My team at Sentinel produces a weekly brief on global risks. Here is the executive summary for this week:

• Geopolitical risks: Thailand and Cambodia exchanged fire, and Thailand sent some F-16s into Cambodia, but a ceasefire has already been reached. Israel is being accused of causing mass starvation in Gaza by the WHO and others. Zelensky faced his first large-scale domestic protests over a law weakening the independence of anti-corruption bodies.
• Economy and trade: The US and the EU reached a trade deal that would tariff EU goods coming into America at 15%. A similar agreement was reached between the US and Japan, which will now also face 15% tariffs on its goods exported to the US.
• Biorisks: A paper reports a method for genetically editing mosquitoes to make them resistant to malaria in a way that can be transmitted to subsequent generations, and a method for driving such mutations into mosquito populations. H5N1 continues to pose an ongoing risk.
• AI: The White House published an extensive AI Action Plan, which seeks to accelerate investment and buildout and espouses a framework of a race against China, while also recognizing some potential dangers from AI and including some recommendations for research on controllability and safety. Anthropic is embroiled in a new trial over pirating IP; forecasters assign a 68% chance that the case will reach a jury trial, and a 56% chance that $1.5B damages will be awarded conditional on losing a trial, but only a 1.1% chance that Anthropic will be out of business by the end of 2026 as a result of the trial.
• Gray swans: None detected this week.

Our status is at green, representing that we aren’t seeing signals of incoming catastrophic risks over the short-term.

You can read the rest of it & sign up here.

Jamie Joyce seems like a better than replacement candiate for the 12th California Congressional district

Jamie Joyce seems like a better than replacement candidate for the 12th California Congressional district, which includes Oakland and Berkeley. She came second in the primary, meaning that she will be in the ballot in November. Here is her Ballotpedia page.

I know her from the FLF fellowship last year. She is not a candidate which embodies rationalist mores, but is rather more bubbly and normal. However, worries about AI are at or near the top of her policy agenda (framed as a proposed act of Congress named the “MAD act”), and she is generally able to integrate worries about AI safety with other mainstream issues. That is, she is not a rationalist candidate, but is one cadidate which rationalists could form part of a coalition to elect.

Her opponent is a very mainstream & uninspiring Democratic politician.

Unlike Alex Bores’ campaign, which is far away in NYC, California’s 12th Congressional district is right in the seat of the rationalists’ power. This means that acts like canvassing, putting up posters, etc. would be much cheaper to do. Shooting from the hip, maybe I’d give her a 10% chance of success on her own, and a 20% chance if the rats meaningfully back her (e.g., she gets 30% as much effort thrown into her campaign as the Alex Bores has). But I haven’t thought deeply about these probabilities.

Doing some quick math, if there are 2k to 10k rationalists in Berkeley, and they each can convince 2 to 10 friends, that’s 10k to 70k votes. This would be a big chunk of the overall population of 750K inhabitants, and a pretty big chunk of the 150K-190K votes one would have needed to win the 2024 election for that same district. So again, rationalists alone would not be able to win the election, but could as part of a coalition.

Per the Ballotpedia page, Jamie raised $14K so far and spent 10k on the primary, winning 28K votes (!!), meaning getting a vote for 36cts, which seems fairly efficient. In contrast, the incumbent spent $861K for 153K votes, or $5.6 per vote. I.e., the incumbent probably overspent on the primary, in a show of strength but also of inefficiency. The incumbent has $687K on hand.

If one needed to get 150K-190K votes to win past years, and Jamie could keep up the 36cts per vote, she’d need 40-60K to win the general election. This seems too low. On the other hand, $350K probably does moves her chances of winning by 10%, which would be 1.75M for a representative, which still seems like a very attractive proposition. My general sense is that another $100k would probably be fairly helpful. For contrast, Bores has raised $3.6M; this makes me think that Jamie is competitive for the marginal $300K from an EA/rat/safety perspective.

I’ll send this to a few people in the EA/rat politics scene (e.g., Eric Neyman), but in general my sense is that people excited about Bores should also view this as an interesting opportunity. Jamie can be contacted rep@jamiejoyce.com, or on her LessWrong account, and is probably up for having conversations with people here.

While researching past technological discontinuities, I came across some interesting anecdotes. Some follow. I also looked at technology readiness levels, but this didn’t prove fruitful.

Anecdotes and patterns

Watt's safety concerns

As the 18th century advanced, the call was for higher pressures; this was strongly resisted by Watt who used the monopoly his patent gave him to prevent others from building high-pressure engines and using them in vehicles. He mistrusted the boiler technology of the day, the way they were constructed and the strength of the materials used. ...Oliver Evans in his turn was in favour of "strong steam" which he applied to boat engines and to stationary uses. He was a pioneer of cylindrical boilers; however Evans' boilers did suffer several serious boiler explosions, which tended to lend weight to Watt's qualms. Source

This narrative has been later disputed, however, it does seem possible that awareness of the high risk of high-pressure engines, which were more susceptible to boiler explosions, delayed their development and adoption somewhat. In any case, the anecdote might be interesting for those who may seek to delay development of the capabilities of AI until safety technologies have been developed, either as a historical example or as a talking point.

