An eye on AI

The largest AI models should be monitored by a third-party auditor - who would basically check an AI system to ascertain its capabilities and the risk it poses. Both OpenAI and Anthropic - two of the AI labs with the most advanced systems - commissioned a third-party auditor called ARC Evals to act as a ‘third-party evaluator to assess potentially dangerous capabilities of today’s state-of-the-art ML models.’

A safety evaluation of an AI system, known among AI labs as an ‘eval’, checks an AI system’s capabilities to ensure that pre-deployment they are developed and deployed responsibly and with human interests in mind. When ARC Evals stress-tested OpenAI’s pre-aligned GPT-4 it did so in a controlled environment and in essence tried to make the model misbehave.

They managed to make GPT-4 lie to a human and get that same human to perform a task for them on TaskRabbit, make long term strategic plans, and write and run code: ‘As AI systems improve [...] It is important to have systematic, controlled testing of these capabilities in place before models pose an imminent risk, so that labs can have advance warning when they’re getting close and know to stop scaling up models further until they have robust safety and security guarantees.’

ARC Evals is particularly worried that future and more advanced systems might exploit financial arbitrage, create new pathogens, and impersonate online humans. 

With this in mind, British-based Deepmind got together a very stellar cast of AI researchers including Turing Award winners, to hash out *exactly* how one monitors the risks from the increasingly advanced and potentially more dangerous AI models. They find that: ‘Current approaches to building general-purpose AI systems tend to produce systems with both beneficial and harmful capabilities [and that] Further progress in AI development could lead to capabilities that pose extreme risks, such as offensive cyber capabilities or strong manipulation skills.’

And because of this, they go on to explain ‘why model evaluation is critical for addressing extreme risks [...]. These evaluations will become critical for keeping policymakers and other stakeholders informed, and for making responsible decisions about model training, deployment, and security.’

Model ‘evals’ to uncover the risk of extreme risks of catastrophe and existential risk ‘should be a priority area for AI safety and governance.’ Major labs such as Google Deepmind, OpenAI and Anthropic have perhaps the biggest responsibility in the whole AI ecosystem, as they are the ones developing the model - which can be used for great good or for great (even unintentional) ill. Perhaps an International Atomic Agency for AI would be a suitable house for such a system of monitoring.

The Adam Smith Institute’s paper will be released next month and I cannot wait to share with you the fantastic innovation-led policies for the safe deployment of AI that we have cooked up.

Don't use planning, use prices

A certain chortle here as French Greenpeace gets very upset:

Greenpeace France said the halted flights represented only three among 100 domestic routes, which was “extremely insufficient”. However, with other campaign groups it called France’s action important as it set a precedent that could be expanded across Europe. France plans to add more routes to the ban when it reviews its list in three years.

This is about that French plan to ban flights between places that can be reached by train. To which one response is well, having spent all that money on high speed trains they’ve got to do something to get passenger numbers up, don’t they?

What’s annoying Greenpeace (and, joy, it’s the French division of) is that this carefully crafted political plan only stopped those three flight routes - which Air France had actually stopped running already as people voluntarily took to the trains instead. So the actual effect here was, umm, nothing.

Now, we’re not sure that flights need banning, curtailing or even diminishing. We’re even less certain that domestic flying is the abomination some think it is. But there is still some advice possible.

Think, for a moment, about Air Passenger Duty. This was just, last month, halved for UK domestic flights. On the grounds that the rate was discouraging too many people from flying. That is, prices work in curbing domestic flights. Planning doesn’t, as in the French example. So, therefore use prices not planning.

As we say, we’re deeply uncertain (in fact, to the point of not caring) whether domestic flights should be curbed. Just as there are many other things demanded by all sorts of people that we’re either not sure or uncaring about. But the advice we want to pass on is that if you do actually want to change something then use prices, not regulation and planning. Because prices work and regulation and planning don’t work.

Of course, this advice is useless in this specific example because it’s difficult to get a Frenchman to believe that the Anglo-Saxonism of markets will cure anything and absolutely impossible to get anyone from Greenpeace to believe prices are important. But for those few rational out there - use prices, they work.

