Predictions made by Mati Roy in early 2020

Author: Mati Roy | Created: 2020-01-11 | Up­dated: 2020-03-01 (Ad­justed: 2020-11-07) | Pub­lished: 2020-11-07

Qual­ity: Those are notes I took for my pre­sent and fu­ture selves; they weren’t taken with the pur­pose of in­form­ing oth­ers.

Im­por­tance: 310. I don’t think all those pre­dic­tions are im­por­tant /​ trans­for­ma­tional. It’s likely not im­por­tant for you to read this.

Epistemic sta­tus: My cre­dences are already quan­tified for each pre­dic­tion. But my pre­dic­tions are not in­de­pen­dent of each other. For ex­am­ple, shorter AI timelines could mean shorter timelines on all those pre­dic­tions, and vice versa. I might also be mis­cal­ibrated on long-term pre­dic­tions — I don’t have first hand ex­pe­rience with that hav­ing only been born re­cently.

Con­text: I wrote some pre­dic­tions in Jan­uary-Fer­bru­ary 2020 (some with links to Pre­dic­tionBook and Me­tac­u­lus with the ex­act date available on those plat­forms). I hadn’t pub­lished it at the time be­cause there were a bunch of other top­ics I wanted to make pre­dic­tions on. They are mostly pre­dic­tions made on a 10-100 years time hori­zon.

Value to me: Mak­ing this re­search was much more time con­sum­ing than I thought. It was an in­ter­est­ing ex­er­cise, and will likely be in­ter­est­ing to check back on those pre­dic­tions in the fu­ture. Here’s how I felt go­ing down this rab­bit hole.

Medium: Ideally, I would like a pre­dic­tion plat­form where I could post my com­ments and pre­dic­tions on pub­lic ques­tions, like Me­tac­u­lus, but with the ad­di­tional op­tion of hav­ing a per­sonal pub­lic page au­to­mat­i­cally ag­gre­gat­ing all my pre­dic­tions and com­ments, similar to what I did here.

Cel­lu­lar agriculture

Why care?

50 billion an­i­mals are raised and kil­led in fac­tory farms ev­ery year. Most ex­pe­rience ex­treme lev­els of suffer­ing over the course of their lives due to in­tense con­fine­ment and the re­moval of body parts. The meat in­dus­try is also one of the largest con­trib­u­tors to cli­mate change, with 14.5% of global green­house gas emis­sions. (source)

A re­duced pop­u­la­tion of farmed an­i­mals would also likely re­duce the risk of pan­demics.

It would also likely al­low us to cre­ate meat for cheaper.

It might also provide a more ro­bust food sup­ply.

It would also al­low us to cre­ate healthier meat, with less con­tam­i­nants.

It would also al­low us to cre­ate a wider range of food, such as syn­the­siz­ing new forms of meat.

It might also make it eas­ier to sus­tain hu­man life on other ce­les­tial bod­ies.


85%: Meat and/​or dairy prod­ucts from cel­lu­lar agri­cul­ture will be com­mer­cially available by Jan­uary 1st, 2030. (Pre­dic­tionBook)

75%: A com­pany with its main ac­tivity be­ing cul­tured meat and/​or dairy prod­ucts will be val­ued at >= 1bn 2020USD by Jan­uary 1st, 2030. (Pre­dic­tionBook)

52%: The U.S. plant-based and cul­tured-based re­tail mar­ket will be worth more than 9bn 2020USD through­out 2029. (Pre­dic­tionBook)

Pre­dic­tions us­ing Elicit: https://​​​​builder/​​nUEGbEPgb (h/​​t Jung­won)


2020-01-22: Mem­phis Meats re­ceives 161 mil­lion USD in its Series B fundrais­ing, more than the to­tal in­vest­ment in cul­ti­vated meat com­pa­nies so far which was of 155 mil­lion USD. (source)

“In the next few decades I be­lieve that cell-based meat will be­come a ma­jor part of our­global meat sup­ply.” -Richard Bran­son, in­vestor (source)

There are now at least 43 cul­ti­vated meat com­pa­nies on five con­ti­nents pro­duc­ing 15 types of meat, plus a hand­ful of B2B com­pa­nies. BlueNalu, Fu­ture Meat Tech­nolo­gies, Fin­less Foods, Wild Type, and Aleph Farms all raised $10 to 20 mil­lion Series As in 2019. Mem­phis Meats, how­ever, is the first com­pany to raise a Series B round. (source)

The Good Food In­sti­tute shows U.S. re­tail sales of plant-based meat gen­er­ates an es­ti­mated $801M. (source)

The U.S. meat and poul­try in­dus­try ac­counts for $1.02 trillion in to­tal eco­nomic out­put or 5.6 per­cent of gross do­mes­tic product (GDP). (source)

The plant-based meat cat­e­gory alone is worth more than $800 mil­lion, with sales up 10 per­cent in the past year. Plant-based meat now ac­counts for 2 per­cent of re­tail pack­aged meat sales. (source)

Or­ga­ni­za­tion: https://​​www.cel­​​work/​​pro­ject-cmf/​​

In­vest­ment trends: https://​​​​industry

Other pre­dic­tions: Clean meat se­ries, Alt-meat party

Timeline: https://​​en.wikipe­​​wiki/​​Timeline_of_cel­lu­lar_agriculture

More info: https://​​en.wikipe­​​wiki/​​Cel­lu­lar_agriculture

Quan­tum computers

Why care?

