Thanks Ryan. I also found this very interesting. I’m looking forward to your foreshadowed future post on whether algorithmic approaches can give us definitive answers to the questions we have that are as reliable as knowledge. I’d particularly like to understand what you mean by knowledge in that context, as it seems to me that algorithmic approaches do provides us with some (perhaps the best) form of knowledge that we can use to effect beneficial outcomes for ourselves and others. For example, while reading this I couldn’t help wondering what’s wrong with getting an answer from a website about the best insurance policy or gadget to buy if you have a clear set of criteria by which you want to make your decision (price, features, availability etc). In those cases the algorithm is guaranteed to find the solution to your problem by identifying the cheapest and best featured (for you) available policy or gadget. What would be a better method for deciding? (In asking this, I’m outing myself as one of those readers you referred to in your previous post as finding algorithmic approaches to solving problems as intuitive and natural.) More specifically, how, in practice, would “human skill, ability or insight” be better applied in such cases given the inherent subjectivity — and often unreliability — of such attributes? Even the election question could in fact be usefully solved for you by an algorithm if you happened to have simple criteria by which you wanted to make your decision (e.g. which party has the policy that would increase my after-tax income). Algorithms can’t help in decision making if you’re unable to articulate a set of criteria by which you intend to make the decision, but I suspect there is some kind of checklist process going on in our brain for most complex decisions we make, even though we might not admit to it (for sound evolutionary reasons we automatically look for certain criteria in, say, prospective mates even if we can’t articulate them). In deciding whether to marry someone, for example, maybe an effective algorithmic approach is often applied sub-consciously along the lines of answering just a few simple questions like: “Do I want to get married now? Do I want to marry this particular person [which might be broken down into other questions relating to feelings of love, trust, respect, etc)]? Do they want to marry me? Are there any reasons not to marry them [again, this could be broken down into things like current marital status, criminal history, parental disapproval, etc]?” I wonder what a better way to make the decision would be. I suppose you could toss a coin, or consult an astrologer, or ask someone else to make the decision for you, but most of us don’t make significant decisions in that way.
Also, I didn’t understand how there can be “a dominant cultural mindset … that presumes we can control the world” if, as you suggest, it may well be a mindset held by only a minority of the population. In any case, I’m not convinced that there is a dominant cultural mindset that we can control the world and I think most people (operating well within the Overton Window) would agree that the unpredictability and potential influence of human agents (presidents of Russia, say) inevitably rules out orderly control of human affairs, and that controlling complex systems operating in the world like weather and climate is now and always will be just a pipe dream (my understanding is that chaos theory would preclude this in any case). However, I wouldn’t be surprised if most people believe that we (i.e. humanity) have a fair amount of control. Indeed history demonstrates that we have been able to exert a great deal of influence (albeit not absolute control) over things that we care about that can affect our experience of the world (having enough food to eat, improving our health, minimising non-consensual encounters with sharp-toothed animals, etc). And we have got there often because we have rightly trusted the science and we have wisely listened to the experts applying the right techniques. I expect that as our technology and knowledge improves, we will have more and more influence over the world and our experience in it, while always falling short of absolute control. Personally I wasn’t angry or furious about corona virus, nor was I particularly surprised by the outbreak of a pandemic, but I was reasonably confident that (as in fact happened — and surprisingly quickly) science would employ data and facts to find a way to significantly minimise its harm (while not gaining complete control). Is that what you mean by possessing epistemic confidence?
I should emphasise that the key question in my mind is whether we think that an algorithmic approach is enough to answer the question. It is very often helpful. On that, I'd note your phrasing that 'even the the election question could in fact be usefully solved for you by an algorithm if you happened to have simple criteria'. If your intuition is that an algorithmic approach would only work in limited circumstances, I'd fully agree with that.
On the issue of control, I'd point out that there was a vocal and influential group of people who were pushing for 'Zero Covid' - or at least the permanent elimination of covid from Australia (and other countries). This strongly influenced government decision making in a lot of places and is an example of a belief in some kind of absolute control.
On minorities and majorities, there has been some research that suggests that if 10% of the population are strongly committed to something and won't compromise, and the rest of the population isn't particularly opposed - the 10% normally wins out. The idea is that is independent of the decision making approach used, it is purely about social dynamics.
That's a fair point about the 10% minority winning out in relation to certain issues.