It's not that we told you so but we did tell you so

We did suggest that this would happen:

Post-Brexit curbs on immigration are proving a boon for restaurant workers as wage increases outstrip the national average.

Pay for staff in the hospitality sector has risen by 9.5pc – compared with the national average of 6.6pc – over the last year, new research shows.

Actually, we took to the pages of the national press to insist that this would happen:

Brexit is about to give us a problem with this, though. Karl Marx was right: wages won’t rise when there’s spare labour available, his “reserve army” of the unemployed. The capitalist doesn’t have to increase pay to gain more workers if there’s a squad of the starving eager to labour for a crust. But if there are no unemployed, labour must be tempted away from other employers, and one’s own workers have to be pampered so they do not leave. When capitalists compete for the labour they profit from, wages rise.

Britain’s reserve army of workers now resides in Wroclaw, Vilnius, Brno, the cities of eastern Europe. The Polish plumbers of lore did flood in and when the work dried up they ebbed away again. The net effect of Brexit will be that British wages rise as the labour force shrinks and employers have to compete for the sweat of hand and brow.

It is, of course, entirely possible to argue that you’d prefer the cost of labour to not rise. But that is the same thing as arguing that you want wages to remain low. A higher cost of labour, rising wages, higher incomes, they’re all exactly the same thing.

We're really, really, not about to run out of elements

A startling piece of misinformation is circulating again. The thing that makes us sad about this - no, really sad - is that it comes from the American Chemical Society. You know, one of those official bodies that we’d hope would actually be informative. Even, possibly, attempt to be factually correct. You can see their warning about the elements we’re going to have significant supply problems with here. We’ve covered this subject at book length here (it’s free!).

But to cut to the short version. The ACS says that there’s a serious threat of running out of gallium, germanium and hafnium in the next 100 years. All of which is a heck of a surprise to any geologist or anyone actually in the metals business. So it’s difficult to grasp why the chemists are in such a tizzy about it.

Gallium is extracted from a Bayer Process plant. That’s the stage of the aluminium business that turns bauxite, the ore (and possibly the most common component of the Earth’s surface) into alumina, the oxide. The gallium in the bauxite goes into solution and with the right little doohickey it can be extracted. Which is what is done. We have - already mapped out, ready to roll - at least a 1,000 year supply of bauxite. We’re not going to run out of gallium.

Germanium has two sources, a byproduct from the mining of zinc from spharelite and another process which extracts from fly ash. That’s the waste left over from the burning of coal. There are hundreds of millions of tonnes of fly ash lying around the countryside in vast ponds. We’re simply not going to run out of germanium.

Zircon is a common enough mineral, the world uses perhaps 600,000 tonnes a year, there are millennia of it out there at least. All zircon is 2 to 4% hafnium. We usually don’t bother to extract it as for near all uses the two are so similar that we don’t care. Sometimes we do care and so we extract the Hf from the Zr to use them separately. We extract perhaps 500 tonnes a year of the 20,000 tonnes of hafnium already incorporated in the zircon/zirconium moving through the system. We’re not going to run out of hafnium.

All of which is bad enough, the official sources being so horribly out of whack with reality (and as an aside, one of the reasons that state planning works so badly, it so often starts from such lack of knowledge as this). But Ga, Ge, Hf, they’re not exactly at the top of the worries list for most people.

But think on it, if they’re this wrong about these simple things then what else is wrong in all of the other things they’re saying to us?

Peter Thiel is wrong about technological advance here

It’s a common complaint but an incorrect one:

For example, Thiel points out, air travel has hardly improved in decades. In fact, with the demise of Concorde, we now have less access to high-speed air travel than we did 30 years ago.

That is to assume that the only possible improvement to air travel is greater speed. As if the only possible improvement from the agricultural revolution were more turnips. Or the computing revolution meant faster calculators.

