Knowing The Future
We all want to know the future. We are desperate to know: will the Stockmarket go up or go down? Will the insurance pay my claim? What are next week’s lotto numbers? Is what this Aussie is about to say going to be worth my time?
Karl Popper called this our individual “problem situation”. The unique circumstance we find ourselves in by virtue of the fact we are individuals. We want to know the outcome. We want to know the future. And there are a lot of people out there who want to tell you what the future will be. Of course we’ve always had soothsayers and fortune tellers, and all cultures still do. An appeal to superstitious ways of tapping into a secret knowledge or way of knowing denied to the lay person. It was recently posited that the Oracle of Delphi, one of many advisors to Alexander the Great was living in a cave whose fumes may just have been hallucinogenic. In any case when the most powerful man in the world at that time asks you if he is divine, I guess you will answer in the affirmative hallucinogen or not. But she didn’t know that. She could not have known that. And this will be a theme of what I want to say to you today: there is much we cannot know and most of that is yet to happen. Prediction. Prophecy. Between these two ways of speaking about the future is a chasm of science, rationality and reason broadly and I want to show you that gap because I am here to tell you: it cannot be bridged.
At the risk of stealing my conclusion almost before I have commenced let me just say this: there is a difference between knowing - which is to say possessing a good explanation that allows one to calculate what is going to happen precisely when given the laws of physics (there’s your prediction) versus lacking such knowledge or by ignoring the very thing that creates this amazing diverse world around us - and that’s prophecy. Prophecy is ignoring how knowledge yet to be created is yet to affect the world. In short, it is ignoring what choices people both individually and collectively as governments or businesses will make. For deep reasons I will come to, those choices cannot be predicted before they are made. And when we talk choice we mean decision. So in a real way decision making - the very heart of what our conference is about is decision creating or choice creation.
But don’t despair. While it might be true in most cases we cannot know the future, we can still have a commitment to rationality: to science, mathematics, epistemology and yes even intuition. Our own sense that something is or isn’t a good idea whether we can articulate why that is the case or not. None of this changes the fact we still want to know what the future will hold. Our decisions made moment to moment from the minute to the momentous depend on us guessing about the future. Hoping or dreading as the case is what the outcome might be.
But the problem is that while we know a lot - our knowledge is always finite. But our ignorance? Well that’s infinite. The more we discover, the more we find out we do not know. We discover one planet orbiting a distant star (I’m still old enough to remember that - back in 1992 - I was in year 9 - and finally - a planet beyond our solar system had been found. Few doubted it would be eventually but when it was the old debates were over. We knew about one. But also suddenly we knew there must be literally trillions upon trillions of planets out there we do not know yet the locations of. All because we discovered one. One. One discovery opened a window on a new infinity of ignorance and astronomers have been engaged ever since in a rush to find more and more Earth-like planets that we know are out there, but are ignorant on exactly where. But we do explain what we see - a few exoplanets here and there in terms of universal laws of physics we do not see that govern those trillions of other planets we do not see. We explain the seen in terms of the unseen.

But ignorance is not merely of that kind - the Socratic kind. Socrates conceded he was the wisest man in the world only insofar as he understood he knew so very little. Ignorance in this world runs deeper still. We now know, unlike Socrates, that it is woven into the very laws of physics. The laws of physics themselves bound our capacity to know in advance what is going to happen moments from now much less in a month or more. This is not merely a problem of intractability - a fancy term describing how some systems get so complex that it is infeasible to try calculating what all the particles will do from one moment to the next, but also for any single particle there are situations where we simply cannot predict what path it will take. I will come back to this.
And our ignorance is also borne out of what we are as human beings: creative entities. I mean that literally: we create things. Something wasn’t there before: it is there now. It’s not a mere recombination of existing knowledge - although it can be that too - but there is no denying people generate newness in this world. And fundamental physics, powerful though it is at prediction at times, in prescribed places, cannot predict the products of creativity.

The philosopher Karl Popper and his intellectual descendent David Deutsch have written and spoken extensively on what we know about knowledge - what it is, how it is tested against reality, what works, when to move on and the significance of the field that studies it - epistemology - for everything else. What you know about how you know affects your very own psychology. It underpins how science works, mathematics, history, all our academic disciplines and the day to day way we encounter the world in solving our problems. In what follows I credit both Popper and Deutsch up front for inspiring much of this, with some light embellishment from myself as we talk about the possibility of prediction in an age that is becoming increasingly uncomfortable with uncertainty. 
I don’t want people to be uncomfortable with uncertainty. I want them to instead relish in not seeking certainty. For to be certain, truly certain that something is true or something must happen is dogma. And people tend to go to wars, metaphorical and real, over dogmas. We want to resist certainty. Knowledge is good. Plain old uncertain, but good, explanations of this world.
So as you may guess my area is originally physics and physics is pre-eminent, shall we say among the sciences for making highly precise predictions. That’s not all it does, and we’ll come back to that too, but famously if you give me the equation and the height at which you are holding some ball, I will tell you down to the millisecond exactly when it will hit the ground when dropped.
That’s a prediction. And it gives rise to what has sometimes come to be called “physics envy” in other fields. A pejorative phrase because, for one thing, physics just is not mainly about prediction anyways. Incidentally chemistry does just as well. Take hydrochloric acid and mix with sodium hydroxide and you will get sodium chloride and water but no one speaks of chemistry envy.