On Covid control, it's certainly true that there was initially a lot of support for an elimination approach, but ultimately the nature of corona virus made that impossible and it no longer has widespread support or influence over policy. However, an elimination strategy has been successful for other pathogens (smallpox and polio for example), so it wasn't necessarily a vain hope in the early stages of the pandemic. In the end control was achieved by mass vaccination - not as effective as elimination (which proved to be impossible), but sufficient for us to be able control the pandemic.
As for my intuition about algorithmic approaches only working in limited circumstances, I'll be provocative and go further than that and suggest that, while they're not perfect, they're usually better than any other approach. In that regard I think Sean in his comment is pointing out that alternatives to decision making rules created by algorithms are themselves prone to problems, which I certainly agree with.
Is there merit in exploring the difference between gaining and having knowledge and using knowledge? Even if you believe the world is knowable in theory, few would argue that it is known today. This leads me to wonder whether there is a form of certainty / uncertainty boundary or frontier individuals and society is constantly navigating, which necessarily requires responses to point in time unknowability.
I wonder also whether the above becomes important as we consider the role of algorithmic approaches. One use of algorithms is to predictably manage the certainty / uncertainty boundary by creating decision making rules (a heuristic). In essence these rules are designed for the many (the average) but are often applied to the one (the individual). Mistakes are inevitable - both systemic (bias - defined broadly) and random. The alternative, however, individualised decision making based on experience and judgement has the same problem (see Kahneman et al Noise).
Interesting point about mistakes. I think one of the attractions of algorithmic approaches is that we believe (or hope) they will reduce or eliminate mistakes. In many cases they probably will, but they probably also change the types of mistakes we make. If we accept that mistakes are inevitable, the question may then be which types of mistakes we are happiest to accept. There may be different answers to that in different contexts. For aviation safety, I think we would prefer the algorithmic approach. For others, we might want a more individualised approach.
The "which mistakes" question is a really good one. Agree that the answer will depend on circumstances, only some of which is captured in a normal risk management framework. A subtlety involves the difference between machine based and human based algorithms, versus truly individualised decision making. Medical decisions are a great example of interplay between rules and individual based decision modes.
Thanks Ryan. I also found this very interesting. I’m looking forward to your foreshadowed future post on whether algorithmic approaches can give us definitive answers to the questions we have that are as reliable as knowledge. I’d particularly like to understand what you mean by knowledge in that context, as it seems to me that algorithmic approaches do provides us with some (perhaps the best) form of knowledge that we can use to effect beneficial outcomes for ourselves and others. For example, while reading this I couldn’t help wondering what’s wrong with getting an answer from a website about the best insurance policy or gadget to buy if you have a clear set of criteria by which you want to make your decision (price, features, availability etc). In those cases the algorithm is guaranteed to find the solution to your problem by identifying the cheapest and best featured (for you) available policy or gadget. What would be a better method for deciding? (In asking this, I’m outing myself as one of those readers you referred to in your previous post as finding algorithmic approaches to solving problems as intuitive and natural.) More specifically, how, in practice, would “human skill, ability or insight” be better applied in such cases given the inherent subjectivity — and often unreliability — of such attributes? Even the election question could in fact be usefully solved for you by an algorithm if you happened to have simple criteria by which you wanted to make your decision (e.g. which party has the policy that would increase my after-tax income). Algorithms can’t help in decision making if you’re unable to articulate a set of criteria by which you intend to make the decision, but I suspect there is some kind of checklist process going on in our brain for most complex decisions we make, even though we might not admit to it (for sound evolutionary reasons we automatically look for certain criteria in, say, prospective mates even if we can’t articulate them). In deciding whether to marry someone, for example, maybe an effective algorithmic approach is often applied sub-consciously along the lines of answering just a few simple questions like: “Do I want to get married now? Do I want to marry this particular person [which might be broken down into other questions relating to feelings of love, trust, respect, etc)]? Do they want to marry me? Are there any reasons not to marry them [again, this could be broken down into things like current marital status, criminal history, parental disapproval, etc]?” I wonder what a better way to make the decision would be. I suppose you could toss a coin, or consult an astrologer, or ask someone else to make the decision for you, but most of us don’t make significant decisions in that way.
