Editor's pick: Philosophy can describe science, not define
Trevor Hussey warns scientists that methods change, which is undeniable (Letters, 14 April). But he suggests we need to discuss with philosophers not only methods, but also our objectives. Isn't this putting the cart before the horse? The philosopher David Hume understood philosophy as . This can describe science, but not prescribe its goals.
In my experience, most scientists don't have grand philosophical objectives: we are driven simply to Explain Things. In this respect we are no different from Aristotle, though in 2500 years we have become better at recognising a good explanation.
Moral philosophy, on the other hand, is at heart an inquiry into how we should live (and, by inference, think). Tellingly, its “answers” have varied over time: Thomas Hobbes that the sovereign's judgement was absolute.
Philosophy may be helpful in understanding a society's scientific priorities, but I would be amazed if any scientist ever defined their motivation by any philosophical “should”.
What work might robots take away from humans? (1)
Sally Adee reports that robots aren’t coming for our jobs and indeed they may be creating more work (14 April, p 9). So what is the point in having them?
Ask anyone who did housework in the 1950s about vacuum cleaners and washing machines. They gave people leisure time. They removed drudgery. They improved the quality of life. That’s what robots should be doing for us now, not sending us into tedious office jobs. What we need is a new way to distribute wealth.
What work might robots take away from humans? (2)
Adee argues that automation anxiety is overblown, but conflates current automation with the future potential for automation. She cites my research with Michael Osborne, which estimates that roughly 47 per cent of US jobs are exposed to automation technologies 鈥� but argues that this is overstated because recent surveys of the workplace in Germany suggest that not very many jobs have been automated away so far.
The tractor had virtually no impact on jobs before 1920: should this have been taken as evidence that it was an unimportant technology? It took almost half a century for its full effects to materialise.
First class post – 5 May 2018
“… says the builder who accidentally hit the solar panel with a hammer” Heather Tweed to the news that poking tiny dents into solar panels makes them work better (28 April, p 9)
How sex differences may lead to extinctions
You report researchers finding that the species of ostracod crustaceans with the largest differences between the sexes were also the most vulnerable to extinction (21 April, p 9). They speculate that the cause may be related to devoting resources to sexual display at the cost of other survival functions, much like the peacock's burdensome tail.
This only works to explain extinctions if the burden affects individuals before they are able to reproduce; and that would inevitably select out both the male characteristic and the female preference for it.
A better explanation is that selection for extreme characteristics can easily lead to a genetic bottleneck based on a very narrow definition of “fitness”, with the population rapidly becoming dominated by the descendants of fewer and fewer individuals with the most extreme features. When the environment changes, species survive and evolve by having a diverse enough genetic pool that some individuals can endure, even if they were not the “fittest” under the previous conditions. Consequently, it seems plausible that selection based on unhelpful secondary characteristics can create a bottleneck and leave an entire species vulnerable to changes in their environment, explaining the extinction patterns seen by the researchers.
More challenges of screening for cancer (1)
H. Gilbert Welch makes a very good point about the danger that screening will find cancers that aren’t going to do us any harm (7 April, p 44). However, as a doctor who has investigated many people for cancer, and who has been investigated for cancer myself, I think the real challenge for medicine is to determine which tumours will progress dangerously, and which will not.
More challenges of screening for cancer (2)
I accept that screening for prostate-specific antigen (PSA) doesn’t appear to improve prostate cancer survival rates. But I see a paradox. PSA screening followed by checks for symptoms can lead to unnecessary invasive procedures. But when symptoms appear they are generally followed by PSA testing 鈥� and the same invasive procedures. Why does it matter what sequence the evidence arrives in?
The editor writes:
鈥� It’s all about the effect of screening large numbers. Testing for PSA levels in men who have no symptoms will lead to many more false positives than testing it only in those who do, so it will subject more to unnecessary treatment.
In whom do we put our trust for harm reduction?
I quite agree with the boycott signatories demanding that autonomous weapons lacking meaningful human control shouldn't be developed (14 April, p 24). But the overwhelming issue with guns is leaving them unmonitored in the hands of humans who can't be trusted to behave reasonably and rationally.
History is littered with war and other crimes, not to mention the epidemic of attacks on schools in the US and elsewhere. So I suggest that AI be built into all guns to actively monitor their human users. The weapons should be able to fire only when both the human user and the AI agree that it is the right and proper thing to do so.
Not all casualties in war are soldiers
You report Aaron Clauset classifying the size of a war by the number of soldiers killed (3 March, p 15). Civilians aren't included. This criterion leads to the bizarre consequence that a war (nuclear or otherwise), fought by artificial intelligence and killing few soldiers but millions of civilians, would be unworthy of recognition as a “large war”.
Artificial intelligence for better or for worse-ish (1)
Timothy Revell suggests that watchdogs could go through the code of an artificial intelligence “line by line” to understand the decisions it makes (14 April, p 40). But it isn’t the code that matters: it is the data set on which the code is trained. A classic example is the automatic tap which, trained on white hands, didn’t turn on for brown hands.
The idea of testing systems for bias by adjusting inputs and seeing if outputs change works only for factors you choose to test. If you don’t realise that short people literally have a different worldview, you will never see a bias against them.
Artificial intelligence for better or for worse-ish (2)
Timothy Revell’s discussion of the need for transparency in computer systems that decide whether or not you get a loan, or how long you will spend in jail, was most welcome. But can we stop calling them “algorithms”?
An algorithm, such as the rules for long multiplication, is precise. All the systems that the article discusses are like the nearest-neighbour method for solving the travelling salesman problem.
This is a “heuristic”: it almost never gives the best answer; often gives a good answer; sometimes gives a bad answer; and sometimes completely fails.
The current AI explosion is entirely heuristic.
The editor writes:
鈥� This distinction is no longer widely made by practitioners of the craft, and we have to follow language as it is used, even when it loses precision.
Batteries can help keep the lights on
Garry Trethewey seems to think that renewable energy sources weren't behind South Australia's state-wide blackout in 2016 (Letters, 7 April). As the sets out, there were two issues.
Why did 450 megawatts of generating capacity drop out? This was connected with over-sensitive safety settings on wind farms. And why did it lead to a state-wide blackout?
There were three similarly sized capacity losses before 2016, and none led to a state-wide blackout. That's because the inertia of large spinning generators kept the electricity supply frequency up for the time it took to “shed load” – to cut off large customers, for example – and keep the system running. In 2016, the low inertia of renewable sources couldn't do this.
The 100-megawatt Hornsdale battery will replace some of this inertia effect. So could the interconnector that Trethewey mentions the new state government promising. Though the interconnector may have more to do with wanting cash from exporting the excess wind power that is currently wasted, a vastly expanded transmission network is a necessary part of any energy system with high levels of wind and solar generation.
An earlier case of making tools from stools
Making tools from stools is neat (14 April, p 15). But it's not new.
In his , the Danish explorer Peter Freuchen writes about the time he took shelter from a blizzard under his sledge in Greenland and realised the snow that had drifted over him had frozen, entombing him. He couldn't dig himself out as his gloved hands couldn't get a purchase and taking his gloves off to dig would mean losing his hands to frostbite.
His solution? He passed a stool and while it was still soft shaped it into a chisel, which he allowed to freeze and harden. He then dug away the snow and freed himself.