In any case, science in the main is not about prediction. It’s about explanation. This distinction matters more than most seem to understand. Most of what we are doing day to day is trying to understand - to explain. Not to predict. And sometimes our explanations lead us to understanding that we cannot predict some things even in principle. Because of what it is we are: people. The creators of knowledge. And when in some fields we cannot make physics-like precise predictions this is no value judgement. After all the very thing I am talking to you about right now: epistemology - how knowledge is created, what it can do and its limitations, cannot predict anything like the content of any future explanation of epistemology say. 
But epistemology as it is explains quite a lot. So when I say, as I will, some things are inherently unpredictable this does not mean they are forever incomprehensible and ultimately that is what we need. An understanding. A good explanation. But let us deal with prediction first and what cannot be known as a matter of the most fundamental physics we know to date.
We have known about subatomic particles since the 1930s. Subatomic. Things smaller than atoms. We can take half-silvered mirrors - called so for unsurprising reasons - they have half the silver of regular mirrors and so instead of seeing a lovely almost perfect reflection, what one sees is only half the light bouncing off the mirror. The rest goes through - is transmitted as light tends to through glass. In fact a half silvered mirror is also known I some applications as one-way glass. You see it in those detective movies. It’s the glass in an interrogation room - like this. The room brightly lit can be seen into. The occupants of a darkened room outside cannot be seen. Not enough light is able to get through.


Now the thing about light is that it is made of packets of energy called photons and it was photons among other things that gave Einstein his Nobel prize. It wasn’t relativity - it was work on the photo electric effect. A demonstration of how light is quantised - comes in particles. So now I’m going to tell you, using half silvered mirrors and photons of light why there is a kind of inherent unknowable aspect to the world at the most fundamental level known.
We are able to fire single photons at half silvered mirrors. I said before if you just shine light at a half silvered mirror half is reflected, half is transmitted. Ok, well let’s turn down the intensity of our light beam until we have just one particle of light - a single photon - travelling towards our half silvered mirror.
Does it - the very one you choose to fire now - go through the half silvered mirror or does it bounce off? Now it’s no use saying that: well half of the time it goes through and half of the time it bounces off. 50% transmission, 50% reflection. That’s true. But you want to know about your photon.The one you are about to fire - what will it do? We literally cannot say. At heart the laws of physics are subjectively unpredictable. 
The laws are deterministic - it is absolutely determined that repeating the experiment over and over again the data will converge on 50% go through and the other 50% bounce off. But that is of no use if you are interested in your photon which may determine what you will do next for whatever reason. From your point of view you just don’t know. And not only don’t you know, the very laws of physics say it’s impossible to know. That’s baked into reality. There are many such examples. But this is just the start of it. There are some systems so simple: single particles and little mirrors and we cannot even know what will happen with certainty something as simple as whether it will bounce off or not.
To speak in terms of the motion of particles when discussing something like the topic of our conference here - decision making - is to commit the fallacy of reductionism. Of course it is. Now it’s not that anything I said about particles is false or entirely irrelevant - but rather it misses the point. You can’t base important life decisions on whether a particle of light will bounce off a mirror or not. 
But we are talking, if we are talking decision making, about guessing at the future. Wanting to be able to predict. So we should want to know when that’s on the table and when it’s not. When we make a choice we are choosing among options. Which options? Well the options we already know, yes, ok, that’s part of it. You’re in a foreign country. They have McDonalds and KFC - you know about them. They’re on the table. They’re options.
But also there are options you are yet to discover or better yet: create. And this idea of knowing and creating new knowledge is the domain of knowledge creation. And that’s really where I am going with this. So let’s begin with the good news.
You cannot predict the content of future knowledge. Well that’s just great. You cannot know what the content of the next theory yet to be created will be. If you could predict the content of the next scientific theory then, well you’d have that prediction of what’s in the theory to be discovered next year now. So, you’d know it now. You’ll have just contradicted yourself. 
I mean every physicist would love to know how to improve on Einstein. If any scientists in this world think they can predict stuff it’s physicists. So why aren’t they just calculating their way to whatever might replace Einstein’s General Relativity? Well, again: they cannot. We cannot predict the content of theories yet to be created. If they could do that, they’d already have the replacement. And presumably the Nobel Prize. 
Knowledge is hard won.
Something remarkable goes on when creating knowledge. And it’s all there in the “creating” part. Actual creation is going on in the minds of people even if some try to deny this. Especially with the existence of large language models - ChatGPT and it’s rivals which are all the rage right now. People in the tech world and allied areas are coming to say humans are just like large language models of a kind. Recombining things already stored in the brain in new ways. 
Sounds reasonable. But it’s also false. Recombination is not creativity however much the two at times might resemble each other. Sure recombination can give novelty - the illusion of creativity. But taking all the information across the world and putting it into a database that exceeds the size of any database until now mind you and having a clever algorithm recombine that information in novel ways is impressive, sure. Super impressive. And super useful. But it is not creativity. It is deriving things from that library. An infant child on the other hand - or any person for that matter, is achieving more remarkable things not by referring to vast libraries but rather given very scant information. They guess at the nature of reality with just a few tiny clues and are curious about it. Prodding it - prompting it to use the technical jargon. 
ChatGPT4o - wonderful as it is, hasn’t ever demonstrated curiosity about anything. It is a perfectly obedient chatbot. I don’t know if you know any real people - they aren’t in the main perfectly obedient. So we have a chasm of difference there. Recombination of existing knowledge versus true knowledge creation. It is knowledge creation I want to say that underpins human decision making. Knowledge creation is what separates a person from the tools that people use whether they be pocket calculators or large language models.
But to understand knowledge creation we need to understand knowledge - what is this thing knowledge? Over the years philosophers have often tried to explain some version of knowledge as being justified true belief. JTB: Justified True Belief. That has been said to be knowledge. 
Sometimes the justification is more or less strong - it’s justified to some extent people may say. So it’s “probably” justified they say or justified to a certain level of credibility or credence and this level of credibility can be actually quoted in precise numerical terms. That is known as Bayesianism.
Now I’m here to tell you that Bayesianism is wrong - root and branch. But before we get there we need to refute its ancestor: justified true belief. I can do that with a couple of examples.
First the so-called Gettier problems. They’re really Gettier refutations. Here’s one version - this by Bertrand Russell.
What time is it? You look up at the wall and there we have it an old style clock. 



