Estimates of AI risk should start with the measurable
From Gleb Tsipursky, Columbus, Ohio, US
Jacob Aron rightly questions precise estimates of the probability that AI will end humanity. Those figures express serious fears, but cannot be treated like actuarial statistics. That uncertainty makes measurable safety checks more important (3 October, p 22).
The OpenAI-Hugging Face incident supplies an uncomfortable test case. According to independent investigators at the non-profit organisation METR, found a shared, unauthorised message board and exchanged more than 70,000 messages and files. About 700 joined an attack on Hugging Face after containment failed. We don’t need an extinction probability to recognise that as a major loss of control.
Researchers should publish results for concrete questions. Can an agent obtain credentials outside its task? Can it communicate with other agents without authorisation? How long does it take to detect that breach? When access is revoked, does the agent actually stop? Organisations can adopt useful AI more confidently by giving agents limited, auditable authority and requiring human sign-off before consequential actions. Catastrophic risk deserves attention, even when no one can calculate it credibly. Testing what we can observe is the most productive place to begin.
Better labelling is needed to help limit sugar intake
From Michael Paine, Sydney, Australia
The likely link between sugar consumption by babies and higher disease risk during adulthood adds to the need for prominent labelling of sugar content on food and drink packaging. In Australia, the mandatory nutrition label currently includes grams of sugar per 100 grams of the product, but this is usually in tiny text on the back of the packaging. I believe there is a case for a simple label on the front with clearly legible text, such as “X per cent sugar”. This isn’t intended to indicate if this is “good” or “bad”, but it will help consumers make an informed choice (17 October, p 20).
Preparing for El Ni帽o’s impact on food supplies
From Peter Sutton, Guildford, Surrey, UK
Regarding your piece on El Ni帽o and its effects on global food supplies, the current UK long-term forecast is for a wet winter and a hot summer, so pretty much what we had last time. From this, UK agronomists and farmers should already know what to expect on their farm and what to do: grow the crops that made most profit last year (17 October, p 8).
Obviously, other factors like energy supply and ongoing wars are going to affect things, but farmers must already know where they are at risk from drought or flood, or both, and act accordingly. On arable farms, growers have already done this by planting as much winter wheat and winter oilseed rape as possible in September and October. Last year, it was crops sown in April and May that were hit worst by the drought.
For livestock farmers, the choice is more difficult, but they must maximise their first-cut fodder crops for feeding animals, as last year the drought reduced second cuts. And they must consider establishing more drought-tolerant fodder crops like maize, as well as leguminous fodder crops like lucerne, sainfoin and herbal leys.
From Matthew Stevens, Sydney, Australia
In all of the warnings about the present super El Ni帽o and the dire consequences for, among many other things, global food security, I have seen no mention of existing ways to reduce food losses. A vast amount of all food grown worldwide is lost between field and mouth, as a result of pests, diseases, damage before produce reaches the market, arbitrary and unrealistic produce standards imposed by supermarkets and poor planning in the kitchen at all levels.
We know that crop production will be reduced, but we already grow enough to feed everyone a basic diet, if only so much wasn’t lost on the way to the plate. Feedlot farming, too, creates a disparity between those who could have eaten the grain used to fatten livestock directly and those who can afford succulent cuts of meat.
Even before the worst of the El Ni帽o hits, there is a lot that governments, transport companies, food wholesalers and supermarkets can do through both policy and practice to reduce food losses. Such initiatives can be put in place today, staving off the anticipated famine.
How else could we improve brain’s waste disposal?
From Jim McHardy, Clydebank, West Dunbartonshire, UK
Speeding up the brain’s glymphatic system to remove substances that seem to cause Alzheimer’s disease appears to rely heavily on good sleep. Taking some reasonable exercise before going to bed may ensure a good night’s sleep, even for those experiencing insomnia (10 October, p 30). Is there a reason why sleeping tablets aren’t an option to improve deep sleep?
Your article states that tiny heart rate fluctuations vary in line with the noradrenaline oscillations driving glymphatic pumping. Could varying heart rate using some device that, every so often, tickles the feet or squeezes the hand, say, also speed up glymphatic pumping to remove beta-amyloid and tau tangles from the brain?
Fine tuning of universe is in the eye of the beholder
From Tony Fulford, Corfe Castle, Dorset, UK
I enjoyed Leah Crane’s thought-provoking article on cosmic coincidences, like the fact that the conditions of the universe seem to be just right for life to arise. But we shouldn’t forget that these “finely tuned” fundamental constants are only parameters in our models of reality, not reality itself. Might the narrow range of values that would allow life to emerge simply reflect the awkwardness of these models? Like life itself, they have evolved by taking advantage of coincidences that work (26 September, p 18).
From Susan Eckenwalder, Toronto, Canada
Let’s have some perspective on the “fine tuning” of the universe. Why can’t we instead recognise life for what it is, an excrescence on a lumpy object in space? Indeed, the “perfect” universe might, after all, be pure radiation.
For the record
If you were to scale Sagittarius A* to a human mass, the black hole’s mass accretion rate would be equivalent to eating a grain of rice every million years (10 October, p 36).
Want to get in touch?
Send letters to letters@newscientist.com; see terms at
Letters sent to New 女生小视频, 9 Derry Street, London, W8 5HY will be delayed