Monetary bounties to incentivize progress

In April 1900, Henri offered the Deutsch de la Meurthe prize, also simply known as the Deutsch prize, of 100,000 francs to the first machine capable of flying a round trip from the Parc Saint Cloud to the Eiffel Tower in Paris and back in less than 30 minutes. The winner of the prize needed to maintain an average ground speed of at least 22 km/h (14 mph) to cover the round trip distance of 11 km (6.8 mi) in the allotted time. The prize was to be available from May 1, 1900, to October 1, 1903.[7]

To win the prize, Alberto Santos-Dumont decided to build the Santos-Dumont No. 5, a larger airship than his earlier craft. On August 8, 1901, during one of his attempts, the dirigible began to lose hydrogen gas. It started to descend and was unable to clear the roof of the Trocadero Hotel. Santos-Dumont was left hanging in a basket from the side of the hotel. With the help of the Paris fire brigade, he climbed to the roof without injury.[8]

On October 19, 1901, after several attempts and trials, Santos-Dumont launched his Number 6 airship at 2:30 pm. After only nine minutes of flight, Santos-Dumont had rounded the Eiffel Tower, but then suffered an engine failure. To restart the engine, he had to climb back over the gondola rail without a safety harness. The attempt was successful, and he crossed the finish line in 29 minutes 30 seconds. However, a short delay arose before his mooring line was secured, and at first the adjudicating committee refused him the prize, despite de la Meurthe, who was present, declaring himself satisfied. This caused a public outcry from the crowds watching the flight, as well as comment in the press. However a face-saving compromise was reached, and Santos-Dumont was awarded the prize. In a charitable gesture, he gave half the prize to his crew and then donated the other half to the poor of Paris.[9]

The first carriage-sized automobile suitable for use on existing wagon roads in the United States was a steam-powered vehicle invented in 1871 by Dr. J.W. Carhart, a minister of the Methodist Episcopal Church, in Racine, Wisconsin.[1][11][self-published source] It induced the State of Wisconsin in 1875 to offer a $10,000 award to the first to produce a practical substitute for the use of horses and other animals. They stipulated that the vehicle would have to maintain an average speed of more than 5 miles per hour (8 km/h) over a 200-mile (320 km) course. The offer led to the first city to city automobile race in the United States, starting on 16 July 1878 in Green Bay, Wisconsin, and ending in Madison, Wisconsin, via Appleton, Oshkosh, Waupun, Watertown, Fort Atkinson, and Janesville. While seven vehicles were registered, only two started to compete: the entries from Green Bay and Oshkosh. The vehicle from Green Bay was faster, but broke down before completing the race. The Oshkosh finished the 201-mile (323 km) course in 33 hours and 27 minutes, and posted an average speed of six miles per hour. In 1879, the legislature awarded half the prize.[12][13][14]

After the Scilly naval disaster of 1707 where four ships ran aground due to navigational mistakes, the British government offered a large prize of £20,000, equivalent to millions of pounds today, for anyone who could determine longitude accurately. The reward was eventually claimed in 1761 by Yorkshire carpenter John Harrison, who dedicated his life to improving the accuracy of his clocks. In 1735 Harrison built his first chronometer, which he steadily improved on over the next thirty years before submitting it for examination. The clock had many innovations, including the use of bearings to reduce friction, weighted balances to compensate for the ship's pitch and roll in the sea and the use of two different metals to reduce the problem of expansion from heat.

Non-neurotypical pioneers

Pioneers show a wide range of motivations. In particular, throughout the historical record, (lone) a non-neurotypical pioneer will occasionally obsess over a technology, refine it, make some progress, or leave a record of failed approaches. This may prove a hurdle for some safety approaches.

The following quotes are taken from their respective Wikipedia pages:

In the 1820s British inventor George Pocock developed man-lifting kites, using his own children as guinea pigs

In 1801, the French officer André Guillaume Resnier de Goué managed a 300-metre glide by starting from the top of the city walls of Angoulême and broke only one leg on arrival

The Jesuits were another major contributor to the development of pendulum clocks in the 17th and 18th centuries, having had an "unusually keen appreciation of the importance of precision". In measuring an accurate one-second pendulum, for example, the Italian astronomer Father Giovanni Battista Riccioli persuaded nine fellow Jesuits "to count nearly 87,000 oscillations in a single day". They served a crucial role in spreading and testing the scientific ideas of the period, and collaborated with contemporary scientists, such as Huygens.