Hav­ing a fast, gen­eral-pur­pose quan­tum simu­la­tor could help a lot in de­sign­ing new drugs, ma­te­ri­als, so­lar cells, high-tem­per­a­ture su­per­con­duc­tors, chem­i­cal re­ac­tions for mak­ing fer­til­izer, etc.

Break­ing ex­ist­ing pub­lic-key cryp­tog­ra­phy.

Op­ti­miza­tion and ma­chine learning


Ap­pli­ca­tions: http://​​www.the­​​~preskill/​​talks/​​Preskill-Q2B-2019.pdf


I put ~15% prob­a­bil­ity that noisy in­ter­me­di­ate-scale quan­tum will solve at least one use­ful prob­lem and/​or be eco­nom­i­cally vi­able by Jan­uary 1st, 2030. I’m not sure how to prop­erly op­er­a­tional­ize this ei­ther. Ex. of ap­pli­ca­tions: cer­tified ran­dom­ness.


Quan­tum supremacy (ie. be­ing able to do some­thing which we can’t do with clas­si­cal com­put­ers) is a spec­trum, but we’re cur­rently tran­si­tion­ing to it. (Me­tac­u­lus)

I’m not sure how to op­er­a­tional­ize this; maybe the ra­tio of en­ergy it takes for a clas­si­cal com­puter to solve the same prob­lem as a quan­tum com­puter? Here’s an­other idea: “Will it cost less than $1000 to calcu­late the pri­vate key of a 2048-bit RSA pub­lic key in 2030?” (Me­tac­u­lus).

Richard Haughton says:

Quan­tum-com­put­ing ex­perts are now find­ing them­selves obliged to re­peat a con­stant re­frain: it’s not just about the num­ber of qubits. One of the main mea­sures of the power of a quan­tum cir­cuit is its so-called depth: in effect, how many log­i­cal op­er­a­tions (‘gates’) can be im­ple­mented in a sys­tem of qubits be­fore their co­her­ence de­cays, at which point er­rors pro­lifer­ate and fur­ther com­pu­ta­tion be­comes im­pos­si­ble. How the qubits are con­nected also mat­ters. So the true mea­sure of the power of a quan­tum cir­cuit is a com­bi­na­tion of fac­tors, which IBM re­searchers have called the “quan­tum vol­ume

But I couldn’t find any graphs of the max­i­mum quan­tum vol­ume achieved over time.

It would also be nice to op­er­a­tional­ize how much noise a quan­tum com­puter has.

John Preskill says:

we might suc­ceed in build­ing large-scale quan­tum com­put­ers af­ter a few decades of very hard work. (source)

[noisy in­ter­me­di­ate-scale quan­tum] might also have other use­ful ap­pli­ca­tions. But we’re not sure about that. (source)

Scott Aaron­son:

given the spec­tac­u­lar progress by Google and oth­ers over the last few years, my guess is that we’re at most a decade away from some small, spe­cial-pur­pose quan­tum com­put­ers (with ~50-200 qubits) that could be use­ful for quan­tum simulation

a shift from pure aca­demic re­search to ven­ture cap­i­tal­ists and in­dus­trial efforts just within the last 4-5 years [from 2018-06-06] (source)

Note: It could be eco­nom­i­cally vi­able, yet not eco­nom­i­cally benefi­tial (ex.: if peo­ple only buy quan­tum defenses to se­cure them­selves against quan­tum at­tacks).

China’s Na­tional Lab­o­ra­tory for Quan­tum In­for­ma­tion Sciences is a US$10 billion pro­ject due to open in 2020. (source). Europe’s Quan­tum Tech­nol­ogy Flag­ship is a €1 billion eu­ros pro­ject that opened in 2016. There are ~7000 re­searchers pub­lish­ing ~8,500 ar­ti­cles per year [as of 2014] (source).

Longest quan­tum en­tan­gle­ment dis­tance: Fe­bru­ary 2020 record: 50 km; pre­vi­ous record was 1.3 km (source).

In­vest­ment trends: https://​​www.cbin­​​re­search/​​re­port/​​quan­tum-com­put­ing/​​

Pop­u­lar­ity trends: https://​​www.cbin­​​re­search/​​quan­tum-com­put­ing-startup-ecosys­tem/​​

Other trends: https://​​quan­tum­com­put­​​2018/​​02/​​quan­tum-com­puter-in­vest­ing.html

Other pre­dic­tions: Metaculus

Timeline: https://​​­​​wiki/​​Timeline_of_Quan­tum_Computing

More info: https://​​­taaron­​​blog/​​

Nu­clear fusion

Why care?


Eco­nomic pros­per­ity of hav­ing cheap and abun­dant en­ergy. En­ergy is a $8.5 trillion en­ergy in­dus­try.

It takes roughly 55,000 bar­rels of oil to heat 10,000 homes for one year. With fu­sion en­ergy, it would take one liter of deu­terium and tri­tium, ex­tracted from wa­ter, to power those 10,000 homes.

That limited resi­due has a half-life of a few hun­dred years, not tens of thou­sands of years, and hence is more man­age­able than nu­clear fis­sion waste.