You’ve lost your phone. You don’t wear a wristwatch. You’re in the middle of London and look up. What time is it? Midday. Now you know it’s midday. But what if I told you that clock is broken and has been showing midday for the last week as it undergoes maintenance. 
Ok, now what if I further tell you that by remarkable coincidence when you do look up at it, it really is midday. Do you actually know it’s midday? You’re not justified in thinking it’s midday, right? The clock is broken. But it is true it’s midday. But you don’t know that. So this is a problem for justified true belief. 
Another example and this is one I liked to use with my own students. Farmer Joe has a dairy farm and a large family. His favourite cow Daisy is a black and white beast he checks on every morning to see if she is near the farmhouse or has wondered a long way off to a distant corner of the property. He looks outside and sure enough: the familiar outline and unmistakable markings of Daisy confirm for him he now knows that Daisy is there.
Here’s the rub. Farmer Joe’s children happen to be participating in the local pantomime that week and little Jack is in the head and Jill is down the back inside a wonderfully well made material replica of Daisy. Daisy is just behind them out of sight. Now does Farmer Joe know Daisy is nearby? Again: it’s true she is, but he can’t know that. He never saw her. 
These so called Gettier cases are cut and dried refutations of this way of viewing knowledge as being justified true belief. But I want to take things up a notch.
I prefer real life examples not imaginary cows and broken clocks. My own go to example on all this is Newtonian Gravity. So Newtonian gravity is used in all sorts of ways to this day. It’s used in calculating rocket trajectories. It’s used in calculating tides. It’s even used in calculating the position that the the moon will be in as it orbits the Earth. And it’s historically used in calculating the positions of all the planets in the solar system. This is Newton’s so called universal law of gravitation here on the screen now. Simple Elegant. 
And as it turns out - completely, strictly speaking, false. First the F stands for Force and there is no such force of gravity. Now we know all this because as I say it was historically used to calculate the position of all the orbits of the planets. But there was one planet it just never quite worked for: Mercury. This was a problem for Newtonian gravity. For ages in the late 19th and early 20th century astronomers tried in vain to account for the small deviation that existed but grew worse year on year in calculating the position of mercury. It was known as the problem of the precession of mercury’s orbit. Basically the long axis of the orbit moved over time. Why?
(Insert image).
Well it took the genius of Albert Einstein to solve this. He created general relativity, Created I say and not discovered for reasons I will come to. He created an explanation - a scientific theory that there was no force but rather it was the curvature of spacetime. His equation looks, well shall we say somewhat more complicated than Newton’s. And it is. But it allows for more precise predictions. It gets Mercury’s orbit correct - but Newton’s doesn’t. So that rules out Newtonian gravity in favour of general relativity. Newton’s theory is falsified by experiment. Yes: Karl Popper was right - this is how it works. You take two theories. You perform an experiment and you rule one out - that’s called a crucial test. Yes, the test can go wrong - Popper was right about that too. In 2011 for example, at the Large Hadron Collider - the particle accelerator at CERN in Switzerland - particles called neutrinos were apparently observed by instruments travelling faster than the speed of light. This violated Einstein’s theory. People said at time in breathless headlines - Einstein proved wrong. Of course no such thing had happened. There was no alternative theory against which Einstein’s theory could be proved wrong. What would we replace it with? Not newton’s - that was already known to be wrong. Not just mercury but Eddington’s experiment where light was bent during an eclipse by just the right amount according to Einstein’s theory and not Newton’s. And there was more, but we don’t need that too to be convinced. In the large hadron collider experiment we simply did not have two theories to decide between. There’s genuine decision right? Which of these two theories. Well we only had one. And as it turned out it was the experiment that was flawed. Popper understood this was a possibility. It came to be known as the Duhem Quine hypothesis. Either the theory is wrong or the experiment is wrong. Well in this case at the large hadron collider the experiment was. They mis-connected a cable if you can believe it and so the calculations were all incorrect. 
It turned out after checking that neutrinos did indeed travel slower, if ever so slightly slower, than photons - particles of light.
So Einstein’s theory not only makes more precise predictions - that’s one thing - but it does another more important thing that science and all knowledge generally does: it explains stuff. It explains what exists and what it does. So in the case of Einstein’s theory of general relativity it explains that what exists is a fabric of spacetime. And what it does is bend and weave and curve whenever there is mass present. The mass moves according to where the space bends. Space tells matter how to move. Matter tells space how to curve. That was the way the great General Relativist - John Wheeler explained things. There is no force as no force is needed. Technically in science a force is mediated by a particle called a boson. But there is no boson here. No particle is needed to achieve this. Why does it happen? As Wittgenstein said: our space is turned at this point. In other words, we can dig no deeper. The theory is silent on that. We need a better theory we can test. 