Items - goods, services - can improve by many metrics. Along different spectrums if we like. Which they often do in fact. They get to “good enough” along one direction, toward one end of a spectrum, then stagnate in that direction. Cars, say. The first few decades increased speed - safe speed that is - as something that was developed as being able to cruise at 70 mph, or 80, was a great improvement over doing so at 20. But the improvement from 80 to 130 - say - was of must less value. Partly because of the reaction times of the wetware controlling it but also because most to many of the things we might want to use a car for can be done at 70 or 80 and 130 doesn’t add much value. Cars are much safer than they used to be, much more reliable, vastly cheaper for the level of performance than they used to be. All of which are technological improvements. Anyone claiming that car technology hasn’t improved in the last 50 years would be laughed at as a ludicrously ill-informed rube.

Airplanes have hugely advanced in that same past 50 years. Not in speed, true - 520 mph seems to be good enough for many uses - but ‘planes are very much safer than they used to be. Less noisy too. Also, the use of a ‘plane is hugely, vastly, cheaper. Roughly, you understand, just roughly, a trip from Britain to Europe, 50 years ago, cost about a week’s wages. Today it can usually be done for a day’s. And that’s not because wages have risen that much.

To claim that’s not technological advance is, well, rube-ish.

Of course, this becomes important when we consider train travel. Sure, the average speed hasn’t increased in 50 years. This does not mean we require high speed trains. We’ve developed the internet that works on trains now, so time spent on one is not time that is subtracted from working hours. That’s what kills the HS2 cost benefit analysis as it was originally done. And really, anyone want to claim that adding the internet to trains is not a technological advance? As, you know, has also recently been done to ‘planes?

But, but, we must have common standards to have free trade!

A common enough insistence these days. We must all accord to EU standards in order to be able to have free trade with the EU. Or, vice versa, in order to have EU goods come here we must make sure that our standards are the same as their.

What amuses is that this insistence is being applied to the car industry. Yet cars used in the UK are right hand drive, those on the Continent left hand drive. European factories do make RHD cars for us to drive, British do make LHD for them.

So the insistence becomes that we must have common standards for an industry which actually builds to entirely different designs in the first place. In order to enable the trade of those things built to those entirely different designs we must all agree upon the one single design that is.

Someone isn’t thinking logically here and we’re absolutely certain that it’s not us in error.

That other people build to different standards is something that exporters deal with routinely. As the car industry proves. Therefore we do not require commonality of standards in order to be able to have trade. QED.

Just think on this wealth tax idea

One of the more absurd ideas floating around is that we must tax wealth more. Some are even suggesting that a rise in an asset value should pay income tax, even if that asset is not sold and the value not crystallised. There really has been a bill put together in the US Congress along those lines - although it’s still a bit too silly even for them to get beyond the thinking about stage.

One reason for the silliness is that of course refunds would have to be paid if asset values fall:

Rishi Sunak’s rise to the top job in British politics has coincided with a colossal fall in the fortune shared by him and his wife, Akshata Murty.

This year’s Sunday Times Rich List, published today, reveals that the couple’s wealth has fallen by more than half a million pounds a day over the past 12 months. Their fortune is now estimated at £529 million.

The couple’s most valuable asset has long been a shareholding in Infosys, an Indian IT giant set up by Murty’s father. When the pair made their debut on the Rich List in 2022 this stake had a value of £690 million.

We’d love to see someone trying to defend a substantial tax refund to someone now only - only - worth a half billion.

But beyond the silliness there’s also another problem - such a tax wouldn't raise much money. The overall level of wealth doesn’t change all that much. Who has it is what changes, certain fortunes rise, others fall. Which means that a very large portion of the receipts will have to be paid out again in refunds. The net, well, the net might well not even be positive for large periods of time.

Adding a tax which gains negative revenue some to much of the time really just doesn’t sound all that sensible.

Fertility rates and the conflict with the liberal vision

More worrying about the fertility rate:

Politics works on the basis that society will go on and on into the future. The intense focus on climate change – for good or ill – takes for granted that there will be many future generations whose welfare is threatened by damage to the environment. But can we really be so sure that there will be future generations to benefit from our endeavours – or even to look after us in our old age?

The short answer is no: the evidence suggests that the very existence of future society hangs in the balance. The current UK fertility rate – the average number of children per woman – stands at 1.6. This is significantly below the “replacement rate” of 2.1, and continues to fall.

