Super Technology of the 2020s: Narrow Super AI, Molecular Electronics, Atom Technology

There has and will be substantial development of major super technologies this decade.

There have been several major things with AI. We are in the age of narrow superintelligence.

Here are some definitions for different categorizations of AI.
Artificial narrow intelligence (ANI) has a narrow range of abilities.
Artificial narrow superintelligence (ANSI) has a narrow range of abilities but those abilities are well beyond human level.
Artificial general intelligence (AGI) is a broad range of abilities that are comparable to the range of intelligence for a human.
Artificial general superintelligence (AGSI)is far more capable than a human across all domains of human intelligence.

Google Assistant, Google Translate, Siri and other natural language processing tools are Narrow AI. They are classified as “Weak” AI because they are nowhere close to having human-like intelligence.

Ray Kurzweil calls “fast thinking” AI intelligence – “weak superhumanity”. Google Search is weakly superhuman. It is far faster than human for the task of search.

There have been one million times more computing power devoted to the protein folding problem and drug discovery. Deep Mind has made significant progress in solving protein folding. Alphafold 2 will certainly help to advance biology. It can generate folded structure predictions that can then be used to solve experimental structures by crystallography (and probably other techniques). So this will help the science of structure determination go a bit faster in some cases. This is still short of the 0.3 Angstrom precision needed to be highly useful for drug discovery. However, the work and progress is very promising and it now appears to be a matter of when and not if the problem will be solved in major ways.

The Google Research next-generation AI architecture is called Pathways. It will enable a single model to learn millions of things instead of just one thing. This will be a massive phase change and up-leveling of neural network and AI capabilities.

Computer vision, computer speech, search, language processing, language understanding, translation, protein folding prediction and analysis, self-driving, Teslabot like systems and much more are happening. We will get deeper into the multi-trillion and quadrillion dollar age of super narrow AI dominance. The super-narrow AI will interact. We will be able to order chatbots for delivery or rides or supply chain via self-driving and robotic loading and unloading.

The data advantages of each dominant super-narrow AI will be difficult to displace from a dominated market niche. Imagine the difficulty of a superior search engine trying to displace Google in search. Google has not just search but the advertising ecosystem.

Any new entrant Super AGI or broader super AI will need to displace any market entrenched Narrow Super AI. The owner of a Narrow Super AI can broaden or improve its Narrow Super AI. Big Tech companies can also try to buyout or use mergers or partnerships with entrenched Narrow AI. Narrow AIs that can be broadened or expanded by strong incumbents (like Tesla or Google) will have better chances of increasing reach and displacing weaker competition.

Tesla and other companies are spending billions every year developing self-driving cars. Tesla has 2.5 million cars on the road and each has $3000 in cameras and other hardware to enable full self-driving and will get over the air software updates as the software improves. By 2025, there could be 20-40 million Tesla full self-driving cars on the road. Full self-driving would be considered a narrow superintelligence. I would view FSD performance and capabilities based upon the number of standard deviations better than the average human driver on safety statistics.

Tesla FSD Beta has been made available to over 60,000 drivers. FSD will likely be fully available to over 240,000 prepaid US customers by mid-2022. FSD should reach robotaxi level by 2025.

Molecular Electronics

Will the AI software for protein folding create more value and a dominant position or will the value be from molecular electronics that enables thousands of times faster DNA, RNA and protein reading and synthesis? Roswell Biotechnologies has released a molecular electronics chip that integrates molecules with CMOS. I believe the power will reside with the vastly superior molecular nanotechnology providers and users instead of Deep mind Alphafold software.

Billions and trillions of molecule systems on integrated CMOS means molecular reading and synthesis at trillions of times the speed and scale of today. Rapid, low cost, mobile detection systems for diverse biomarkers. Enabling powerful, in-the-field pathogen detection, infectious disease monitoring, environmental monitoring, and identification of bio-specimens, species or individuals.

By 2030, Roswell Biotech should scale DNA, RNA and protein reading to be billions of times faster instead of thousands of times faster than today.