But in any case Einstein’s theory is the best explanation thus far. But it is not complete. It does not mesh well with quantum theory for one thing for reasons I won’t go into now  - perhaps you can ask me during a break we have. But this is why Einstein’s theory of gravity was a genuine act of creation as a composer would create music. It’s not a discovery. Or more precisely it creates knowledge - general relativity - about something to be ever more precisely discovered - the phenomenon of gravity. Both Einstein and Newton (not to mention their predecessors back to Galileo and Aristotle) agreed gravity existed. They just disagreed on what it was. They had to create knowledge and someone will create better knowledge. We make objective progress and we know this because our prediction become ever more precise. We say: look the planet is there where I calculated and there it is. More or less.
So come to today and as I said Newton’s law of gravity says there is a force - but there is no force. And very strictly speaking it gets the prediction wrong. It’s not as bad as Ptolemy and other earlier physicists before Newton - but it is wrong. And looked at close enough it is completely wrong. 
So it’s not true.
And nothing about Newton’s theory can justify it as true.
So no one should believe it.
But it does constitute knowledge. I know it. Any physics student knows it. Now you know it too if you didn’t know it before. It’s not justified, it’s not true, so don’t believe it. Everything about knowledge = justified true belief is false.
So why does it count as knowledge? I said I’d come back to this.
It’s knowledge because it does solve a problem. Have a problem of calculating tides? Use Newton. Have a problem of calculating how long it will take for some rock to fall to earth when you drop it? Use newton. Want to calculate the trajectory of a rocket? Use newton. No one is tempted to use Einstein’s equation in this situation because although they technically could, it’s too laborious and the difference between Einstein and newton in these situations is only found in the 5th decimal place or something like that. IN other words it makes no practical difference. Unless the gravity is strong like in cases where a planet is super close to it’s host star as mercury is to the sun.
So what is knowledge? It’s important: it’s information that helps us make decisions. It is literally what helps us choose between things like Newtonian Gravity and Einstein’s relativity.




So knowledge solves a problem. Because it solves a problem it is useful information. And because it is useful information it tends to get copied.
As David Deutsch eloquently put it: it is information that once instantiated in a physical substrate tends to cause itself to remain so.
Or as Chiara Marletto a researcher into constructor theory says: it is resilient information.
Ok, so I’ve mentioned information there many times and implied knowledge is a kind of information. So then, what is information? This is a physics question - so let’s avoid that for now and stick to what concerns us here.
Information might be useful - that’s knowledge or it could be useless. And if it’s useless that gets discarded. It’s the errors - so a composer fills up their waste paper basket with all the attempts that do not meet their criteria for what is harmonious or melodic, let’s say. That gives them some information about what does not work.
Einstein didn’t get general relativity right on the first go. He got information from his mistakes. He guessed at what was really going on, as we all do, all of the time, and rather most of the time, no surprise, we’re wrong.