Also, I didn’t understand how there can be “a dominant cultural mindset … that presumes we can control the world” if, as you suggest, it may well be a mindset held by only a minority of the population. In any case, I’m not convinced that there is a dominant cultural mindset that we can control the world and I think most people (operating well within the Overton Window) would agree that the unpredictability and potential influence of human agents (presidents of Russia, say) inevitably rules out orderly control of human affairs, and that controlling complex systems operating in the world like weather and climate is now and always will be just a pipe dream (my understanding is that chaos theory would preclude this in any case). However, I wouldn’t be surprised if most people believe that we (i.e. humanity) have a fair amount of control. Indeed history demonstrates that we have been able to exert a great deal of influence (albeit not absolute control) over things that we care about that can affect our experience of the world (having enough food to eat, improving our health, minimising non-consensual encounters with sharp-toothed animals, etc). And we have got there often because we have rightly trusted the science and we have wisely listened to the experts applying the right techniques. I expect that as our technology and knowledge improves, we will have more and more influence over the world and our experience in it, while always falling short of absolute control. Personally I wasn’t angry or furious about corona virus, nor was I particularly surprised by the outbreak of a pandemic, but I was reasonably confident that (as in fact happened — and surprisingly quickly) science would employ data and facts to find a way to significantly minimise its harm (while not gaining complete control). Is that what you mean by possessing epistemic confidence?
Thanks. Hopefully the next post helps!
I should emphasise that the key question in my mind is whether we think that an algorithmic approach is enough to answer the question. It is very often helpful. On that, I'd note your phrasing that 'even the the election question could in fact be usefully solved for you by an algorithm if you happened to have simple criteria'. If your intuition is that an algorithmic approach would only work in limited circumstances, I'd fully agree with that.
On the issue of control, I'd point out that there was a vocal and influential group of people who were pushing for 'Zero Covid' - or at least the permanent elimination of covid from Australia (and other countries). This strongly influenced government decision making in a lot of places and is an example of a belief in some kind of absolute control.
On minorities and majorities, there has been some research that suggests that if 10% of the population are strongly committed to something and won't compromise, and the rest of the population isn't particularly opposed - the 10% normally wins out. The idea is that is independent of the decision making approach used, it is purely about social dynamics.
That's a fair point about the 10% minority winning out in relation to certain issues.
On Covid control, it's certainly true that there was initially a lot of support for an elimination approach, but ultimately the nature of corona virus made that impossible and it no longer has widespread support or influence over policy. However, an elimination strategy has been successful for other pathogens (smallpox and polio for example), so it wasn't necessarily a vain hope in the early stages of the pandemic. In the end control was achieved by mass vaccination - not as effective as elimination (which proved to be impossible), but sufficient for us to be able control the pandemic.
As for my intuition about algorithmic approaches only working in limited circumstances, I'll be provocative and go further than that and suggest that, while they're not perfect, they're usually better than any other approach. In that regard I think Sean in his comment is pointing out that alternatives to decision making rules created by algorithms are themselves prone to problems, which I certainly agree with.
Fascinating as always.
Is there merit in exploring the difference between gaining and having knowledge and using knowledge? Even if you believe the world is knowable in theory, few would argue that it is known today. This leads me to wonder whether there is a form of certainty / uncertainty boundary or frontier individuals and society is constantly navigating, which necessarily requires responses to point in time unknowability.
I wonder also whether the above becomes important as we consider the role of algorithmic approaches. One use of algorithms is to predictably manage the certainty / uncertainty boundary by creating decision making rules (a heuristic). In essence these rules are designed for the many (the average) but are often applied to the one (the individual). Mistakes are inevitable - both systemic (bias - defined broadly) and random. The alternative, however, individualised decision making based on experience and judgement has the same problem (see Kahneman et al Noise).
Not sure where any of this leaves us of course.
Thanks. Will think about that distinction.
Interesting point about mistakes. I think one of the attractions of algorithmic approaches is that we believe (or hope) they will reduce or eliminate mistakes. In many cases they probably will, but they probably also change the types of mistakes we make. If we accept that mistakes are inevitable, the question may then be which types of mistakes we are happiest to accept. There may be different answers to that in different contexts. For aviation safety, I think we would prefer the algorithmic approach. For others, we might want a more individualised approach.
The "which mistakes" question is a really good one. Agree that the answer will depend on circumstances, only some of which is captured in a normal risk management framework. A subtlety involves the difference between machine based and human based algorithms, versus truly individualised decision making. Medical decisions are a great example of interplay between rules and individual based decision modes.