WEBVTT 00:00:00.000 --> 00:00:02.439 Have you ever stopped and really thought about 00:00:02.439 --> 00:00:06.500 how crazy it is that your brain just figures 00:00:06.500 --> 00:00:10.679 out the world? Seriously, there's this constant 00:00:10.679 --> 00:00:13.779 flood of information hitting you from all your 00:00:13.779 --> 00:00:17.800 senses, sights, sounds, smells, everything, right? 00:00:18.399 --> 00:00:21.039 But you don't have to consciously think about 00:00:21.039 --> 00:00:23.170 putting it all together. It just like. happens 00:00:23.170 --> 00:00:25.609 automatically. It really is remarkable when you 00:00:25.609 --> 00:00:27.649 put it that way. Right. So how does that even 00:00:27.649 --> 00:00:29.350 work? Like, how does the brain take all that 00:00:29.350 --> 00:00:32.170 raw input and just create this like this whole 00:00:32.170 --> 00:00:34.789 reality that we experience? Yeah, that's a question 00:00:34.789 --> 00:00:36.829 that neuroscientists have been grappling with 00:00:36.829 --> 00:00:40.429 for, well, I mean, forever, really. And it feels 00:00:40.429 --> 00:00:42.929 so effortless for us, doesn't it? Totally. And 00:00:42.929 --> 00:00:44.950 speaking of this whole like figuring things out 00:00:44.950 --> 00:00:47.310 thing, there's this theory I came across recently 00:00:47.310 --> 00:00:49.359 that I just like. had to dive into. It's called 00:00:49.359 --> 00:00:52.240 the free energy principle or FEP. Oh yeah, I've 00:00:52.240 --> 00:00:54.859 heard of that. Yeah and get this was developed 00:00:54.859 --> 00:00:57.960 by this like brilliant neuroscientist Dr. Carl 00:00:57.960 --> 00:01:00.119 Friston and the really cool thing is there's 00:01:00.119 --> 00:01:03.359 been some like really solid experimental validation 00:01:03.359 --> 00:01:06.400 of it recently. So like this deep dive we're 00:01:06.400 --> 00:01:08.920 going to try to unpack what this FEP is all about 00:01:08.920 --> 00:01:11.599 and really dig into how it's like changing the 00:01:11.599 --> 00:01:14.019 way we think about how our brains learn. It's 00:01:14.019 --> 00:01:16.620 a timely discussion, for sure. I mean, for years 00:01:16.620 --> 00:01:19.599 now, we've seen these huge advances in artificial 00:01:19.599 --> 00:01:21.540 intelligence, particularly in machine learning. 00:01:22.019 --> 00:01:24.640 But they've kind of hit a wall in some fundamental 00:01:24.640 --> 00:01:26.700 ways, wouldn't you say? Totally. Like, think 00:01:26.700 --> 00:01:29.900 about it. These AI systems, they need just mountains 00:01:29.900 --> 00:01:32.700 of data to work, right? And then even when they 00:01:32.700 --> 00:01:35.379 work well, it's often really hard to understand 00:01:35.379 --> 00:01:37.480 how they actually got to their conclusions. Like, 00:01:37.599 --> 00:01:39.519 we call it the black box problem. Yeah, exactly. 00:01:39.599 --> 00:01:41.780 And there's this ongoing debate, you know, is 00:01:41.780 --> 00:01:44.540 it truly intelligent, or is it just really sophisticated 00:01:44.540 --> 00:01:46.340 pattern recognition? It's not quite the same 00:01:46.340 --> 00:01:48.519 thing, right? Right. And that's what makes this 00:01:48.519 --> 00:01:51.140 FEP approach so interesting. It's a totally different 00:01:51.140 --> 00:01:54.180 way of looking at intelligence, both for us and 00:01:54.180 --> 00:01:58.500 for potential AI. You know, so in this deep dive, 00:01:58.640 --> 00:02:00.760 our goal is to really get to the heart of what 00:02:00.760 --> 00:02:03.739 the FPP is all about and understand why this 00:02:03.739 --> 00:02:06.439 recent proof is such a big deal, not just for 00:02:06.439 --> 00:02:08.699 understanding our own minds, but like for the 00:02:08.699 --> 00:02:11.599 whole future of AI too. And, you know, this deep 00:02:11.599 --> 00:02:13.939 dive was actually inspired by some awesome resources 00:02:13.939 --> 00:02:16.460 that you, our listeners, shared with us. So thank 00:02:16.460 --> 00:02:19.080 you for that. Absolutely. So to kick things off, 00:02:19.439 --> 00:02:21.539 let's think about the sheer scale of what our 00:02:21.539 --> 00:02:23.159 brains are dealing with every single second. 