which is simple, which is black and white, good or bad. You know, the complexity of the world, I have no doubt whatsoever that there is an essence of truth in the fact that through the amount of AI generated content on the internet, there are going to be consequences about how people interact with it, about what the internet as a whole becomes because you are changing the population.

These are complex systems. So I have absolutely no doubt that there are elements of this which are noteworthy and worthy of discussion. But taken to its extreme like this, then it’s no longer a discussion point, it is no longer interesting.

Santiago: Okay, let me grab one of those specific cases then. There was this, lawyer in the US that worked very much with case law who asked some large language model for cases that would help their client. And it came up with [00:12:00] two cases with case numbers and reference and everything, and it was all rubbish.

And they presented their defense and the judge said, hang on a second, I need to look into this case. And they looked into that case and said, what you’re saying is complete nonsense. 

David: This is the same sort of issue has happened in a number of cases, where there have even been consequences of these, which are extremely serious, where these fictitious cases, phantom cases, have appeared in different ways. Absolutely. And these are things where I would argue it becomes a training issue.

We need to help train lawyers, for example, to actually use not just chat GPT or large language models, to do their research for them, but to understand that there is a responsibility to understand what sources of truth are there. There were [00:13:00] repositories of cases, he’d have just needed to go and look up that case number in a sensible repository and job done. He’d have found it doesn’t exist. That’s, I’m afraid, on the lawyer in my opinion.

Santiago: I agree. I agree completely. But as we said at the start, I just come back from eight months in a school and you can see teachers and students using large language models incorrectly in that way. That’s I think why I like our responsible AI courses so much. I think that is so needed in society in general, but in the professional context, that lawyer, as you say, could have very easily verified. 

David: And this is the thing. It is not to say that using the large language models to ask that question is irresponsible in its own right. But using it and then trusting it [00:14:00] immediately without checking, without doing your due diligence, as a lawyer, surely you should know there’s a lawsuit waiting there for you.

Really, such a person should be sued for not doing their job properly. A lawyer should know that a case file is something which exists and which you can look up and you can check and you should check. You shouldn’t just trust a random large language model. That seems to me to be obvious.

And if in our society we don’t take an approach where that is the minimum that is asked of you to actually find a reliable source to understand what a reliable source is in your domain, let’s be clear here, this is an expert, a lawyer is an expert in the law, and they are not checking the case files. That to me is simply an expert not doing their job.

Santiago: Yeah. [00:15:00] Yeah. Agreed.

Now of course you challenged my premise of the dead internet.

David: Yep.

Santiago: Quite successfully, I think quite sensibly. It’s not that I believe in that theory anyway, it’s just something interesting that I came across, but there was another similar theory with a silver lining that I found which relates to this checking and doubting and having that doubt in your head, that questioning, that the more we use these resources, the more we will realise that sometimes they just don’t know and give you gibberish, that we will end up, or there is potential for people to rediscover [00:16:00] questioning in many ways similar to the scientific method.

David: Okay. For individuals, there may be cases where that is true. But that’s not the interesting question. The interesting question is, at a societal level, is that likely to be true or not? And my argument would be no. On average that is not the response I would expect from individuals. And I’m not even interested on average, because I guess it doesn’t matter whether it’s 50% of your population or 20% of your population that is getting caught up in falsehoods and misinformation. That is a serious threat to a well-functioning society.

And so actually, can this be done at the scale? It will not happen automatically. Could [00:17:00] we change our education systems so that this is an important part of the education system and we can make a lot of the population large language model literate, ooh, that’s an interesting challenge.

The success of standard literacy is already difficult across the world. You then get to mathematical literacy, again, that’s even harder. You get to data literacy, data literacy is a disaster across the world, and this is further than data literacy because really you need good data literacy to be able to do this.

So can we, in our educations, get better literacy? By literacy in this context, I’m thinking of literacy of the large language models so that people know how to use them responsibly. That is actual work that somebody needs to do. This will not happen by magic, and this is where these things are in my mind…

Santiago: [00:18:00] Naive.

