Adam Russell: It's the Institutions, Stupid | Intelligence at the Frontier 2026

My my real job here is to fill time until our true guest, Tom Kil, shows up, who's uh

Yeah, thank you. I don't I don't use the word very very lightly, but uh yeah, my service is appreciated. Uh Tom Tom's going to come and I'm going to interview Tom. Uh which is super exciting for me because uh if you think I know anything about government, like to Tom Tom, we can't blame him for it, but he he knows a lot about government in that sense. Uh so thank you for the introduction, Trisha.

Uh man, I think I've expended all my extraversion points today. Uh this is the problem with coming this late at the end of the day and having so many like creative, incredibly engaging talks and people uh that I should probably just turn around as an introvert and stare at the wall and talk to this way and then the effort requested. Uh I'm not normal so I'm going to try something I hadn't done before. Uh thank you. Um you there there are no IRB approvals here by the way.

There's no inform you you sat down was good enough for me. Uh which means you can get up. So the opening question is how how did you get here? Why are you how did this happen? Uh you'll see I have some ask slightly bigger questions.

So really it's like how did we get here? And then what I really care about is how do we kind of get past this what whatever this is as it were. Uh so the first thing to know uh is not that. So the good news is I come from DARPA where the whole principle is uh failure. You you you embrace failure.

So there we go. I' I've failed the first time. Uh let me uh start by saying I'm an anthropologist by training. I usually say this right now. Okay.

That's not the response I usually get. So, I want to give people time to leave. If you, oh, you're an anthropologist, see you later. Um, the reason I bring this up in the front is because uh, one, it means I have a very particular uh, perspective. Two, it means I'm very unimpressed by human individual brains, right?

We think our brains are these like amazing uh, centers of like reason and logic. And actually, our superpower is each other, right? You take me and a chimp and you put us on a a desert island for two weeks. Who do you think is still going to be there? it is not me.

Right? So, uh, individual human brains are very underwhelming, but collectively it's pretty impressive what we can do. The other reason I mentioned this is because, uh, in anthropology, there's a guy named Saber, well, there's a saber warf hypothesis. Anybody heard of this Edward Sephir, which basically specifies in its weakest form that your language kind of shapes your reality. Um, that takes on a new meaning entirely as we watch LLMs like literally embody that in some sense.

Uh but the point being like if you have words for things it's easier for you to think about them. I have abused that principle uh because now I actually uh have taken on a habit of building neogisms. Uh the way I I did not I think originate this uh but I did from a podcast in 2021. I was on a career thinking podcast and someone was asking me about my opinion like how do you feel about AI? And the best I could offer was I'm an apocalyptus.

It could be amazing or it could all go to right? So I and and the problem is I don't I don't think there's an in between in some weird way. Uh by the way, there's a movie coming out soon called AI doc or how I became an AI apocalyptist. Uh I think that word came in through my best friend who's interviewed in that movie. I didn't trademark it so I won't be retiring early.

Uh but it's I think it's an example of like when you have a term like this uh one it gives you room to think uh differently slightly maybe try to try to address the binary distinctions of it's either or uh but I think it also is important because you're going to confuse AI uh in some sense. Um okay so let's start from big uh big violation of basic anthropological rules. I'm gonna talk about society which would normally get me excluded from most anthropological discussions because it is too big and I totally accept that. But let me start with the notion that why we have society. What I mean why do you even need it subsets?

Am I urbanized? I saw a lot of literature by the way is that uh the reason society is because societies have to solve for key functions. So, this is the only math you're going to see uh on this slide. And as you can tell, it's legit. This is hardcore math here.

Okay? So, I do I'm going to unpack it for you. Uh so, societies have to figure out how to solve for things like coordination. Big word we've heard all the time, but they have to do it at coordination at scale. Right?

So, every society has to solve this. And they all do it in vaguely different ways. uh every society has to solve for this function of knowledge not just creation but accumulation and there are examples of societies that didn't solve for this as you went backwards technologically so this is not not trivial every so society has to solve for legitimate authority every so society has to solve for resource production distribution ultimately allocation and resources here of course include not just mut resources but labor right in a very real way and every society has to figure out how to solve for this. Every society has to solve for conflict. That is to say, organize to conduct it.

