middle management is what the other guy does

climbing the ladder of discomfort

I am back to thinking about artificial intelligence, which means that I am once more considering the Stafford Beer joke which is one of the central set texts of this blog:

“When businesses adopted the computer, they tended to use it to automate their existing processes. It was as if they had been given the opportunity to hire Mozart, Einstein and Galileo, then put them to work memorising the phone book so that executives could look up numbers more quickly”. Discuss[1]

This is definitely a worry; if you’ve heard me on a podcast telling this joke, you’ve heard me noting that if you believe most articles in the business press about AI, we’re on the verge of creating a godlike superintelligence which we will use to send more emails. The problem is not so much that sceptics aren’t taking AI seriously, but that too many people want a revolutionary, disruptive technology which doesn’t really change anything. What would management look like if, rather than going “an even cheaper and worse way to do customer service!” or “it’s like junior analysts, that don’t need sleep!”, we started thinking – what management tasks could potentially be replaced by AI? As Baldur Bjarnson (thanks to Dave Gerard for the link!) says, it’s hard to take management’s embrace of AI seriously when they appear to only want it to change the world for other people.

I think this is interesting partly as a bit of futurism fun, but also because it might sharpen up our understanding of what management really is, and what are the things which managers do that aren’t really important, but just have to be done by managers because nobody else will do them. These thoughts are somewhat influenced by Henry Farrell and his gang’s essay on social technology[2].

(I’ve occasionally argued in the past that we used to think that memorising long texts was an important part of being a poet, or that accurate addition was an important part of being a mathematician; as technology improved, so did our insight into what these things were really about. The same point, about accurate representation of images and its relationship to art, is now a debate that only fogeys even have).

So, here’s a sort of “ladder of discomfort”. If any of you guys have more LinkedIn clout than I do (practically zero), do try this out on your networks and report back. I’ve thought of five managerial-type things which might be AI-substitutable, and arranged them in ascending order of how uncomfortable I think people would be in actually doing them:

---

1. Would you be comfortable in using an LLM to draw up your marketing strategy with a goal of increasing share by 10%?

I am pretty sure that a significant percentage of relevant managers are already doing this, at least “as a first step” or “just to help come up with ideas”.

2. Would you be prepared to allow an LLM to make hiring decisions for hourly-paid employees? For junior managerial employees? For senior management?

Obviously this is also being done right now for very junior posts, and I think you might get surprisingly far up the tree before people started to demur. If only as a cybernetic teddy bear, or as a way of rationalising decisions already made, but there are plenty of human managers in employment who also have this role.

3. Would you let an LLM agent handle your procurement, dealing with suppliers and negotiating prices autonomously? How much supervision would you feel the need to provide, in the best case?

My understanding is that the current state of agentic AI is not up to this task yet, but it’s definitely a live work programme. There is also quite a lively problem of security at this stage. Even with those problems assumed solved (and I think there would be a substantial “reject premise” element to the survey), I think quite a material proportion of managers would be reluctant to outsource this.

4. Rather than having a defined organisation chart, roles and divisions, would you be interested in a model where AI responded to strategic priorities by assigning people to functional and problem-solving teams on the basis of its assessment of their abilities, compatibility and the urgency and importance of the tasks?

In my view, this is table stakes for “are we really going to be able to use AI as a social technology for management”? but I don’t think many actual managers would really be comfortable with it. Although I have a suspicion that quite a few of them, answering a survey on LinkedIn, would recognise this as the trick question it kind of is, and give an insincere answer.

5. Imagine a new AI-enabled email tool, which doesn’t have a “To:”, “CC:” or “BCC:” bar. You just type in the email or report that you want to produce and press send. The LLM, trained on all of your company’s data, decides who needs to see your email, with what priority and decides who should be told who else it has been sent to. Assume for the purposes of this thought experiment that the LLM is much more advanced than any current chatbot.

My gut feeling is that almost nobody would go along with this. And now, I think, we have learned something about the essence of management.