We must decide between what is useless information and what is actual objective knowledge - by which I mean a good explanation. And by good explanation I mean “hard to vary” and that means every part of the explanation has causal power and cannot be changed even slightly without ruining the explanation. Our explanation of what matter is comes down to atoms and the atoms themselves are constituted of smaller particles still. For example the electron. The electron is an essential part of the explanation - exceedingly hard to vary. For example it has a particular mass and a particular charge. We cannot arbitrarily change, swap or decide that some other particle will just as well fulfil that role. Indeed the existence of the electron is forced upon us by this grand explanation - which actually has a name in physics - the standard model of particle physics. 
By the way I did say I would come back to Bayesianism. This is an alternative epistemology that says we can assign probabilities to the options before us and calculate which is most likely - likely to be true presumably. But in fact as we have already seen, it’s rare to have many options on the table much less be able to assign probabilities to them. In the debate over gravity between Newton’s theory and Einstein’s no one was making actual bets with literal probabilities. They were performing experiments which were decisive. Crucial tests. Those tests simply ruled out Newton’s theory from being true. Now whatever the probability of it being true is now known. It’s zero. But it was always zero - it never was true. Yes, it contained some truth but we do not know how to quantify truth. It was just saying something qualitatively right about reality: there is this thing called gravity that exists between masses and so long as those masses are small this formula approximately calculates the magnitude of that influence. But it is and always was strictly false. So too with General Relativity. But both of them have been good explanations and are good explanations given a context. They are hard to vary. Newton’s here is an inverse square law. You can’t just change it to an inverse cubed law. That does not work. It’s hard to change it. And the only change we know of that works is to dump it altogether in favour of Einstein’s. An even harder to vary explanation. The Enlightenment itself - which we are still living through - is defined by a civilization’s search for good - hard to vary - explanations.
A tradition of criticism. Criticism aimed at finding flaws with existing knowledge with the aim of finding something better. Improving things.
Science and experimental testability is just a special case - it is a special kind of criticism that applies in science to scientific theories. And when a scientific theory is testable - that is a necessary but not sufficient criterion for a theory to be scientific - see here with David’s “grass cure for the common cold” thought experiment - a testable scientific theory makes it especially hard to vary. After all, its precise predictions of what happens in the physical world are very hard to vary. That makes the scientific theory of any phenomenon typically very good indeed.
So then, what’s an easy to vary explanation? Magic. Anytime something happens anywhere for which you do not yet know the answer just say: I explain it by magic. Magic did that thing. Don’t know how thunderstorms work? Lightning is magic. It’s an all-purpose explanation that can be varied from phenomena to phenomena never actually explaining anything because ultimately it’s contentless. We could swap out magic for miracle. Or sorcery. Easily varying the word we are using to describe such supernatural forces at work manipulating the physical world behind the scenes as it were. It is for this reason we say all appeals to the supernatural in this way are bad explanations. And I mean this in a technical sense. I do not mean “false” - although they may be that - I mean bad: too easily varied. If you’re happy with one supernatural explanation (say: The God of Thunder Thor causes electrical storms, then why would you be dissatisfied with a similar explanation that not the Norse God but rather the Hindu God Indra caused the storm? Incidentally Indra is the God of Thunder in Hindu. Gods of Thunder are easy to vary. Maybe they’re all the same thing. But whatever the case: magic or miracles did it is no explanation. We want to know how. And why? And more besides. We want to know how to test those explanations - how to sift, as it were, the good explanations from the not so good. And that is along this hard to vary criterion.

Now there are very rare cases where we have two good explanations to choose from - especially in science as we have already mentioned. Sometimes in real life we do: what to have for dinner. All options might seem good. Sometimes none do. And the decisions we have available to us in all these cases are actually infinite. Why infinite?
Because we are not machines. We are people. And people are not mechanisms just following deterministic laws. We are not dumb computers - we are something more. We are creation engines. We create. We bring new things into the world. New options that were not there before. Before anyone discovered nuclear physics no one could choose to have nuclear energy as an option on the table when it comes to energy production. There is a big debate agitating the political class in Australia right now: should we build a nuclear reactor or not? Never mind the answer - for our purposes here the fact that we can even ask the question is proof positive some knowledge was literally created. It is a question that could not have been asked in 1900.
So when we make a decision we are choosing not only among options we already know but among options we also create often on the spot AND we are choosing among the various explanations as to WHY we are doing what we do.
Again, I have to emphasise, this choice to do what we do often in real life, indeed usually, means rejecting all known options - everything known to that point and their associated explanations and creating something new.
The great mystery at the heart of what it is to be a person is we do not know in fine grained detail how this works. We know we do it - we create - we especially create explanations that put new options on the table - but we do not know how we do it. Objectively in terms of science, it is a mystery and it is the mystery that needs to be solved before we have artificial general intelligence. A computer system that would be a person. 
But also subjectively: you speak a sentence and you do not know how you get to the end of it. The words just flow. An idea pops into your head: that’s you doing that. You had the idea. You created it. But how? We don’t know. Oh sure, we can use words like subconscious or the great unconscious or instinct or perception, nurture and nature but these are just to label yet other partial aspects of the central mystery. Our creative capacity to solve problems and make progress and build things like this. And our choice to make things better or fail to make that choice at times comes down to that thing which we all share. The capacity to create something new. To have an idea. This mystery may one day be solved but it has not been solved yet. Although some think we are indeed on the way to solving it. That the super intelligence is around the corner. After all isn’t ChatGPT a sign of this? 