And whereas those 55,000 bar­rels of oil would re­lease 23,500 tons of car­bon diox­ide, fu­sion pro­duces no emis­sions and will have a life­cy­cle car­bon in­ten­sity lower than so­lar or wind (as mea­sured in CO2 from all con­struc­tion, man­u­fac­tur­ing and op­er­a­tions per kWh pro­duced). (source)

pro-fu­sion ad­vo­cates also say that fu­sion re­ac­tors would be in­ca­pable of gen­er­at­ing the dan­ger­ous run­away chain re­ac­tions that lead to a melt­down (source)


But fu­sion re­ac­tors have other se­ri­ous prob­lems that also af­flict to­day’s fis­sion re­ac­tors, in­clud­ing neu­tron ra­di­a­tion dam­age and ra­dioac­tive waste, po­ten­tial tri­tium re­lease, the bur­den on coolant re­sources, out­size op­er­at­ing costs, and in­creased risks of nu­clear weapons pro­lifer­a­tion. (source)

Also see: Pure fu­sion weapon — Wikipedia


When will the fu­sion en­ergy gain fac­tor of a fu­sion re­ac­tor reach a Q >= 1?75% be­fore 2044

When will a fu­sion re­ac­tor reach ig­ni­tion?50% be­fore be­fore 2051

When will a nu­clear fu­sion re­ac­tor hit an eco­nomic breakeven?50% be­fore 2057


Will rad­i­cal new “low-en­ergy nu­clear re­ac­tion” tech­nolo­gies prove effec­tive be­fore 2019?

Will Gen­eral Fu­sion [meet] its plasma tem­per­a­ture tar­gets by mid-2020?

When will the first 100 Me­gawatt fu­sion-based elec­tri­cal gen­er­a­tion fa­cil­ity come into ser­vice?

A re­vival of in­ter­est in muon-cat­alyzed fu­sion?


Now [2019], pri­vately funded fu­sion en­ergy start-ups like TAE Tech­nolo­gies, Com­mon­wealth Fu­sion Sys­tems and Gen­eral Fu­sion [...] are leav­ing the lab­o­ra­tory and start­ing to build pro­to­type demon­stra­tion plants. (source)

The JET (Joint Euro­pean Torus) is the world’s largest op­er­a­tional mag­net­i­cally con­fined palsma physics ex­per­i­ment. It started its op­er­a­tion in 1983. (source)

ITER is a large in­ter­na­tional pro­ject to build a mag­netic con­fine­ment plama­physics ex­per­i­ment 10 timers big­ger than the cur­rent biggest. The pro­ject was offi­cially ini­ti­ated in 1988. The build­ing costs were over US$14 billion by June 2015, and the con­struc­tion of the fa­cil­ity is ex­pected to be com­pleted in 2025. Ini­tial plasma ex­per­i­ments are sched­uled to be­gin in 2025, with full deu­terium-tri­tium fu­sion ex­per­i­ments start­ing in 2035. Since the 1950, more than 100 fu­sion re­ac­tors have been built. (source)

DEMO, ITER’s suc­ces­sor, is planned to be con­structed in 2024-2033, and have a first phase of op­er­a­tion in 2033-2038. After which, it would be ex­panded and up­dated, with the sec­ond phase of op­er­a­tion start­ing in 2040. In 2012, the roadmap aimed to have an elec­tric­ity gen­er­a­tion demon­stra­tion in 2048. (source)

PROTO, DEMO’s suc­ces­sor, is planned for af­ter 2050. PROTO would act as a pro­to­type power sta­tion demon­strat­ing elec­tric­ity gen­er­a­tion on a com­mer­cial ba­sis. (source)

From fu­sion en­ergy gain fac­tor:

The fu­sion en­ergy gain fac­tor, usu­ally ex­pressed with the sym­bol Q, is the ra­tio of fu­sion power pro­duced in a nu­clear fu­sion re­ac­tor to the power re­quired to main­tain the plasma in steady state. The con­di­tion of Q = 1, when the power be­ing re­leased by the fu­sion re­ac­tions is equal to the re­quired heat­ing power, is referred to as breakeven[.] [...] With typ­i­cal fuels, self-heat­ing in fu­sion re­ac­tors is not ex­pected to match the ex­ter­nal sources un­til at least Q = 5. If Q in­creases past this point, in­creas­ing self-heat­ing even­tu­ally re­moves the need for ex­ter­nal heat­ing. At this point the re­ac­tion be­comes self-sus­tain­ing, a con­di­tion called ig­ni­tion. Ig­ni­tion cor­re­sponds to in­finite Q, and is gen­er­ally re­garded as highly de­sir­able for prac­ti­cal re­ac­tor de­signs.

Oper­at­ing above en­g­ineer­ing breakeven, a ma­chine would pro­duce more elec­tric­ity than it uses and could sell that ex­cess. One that sells enough elec­tric­ity to cover its op­er­at­ing costs is some­times known as eco­nomic breakeven. Ad­di­tion­ally, fu­sion fuels, es­pe­cially tri­tium, are very ex­pen­sive, so many ex­per­i­ments run on var­i­ous test gasses like hy­dro­gen or deu­terium. A re­ac­tor run­ning on these fuels that reaches the con­di­tions for breakeven if tri­tium was in­tro­duced is said to be op­er­at­ing at ex­trap­o­lated breakeven.