David: Well, yes, the world is a complex place. It isn’t because there’s misinformation out there that people learn how to deal with misinformation. However, if there is a lot of misinformation out there and there’s good training around looking at this and doing this and thinking critically, then in many ways, and I want to be clear that I’m not trying to equate these large language models type of misinformation with all types of misinformation. There are separate problems at play here.

Santiago: The article I saw, I’m not sure if it was a formal article, it was not just based on large language models, it was on deliberate misinformation. But I do see your point that in order to understand misinformation, you need to have all sorts of literacies and life experience in order to be able to discern from actually this is something I can trust and, no, maybe I need to check this a bit [00:19:00] better.

David: The thing which I find so interesting about this at the moment is that it’s not just, this is something I can trust. It is also the real problem of this is someone I can trust. And this is actually a really interesting societal issue that we’re having at the moment, I believe, because the difference between trusting individuals and trusting institutions is huge.

And when trust in institutions is lost, that means that trust tends to pivot towards individuals. These are societal processes, which I think are complex, but very interesting, that the difference between actually putting your trust in individuals and putting your trust in institutions is very different.

And in the world that we have at the moment, I feel there’s a tendency of taking, of trusting institutions less and individuals more. And I feel that is ripe [00:20:00] for then areas of misinformation because within an institution you can have layers of checking. So if the institution makes that statement, it isn’t just the point of view of an individual in the institution, it’s gone through layers of review to the point where the institution is willing to take responsibility for it.

Santiago: Can I just clarify? You would, for example, class open AI as an institution.

David: So this is an interesting question. By institutions, the standard institutions of trust in this sort of area do tend to be governments. They tend to be media sort of… 

Santiago: The BBC for example.

David: The BBC. You know, that is a media institution, there’s many other news outlets in different ways. The point is that these institutions, the UN institutions, all these institutions have structures in place which limit the freedom of expression in some form or other of [00:21:00] people who speak on behalf of the institution. And so the comments which are made by the institution have gone through a process.

Now this is exactly the point and why some people have lost trust in those institutions because they fear that process is biased. And therefore the information coming outta the institution is biased. And therefore there’s been a loss of trust in the institutions in certain ways.

I’m not wanting to comment on this point about whether that loss of trust is correct or not, because actually that’s a really complicated issue. But what I do believe is that if you have institutions, and open AI I’m not including in that same way because the CEO of Open AI makes comments on behalf of Open AI as an individual, and those are Open AI comments.

And so that’s not gone through the same processes, you know, it doesn’t have the same structures in place to make sure that the information coming out of the institution represents the values, the [00:22:00] consensus of the institution, rather than the opinions of an individual.

Santiago: And it doesn’t necessarily represent, well, it represents their shareholders as well, in some ways.

David: Well, the BBC doesn’t have shareholders.

Santiago: No, no no, I meant Open AI.

David: Well, but that’s a whole different problem, and this is why it is not an institution, which is around the news. There are institutions which are around giving news, which are representing private interests. And there, that process, I think part of the debate is the fact that actually it is true that it is possible to have institutions which have these processes in place and are deliberately inserting a bias.

So I’m saying people are not wrong to challenge and to question institutions, what are the biases that may exist within the institution, and this is something that is a genuine issue. So I’m not trying to [00:23:00] diminish that issue. What I am trying to say is that it worries me that we as a society seem to be moving away from trusting institutions to trusting individuals more.

And once you are with an individual, you have an opinion. I have an opinion, you have an opinion. What the podcast represents in an interesting way, and we don’t have strong review processes at this point in time to make sure that there is an institutional structure behind this, but it is something which over time I hope the IDEMS podcast, this is something I’m happy for IDEMS as an organisation, as an institution to take responsibility for the content of it.