The rather abrupt appearance of the Large Language Model on the scene was a surprise to many. That remarkable piece of software I talked about earlier manages to achieve a level of language ability in an otherwise dumb computer few saw coming so soon. One moment it was obvious that machines were not able to pass the Turing Test and the next moment - well there it was. ChatGPT4 and its descendants and offshoots are able to have quite passable conversations with people. So, ok, it’s not perfect. We can still tell. But it’s a great leap upon what existed before. Now what explains this remarkable increase in functionality? Is it about to achieve what we have? Is it a stepping stone on the road to becoming an actual thinking creative person?
There are two parts to a large language model I want to highlight. First is the large part. ChatGPT4o has had loaded into it the largest library of information ever gathered together in one place on the planet. It is that huge. Almost all the books, the translations, the corpus of the existing internet and other libraries and it was trained on the works of human civilisation assembled over thousands of years and it can, at near the speed of light access it all from hard drives and well that’s the large part of the LLM, so to the second part of all that - the model:
The model is able to recombine all that information in ways people may never have seen before. But that is, from a God’s eye perspective so to speak, what it is doing: taking a library larger than any other that has existed before and using an algorithm (this is known as the transformer architecture for what it’s worth, I won’t attempt an explanation of that now - references here on my website for that) to recombine all of that information to provide useful answers in an, as I will again emphasise - an obedient way to prompts given by people. It does not self prompt. It does not have interests. It’s not creating new explanations even if it does generate novelty. 
What’s the difference? Well novelty is just something not seen before. And if your special language calculator has access to the entire corpus, more or less, of human knowledge and clever rules about how to recombine it in a sensible way you will get novelty: things not seen before. A new limerick, let’s say. A new script for a never before made episode of Seinfeld. But as I have already observed - recombination is not creativity because you are drawing from inside a finite library even if it is large. But we people can step outside of the library. We can think of something new that is not a mere recombination but a creation. It may also involve recombination but it has something else, some other spark of originality. A true innovation. This is what people can do, they can disobey. They can look at all the otherwise good explanations and reject them all and say: no. I’ll make something better. And why?
Well just for fun might be one reason but another could be what is animating us here and now. Decision making. What should I do next? How can I know what will work out? Or what might fail? But also because we want to solve our problems. All life is problem solving in the hope we make the future a better place.
So we want to predict - to know what that future holds. These days we even have whole professions devoted to it: futurists they are sometimes called. Or forecasters. And some even superforecasters! They’re all claiming to know the future. Can they? Well if they have a good explanation they might be able to.
A physicist as I have already mentioned can drop something out of a window and can calculate to a very high precision exactly how long it will take for that object to hit the ground given the height of the fall. But here’s the rub. This must ignore anything a human might do to intervene. If I am on level 8 of an apartment block and I drop a ball down to the ground floor, I might have a computer to help me with the calculations by correcting for air resistance and the shape of the ball and so on and it might well lead to the prediction that the ball will hit the ground exactly 2.21 seconds after I let it go. But all that precision and the very laws of physics are never going to give me the correct answer if the truth is that on level 4, Jeremy reaches his hand out the window and catches the ball before it ever hits the ground. The actions of a human being - the inherently unpredictable actions of a person - have intervened and changed the outcome. And that is for so simple a system as balls falling to the ground. This leads to a great dichotomy in forecasting. The distinction between prediction and prophecy and if we want to make rational decisions we should understand how this distinction plays out.
I have mentioned prediction a lot already. And predictions are indeed possible as I have said. The physical sciences are wonderful with prediction. Medicine is becoming increasingly good at predicting (take this antibiotic, your infection will be cured at some point in the future. Take this vaccine you will be less likely to suffer symptoms and so on. Predictions are possible. But long term and across civilisations they become increasingly difficult, intractable an indeed inherently impossible. Because, again, the one thing we cannot predict is the growth of knowledge. We cannot predict the content of our future explanations. Those things yet to be created. Trying to predict what the population of the planet Earth will be in 100 years assumes we know what choices people will make in 5 years. Then in 50 and so on. What life extension technologies might be invented. What natural disasters might occur. What cultural shifts happen. How trends might change. In other words a forecast about the future population of the globe is not a prediction: it is a prophecy. It cannot be known because it depends on what problems people are yet to encounter (that’s very unknown) and what solutions they might propose to overcome those problems (that’s doubly unknown). But people have a deep need, a hunger for knowing the future. 
Because they want to decide what to do today so they can best be prepared for tomorrow. Or next year. Or 10 years or more from now.
And there are all too many people with a vested interest in telling you they know the future. So we have new ideologies arising. And they share two things in common. The first is they claim to be predicting the future. And the second is: it won’t be good.

Some claim Artificial Intelligence today portends the end times. That the AI will become superintelligent and regard us as a threat and kill us all. Those are the AI doomers.

Some claim that the next virus to escape a laboratory or a wet market will be impervious to our attempts to curtail it and the population will collapse under the pressure of a spreading pathogen. Those are the disease doomers.
In my own area, astronomy: we can already point to asteroids and comets that have hit our planet over the course of it’s 4.5 billion year history that could strike the earth and wipe out almost all species. It’s coming again we are warned. Cosmological doomers. 