As of 2017, the record for Q is held by the JET toka­mak in the UK, at Q = (16 MW)/​(24 MW) ≈ 0.67, first at­tained in 1997. ITER was origi­nally de­signed to reach ig­ni­tion, but is cur­rently de­signed to reach Q = 10, pro­duc­ing 500 MW of fu­sion power from 50 MW of in­jected ther­mal power. The high­est record for ex­trap­o­lated breakeven was posted by the JT-60 de­vice, with Q<sub>ext</​sub> = 1.25.

Op­ti­mists be­lieve that we are five years away from the first proof-of-con­cept fu­sion en­ergy demon­stra­tion plant and 15 years from a com­mer­cial roll-out at scale.

I men­tioned three pri­vate start-ups in the be­gin­ning of this piece be­cause they all are on vi­able paths to com­mer­cial­ize fu­sion in the next five to 15 years. If any of them win, we all win. But they all are un­der­funded.

TAE Tech­nolo­gies (formerly Tri Alpha En­ergy) has raised $800 mil­lion over the past 20 years and earned no­to­ri­ety for be­ing backed by Microsoft co-founder Paul Allen. TAE has been ex­per­i­ment­ing on a $100 mil­lion re­ac­tor and talks about com­mer­cial­iz­ing the tech­nol­ogy in five years, for which they are ex­pected to start rais­ing ad­di­tional cap­i­tal soon.

Com­mon­wealth Fu­sion Sys­tems is backed by the Ital­ian oil com­pany Eni and the celebrity billion­aires of Break­through En­ergy Ven­tures. They ex­pect to add fu­sion power to the grid in 15 years but have raised only $115 mil­lion so far. Again, too lit­tle.

Gen­eral Fu­sion, based in Burn­aby, Bri­tish Columbia, is cur­rently en­g­ineer­ing a $350 mil­lion demo plant, set to go live in five years. If that demo plant op­er­ates well, Gen­eral Fu­sion could have its first com­mer­cial plant built in 10 years from now. To date, Gen­eral Fu­sion has raised funds from the Cana­dian fed­eral gov­ern­ment, Khaz­anah Na­sional (Malaysia’s sovereign wealth fund), Cen­ovus En­ergy, Brae­mar En­ergy, Jeff Be­zos and Chrysalix, among oth­ers.


Th­ese are three of about 22 pri­vate com­pa­nies in the fu­sion game, so of course, there’s in­tense com­pe­ti­tion for the few in­vestors will­ing to back these high-risk and cap­i­tal-in­ten­sive but very high-re­turn in­no­va­tions.


Google’s US$10-mil­lion pro­ject aimed to test the cold-fu­sion claims rigor­ously in a field that lacked cred­ible sci­en­tific data, says Tre­vithick. (source)

Other pro­jects in­clude: Wen­del­stein 7-X, HiPER, Na­tional Ig­ni­tion Fa­cil­ity.

The main source of tri­tium is fis­sion nu­clear re­ac­tors. (source)

Graph of US in­vest­ment: https://​​i.huff­​​gen/​​2470392/​​origi­nal.jpg

Pro­jec­tion for lead­ing fu­sion pro­jects: https://​​2oqz471sa19h3vbwa53m33yj-wpeng­​​wp-con­tent/​​up­loads/​​2017/​​07/​​fu­sion-timeline.jpg

Kurzge­sagt video: https://​​​​watch?v=mZsaaturR6E

Timeline: https://​​en.wikipe­​​wiki/​​Timeline_of_nu­clear_fusion

Space mining

Why care?

As of Septem­ber 2016, there are 711 known as­ter­oids with a value ex­ceed­ing US$100 trillion.

Although part of the wealth is at cur­rent mar­ket value, which would drop a lot with a higher offer.

Un­like Earth, which is pro­tected by its mag­netic field, the Moon has been bom­barded with large quan­tities of Helium-3 by the so­lar wind. (source)

This could prove very use­ful if we de­velop vi­able nu­clear fu­sion.


When will space min­ing reach prof­ita­bil­ity?

50%: Space min­ing will be prof­itable be­fore 2090, con­di­tional on no ex­is­ten­tial catas­tro­phe or win


On April 24, 2012 a plan was an­nounced by billion­aire en­trepreneurs to mine as­ter­oids for their re­sources.

More info: https://​​en.wikipe­​​wiki/​​As­teroid_mining

Self-driv­ing cars

Why care?


Nearly 38,000 peo­ple died in car crashes in 2016, ac­cord­ing to the Na­tional High­way Traf­fic Safety Ad­minis­tra­tion, and most se­ri­ous crashes are due to hu­man er­ror. (source)


Au­tonomous ve­hi­cles could add $800 billion [per year] to the US econ­omy, a new re­port says.


In 2015, some 15.5 mil­lion work­ers in the U.S. worked in jobs re­lated to driv­ing, ac­cord­ing to an Au­gust 2017 re­port from the Depart­ment of Com­merce’s Eco­nomics and Statis­tics Ad­minis­tra­tion.