But also, how do you resolve it? Right? The war part in some cases, the the violence part is the easy part for species. It's how do we resolve that in a way that we sort of like continue on as it were. Uh, every society has to solve for risk pooling.

Uh, this is going to be increasingly interesting with AI. We have to figure out how any given one of any individual cannot tolerate the levels of risk that industry the donation other people are buying. So we have to figure out how do we pull that level risk. Uh and this is where uh again my physicist friends begin to roll their eyes. Uh every society has to solve for identity uh or identities in some sense.

And the problem with that though is that in some societies I can solve that by telling you what your identity is. So the best ones solve for not just identity but also belonging a sense that yes this is an identity that I recognize or identities I recognize and that gives me meaning okay what's with the grand scope uh discuss this whole thing it's because I want to propose that if we step back and think about the term technology and and and deconstruct it slightly we tend to think about technology in very specific ways less so this crowd fortunately uh but that actually references like systematic knowledge and treatment of practical knowledge to solve problems. And I want to argue that institutions are our as you said OG technology. It is the institutions the formal semiformal informal that we create that solve for those functions in a very real way. And the implication then is that nowadays we talk about institutions versus technology.

We distinguish those things. And I see this certainly in DC for example the idea we're going to build an institute to help address challenges with technology. But again my argument is institutions are technology and the implications I think I want to pull on that thread a little bit and I would love to engage a little bit further about what that means. Uh this is a slide I'm going to pronounce very carefully SDTS uh which stands for uh Hopter's 2022 article about socially disruptive technologies. uh and the reason that's relevant is because I think as we think about AI across these criteria uh our institutions are it's an open question whether we uh the degree to which this SDT is going to challenge our ability to solve for these functions.

So the first thing that he talks about is the depth of impact right so pick any discipline any industry any silo and look how deep that technology is going to run I don't have to give it to you guys that's happening uh there's also the range of impact right that general purpose technology it's called general for a reason uh but that is uh that is a challenge when our institutions are subject to very different things across the silos yet this technology is got such a wide range uh there's anybody know anybody who doesn't have strong than anybody have a sincere Oh yeah question. Yeah. So uh I have one person in my life who does not have a sincere or strong opinion about AI and I think it's my dog but I'm not sure. The dog may still have stronger opinions. Uh the veilance of the psychology is inherently sort of socially disruptive as you could tell by the fact that I am apocalyptivist.

Uh the ethical salience um is also super important. An example of uh previously socially disruptive technology is the mechanical ventilator uh which doesn't sound all that disruptive except that previously our definition of death had been when people stop breathing. Now you invented technology that allows people to continue to breathe and we have to reformulate like what does death mean. And the good news is our institutions were able to solve for that right they could now move to like brain death. There was discussions around insurance legalities etc.

Fantastic. I do not see that happening right now to be honest. partly because of the uncertainty the uncertainty of of the technology means that uh that it's virtually impossible uh I think to write meaningful policy in a space where this stuff is changing so quickly uh at least with the given current uh institutions and that speed part is also again stuff you all know already uh but it's not just the speed of diffusion as we're aware it's the speed of self-improvement uh I actually was a director of the AI division at one of the more famed institutes actually had a couple co-authors on the 2017 attention is all you need paper. Uh, and it is still amazing how often we have conversations about how AI will never do X and then it does it. Uh, which means that this stuff is pervading and another element of this is just the irreversibility of it.

Now, I don't like the expression you can't put toothpaste back in the tube because I have two kids and I know for fat you can. That's that's you can put toothpaste back in the tube and that's a valuable skill if you have children. Uh but I think the irreversibility of this is pretty clear and that's evident by the fact for example that you go and you ask people if you had to pay if we took these tools away from you how much would you pay to get them back and it's pretty clear that even the biggest skeptics would pay a lot of money to keep this. So this is true for now right AI what I call AI now it's probably going to be more true for what I call AI next. It's probably even more true for what I call AI after next.