00:02:23.310 --> 00:02:25.669 You know, you mentioned this flood of information, 00:02:26.110 --> 00:02:28.689 and it truly is like an ocean, just this massive 00:02:28.689 --> 00:02:30.930 amount of data constantly hitting our senses. 00:02:31.110 --> 00:02:32.629 It's kind of overwhelming when you think about 00:02:32.629 --> 00:02:34.949 it like that. Right. And our brains somehow manage 00:02:34.949 --> 00:02:37.270 to filter through all of that, extract what's 00:02:37.270 --> 00:02:39.639 important, and make sense of it all. So like, 00:02:39.680 --> 00:02:41.939 think about trying to follow a conversation in 00:02:41.939 --> 00:02:44.960 a really noisy crowded room, or like trying to 00:02:44.960 --> 00:02:47.539 pick out a friend's face in a huge crowd. That's 00:02:47.539 --> 00:02:49.120 what our brains are doing all the time, right? 00:02:49.479 --> 00:02:52.599 Exactly. And it's not a simple task at all. It's 00:02:52.599 --> 00:02:55.300 like trying to solve this incredibly complicated 00:02:55.300 --> 00:02:58.680 puzzle in reverse. The world throws all these 00:02:58.680 --> 00:03:01.759 effects at us, the sounds, the sights, and our 00:03:01.759 --> 00:03:03.860 brains have to figure out the hidden causes. 00:03:04.420 --> 00:03:07.400 What's making that specific sound? What's causing 00:03:07.400 --> 00:03:09.419 that particular pattern of light that our eyes 00:03:09.419 --> 00:03:11.740 are picking up? It's like being a detective, 00:03:12.120 --> 00:03:13.919 but you only have the clues and you have to work 00:03:13.919 --> 00:03:16.180 backward to figure out the crime And to make 00:03:16.180 --> 00:03:18.479 it even more complicated, the same underlying 00:03:18.479 --> 00:03:20.879 cause can produce wildly different effects depending 00:03:20.879 --> 00:03:24.020 on all sorts of things, right? Precisely. Like, 00:03:24.020 --> 00:03:26.280 think about recognizing someone's face. That 00:03:26.280 --> 00:03:28.039 same face can look totally different depending 00:03:28.039 --> 00:03:29.879 on the lighting, their expression, the angle 00:03:29.879 --> 00:03:31.680 you're seeing them from, all sorts of variables. 00:03:32.439 --> 00:03:34.659 But our brains still manage to, like, instantly 00:03:34.659 --> 00:03:36.879 recognize that it's the same person. It's mind 00:03:36.879 --> 00:03:39.479 -blowing. It is. And this is where the free energy 00:03:39.479 --> 00:03:42.460 principle comes in. This theory proposes that 00:03:42.460 --> 00:03:44.879 our brains aren't just passively absorbing all 00:03:44.879 --> 00:03:48.180 this sensory data like a sponge. Instead, it 00:03:48.180 --> 00:03:51.060 suggests that our neurons are constantly generating 00:03:51.060 --> 00:03:54.139 these what we call top -down predictions about 00:03:54.139 --> 00:03:56.020 what they're going to encounter. It's like our 00:03:56.020 --> 00:03:58.060 brains are actively trying to guess what's going 00:03:58.060 --> 00:04:00.680 to happen next, to explain away the incoming 00:04:00.680 --> 00:04:03.939 sensory data before it even fully arrives. So 00:04:03.939 --> 00:04:06.180 it's like our brains are making these mini hypotheses 00:04:06.180 --> 00:04:08.819 about the world all the time. Exactly. And those 00:04:08.819 --> 00:04:10.580 predictions aren't always going to be accurate. 00:04:10.699 --> 00:04:13.379 Right. And that's where learning comes in. When 00:04:13.379 --> 00:04:15.960 there's a mismatch between our predictions and 00:04:15.960 --> 00:04:18.740 the actual sensory input we receive, that creates 00:04:18.740 --> 00:04:20.959 what's called a prediction error. A prediction 00:04:20.959 --> 00:04:23.319 error. Yeah. And those prediction errors act 00:04:23.319 --> 00:04:26.339 as a signal to the brain that its internal model 00:04:26.339 --> 00:04:29.420 of the world Its current best guess about how 00:04:29.420 --> 00:04:32.240 things work isn't quite right. So the brain uses 00:04:32.240 --> 00:04:34.779 those errors to learn and improve its predictions. 