Only 3.8 mil­lion of those work­ers op­er­ate mo­tor ve­hi­cles such as a truck or taxi

The 2017 me­dian pay for heavy and trac­tor-trailer truck drivers was $20.42 per hour, which adds up to $42,480 per year, ac­cord­ing to the Bureau of La­bor Statis­tics.

It is es­ti­mated that au­tonomous cars could elimi­nate 300,000 driv­ing jobs a year, ac­cord­ing to a May 2017 re­port from Gold­man Sachs. But that won’t hap­pen right away; the re­port es­ti­mated that from 2025 to 2030, au­tonomous cars will be 20 per­cent of car sales.

Nearly 38,000 peo­ple died in car crashes in 2016, ac­cord­ing to the Na­tional High­way Traf­fic Safety Ad­minis­tra­tion, and most se­ri­ous crashes are due to hu­man er­ror.

self-driv­ing cars could con­tribute to ma­jor gains in fuel effi­ciency, lower trans­porta­tion costs to the con­sumer and in­crease ac­cess to ru­ral ar­eas.


Car own­ers will be able to put their cars to work as a taxi when they aren’t us­ing it. In fact, most peo­ple won’t need to own a car be­cause of that.

Given that fleets of self-driv­ing cars are bet­ter co­or­di­nated, traf­fic will also be re­duced.

Given that on-de­mand cars will be as cheap as own­ing a car, and that these cars will be safe, smaller cars (for 1 or 2 pas­sen­gers) will make sense.


This added effi­ciency will also re­sult in less cars be­ing built and less green­house gases per dis­tance.

  • Although given it will be cheaper, more dis­tance might be driven.

There’s one small study sug­gest­ing adap­tive cruise con­trol im­proves effi­ciency a lit­tle (5 to 7 per­cent), but there’s lit­tle else be­yond that. (source)


My predictions

50% Level 3 au­tonomous cars will be com­mer­cially available by 2030. (Pre­dic­tionBook)

50% Level 4 au­tonomous cars will be com­mer­cially available by 2036. (Pre­dic­tionBook)

50% Level 5 au­tonomous cars will be com­mer­cially available by 2042. (Pre­dic­tionBook)

50% Level 5 self-driv­ing air ve­hi­cles will be available by 2030. (Pre­dic­tionBook)

Clar­ifi­ca­tion: Needs to be available for com­mer­cial use on pub­lic roads in at least one city (not just on the high­way).


When will Level 3 self-driv­ing cars be available com­mer­cially? (Me­tac­u­lus)

When will Level 4 self-driv­ing cars be com­me­cially available com­mer­cially? (Me­tac­u­lus)

When will Level 5 self-driv­ing cars be available com­mer­cially? (Me­tac­u­lus)


In what year will half of new cars sold in the US be fully au­tonomous? (Me­tac­u­lus)

[2018-04-17] Re­searchers fore­cast that by 2025 we’ll see ap­prox­i­mately 8 mil­lion au­tonomous or semi-au­tonomous (L3, L4, L5) ve­hi­cles on the road. (source)

Key Volk­swa­gen Exec Ad­mits Full Self-Driv­ing Cars ‘May Never Hap­pen’ (source)

Past prediction

In 2020, you’ll be a “per­ma­nent back­seat driver,” the Guardian pre­dicted in 2015. “10 mil­lion self-driv­ing cars will be on the road by 2020,” blared a Busi­ness In­sider head­line from 2016. Those dec­la­ra­tions were ac­com­panied by an­nounce­ments from Gen­eral Mo­tors, Google’s Waymo, Toy­ota, and Honda that they’d be mak­ing self-driv­ing cars by 2020. Elon Musk fore­cast that Tesla would do it by 2018 — and then, when that failed, by 2020. (source)


  • Dis­tance driven (ie. train­ing data ac­quired)

  • Num­ber of dis­en­gage­ment per kilometer

  • Num­ber of ac­ci­dents /​ deaths per kilo­me­ter (no­tably com­pared to hu­mans)

Hu­man driv­ing pro­duces one fatal ac­ci­dent in ev­ery 100 mil­lion miles driven. (source)

  • Laws

A quick primer on the now-in­dus­try-stan­dard SAE In­ter­na­tional rules on how to dis­cuss self-driv­ing abil­ities: Level 0 is no au­toma­tion what­so­ever. Level 1 is par­tial as­sis­tance with cer­tain as­pects of driv­ing, like lane keep as­sist or adap­tive cruise con­trol. Level 2 is a step up to sys­tems that can take con­trol of the ve­hi­cle in cer­tain situ­a­tions, like Tesla’s Au­topi­lot or Cadillac’s Su­per Cruise, while still re­quiring the driver to pay at­ten­tion.

Get past that and we en­ter the realm of spec­u­la­tion: Level 3 promises full com­puter con­trol with­out su­per­vi­sion un­der defined con­di­tions dur­ing a jour­ney, Level 4 is start-to-finish au­tonomous tech limited only by vir­tual safe­guards like a ge­ofence, and Level 5 is the to­tal hands-off, go liter­ally any­where at the push of a but­ton ex­pe­rience where the ve­hi­cle might not even have phys­i­cal con­trols.