---

[1] It should be noted here, that this quote in this form is very much apocryphal, and summarised by me! On Friday, I’ll try to reproduce the accurate quote (or at least, the most conveniently available online version of it, as Stafford used it in several lectures), along with some of the great man’s gloss on it, in the latest episode of the “Beer Tasting Notes” series.

[2] Henry has the charming habit of saying at conferences that the audience should credit his co-authors with all the great insights and attribute any mistakes to him. I don’t believe for a moment this is true, so I invite my audience to do the opposite. Unless it’s something co-written with me of course, if someone’s prepared to share the blame for my crap I need all the help I can get.

subtle and unknowable but quite important

exercises in practical epistemology

A short one for Friday, that’s been on my mind for a while. Let’s take a simple formula for estimating a risk exposure:

Probability of Loss x Severity of Loss = Expected Loss

Now let’s get a little bit more philosophical. Assume that you’ve made the best possible estimates you can, on the basis of all the data that you have, of the probability and severity of loss. But you’re not satisfied…

… because you remember that Donald Rumsfeld quote, and you think to yourself that there are “novel risks” out there. Things like climate risk, geopolitical risk, cyber risk and such like, which aren’t in your data set (or at least, aren’t in it with sufficient weight) because they haven’t happened before (or at least, haven’t happened as often as there’s reason to think they will happen in the future).

This kind of thing is the problem that’s been on my mind forever; the extent to which things don’t follow actuarially well-behaved processes, the future doesn’t always resemble the past and the distribution of the outcomes of a random variable over time don’t have to be the same as the probability distribution at any one moment. So what do you do?

Obviously, you apply a margin of safety. But how? There are at least three possibilities:

1. Bump up your estimates of both probability and severity of loss

2. Bump up your estimate of one of the two parameters but not the other

3. Leave the parameters (which are, to reiterate, your best estimates given the data) the same, but bump up the final number

It’s on my mind because it’s a concrete example of the offhand remark I made on Wednesday about analytic philosophy potentially having a lot to contribute to accountancy. My intuition is that option 3 above is the one that does the least damage, and that the correct place to put margins of safety is right at the end of the process. But this is not the accounting standard! IFRS9 (for it is she) requires that “model overlays” of this sort have to be carried out in such a way as to affect the “staging” of a loan portfolio. (It would be tedious to explain in detail what staging is, but basically, it matters for accounting purposes whether something has experienced a “significant increase” in the probability of loss in the last year).

The European bank supervisors wrote about this last year, reminding me that my intuition is wrong, but also reminding me that even though option 3 is not allowed under the relevant standard, it is still the most common business practice for the banks they regulate. There is no real conclusion to this post; I just thought you might find it interesting.

analysing the explosions

reality, and its occasional legal implications

I’ve argued in the past (it was even occasionally an advertising slogan for this ‘stack) that management science is a branch of analytic philosophy and that it is a shame and a problem that neither philosophers nor management scientists have any interest in progressing this way. In general, philosophy ought to pay a lot more attention to accountancy (which is the major field of human activity where people spend a lot of time on conceptual definitions, the categorisation of events and the nature of control) and to tax (which, along with theology, is the major field in which large amounts of effort are put into technical compliance with a complicated body of arbitrary rules while undermining their intention).

Here's an example of the sort of thing I’m talking about – a couple of years ago in 2021, my local university lost a couple of blocks of student flats due to the discovery of an (at that time) unexploded German bomb, which had to be detonated to make it safe. The insurance claim was rejected, because the policy did not cover “acts of war”.

It went to court, with the two sides making respective cases of “well it didn’t explode in a war did it” and “well it was dropped in a war wasn’t it”. Exeter University ended up £10m poorer because the judge decided that, indeed, the controlled explosion at the building site on Glenthorne Road in 2021 was an event which had begun during the “Baedeker Raids” in 1942.

Or at least, it was partly. Looking at the judgement, the actual decision was that there were two proximate causes of the student flats being blown up – the dropping of the bomb in 1942, and its explosion in 2021. Neither cause could have destroyed the flats without the other (if the bomb had exploded in 2021 in a museum in Dresden or whatever, it wouldn’t have triggered this insurance policy). And the rule for insurance cases is that when there are two causes, one of which is excluded and one which isn’t, the exclusion prevails.