Atom Technologies – Better and smaller Atomics clocks, QuantumRF and More

ColdQuanta is unlocking atom technologies with laser-trapped atoms for quantum computers with millions of error-corrected qubits. There will be millions of ultra-precise atomic clocks which will transform GPS locations’ precision and master clocks for data centers. This will mean that Google’s Spanner database would be enabled to have hundreds to thousands of times greater speed or scale. They will have QuantumRF which will transform the energy efficiency and speed of communication.

Reusable Rockets

The SpaceX Super Heavy Starship will reach orbit by the end of this year.

In the most recent SpaceX Starship presentation, Elon Musk stated that the SpaceX Starship can become lower cost than airplanes for long haul cargo delivery.

I have some detailed calculations. Starship will double the payload capacity. Full fueling costs are about the same for the 1200 tons of fuel for Starship versus 50,000 gallons of jet fuel for a long-range cargo plane for each flight. However, twice payload means half the cost. Non-fuel costs for the vehicles and the frequency of flights are hugely in Starship favor. Starship should have twice the payload.

Starship starts off at half of the loaded cost per flight and four times cheaper per ton of payload and then will proceed to be ten to fifteen times cheaper. Plus they will have some initial new no-competition markets for ultra-fast delivery.


There are over 170 companies working aging damage reversal. First and second-generation versions of these treatments against each of the seven kinds of aging damage will be widely available and in use by 2030.

Aging damage repair treatment will be mostly periodic injections. If there are many companies each with their own injectable aging damage repair treatment then it will be like COVID vaccinations where there are many treatment options.

Currently, there are about 130 million births per year and about 60 million deaths per year. About 60% of the deaths are from old age-related conditions. About 18 million deaths (30%) per year are from diseases and conditions related to global poverty. About 2-3 million deaths are related to various kinds of accidents.

It would take time for successful antiaging treatments to get broadly deployed and to alter the amount of deaths per years by the millions. If there was a 5% drop in age-related deaths every year from 2025 onwards. This would mean 1.5 million fewer deaths in 2025, 3 million in 2026 and 9 million fewer deaths in 2030. This scenario would require major progress and success and against all seven types of aging damage.

Limited Aging damage reversal success (that would still be substantial) would mean the curing or substantial reduction in arthritis or aging eye diseases and other isolated aging conditions.

SOURCES – Roswell Biotech, ColdQuanta, SpaceX, Tesla
Written By Brian Wang,

51 thoughts on “Super Technology of the 2020s: Narrow Super AI, Molecular Electronics, Atom Technology”

  1. clean air, clean water, food, shelter OR clothing

    Good enough clothing and you don't need shelter, good enough shelter and you don't need clothing.

  2. Thanx! So, it is pretty close! I clearly was looking at a Webb L2 illustration and it was *not to scale*, as usu with the solar system. The L1 distance is very small, but even so the shade could not be a whole lot farther out.

  3. Well, sure: The Sun subtends about half a degree at Earth's orbit, which means that the shadow generated by the shade also spreads out by half a degree, and once it is far enough from the Earth that the Earth doesn't subtend a half degree, some of that shadow misses the Earth, you're obstructing light that would never have hit the earth.

    The way it works out, if the shade is further than about 1.5M km from Earth, some of the shadow is wasted. (Assuming a point sized shade, which wouldn't do you any good.)

    Happily, the L1 point is only 1M km from Earth, so a shade could be about 1/3 the diameter of Earth, and if optimally placed, none of its shade would be wasted. OTOH, you need to be further away than that, so that there's some solar gravity left over to aid in negating the thrust from the sunlight intercepted. The exact distance depends on the amount of light intercepted and the sectional density of the shade, but since we only would have to intercept a small fraction of the sunlight to put Earth back into energy balance, I think you could figure that all the intercepted light would have arrived at Earth. At least it's a reasonable approximation.

    This site explains it in terms of lunar eclipses, but you can see how it applies to the artificial solar eclipse we're trying to generate.

  4. The thing is, the optimal light sail will have 10 times the total solar gravity, without a load. So the shade must be intentionally heavy, more reason to use lunar resource. Also, seems like, but I have never seen an explanation, that the shade would/could be too far from Earth for the shadow to fall completely on the Earth, missing the point to some extent. This is because the Sun is not a point source. ???