Blyth's 10 m high, cloth-sailed wind turbine was installed in the garden of his holiday cottage at Marykirk in Kincardineshire and was used to charge accumulators developed by the Frenchman Camille Alphonse Faure, to power the lighting in the cottage, thus making it the first house in the world to have its electricity supplied by wind power. Blyth offered the surplus electricity to the people of Marykirk for lighting the main street, however, they turned down the offer as they thought electricity was "the work of the devil." Although he later built a wind turbine to supply emergency power to the local Lunatic Asylum, Infirmary and Dispensary of Montrose, the invention never really caught on as the technology was not considered to be economically viable

In 1941 the world's first megawatt-size wind turbine was connected to the local electrical distribution system on the mountain known as Grandpa's Knob in Castleton, Vermont, United States. It was designed by Palmer Cosslett Putnam and manufactured by the S. Morgan Smith Company. This 1.25 MW Smith–Putnam turbine operated for 1100 hours before a blade failed at a known weak point, which had not been reinforced due to war-time material shortages. No similar-sized unit was to repeat this "bold experiment" for about forty years

In 1741 de Vaucanson was appointed by Cardinal Fleury, chief minister of Louis XV, as inspector of the manufacture of silk in France. He was charged with undertaking reforms of the silk manufacturing process. At the time, the French weaving industry had fallen behind that of England and Scotland. During this time, Vaucanson promoted wide-ranging changes for automation of the weaving process. In 1745, he created the world's first completely automated loom, drawing on the work of Basile Bouchon and Jean Falcon. Vaucanson was trying to automate the French textile industry with punch cards - a technology that, as refined by Joseph-Marie Jacquard more than a half-century later, would revolutionize weaving and, in the twentieth century, would be used to input data into computers and store information in binary form. His proposals were not well received by weavers, however, who pelted him with stones in the street and many of his revolutionary ideas were largely ignored.

David Unaipon, Australian inventor, had a lifelong fascination with perpetual motion. One of his studies on Newtonian mechanics led him to create a shearing machine in 1910 that converted curvilinear motion into straight line movement. The device is the basis of modern mechanical shears.

Shockley was upset about the device being credited to Brattain and Bardeen, who he felt had built it "behind his back" to take the glory. Matters became worse when Bell Labs lawyers found that some of Shockley's own writings on the transistor were close enough to those of an earlier 1925 patent by Julius Edgar Lilienfeld that they thought it best that his name be left off the patent application.

Due to Shockley's earlier work on FETs and the existence of the Lilienfeld patent, Bell Labs left Shockley off their patent on the point-contact design. Shockley was incensed and decided to demonstrate who was the real brains of the operation. Only a few months later he invented an entirely new type of transistor with a layer or "sandwich" structure. This new form was considerably more robust than the fragile point-contact system, and would go on to be used for the vast majority of all transistors into the 1960s. It would evolve into the bipolar junction transistor.

Technology Readiness Levels

Technology readiness levels (TRLs) are an interesting way of measuring the maturity of a given technology. They originated at NASA to be used in the context of the space program, and then generalized to the point of uselessness, after which they were implemented in some projects of the European Union.

Originally I thought that trying to map out something akin the technology readiness level of each technology in my list would be worth doing, and created the following scale:

• L1. Earliest reference
  • L1-a: In a fictional story.
  • L1-b: Somewhere else.
• L2. Concept rigorously formulated.
• L3. Creation stage.
  • L3-a. First person working on the area towards a prototype.
  • L3-b. Proof of concept or prototype: A hobbyist has built the technology in their own laboratory, garage or basement, or as a private demonstration, or as a toy, but perhaps with no further intention of bringing it to market.
  • L3-c. Further development: The technology starts to be invested on and improved; there may be concrete plans about bringing it to market.
• L4. Technology is available.
  • If commercial: The product can be bought, even if it's expensive. This does not ask whether a sufficiently motivated billionaire could buy it, but rather whether the product is being produced in order to be sold.
  • If military: The product can be employed in action. One could not buy a nuclear weapon, but it was still available to the USA by 1945.
  • L4-a: Technology is available and cheap.
• L5. Technology is available and decent among the relevant dimensions
  • L5-a: Technology is decent and cheap.
• L6. Technology is pretty good.
  • L6-a: Technology is pretty good and cheap.
• L7. Technology is really good.
  • L7-a: Technology is really good and cheap.
• L8. Widespread adoption.
• L9. Societal influence. The invention has influenced society.

Although I did gather the information, this proved to be an unfruitful idea. The data was often not available, there were category errors, the scale asks for absolute quality when I mostly have a comparison with current levels, etc. I did get the insight that by the time a product is cheap, it's often of much better quality than early versions of the same product. Thinking about TRLs might have clarified my thinking about the evolution of some technologies, but overall I'd say it probably wasn't worth it.