00:04:35.279 --> 00:04:37.899 Exactly. It's constantly updating its beliefs, 00:04:38.560 --> 00:04:41.480 refining its internal model so that it can make 00:04:41.480 --> 00:04:44.060 better, more accurate predictions in the future. 00:04:44.600 --> 00:04:46.439 You can think of the brain as this incredibly 00:04:46.439 --> 00:04:50.069 sophisticated inference machine, constantly trying 00:04:50.069 --> 00:04:52.810 to figure out the best explanation for the sensory 00:04:52.810 --> 00:04:55.329 data it's receiving. So it's not just about reacting 00:04:55.329 --> 00:04:57.490 to what's happening, but more about proactively 00:04:57.490 --> 00:05:00.029 trying to understand the underlying causes, the 00:05:00.029 --> 00:05:02.509 why behind everything. Nicely. That makes a lot 00:05:02.509 --> 00:05:04.329 of sense. And you know, it's interesting because 00:05:04.329 --> 00:05:07.230 for a while, the FEP was this really compelling 00:05:07.230 --> 00:05:10.829 framework, but there wasn't much direct experimental 00:05:10.829 --> 00:05:13.990 evidence to support it. Like it made sense theoretically, 00:05:13.990 --> 00:05:17.069 but it was hard to prove that this was actually 00:05:16.910 --> 00:05:19.649 how brains were working at a fundamental level. 00:05:19.709 --> 00:05:21.930 Yeah, you need that real -world validation. Exactly. 00:05:22.269 --> 00:05:24.129 And that's where this amazing research out of 00:05:24.129 --> 00:05:26.250 the Riken Research Institute in Japan comes in. 00:05:26.529 --> 00:05:28.990 They actually managed to get some really solid 00:05:28.990 --> 00:05:31.769 experimental proof for the SEP. It was a groundbreaking 00:05:31.769 --> 00:05:34.250 study. Totally. So, like, tell us more about 00:05:34.250 --> 00:05:36.810 the study. What did they actually do? Well, they 00:05:36.810 --> 00:05:39.430 started by creating these miniature controlled 00:05:39.430 --> 00:05:43.310 networks of neurons using cells from rat embryos. 00:05:43.969 --> 00:05:46.230 Essentially, they were working with these tiny, 00:05:46.569 --> 00:05:49.230 simplified versions of brains in a dish. Wow, 00:05:49.269 --> 00:05:52.569 that's wild. It is, right. And then, they delivered 00:05:52.569 --> 00:05:55.009 these patterns of electrical stimulation that 00:05:55.009 --> 00:05:58.089 were designed to mimic auditory sensations. Like, 00:05:58.269 --> 00:06:00.129 imagine you're listening to two different speakers 00:06:00.129 --> 00:06:02.750 talking at the same time. That kind of complex, 00:06:02.949 --> 00:06:05.470 mixed audio signal. They replicated that for 00:06:05.470 --> 00:06:07.350 these little neuron networks. So it was like 00:06:07.350 --> 00:06:09.550 they created a miniature, simplified version 00:06:09.550 --> 00:06:12.170 of a noisy environment for these brain cell cultures. 00:06:12.209 --> 00:06:15.319 Exactly. And initially, the networks reacted 00:06:15.319 --> 00:06:17.180 kind of randomly, which is what you'd expect 00:06:17.180 --> 00:06:18.879 from a system that's just starting to form and 00:06:18.879 --> 00:06:21.920 hasn't learned anything yet. But then, over time, 00:06:22.220 --> 00:06:24.160 something really remarkable happened. What happened? 00:06:24.500 --> 00:06:27.300 These initially chaotic networks started to self 00:06:27.300 --> 00:06:29.860 -organize. They began to selectively respond 00:06:29.860 --> 00:06:32.000 to one of the simulated speakers or the other, 00:06:32.399 --> 00:06:34.019 filtering out the noise from the other speaker. 00:06:34.250 --> 00:06:36.329 It was as if they were learning to tune into 00:06:36.329 --> 00:06:38.730 a single voice in a crowded room, just like we 00:06:38.730 --> 00:06:40.550 were talking about earlier. So they were actually 00:06:40.550 --> 00:06:43.529 able to separate out those mixed sensory signals 00:06:43.529 --> 00:06:46.009 and identify the likely sources, even though 00:06:46.009 --> 00:06:47.930 they'd never encountered anything like that before. 