  5. really? you should see the cubicle jockeys from the remote-work hordes who day-trade all morning, get their day-drunk on at 11a, and then rush to get real work and emails done by 9p in their jams – often only showering every 2nd day – perhaps a 2:15p powernap. The new work world.

  6. It's about choice and transparency. Many will choose not to be sucked into the rat race and will descend to lower class (complaining all the time of course). Tech will enable access not success – which is defined by competition, scarcity, and rank.

  7. The tech will not be well-distributed in any case, even in richer countries. Maybe by choice. Maybe by access. Maybe by station.

  8. The shade actually has to be a bit Sunward of the L1 point, so that the Sun's gravity predominates a bit, to compensate for the thrust due to light intercepted.

  9. yep. if it's good now, it'll be owned elsewhere and trickle-down will make it essentially common-place. There's no short-cut for lazy-based and unemployable people, places, and groups.

  10. Perhaps the hardware and its management need some type of highly-adaptable, epic-process-driven AI system (at its spine) – but most transactions are mobile and on smaller devices…

  11. agreed. tech and information support –and — security will be brudensome, expensive, and specilaized.
    These get, quick, rich and own all the pieces will never work.

  12. Its not about the tech as the users and facilitators.
    Aboriginals with bitcoin??– and a specialy-administered economic district??? What have a lot of the reserves and other special exemptyed areas accomplished? sideshows. casinos. tourist traps. Utter waste and under-accomplishment.
    Very few will be noticeably improved and enlightened by this tech and its decentralized foundation.

  13. I think it quite likely that it will limit itself.

    Suppose, in the simplest possible model, you had to take a pill daily. One of the prescription pills I take daily costs ten dollars per pill, and it is not the only prescription I take.

    But let's say we could get the price of the pill down to five dollars, around what a single drink at Starbucks used to cost. Not too shabby! I would be on it, I would buy it for my whole extended family if I had to. Still under a hundred bucks a day. A lot of us could still do it, would still do it, given its importance.

    But even if you limit your cost to just yourself and your spouse? Ten dollars a day, $3650 dollars per year.

    Now look at the world at large, where the world median income is more like twenty-six dollars a day. There is no way the bulk of humanity is getting it at this price.

    Draw up all the utopian plans you want. We can't even get one country to stay out of another, let alone approach the level of cooperation this would take.

    But this would all change if it was 30 cents per pill, you say? Right. But who is to say it won't be $35 per pill, or a hundred? Since this likely will require quite a bit more than just a pill (regular treatments, diagnostics, etc.) the price is more likely to be higher, rather than lower.

    There are also a lot of people who simply don't have the discipline to obtain and take a pill every day for some distant reward . . . or at all. We still have smokers for crying out loud.

  14. I did hear an interesting point, that these things would blow away with light pressure, as solar sails. The solution proposed was to make the *shades* actually transmit the light, just slightly refract/diffuse it, so most of the light hitting the shade misses the far away Earth. This places little force on the shades, so they don't move as much. It would be an initial attempt, with just very thin imperfect sheets of glass, from the Moon ideally. Your color idea may be accomplished with diffraction gratings, perhaps. Also, absorbing the light in a solar cell only imparts half the light pressure as reflecting it, and gives a power source for station keeping. edit: *fused* silica?

  15. Yeah, I'd misunderstood his point, which was economic, while I was thinking technology.

    But on a technological level I'd assert that the real trick to AGI has to be the automated generation of narrow AIs as needed, and managing them. That's how natural intelligence works, basically: As soon as you've really learned how to do something, you stop consciously reasoning about it, and it gets done automatically in the background. Generating an agent that can handle a task essentially IS what learning skills is all about. That's what practice accomplishes.

  16. I've actually proposed, if needed, that we put thin film spectral filters at the Earth-Sun L1 point; Plants use red and blue light, that's why they're green, due to reflecting the frequencies they don't use. Shifting the solar spectrum just a little bit in the direction of purple would avoid impacting energy available to the biosphere, while reducing insolation.