(source; dis­cus­sion)

Fe­bru­ary 2020 notes


Volo­copter and RTA Dubai co­op­er­ate on eVTOL Au­tonomous Air Taxi (AAT) (source)


You can buy a car that will au­to­mat­i­cally brake for you when it an­ti­ci­pates a col­li­sion, or one that helps keep you in your lane, or even a Tesla Model S (which — dis­clo­sure — my part­ner and I own) whose Au­topi­lot mostly han­dles high­way driv­ing. (source)


In 2018, Waymo drove 1.2 mil­lion miles in Cal­ifor­nia, with 0.09 dis­en­gage­ments ev­ery 1,000 miles.

Com­ing in sec­ond is Gen­eral Mo­tors’ Cruise, with about half a mil­lion miles and 0.19 dis­en­gage­ments per 1,000 miles. (Cruise ar­gues that since it tests its cars on San Fran­cisco’s difficult streets, these num­bers are even more im­pres­sive than they look.)

The com­pany doesn’t re­lease spe­cific figures, but its filings for its IPO last year said that it had driven “mil­lions” of miles.



March 18, 2018, was the first time a self-driv­ing car ran down a pedes­trian.

A re­port from the Na­tional Trans­porta­tion Safety Board im­pli­cated Tesla’s Au­topi­lot sys­tem in an­other lethal 2018 ac­ci­dent; while the driver had his hands off the wheel, the car steered into a con­crete di­vider and crashed, kil­ling him.

A full in­ves­ti­ga­tion hasn’t yet been con­ducted on three more re­cent deadly Tesla crashes.

The prob­lem, ac­cord­ing to NTSB chair­man Robert Sumwalt, is that drivers as­sume Au­topi­lot lets them take their at­ten­tion off the road, when they shouldn’t.



the laws sur­round­ing self-driv­ing cars vary enor­mously by state, with 29 states hav­ing passed leg­is­la­tion.



2014: $167 million

2017: #3 billion

First 3 quar­ters of 2018: $4.2 billion

(CB In­sights)

To­tal so far [early 2020]: at least $16 billion (source)

Tesla has been build­ing an au­topi­lot fea­ture since 2014, and both Ap­ple and [Alpha­bet] are de­vel­op­ing their own self-driv­ing car mod­els.

Tra­di­tional au­to­mo­tive com­pa­nies are also in­vest­ing. Ford re­cently an­nounced that the au­tomaker plans to spend $4 billion on au­tonomous ve­hi­cles by 2023. Gen­eral Mo­tors will pour $100 mil­lion into self-driv­ing cars, and Toy­ota launched a $2.8 billion self-driv­ing car com­pany in Tokyo. (source)

Other notes

it is ex­as­per­at­ingly difficult to get a good es­ti­mate of how long un­til self-driv­ing cars hap­pen for real for the typ­i­cal Amer­i­can, both be­cause no one knows for sure and be­cause com­pa­nies have in­cen­tives to pub­li­cize op­ti­mistic es­ti­mates. The com­pa­nies boast about their progress but don’t pub­lish their mishaps.

His­tory: https://​​en.wikipe­​​wiki/​​His­tory_of_self-driv­ing_cars

More info: https://​​en.wikipe­​​wiki/​​Self-driv­ing_car

Fully ar­tifi­cial kidney

Why care?

End-stage re­nal dis­ease, or chronic kid­ney failure, af­fects nearly 2 mil­lion peo­ple wor­ld­wide, in­clud­ing more than 570,000 in the United States, and costs the na­tion al­most $40 billion each year for treat­ment. The most effec­tive treat­ment is kid­ney trans­plan­ta­tion, but those or­gans are in short sup­ply; last year, only 16,812 kid­neys were available for trans­plant, leav­ing 92,000 pa­tients on the wait­ing list, ac­cord­ing to the Or­gan Pro­cure­ment and Trans­plant Net­work. (source)


50% Fully ar­tifi­cial kid­neys will be pub­li­cly available by 2042. (Pre­dic­tionBook)


This idea was re­searched at the Univer­sity of Cal­ifor­nia, San Fran­cisco (UCSF), lead­ing to a pro­to­type model in 2010* and clini­cal tri­als be­gin­ning in 2017.* (source)

The FDA an­nounced to­day (April 9) that it had cho­sen three re­nal de­vice pro­jects to pi­lot a new reg­u­la­tory ap­proval pro­gram called In­no­va­tion Path­way 2.0, in­tended to bring break­through med­i­cal de­vice tech­nolo­gies to pa­tients faster and more effi­ciently.

The ar­tifi­cial kid­ney pro­ject, which is tar­geted for clini­cal tri­als in 2017, was se­lected for its trans­for­ma­tive po­ten­tial in treat­ing end stage re­nal dis­ease and for its po­ten­tial to benefit from early in­ter­ac­tions with the FDA in the ap­proval pro­cess.

in­vestors and grant­ing agen­cies to fund the ex­pected $20 mil­lion it will cost


Re­lated: https://​​en.wikipe­​​wiki/​​Ex­panded_access

More info: https://​​en.wikipe­​​wiki/​​Ar­tifi­cial_kidney

Global population

8.5 mil­lion peo­ple in 2030 (source)

50% CI: 8.3-8.7

More info: https://​​our­wor­ld­in­​​world-pop­u­la­tion-growth

Male birth con­trol pill


50% A male birth con­trol pill will be com­mer­cially available in the US or EU by 2045. (Pre­dic­tionBook)


Still, 10 years to mar­ket seems op­ti­mistic to him. It’s ex­pen­sive to make an idea into re­al­ity, and lengthy, well-de­signed stud­ies are re­quired to make sure such a con­tra­cep­tive is safe, effec­tive and re­versible.