We are not a news creating institution, but if I were to have a personal podcast, it would be different. Even though I’m quite noisy on the IDEMS podcast, my voice is heard quite a lot. And this is still [00:24:00] different from it being my podcast where I would just be able to air my opinions.

I feel it is really important that in the context that we’re doing, I am framing, and my opinions are framed in the context of the institution. And that’s something I would like to work towards.

Santiago: Okay, let me close this episode because we’re getting towards the time limit. Let me close it by asking where you place universities in there. Because one of the things that you mentioned was limitation or restriction or regulation on freedom of speech, while academic research… 

David: I want to be careful here. Institutions like news media, it’s not that they regulate freedom of speech, it’s that they want what they put out to represent their institution and they might even have an opinions column, which would then not represent theirs, [00:25:00] but it’s visibly something which then represents the individuals. So they try to be clear on distinguishing this is an opinion which somebody has who might be a member of staff or whatever it is, and this is something which has gone through the fact checking and the rest of it, and we are happy as an institution to put it out there.

So I think that’s the distinction. And I think this is really important with universities as well. Universities are balanced the other way. Because of academic freedoms, good universities should have their professors able to express their opinions, whether or not the institution agrees with them. But if the university makes statements as an institution, those have almost always gone through careful processes because they represent the university reputation.

Now, don’t get me wrong, universities have reputational challenges because of professors’ behaviors and opinions all the time. But that’s the nature of a university. It’s bringing together experts who are diverse [00:26:00] and will have their opinions, and whose opinions are valued by society in ways because of their expertise. And their association to institutions does mean that this is fun tension. This is complex.

Santiago: Of course it is, of course it is, and I brought here a very simple idea that of course it was going to be destroyed, in a sensible way. And I thank you for it as always. I didn’t bring it feeling that it was right theory or anything like that, I just thought, I wonder what David thinks about this.

David: This will stimulate discussion. Yes.

Santiago: Yeah. Yeah.

David: Which is great. And I really enjoy these discussions, and being confronted with theories about the world, which I’ve not looked into. And it was interesting to me that when you said, this goes back to 2016, I thought, oh, what have I missed? Where, why am I not paying attention? [00:27:00] I’m now not so worried that I wasn’t paying attention. I think there’s lots of things that I’m not paying attention to in this way.

Santiago: Of course. But yeah, it was good fun.

David: It was fun and thank you for bringing it and thank you for engaging it in a way, which I think, I hope this is, this sort of discussion is what I believe needs to be, how can I put it, needs to be happening where we do challenge extreme. And what I like about this is this is an extremist view. It’s not extremist in the way we tend to think of extremism, such as religious extremism or other forms of extremism.

But it is the idea that there is a valid point with the concerns about how AI is interacting with something which people hold precious, the internet, and it’s taken to an extreme. [00:28:00] And therefore it’s become an extremist view. The idea, the core point of it is extremely interesting. If you take it back from the extreme, you are down to a really valuable discussion. And, it is a concern, is the internet getting better or worse because of AI generated content?

Now that’s a really interesting question, and that’s a complex answer, and I believe, in the vein of our discussion, I believe the internet is getting worse at the moment because of AI generated content. I think there is a concern here and I think these are concerns which then relate to education and they relate to other things.

So I think there are valid concerns in interesting ways. I’m a great supporter of AI as you know in many other contexts, but I do think that the internet as a precious institution, because in some ways that’s what it is, it’s not an institution in the sense of a governmental institution or a news [00:29:00] institution or a university, but it is in some ways a wonderful example of a self-governing institution which has substance to the institution itself.

And it is an institution which I believe is under threat. And I do believe there are concerns to be dealt with. And I think there were people working on this. Not that I think, I know people working on this and so it’s interesting.

Santiago: Perhaps I need to get into that and challenge you again in a more rigorous manner.

David: I look forward to it. And I have to be clear, I know enough to know that I really don’t know enough about the challenges to the internet.

Santiago: Okay. Thank you David.

David: Thank you. [00:30:00]