00:06:48.290 --> 00:06:50.790 Exactly. And the most compelling part, this self 00:06:50.790 --> 00:06:53.699 -organization wasn't just random. The researchers 00:06:53.699 --> 00:06:57.339 compared the way these biological networks structure 00:06:57.339 --> 00:07:00.240 themselves To computer models that were based 00:07:00.240 --> 00:07:02.819 on the free energy principle and guess what a 00:07:02.819 --> 00:07:06.220 match they matched perfectly The biological networks 00:07:06.220 --> 00:07:08.279 were organizing themselves in a way that was 00:07:08.279 --> 00:07:10.779 completely consistent with the predictions made 00:07:10.779 --> 00:07:14.699 by the FEP based computer models They essentially 00:07:14.699 --> 00:07:17.379 reverse engineered the computational models that 00:07:17.379 --> 00:07:19.939 these living neuronal networks seemed to be implicitly 00:07:19.939 --> 00:07:21.939 using. It was like they had cracked the code 00:07:21.939 --> 00:07:24.279 of how these networks were learning and adapting. 00:07:24.439 --> 00:07:26.180 So they could actually predict how the learning 00:07:26.180 --> 00:07:28.839 would unfold just by applying the free energy 00:07:28.839 --> 00:07:32.000 principle. Precisely. And they also saw that 00:07:32.000 --> 00:07:34.319 the mismatches between the network's internal 00:07:34.319 --> 00:07:37.339 state, its expectations, and the incoming sensory 00:07:37.339 --> 00:07:40.480 data were causing changes at the synaptic level. 00:07:40.639 --> 00:07:43.379 the connections between neurons. These synaptic 00:07:43.379 --> 00:07:46.199 changes were essentially fine -tuning the network's 00:07:46.199 --> 00:07:48.459 internal model of the world, making it better 00:07:48.459 --> 00:07:50.259 at predicting what was going to happen next. 00:07:50.459 --> 00:07:52.500 So it was this constant feedback loop, right? 00:07:52.500 --> 00:07:56.060 Prediction, error, update, repeat. Exactly. And 00:07:56.060 --> 00:07:57.779 they didn't stop there. They wanted to test the 00:07:57.779 --> 00:08:00.939 theory even further. So they used these special 00:08:00.939 --> 00:08:03.259 drugs to temporarily change the activity of the 00:08:03.259 --> 00:08:05.600 neurons. They could make them either more or 00:08:05.600 --> 00:08:09.410 less likely to fire kind of mess of their normal 00:08:09.410 --> 00:08:10.930 functioning a bit. And what happened when they 00:08:10.930 --> 00:08:13.990 did that? Well, the FEP predicted that these 00:08:13.990 --> 00:08:15.930 changes would affect the learning process in 00:08:15.930 --> 00:08:17.930 a very specific way, and that's exactly what 00:08:17.930 --> 00:08:20.930 they observed. By disrupting the network's internal 00:08:20.930 --> 00:08:23.889 models, they could actually control how the learning 00:08:23.889 --> 00:08:26.649 process unfolded. It was like providing even 00:08:26.649 --> 00:08:28.990 stronger evidence that the FEP was indeed the 00:08:28.990 --> 00:08:30.810 guiding principle behind how these networks were 00:08:30.810 --> 00:08:33.220 learning. So it's not just a theoretical concept, 00:08:33.299 --> 00:08:36.139 it's actually how real brains work. Precisely. 00:08:36.539 --> 00:08:38.679 And one of the lead researchers on the project, 00:08:38.919 --> 00:08:41.379 Takuya Isomura, put it really well. He said, 00:08:41.559 --> 00:08:44.120 I'm quoting here, our results suggest that the 00:08:44.120 --> 00:08:46.500 free energy principle is the self -organizing 00:08:46.500 --> 00:08:49.620 principle of biological neural networks. That's 00:08:49.620 --> 00:08:52.539 a pretty powerful statement. It is. And it has 00:08:52.539 --> 00:08:54.960 huge implications, not just for understanding 00:08:54.960 --> 00:08:57.259 our own brains, but also for the future of AI. 00:08:57.989 --> 00:08:59.789 Absolutely. Because if we can figure out how 00:08:59.789 --> 00:09:03.129 to build AI systems that learn and adapt in the 00:09:03.129 --> 00:09:05.730 same way that our brains do, using these principles 00:09:05.730 --> 00:09:08.529 of the FEP, that could be a real game changer. 