Wait, there's more. the AI after AI after next. Uh notice I have not said super intelligence. I haven't said AGI. I don't even think you need to use terms like that is all it all it involves is just thinking about what what is the AI after next and there's really no reason to believe that it's not something uncertain irreversible deep and broad is impact etc.

Okay. So there's obelith besides this me of the optimism in some sense. Well, let's start with the opposite. The first is that when institutions are faced with these kind of threats, uh they have a couple different strategies. The first one is denying.

I actually got into this business working for a guy whose whole whose whole mission was to help the department of defense avoid what Arthur CR calls the true errors. There's the error of imagination where you just can't see how some logical or mechanical weap occur that makes something that was previously impossible possible. And then there's error of nerve which is you can kind of see it but boy that's inconvenient. Boy that means like the things of the world that we think we understand and all the plans we made is complicated stuff. So we're just going to be sort of the error of nerve in some sense.

I'm seeing a lot of that being committed in both cases, right? The measation and nerve. But the deny strategy is nothing to see here. It is amazing how many people still feel this way and I think signals like the challenges we've got going on. Uh the other strategy of course is defend.

Um at where I sit at university that is the immediate instinct. Let us defend the institution. Let us avoid people using this tool because it's cheating. uh what is what is at the very best let's use it to do what we already do. Um I I think that's less tenable as well to be honest.

Uh so the other I would say uh the third option is is you develop um which unfortunately means change uh and runs into a common uh misunderstanding I think when it comes to resilience that resilience is not stability. You know we we call things resilience when we build them and they defend they don't bend under exact the real resilience actually is when something changes the world and it also can change. So so me resilience isn't fast co-evolution and choose the word evolution very carefully here. It's not just development but evolution because evolution can do something we cannot do yet. We need to figure this out.

And here's what I think it does. Evolution has solved in many ways the dilemma that we face and the dilemma is we can exploit right you can work with the resources you have in front of you you can you can uh you harvest in this area you can the institutions we have now are harbingers of expertise they're harbingers of process they are exploiting and I don't mean this in necessarily the most you know malfeasant way they're exploiting what they were built to do in that ecosystem or we can explore and The problem is we have to build institutions that could do both now and I don't know the actual solution to that. Uh but I think this is an example of a community that like understands the exploration has deep uh roots in exploitation not not in the wrong way. Uh but we need to figure this out. How are we going to build these things?

So my solution in the near term is going to have to save a warf. I'm going to need your help here. Uh we're going to work for what we're going to do. Yes, I worked a little bit with this. So consider explor how about exportion.

Okay. Now how about esportation? For some reason quad really like this one graduate which Yeah. Hey song. All right.

We're good. Uh we offer a moment. We should talk to the guy who knows too. Uh flexionism. It's not bad, right?

You're a fuss fascist. Uh scans sounds risque. Uh venturagene you for venturing while you for agents uh or faginary that's got some right uh query vesting a chromosome was done uh rome steady is not bad are you a romestead right set no shop but also I still still working around you uh this was about entropizing entropy right okay off uh and then finally harvestigating you are harvestigators We need you to know that. Now, if it were for you to understand, you are harvesters. The problem is like you can do this individually, but if you're not surrounded by the institutions that can help support that, you you got an uphill fight as you already have.

So, what are the kinds of things we're actually trying to build? Are they fluxit? Are they exploitions? I think we'll pass that one. Uh, are they evolve institutions?

I don't even know how to pronounce that one. uh are they metitutions which sounds yeah vaguely unfortunately like metastasis which is not the other way. So you can see why clearly I need your help uh to figure out like what is the neoism now to talk to give us the framework of like how to do this uh especially and I will simply offer that like you know with that onwards because the only way around is through uh Tom is here and now I have the job of fireside chatting with him by the way Tom has actually built flexitions even if you didn't know it

that was great Adam so um thank you gives Adam a round of applause.