And some are concerned that we will do nothing to change the course of climate change: ecological doom. 
Some are concerned our political institutions are going awry too fast under the sway of social media influence. Civilisational doom. 
Elon Musk is worried about population collapse due to declining birth rates today. We’ll be extinct because people aren’t making enough babies. Baby doomers.
Nuclear apocalypse. Famine. Pestilence. Locusts. The list is much longer than those pictured. It’s a horror story out there if you pay too close attention to some of these people. People who guess at the future are very very good at imagining the problems we have and are going to have. What they are rather worse at is imagining the solutions to those problems. But we are problem solvers. It’s what we do as creators.
And the solutions are the hard part. The solutions are the scientific theories among other things. And the great feats of engineering. General Relativity was a solution that solved the problem of why Mercury orbited the Sun in just the way it did. Spotting the problem? Any old astronomer could do that. Anyone could point to the problem and get it. But it took Einstein to create the solution. And these days everyone’s mobile phone equipped with Google Maps harnesses the General Relativistic equations of Einstein daily literally saving lives when people would in the past have gotten hopelessly lost.
And so doomers are not predicting but prophesying. They are claiming to know the content of future explanations and in the main they are saying: no solution will come in time. We are doomed. So: make your decisions now given the worst is going to happen. The AI will take over, you’ll get a deadly virus, the sea levels will rise and democracies will fall everywhere. But how can they possibly know any of this? They cannot. They do not. It’s a living though, talking about the future death of civilisation. There’s a dollar in it. Now I concede on the other hand there is an equal and opposite irrational reaction to this. There are such people who say technology will inevitably save us. That the arrow of progress only ever points in one direction. Well that’s false. It’s not as bad - because it’s optimistic at least, but it is naive. There is no inevitability in things getting better in our making progress. We have to continue to choose to. To make the right choice.



So here’s a distinction for you. A prediction follows from a good explanation. It is a logical consequence of a good explanation. It says: ignoring people intervening, such as Jeremy reaching out to catch my ball before it hits the ground, or ignoring people creating the knowledge to reverse climate change and so on, this is what will happen. Or would happen. And explicitly saying: ignoring knowledge creation. That’s a prediction.
But a prophecy. A prophecy does the same thing but hides the ball on human creativity. It says: here’s what’s going to happen, inevitably. No matter what people do. The worst is going to happen. Prophets tend in the direction of pessimism precisely because imagining problems is easy but guessing at solutions is hard. Much harder. But it is what we do in the long run. We the descendants of the Enlightenment I mean. We who inherit what David Deutsch calls a tradition of criticism. We critique our best ideas in the hope of moving forward. And ideas come in many sorts as I have explained before in this podcast - with a focus on the conscious contents of our minds and not the unconscious which are substantial too.
But criticism is how we improve and we want to avoid becoming prophets in our own lives and in public. We want to make rational decisions that are informed by good predictions if and when they are available. Often they are not. Often we are engaging in pure guess work. So is there any way we can we make better guesses? 

One thing I want to push back on and I hope it isn’t heresy to say here is that when making decisions we should weigh the evidence. Ever watched a police or courtroom drama? Ever been in a courtroom and listened to the way those involved in the legal system speak? How does the judge make their decisions? What do they say explicitly about their process? They say they weigh the evidence. They say: on the balance of probabilities. But this entire way of thinking is not and cannot be the way anyone ever makes decisions.
For one thing: how much does evidence weigh? I’m not trying to be clever. If this is just a metaphor, then a metaphor for what? So the trope goes: there is evidence over here supporting theory A and then there is evidence over here supporting theory B.
But in truth? The truth is that in any legal case, say a murder, it’s rare for there to be many theories. Either you’ve got one very good suspect or you’ve got none. A single theory or no theory. If you have good detectives, and good forensic work you may just if you are lucky form a single hard to vary good explanation. And the judge can then decide that such an explanation explains the evidence. He does not weigh the evidence. The evidence is the very thing to be explained. Why were the defendants fingerprints on the knife which had the blood of the victim? The theory that the defendant killed the victim with the knife explains a lot. And absent any other better explanation, that defendant is now going to be a convicted murderer. The judge does not weigh the evidence, they explain it. Who cares about these semantics?
Well the thing is, it’s not semantics. If we do not know the function of evidence and the function of explanations then we are liable to fall into irrationalities when making decisions. And who wants to be irrational?