The promise of a safe and effec­tive oral male con­tra­cep­tive “has been always on the verge of be­com­ing a re­al­ity, but I feel like I’ve been hear­ing that for a long time now,” said Na­jari, who was not in­volved in the new re­search. “I think it’s go­ing to take at least 10 years, if not longer.”


More in­for­ma­tion: https://​​en.wikipe­​​wiki/​​Male_con­tra­cep­tive#Research

Man mis­sion to Mars

63%: Will Mars have a per­ma­nent pop­u­la­tion of 10,000 be­fore the Moon does?

1%: Will NASA land peo­ple on Mars prior to 2030?

Will SpaceX land peo­ple on Mars prior to 2030?

Deep­est hole


50% A hole will be drilled down to Earth’s man­tle by 2090. (Pre­dic­tionBook)

The In­te­grated Ocean Drilling Pro­gram has been in op­er­a­tion since 2003 [...] The effort is ex­pected to take many, if not dozens, of years, and may re­quire $1 billion.

More info: https://​​www.smith­so­ni­an­​​smith­so­nian-in­sti­tu­tion/​​ask-smith­so­nian-whats-deep­est-hole-ever-dug-180954349/​​


24%: Bit­coins thought to be­long to Satoshi Nakamoto will be spent by 2075-04-05. (Pre­dic­tionBook; Me­tac­u­lus)

Notes with­out predictions

Gene editing


At least one other CRISPR baby will be born by Jan­uary 2030.


**Timeline: **https://​​­​​wiki/​​Timeline_of_ge­netic_en­g­ineer­ing_in_humans

Geother­mal power

In its 2016 power pro­duc­tion re­port, the Geother­mal En­ergy As­so­ci­a­tion (GEA) found that only 6 to 7 per­cent of global geother­mal power po­ten­tial has been tapped. How­ever, pro­duc­tion is on the rise and the GEA pre­dicts global geother­mal en­ergy pro­duc­tion will more than dou­ble by 2030. (source)

Elec­tric cars

The main bat­tery tech­nol­ogy used to be lead-acid, then came alkaline, then nickel-cad­mium, and now lithium-ion in many differ­ent chem­i­cal con­figu­ra­tions. In the fu­ture, we ex­pect solid-state bat­ter­ies, and per­haps boxes with nanoscale fly­wheels, flow bat­ter­ies, or su­per­ca­pac­i­ta­tors.

Bloomberg ex­pects 1,000% (a ten­fold) growth be­tween now and 2025, reach­ing 11% mar­ket share for BEVs in the to­tal pri­vate ve­hi­cle (PV) mar­ket by then. It then an­ti­ci­pates an­other 170% growth from 2025 to about 30% in 2030. That is fol­lowed by only 80% growth from 2030 to reach a 55% mar­ket share in 2040. Why Bloomberg ex­pects the growth curve to flat­ten when the prod­ucts get bet­ter and cheaper in­stead of fol­low­ing the nor­mal S-curve is not dis­cussed.

Bat­tery de­vel­op­ment is on a tech­nol­ogy curve. Char­ac­ter­is­tics of this curve are a ~14% de­cline in price ($/​kWh) and a 6% in­crease in den­sity (kWh/​kg). Com­bined with power elec­tron­ics, bat­tery and mo­tor man­age­ment com­put­ers, mo­tors, and gears, a com­bined de­crease of 7% per year is used.

En­vi­ron­men­tal con­cerns call for in­creas­ingly stric­ter ex­haust stan­dards, mak­ing in­ter­nal com­bus­tion en­g­ines and ex­haust-treat­ing sys­tems more ex­pen­sive, re­sult­ing in an ex­pected 3% yearly in­crease.

Drug resistance


  • Num­ber of drug-re­sis­tant infections


The num­ber of drug-re­sis­tant in­fec­tions will in­crease (70 per­cent)

An­tibiotics are great when they work, but be­cause we’ve overused them, more and more in­fec­tions are be­com­ing re­sis­tant to them. 2020 will con­tinue that trend.

I’m bas­ing this pre­dic­tion on two high-level re­ports re­leased last year: one from the Cen­ters for Disease Con­trol and Preven­tion (CDC) and the other from the United Na­tions. The re­ports pre­sented some pretty ter­rify­ing find­ings: Re­sis­tance to sec­ond- and third-line an­tibiotics (of­ten the last lines of defense) is pro­jected to al­most dou­ble be­tween 2005 and 2030. If we don’t make a rad­i­cal change now, drug-re­sis­tant dis­eases could kill 10 mil­lion peo­ple a year by 2050 — up from 700,000 a year now.