Exeter University tried to claim that because the bomb had been sat there for 80 years without exploding, the “agent of change” was the attempt to remove it, which didn’t happen during a war. The judge rejected this:

“On the facts here, I consider that the problem with this argument is that it treats the bomb buried in the ground close to the appellant’s buildings as part of the status quo. In my view, one could equally consider that the status quo was that, immediately before the bomb was dropped, this was farmland on the outskirts of Exeter. The bomb that fell in 1942 was the first change to the status quo, because it introduced a mass of high explosive into the ground where previously it had not been. The potency of that explosive did not dissipate over the years. The second change to the status quo was the construction of the appellant’s halls of residence close to where the bomb lay undiscovered. The third change to the status quo, following the subsequent discovery of the bomb, was the perceived need to attempt a [removal of the bomb]. All events might fairly be described as “agents of change”.

Which just goes to show, in my view, that when you’re asking for “explainability” in a complex system, you’re asking quite a lot. If the question “did this building blow up because of something that happened in 2021, or something that happened in 1942” requires seventeen pages of analysis by the Court of Appeal, then I’m not sure it’s reasonable to require that any output of a machine learning model should be traceable in a couple of sentences to be understood by a layperson.

you can't take it back in a disclaimer

thoughts on the state of opinion research

Over the years, when I have needed to be firm with a representative of the financial services industry (or sometimes, the legal profession, or very rarely an elected representative), I have developed a pompous and annoying little speech to introduce my objection:

“I have to tell you this [contract, disclosure, draft legislation etc] is not clear. I have two degrees in finance and economics, I have worked as a bank regulator and as a financial analyst, I have written an award winning financial newsletter for ten years and three books. I have read this carefully and I do not share your understanding of what it means, and I am not prepared to accept that this is mainly my fault”.

If you’re not planning to meet that person again, and don’t mind coming across as kind of a wanker, both of which are usually true in my case, it often works.

My point here is preparatory to an old-fashioned firendly blog debate! Between me, Andrew Gelman the political statistics expert and (by proxy) Ben Recht and Leif Weatherby, on the subject of whether Ben and Leif’s criticism of the polling industry could fairly have been described as “a Gatling Gun of truth bullets”. In summary, Andrew thinks it’s not, and might better have been described as “extreme positions, [which] other people who should know better applaud them for …which just creates incentives for further hot takes”.

I have too much of a survival instinct to get into a technical debate with Andrew Gelman about statistics! But in my view, he’s doing something familiar to me from being in the banking industry during its periodic failures and outrages; hanging everything on being technically right, the worst kind of right, while missing the big picture. Of course, the definition of a Gatling Gun is that it’s a big machine that shoots lots of bullets indiscriminately and is best used against a large and vulnerable target, so here goes…

The basic shape of the argument is that Ben and Leif wrote that the pollsters screwed up mightily by referring to the 2024 Presidential election as 50/50, too close to call, really knife edge balanced etc when it wasn’t. Andrew says this is completely unfair and gives a number of quotes to back himself up, showing that nearly all of the poll talking guys were warning that 50/50 odds doesn’t mean that it will necessarily be close, that it was entirely possible for one candidate to run through all the swing states, and so on. He also argues that a 3% error is not that bad and everyone expects too much from pollsters anyway.

And my reaction to this is the title of this post – “you can’t take it back in a disclaimer”.

There are some financial products (shared appreciation mortgages, for example, or small business interest rate swaps) which are, in my view, completely impossible to sell without creating a massive legal risk for yourself. The problem is that lots of the value in these products comes from tail probabilities of very large payouts, and it is more or less impossible, not that anyone tries very hard, to get a retail client to understand that the very large payouts in question are going to come from them. The contract might be clear, but time and again, the regulators have found that there’s a higher principle of fair dealing[1] that’s been broken, and the contract gets unwound at the expense of the bank that thought it was being really clever.