00:09:08.750 --> 00:09:11.350 It could. And you know, Dr. Friston, the guy 00:09:11.350 --> 00:09:13.690 who came up with the FEP, is actually the chief 00:09:13.690 --> 00:09:16.149 scientist at a company called Versus AI. And 00:09:16.149 --> 00:09:17.750 they're doing some really fascinating work in 00:09:17.750 --> 00:09:19.769 this area. They're developing what's called Active 00:09:19.769 --> 00:09:23.299 Inference AI. So active inference AI, how is 00:09:23.299 --> 00:09:25.559 that different from the kind of AI that we're 00:09:25.559 --> 00:09:27.059 used to hearing about, like machine learning 00:09:27.059 --> 00:09:29.759 and all that? Well, traditional AI is very data 00:09:29.759 --> 00:09:32.240 -driven. You need these massive data sets to 00:09:32.240 --> 00:09:35.399 train the algorithms. And it can be really computationally 00:09:35.399 --> 00:09:38.059 expensive. Active inference AI, on the other 00:09:38.059 --> 00:09:41.340 hand, is inspired by how biological systems learn. 00:09:42.019 --> 00:09:44.200 It's all about using Bayesian inference, like 00:09:44.200 --> 00:09:47.440 our brains do, to constantly update and refine 00:09:47.440 --> 00:09:49.970 internal models of the world. So it's not just 00:09:49.970 --> 00:09:52.769 about passively processing data. It's about actively 00:09:52.769 --> 00:09:55.169 making predictions and learning from the results. 00:09:55.289 --> 00:09:57.950 Exactly. And it's much more efficient, too. Active 00:09:57.950 --> 00:10:00.649 inference AI systems can learn from much smaller 00:10:00.649 --> 00:10:03.210 amounts of data, and they can adapt to new situations 00:10:03.210 --> 00:10:05.210 more quickly. It's like they're more curious, 00:10:05.289 --> 00:10:07.889 more proactive in their learning. In a way, yes. 00:10:08.009 --> 00:10:09.909 They're not just waiting to be fed information. 00:10:10.250 --> 00:10:13.009 They're actively seeking it out and testing their 00:10:13.009 --> 00:10:14.929 predictions against the world. And that's a big 00:10:14.929 --> 00:10:17.169 part of why people are so excited about the potential 00:10:17.169 --> 00:10:20.220 of active inference AI. So, you know, we've talked 00:10:20.220 --> 00:10:22.500 a lot about these internal models that our brains 00:10:22.500 --> 00:10:25.240 create, but how does that actually work in practice? 00:10:25.399 --> 00:10:27.700 Like, what does an internal model even look like? 00:10:27.919 --> 00:10:30.299 That's a great question. And it's one that researchers 00:10:30.299 --> 00:10:32.659 are still trying to fully understand. But we 00:10:32.659 --> 00:10:35.080 can think of an internal model as a kind of simplified 00:10:35.080 --> 00:10:37.830 representation of the world. It captures the 00:10:37.830 --> 00:10:40.269 relationships between different things, the causes 00:10:40.269 --> 00:10:43.309 and effects, the patterns and regularities, and 00:10:43.309 --> 00:10:45.289 it allows us to make predictions about what's 00:10:45.289 --> 00:10:47.929 going to happen next, even in new and unfamiliar 00:10:47.929 --> 00:10:50.990 situations. So like, if I see a ball rolling 00:10:50.990 --> 00:10:53.950 towards me, my internal model of physics tells 00:10:53.950 --> 00:10:55.970 me that it's going to keep rolling unless something 00:10:55.970 --> 00:10:59.149 stops it. Exactly. And your internal model allows 00:10:59.149 --> 00:11:01.429 you to predict where the ball is going to go, 00:11:01.769 --> 00:11:03.929 how fast it's going to be moving, and even what's 00:11:03.929 --> 00:11:05.690 going to happen if you try to catch it. So these 00:11:05.690 --> 00:11:07.789 internal models are what allow us to navigate 00:11:07.789 --> 00:11:10.730 the world, make decisions, and plan for the future. 