The correct answer there is well, bully for politics then. Because the liberal vision - by which we mean the real one, not the modern - is to maximise choice, freedom and liberty to the individual. Society is then the aggregation of those individual choices. Yes, of course there are limits, things like third party harm and so on. But leave people be and see what happens - that is the design for the liberal society.

This past century has included glorious events - the economic liberation of women for one. The result of that freedom and liberty is fewer children. Oh well, that’s just what humans want to do with their freedom and liberty.

It’s therefore the politics that needs to change, nothing else. For the people have spoken in their most intimate acts and decisions.

It might well be true that some don’t like that aggregate result, the society that results from freedom. But bully for the complaint, not the acts.

We just can't see the terror here

The general political thought these days is that this is some grand terror:

China owns most of the cobalt mines in Congo, which has the majority of the world’s supply of this scarce material needed for the most common type of battery. American companies failed to keep up and even sold mines to their Chinese counterparts.

As a result, China controls 41 percent of the world’s cobalt mining,

By golly this is a horror and gosh darn it something must be done. A reaction that we really don’t understand ourselves. China is doing a large amount of the economic scut work and we’re all worried about that? They work, we get batteries and electric cars and this is a problem?

It’s not as if China is shafting us on the price of cobalt either. Just 6 weeks back an American cobalt mine was just ready to start operations. They didn’t - the cobalt price is currently below production costs. And yes, given the way mining works it is possible for prices to be below production costs for years, possibly even decades. So not only is China doing all that work they’re losing money on it and we’re gaining.

We really can’t see this as a major problem.

Of course, there is that worry that perhaps, once a monopoly position is gained then China will try to rook us. But we’ve seen that too. Back in 2010 one of us pointed out that China’s attempts to restrict rare earth exports weren’t going to work. We’d just open new mines if they tried. Fast forward to 2014 and the problem was over - new mines had opened and prices were below their starting point.

Addendum: Bonus points to Tim Worstall, economist blogger and rare earth dealer, who in 2010 at the height of the crisis pointed out that rare earths were neither rare nor earths and China’s monopoly had been won only by low prices that accrued to our benefit. “If Beijing wants to raise its prices and start using supplies as geopolitical bargaining chips,” he wrote, “so what? The rest of the world will simply roll up its sleeves and ramp up production, and the monopoly will be broken.” Nailed it.

If they try - if - then we know how to react. Our only real necessary action if they do is to make sure the Nimbys (and the BANANAs) don’t prevent planning applications. After that we can just stand back and watch markets work. The cost will be the same as trying to boost our own production right now but we’ll only incur it if we need to.

Behind this is a gross misunderstanding. The world’s really a very large place. There are plenty of places we can dig holes and gain cobalt - or lithium or rare earths, or whatever else people want to worry about. The trick is to only go mining when we need to. If China’s willing to do all that work - often enough at a loss - then leave ‘em be. There’s absolutely no point at all in insisting that we too must make a loss right now.

Suggestions from people who simply do not understand

There is this thing in economics called “economies of scale”. If you do things on a larger scale then it might be possible to have lower unit costs for each of the things done. It’s not a certainty, thus the “might” there. But retailing is one of those things where we do see it.

Cheap stuff is sold in high volumes from sheds on the edge of town. Small quantities in the centre of town are more expensive. That’s just the way the economics work out.

At which point we get this:

Sue Davies, the head of food policy at Which?, urged Rishi Sunak to ask grocery bosses gathering at No 10 on Tuesday to commit to doing more to hold prices down, “including stocking budget lines in convenience stores to ensure easy access to basic, affordable food ranges that support a healthy diet”.

Convenience stores are those small places in the centre of town. Where the volume of sales is small and upon a higher overhead and cost base. Therefore retailers do not sell budget items in them because the margins aren’t enough to contribute sufficiently to the costs of having the convenience store.

This is not something to be overcome by asking eveyone to be nice.

But apparently we’re to solve the problem - if we’ve even got a problem, for we might usefully suggest that those who are budget constrained go to the sheds - by so asking. Which isn’t going to work when facing the basic economics here, is it?