With that in mind, let’s look at some exhibits of how the election forecasts were presented. First, here’s The Economist:

I really don’t see how you could see this as communicating that a Trump landslide was a significant probability. Maybe if you stare and cross your eyes, you might notice that somewhat darker red blob on the right hand side of the line? But to me, this is quite clearly communicating the medians and (via the presentation of simulation results) the expectation. It’s not giving much idea that the distribution of outcomes was bimodal, and I don’t think it is suggesting that the actual result was within what you’d normally think of as the bounds of error.

Five-thirty-eight did it quite a bit better, I think; if you know what to look for, it’s communicating a bit of bimodality in the most prominently presented picture:

Not only that, but the site does warn that “Trump and Harris, our model says, are both a normal polling error away from an Electoral College blowout” in the text.

But I think it’s very hard to communicate how likely this is, because “a normal polling error” is a difficult concept to get across. The 538 election eve post shows maps with outcomes four per cent away from their final average, and a 54/46 race is just really really different from a 50/50 one, psychologically. (I will return to this point later, but I just can’t accept that a 3% nonsampling error is pretty good; it's bigger than one third of the popular vote margins in US Presidential elections since the second world war). Unless you’ve seen the answer and know what you’re looking for, I think you’re visually much more likely to ignore the big spikes in the histogram as outliers and see a broadly normal distribution with the (eventual ex post) true outcome quite far from the centre.

How should the bimodality and uncertainty have been communicated then? I don’t know. But this is the whole problem. I am fond of wryly saying that “if something’s not worth doing, it’s not worth doing properly”. But that joke can be turned around – if something can’t be done properly, maybe it shouldn’t be done at all.

Basically, to adapt the language of a previous post, “cleaning the draining board is part of the job of washing up”. If a forecast is going to be used in public life, part of the job of forecasting is ensuring that the forecast is communicated clearly and directly, without putting excessive cognitive burden on the readers by expecting them to read and understand extremely important caveats that are placed somewhere else. At the end of the day (and in the knowledge that I’m sure this comes across just as badly as in my little speech in the introduction), I understood the pollsters to be predicting a tight election, so did Ben and so did Leif, and I’m not prepared to accept that this is all of our faults for not understanding statistics well enough or not being interested enough in politics.

Technically correct is, in this case, the worst kind of correct. The polling industry, as it exists, is putting things into the public sphere which make it worse. As well as being problematic in their own terms, the election forecasts are, in many cases, the shop window for actual policy advice. That’s really frightening for a number of reasons. First, the communication of the polling results to policymakers is likely to be at least as difficult a problem as the election forecast presentation, probably much more so. Second, the technical problems with polling are going to be worse.

As Ben and Leif point out, response rates are very low, and the efforts to reweight nonrandom samples by subjective brute force have a very significant effect on the outcome. It’s pretty easy to see why this is extremely problematic when the polling is meant to provide objective evidence of what’s popular. Andrew’s response to this issue is to say

“There are essentially no random samples of humans, whether you’re talking about political surveys, marketing surveys, public health surveys, or anything else. So if you want to say “If you can’t randomly sample, you shouldn’t survey,” you should pretty much rule out every survey that’s ever been done.”

And … yeah. Perhaps you should. Or at least, you might regard the bias and unknown error as having grown to a level where it’s acceptable for market research, but not for important policy issues. (In fact, market research has been worrying about response rates for a while, and seems to be much less hung up on traditional survey methodology than it used to be). This is a real problem – technological and social change has made it extremely difficult to get representative samples – and it might be the case that this means that survey research is no longer a viable way to find things out. That’s really unpleasant to consider, but it might be true.

It is a very very hard thing to accept that something can’t be done. Andrew Gelman’s view is a kind of harm-reduction approach; we recognise that there’s a lot of uncertainty and inaccuracy here, but people are going to demand this so let’s try to do it in as technically defensible a way as possible. Ben and Leif are taking more of a prohibitionist approach; things have got bad enough that even the best we can do is still bad, so (in their words) “the survey industry needs a tear-down …the industry is in a crisis that can’t be fixed with better regression analysis”.

Harm reduction and prohibition are, of course, tactics rather than principles; you choose the approach that you think will work best. I think I come down on the side of prohibition. (Not legal prohibition, you daft hap’worth!). The policy community has to wean itself off the use of surveys, and start using harder evidence about what is worth doing. As Ben and Leif nearly put it, the Era of the Data Guy has come to an end.