00:11:11.129 --> 00:11:13.289 Precisely. And that's what makes the free energy 00:11:13.289 --> 00:11:16.409 principle so powerful. It provides a unifying 00:11:16.409 --> 00:11:18.830 framework for understanding all sorts of different 00:11:18.830 --> 00:11:21.570 cognitive processes, from perception to learning 00:11:21.570 --> 00:11:24.879 to decision -making. And it also has huge implications 00:11:24.879 --> 00:11:27.460 for AI, because if we can build AI systems that 00:11:27.460 --> 00:11:30.500 have these sophisticated internal models, they'll 00:11:30.500 --> 00:11:32.879 be able to do so much more than just pattern 00:11:32.879 --> 00:11:35.460 recognition. They'll be able to actually understand 00:11:35.460 --> 00:11:38.159 the world, reason about it, and make intelligent 00:11:38.159 --> 00:11:40.879 decisions. Exactly. And that's why this recent 00:11:40.879 --> 00:11:43.759 experimental validation of the FEP is so significant. 00:11:44.039 --> 00:11:45.899 It suggests that we're on the right track. We're 00:11:45.899 --> 00:11:48.320 starting to understand the fundamental principles 00:11:48.320 --> 00:11:50.159 that underlie intelligence. And we're starting 00:11:50.159 --> 00:11:52.860 to build AI systems that are inspired by those 00:11:52.860 --> 00:11:55.080 principles. It's a really exciting time to be 00:11:55.080 --> 00:11:57.799 following this field. And for our listeners who 00:11:57.799 --> 00:11:59.440 want to dive deeper into this, there are some 00:11:59.440 --> 00:12:02.539 great resources available online. Versus AI, 00:12:02.620 --> 00:12:05.100 the company that Dr. Friston is involved with, 00:12:05.320 --> 00:12:07.399 has a really informative website. You can find 00:12:07.399 --> 00:12:10.600 them at versus .ai. And for those interested 00:12:10.600 --> 00:12:12.639 in the broader context of how this might all 00:12:12.639 --> 00:12:15.100 fit together in the future, the Spatial Web Foundation 00:12:15.100 --> 00:12:17.860 is doing some really interesting work on creating 00:12:17.860 --> 00:12:20.740 this kind of shared, intelligent digital space, 00:12:21.000 --> 00:12:23.539 a sort of metaverse for AI. You can check them 00:12:23.539 --> 00:12:26.320 out at spatialwebfoundation .org. And Denise 00:12:26.320 --> 00:12:28.519 Holt, she's got this awesome blog where she breaks 00:12:28.519 --> 00:12:31.200 down all these complex AI concepts in a way that's 00:12:31.200 --> 00:12:33.080 really easy to understand. You can find her at 00:12:33.080 --> 00:12:36.320 DeniseHolt .us. Oh, fantastic resources. So I 00:12:36.320 --> 00:12:38.139 guess the big question that this all leaves us 00:12:38.139 --> 00:12:40.440 with is, if our brains are essentially prediction 00:12:40.440 --> 00:12:43.500 machines, constantly shaping our perception of 00:12:43.500 --> 00:12:46.289 reality based on our internal models, What does 00:12:46.289 --> 00:12:48.850 that say about the nature of reality itself? 00:12:49.129 --> 00:12:51.649 Like, is reality even objective or is it just 00:12:51.649 --> 00:12:53.809 a construction of our own minds? It's a deep 00:12:53.809 --> 00:12:55.730 philosophical question and one that I don't think 00:12:55.730 --> 00:12:58.529 we have a definitive answer to yet. But I think 00:12:58.529 --> 00:13:01.210 the free energy principle gives us a really interesting 00:13:01.210 --> 00:13:04.490 new perspective on this whole debate. It suggests 00:13:04.490 --> 00:13:07.350 that the reality we experience is, at least in 00:13:07.350 --> 00:13:10.470 part, a product of our own expectations and predictions. 00:13:10.809 --> 00:13:12.570 And it's a pretty mind blowing idea to think 00:13:12.570 --> 00:13:14.850 about. It really is. Thanks for joining us on 00:13:14.850 --> 00:13:16.769 this deep dive into the free energy principle. 00:13:17.149 --> 00:13:19.549 Until next time, keep those brains predicting.