Note that I’m not pre­dict­ing the num­ber of deaths from drug-re­sis­tant in­fec­tions will in­crease rel­a­tive to 2019. It’s to­tally pos­si­ble to make progress on that front. In fact, such deaths have de­creased by 18 per­cent since 2013, in part be­cause more hos­pi­tal pro­fes­sion­als started to heed ex­perts’ warn­ings. But even if we man­age to sta­bi­lize or de­crease the death toll, the num­ber of drug-re­sis­tant in­fec­tions could still rise — and the trend sug­gests that’s likely to hap­pen, be­cause we aren’t ad­dress­ing our overuse of an­tibiotics with any­thing like the nec­es­sary speed. — SS


Fa­cial recognition

Fa­cial recog­ni­tion will be banned in at least three more cities (70 per­cent)

In 2019, we saw a grow­ing back­lash against fa­cial recog­ni­tion tech­nol­ogy. San Fran­cisco, Oak­land, and Berkeley banned it in Cal­ifor­nia, as did three com­mu­ni­ties in Mas­sachusetts: Somerville, Brook­line, and Northamp­ton. In 2020, I pre­dict we’ll see at least three more cities in­sti­tute a ban on the con­tro­ver­sial tech.

To be clear, I’m talk­ing about a ban that ap­plies to city de­part­ments like po­lice; I think out­right bans that would also cover busi­nesses, in­di­vi­d­u­als, and fed­eral agen­cies are way less likely.

I’m partly go­ing off lo­cal news about par­tic­u­lar cities — Port­land is cur­rently de­liber­at­ing a ban, and the west­ern Mas­sachusetts city of Springfield might be next. Last year saw mount­ing push­back against fa­cial recog­ni­tion from AI re­searchers, groups like the ACLU, low-in­come ten­ants in Brook­lyn, and many more. Their protests seem to be grow­ing bolder, not quieter.

I should note that ac­cord­ing to Pew Re­search Cen­ter sur­vey data, most Amer­i­cans are now in fa­vor of po­lice us­ing fa­cial recog­ni­tion. I don’t think a na­tion­wide ban is in the cards for 2020 (sorry, Bernie). But a lot can still hap­pen on the city level, and I think it will. — SS


Cli­mate change

Global car­bon emis­sions will in­crease (80 per­cent) (source)

Aver­age world tem­per­a­tures will in­crease rel­a­tive to 2019 (60 per­cent) (source)

Cal­ifor­nia has a wild­fire among the 10 most de­struc­tive in state his­tory (60 per­cent)

Fu­ture of bets

A re­cent Supreme Court de­ci­sion has cleared the way to bring an es­ti­mated $300 billion in black-mar­ket sports bet­ting into the light. (source)

Pre­dic­tion on pre­dic­tion plat­forms on Foretold


Notable play­ers: Face­book, Microsoft, Neu­ral­ink (source), Nis­san (source)

Notable early ap­pli­ca­tions: con­trol­ler for a VR head­set; com­puter nav­i­ga­tion for hand­i­capped people

Met­rics: words per minute typed; vo­cab­u­lary it can de­tects + accuracy

Cur­rently: In 2013, it was pos­si­ble to iden­tify which sin­gle image was be­ing seen from a set of 120. (source)
More info: Brain-read­ing — Wikipe­dia; Neu­ral­ink — WaitButWhy

Other top­ics I was in­ter­ested in predicting

Tech­nol­ogy /​ politics

In­ter­net speed

Fuel cells cars

An­ti­mat­ter production

Space elevators

Heat re­sis­tant material

3D printed build­ing (note: Dubai has built the world’s largest 3D printed build­ing)

Small nu­clear re­ac­tor (note: Pen­tagon awards con­tracts to de­sign mo­bile nu­clear re­ac­tor — skimmed)

Daily imag­ing of Earth (see: Planet Labs, Inc.)

Mili­tary drones

Re­mote work: Telehealth

Smart cities; pri­vate cities (note: Smart City Will Pri­vately- Owned Towns Drive the Devel­op­ment of Smart Cities? — not read)

Holog­ra­phy (Wikipe­dia)

Vac­cine with­out nee­dles (not read)

Geother­mal energy

En­ergy consumption


Ad­vance au­to­com­plete (ex.: draw­ing)

Taste viewer


FDA av­er­age years to ap­prove drug

As­sisted suicide


World pop­u­la­tion growth

Fer­til­ity rate

Life expectancy

Pop­u­la­tion density

Most pop­u­lous cities

Ge­net­i­cally mod­ified an­i­mals (op­er­a­tional­i­sa­tion: If DNA al­ter­a­tions con­tinue to re­quire FDA ap­proval by de­fault, how many in­ten­tion­ally ge­nomic DNA al­tered an­i­mals will be de­ter­mined as safe to eat, by the end of July 2025? ; Me­tac­u­lus)

Emer­gency called automatically


Real Es­tate Sales To Take Place En­tirely On­line In A Seam­less Way

To­k­eniza­tion And Global Liquidity

Phys­i­cal money doesn’t ex­ist in the US anymore

More money in proptech


Porn with cre­ated from images of peo­ple of your choices

Porn gen­er­ated by AI based on key­words of your choice

Re­in­force­ment learn­ing to gen­er­ate porn tai­lored to your preference


Spo­ken lan­guages,

North Korea and South Korea unifi­ca­tion (Pre­dic­tionBook)

Eco-eco­nomic decoupling

Ve­ganism (Me­tac­u­lus)