So let’s return to probabilities once again. The balance of probabilities. We hear this quite a lot too. So we have Alice and Bob and both are suspects in a murder. Alice had motive - but Bob, he had means and opportunity. After considering all evidence Judge Judith says that on balance of probabilities Alice is the murderer. But how did she balance these probabilities? Seriously? Who told her what the probability was? Is it just a manner of speaking? Well if it’s just a manner of speaking and is synonymous with “given the evidence the only known explanation that fits the facts is that Alice is the murderer” then very well. But again, we have to be careful. Whenever you are told that something is probably the case ask what that probability is. And then the really hard question: how do they know?
All the rage as I say now in my field is Bayesianism and this is a particular hobby horse of mine. This is supposed to be the mathematically sophisticated way to make decisions. We actually have a formula - here is it - Bayes’ Theorem - for making decisions. But these variables here, they need to be replaced by numbers. Actual probabilities - but how can we know what the probabilities are. 
If I roll a dice then to a first approximation (namely assuming an ideal dice and no real physical dice is actual ideal in the mathematical sense) we can say: there is a 1/6 chance of rolling a 3. Or a 2. Or any other number. And how can we say that? Because we know the odds going in. Importantly we know that denominator. There are only 6 numbers. We say a-priori - our prior probability that is — absent knowing anything else in the universe we know that. Dice have six sides each with a different number and each equally likely to come up. Our P(A) in other words in our theorem is 1/6. The denominator is 6 because we know a fair regular die has 6 sides.
But now imagine when you do not know the denominator. What do you do? What is the probability of an Earthquake tomorrow here in Aukland? Even if we can look at historic data - past trends do not indicate future performance - or future movements of the Earth as the case may be. 
How can we make decisions about what to do when the future is inherently uncertain? Well not on the basis of probability, which is unknown, but based upon good explanations. Earthquakes have happened, can happen and will happen - we just do not know when. It is an active area of research in the community of Seismologists. And not knowing when is no argument for assuming they will not happen tomorrow. So choose to build your structures AS IF the earthquake will hit tomorrow. That way there is no need to concern yourself with probability or Bayesian inference. You just rely on what you know to guard against, to the best extent possible, your ignorance of the future. And besides, things do not probably happen anyways. They either happen or do not happen. This renders the entire Bayesian approach misguided from the start - fundamentally. We want to know - want to have a good explanation if - something will happen or not. And if we do not know - if we lack a good explanation - we should admit that. We should say we do not know. That is the rational thing. Hence this maxim in yellow here. It is simply an appeal to modesty in the face of the unknown.
Now many decisions we make each and everyday aren’t quite so life and death as preparing for natural disasters, it must be admitted - but they do affect and involve other people. When making a decision what you want to be ideally is logical. Make a logical decision. But it should also be fair. Who wants to be illogical and unfair? Well I’m here to tell you it is impossible to be both fair and logical simultaneously when making decisions for groups of people and this strange but very deep and true fact about so called rational choice theory has wide ranging consequences.  It is known as Arrow’s theorem named for Kenneth Arrow - there he is - and putting things in rough plain English it says that there exists a proof that it is impossible to simultaneously be both logical and fair when choosing among existing options. 
Now to fully explain this would take a lecture in itself but let me give the abbreviated version. The United States has a congress and it has a certain number of members - 435 in the house of representatives. How many members does California have? It’s 52. Why? Well the law says that the politicians there are allocated from the states according to and in proportion to the population of those states. That’s what’s fair. So California has far more congresspeople than Rhode Island for example. But how many more? Well to make it perfectly fair the number of representatives should be exactly the proportion in the congress that there are people in California. The problem with this is you never get whole numbers. California is entitled to 51.7  representatives to be fair. But there is no such thing as 0.7 of a representative. So what do we do? Well we round. That’s seems right. Why not?  Well now you’re illogical because you round up to 52 and you round for all the other states as well and now you don’t have what you wanted - a 435 seat house of representatives. So why not just change that new number you get after the rounding? Well if you do that it won’t now be fair because the house will be that much larger or smaller and it won’t be fair for California or any other state because now they do not represent 12% of congress but something less - unfair. You cannot be both fair and logical and as I say there is a proof of why this always is the case. Now in the culture wars right now one side does make the point that equality of outcome just is not possible if you give people equal opportunity. But no one ever makes the point that really, if you are being logical - and you want to be logical by having equality of opportunity - we cannot have equality of outcome - or fairness of outcome. As I say there’s a mathematical proof of this. But no one ever mentions this. I guess it’s not expedient to start talking mathematical theorems in political debates. But why not?

But what do we do if we cannot be simultanously logical and fair? What do we do? Well we have to create some solution and they do and it’s not perfect of course because perfection here: perfect fairness and perfect mathematical logic cannot both occur simultaneously. This again shows how decisions are not a mechanistic - just follow the logic, wind the crank of rationality, kind of thing and generate your decision or your prediction out the other side. We must problem solve in the main - not choose between existing options as Bayesianism implies we do.
Rational decision making is not a matter of weighing the evidence or knowing the future. We cannot know the future but we can create it. Decision making is not about calculating the probabilities (when was the last time you took out a calculator to calculate the prior probability of any major life decision. Or any decision for that matter)? 
Decision making is choice creation. 
When people decide or make a decision, the common sense - but false - idea is that they decide among the options they know. This, so we are told, is by weighing the evidence or “on the balance of probabilities” or even “simply refuting all but one idea” - or so we are told. All this is totally wrong or misleading to some extent.
To decide one chooses. But not, typically, among the options already known - things already on the table even if this occurs sometimes. But often we are dissatisfied with everything on offer to some extent. The solution?
We create a new option. And that one, by our own lights, is better than all the others we knew about. We created something new and choose it. *That* is rational choice theory. That is decision making.
So rational decision making is a matter of creating and understanding or indeed creating an understanding and that means having an explanation. Ideally a good explanation. This does not mean it must be true. It just means good to you. By your own lights. You’re choosing this and not that because you can to your satisfaction in your mind say that you have indeed considered the options and ruled them out decisively. You’ve refuted them. And only one remains. The one you act upon. That’s your choice. It’s the only rational thing left to do. So you create the choices you make and you see yet to make. But you don’t yet know what decisions you will make yet given choices you are yet to create so you can predict the future. And that’s you in your own mind predicting you own behaviour. That which you knows better than others. So what hope do you have predicting what others will choose to create let alone civilisation?
Thankyou.