---

[1] In “Lying for Money”, in the context of “market crimes”, I talk about the unusual areas of the law where things are based not on common principles of justice and contract, but on the question of “what is the best rule if we want to maximise trust in this valuable institution?”

songs my portfolio manager taught me (part 6)

another miscellany

This will be a short one, reversing the usual pattern (or at least, the pattern will be reversed if I manage to complete a longer post on Friday, otherwise it’s just a short one). Previous episodes in this series: 5, 4, 3, 2, 1 and a related series – it’s where I reminisce about things people said to me when I was an equity analyst, which have stuck in my mind and which might have more general applicability to financial markets. Either because they were very clever indeed, or because they were dumb in a way that illuminates a particular problem.

“I want at least six high conviction ideas every quarter”. This was very much in the second category, and I will spare the various Heads of Research I used to work for their blushes – guys I am not singling anyone out, you all did it.

The problem is not quite that it’s difficult to generate ideas to order; there are various sorts of structured approaches to creativity and thinking which we were often given training in, and it would have been perfectly possible for a reasonable analyst to come up with a brand new trade idea every single day. The problem is that having high conviction in an idea isn’t something you can force. The stock market isn’t perfectly efficient but it’s often pretty much there or thereabouts; I would doubt whether, in any given quarter, there even exist as many as half a dozen significant mispricings, even for a stock picker who was good enough to spot them all.

So what does this achieve? When this sort of order went round, usually after complaints from the sales desk of not having enough research to sell, everyone did the only thing they could do; they took all the existing crap they had, and lied about their level of conviction. This is actually one of the most destructive things you can possibly do in equity investing; having excessive confidence in a mediocre idea can get you into much worse trouble than simply having a bad idea that you’re aware is not your greatest. And of course, flooding the zone with bogus “high conviction” stuff tends to devalue the brand and make it harder to communication when you actually do have a good idea.

A lot of the bad reputation of sell-side analysts, in my view, comes from this management practice, and its close cousin, “get off the fence, you’ve got too many Hold recommendations”. In my view (which I would have put into practice if anyone had ever been insane enough to promote me to a management role), every research team should have Hold as about 80% of their recommendations; it’s just not credible to claim to have a really strong opinion one way or the other about more than one stock in five in any given sector.

“The poorhouse is full of guys who knew more than anyone on the Street”. A related concept, phrased in this way by a salesman who was watching in horrified frustration as one of his clients rode a bad investment all the way into the ground. The trouble is that you can know a huge amount of detail while missing the big picture. And knowing a lot of detail has a very nasty way of increasing your confidence that you know it all; again, confidence out of proportion with quality is an incredibly dangerous place to be. I think this might be the reason that the “alt-data” and “quantamental” trends seem to have come and gone in hedge fund investment. There is something about a quirky and expensive dataset, like satellite pictures of shopping mall parking lots, which puts the idea in your mind that you know better than the market, and ironically, while informing you much better about some things, makes it much more difficult for you to learn anything important.

“There’s no such thing as the economy of Italy”. A favourite economist pal always used to say this, and it’s of much more general application. In economic statistics, “the GDP of Italy” refers to the sum of North and South, and consequently almost every calculation you might do using that number gives you an answer which refers to a statistical average concept that describes practically no actual place in Italy.

“Savings products are sold, not bought”. This hoary old industry cliché actually has a lot of depth to it, in my opinion, and I might need to do a much longer post (a companion to my pensions piece) about the policy failures that have resulted in the UK from a refusal to understand it. But this “short” piece is already too long, so I’ll just repeat a second hand anecdote about the sales strategy of one of the most successful financial advisers I ever met.

“Fear. That’s what it’s about. I can walk into an office with a master of the universe, a trader who’s a multimillionaire at the age of 26. In the space of half an hour I can have him trembling, buying products to protect himself against fears he didn’t even know he had. The 1970s, mate. The effect of the 1970s and 1980s on the British middle class. That’s what I sell.”