    The planet is only out of energy balance by a tiny fraction of a percent, you'd hardly have to intercept any light at all. Do it right and you could even deflect the green light intercepted into SPS's.

  17. The thing is, effective anti-aging therapies aren't something you'd take once and be done with. You'd be on them perpetually, like a diabetic taking his insulin. And they'd preserve your income potential.

    What's undesirable about that? A drug you have to take every day, that helps you have the funds to pay for it?

    Granted, they'd obsolete a lot of other treatments, but you'd really be hooked on them.

  18. There's still hope for cheap graphene hyper-capacitors. Recharging would be fast, but you might have to have a dedicated substation.

  19. You're right, that displacement doesnt neccessairly need to happen.
    Businesses use virtualize mainframe environment running cobol software written ages ago and have no intention of replacing it because it works. Does the AI in your toaster oven really need to also be a wiz in orbital mechanics, probably not.

    I think the CAIS model of AGI laid out by Drexler is likely to be the winning pathway.
    Humans creating millions of ANSI will be a challenge, the key to Drexler's vision is automating AI r&d and producing new ANSI at AI speeds.

    In short, you dont necessarily need a monolithic AGI, the millions of ANSIs with networked access in the cloud will collectively represent the functionality of an AGI.
    No need to worry about such a setup becoming self aware.

    Reframing Superintelligence
    Comprehensive AI Services as General Intelligence
    K. Eric Drexler
    Technical Report #2019-1

  20. The AI and bots that will make the vast majority of the human race unemployable doesn't currently exists. The current commercial tech isn't going to do it, people confuse the automation they see around them and the aspirational future tech the employment alarmists are actually referring to. Completely different "animal".

    You cant predict the timings of tech breakthroughs, but the only pathway that could definitely happen with evolutionary improvements to existing architectures is ANSI for all valuable human tasks. A integrated systems approach to the concept of AGI, assuming adding high level cognitive facilities to AI isn't viable in the near term.

  21. Bezos is the one they really hate, all socialists. He wants to make more land to escape to, and make a living. Trump was going to let Musk skate this. He did not even apply for a launch permit, iirc.

  22. "at a time when people seem strongly against breeding more people, because that's bad for mother Earth or something." Gerard K. O'Neill in "The High Frontier" 1976 gives the perfect reason to re-produce, unlimited resources. I have been telling people to have more kids all this time, but the people who don't tout O'Neill are overwhelming in number.

  23. "our decisions are made before we are consciously aware of them, and that
    our consciousness just gives us a rationalization for why we made them
    and, possibly, looks for lessons learned." And that is before you consider Repression and the patterns of the defense system, unique to humans in the developed form we support. "Once understood, the System must be destroyed." Janov 1970

  24. Effective antiaging therapies would devestate the existing medical establishment's profit models. Almost by definition, these therapies will be actual cures for chronic disease, like type two diabetes, and various autoimmune difficulties. Then there is end of life care.

    I can't wait to see how the politicians try to stop the whole thing.

  25. I suspect SpaceX, and other Musk associated corporations are the victims of political persecution. They are based in Texas, are non-union, hires the best candidates for technical position without bias, and as far as I can tell do not virtue signal to anyone.
    The FAA decided starship was too big for environmental reasons. When did an airliner ever face being refused clearance for it's size?

    The FAA keeps delaying it's "environmental" decision, Biden does not invite Tesla to BEV event. Tesla vehicles get lower tax credits as punishment for being nonunion.

    How obvious can it be, and the MSM still pretends not to notice?

  26. Following the long standing trend of pornography pushing early adoption of consumer technologies, I predict narrow superhuman AI dirty talking software. It will far exceed the naughtiness of human based dirty talking for common paraphilias. However, there will always be a niche market for human dirty talkers.

  27. You do know that the only reason anybody proposes Sulphur dioxide for that purpose, is that it's what volcanoes inject into the upper atmosphere, right?

    That doesn't mean it's the best compound for the purpose. Might be something a lot better, like TiO2, or fumed silica.

  28. I modified that statement to address your point and to try to be more clear about my position. thanks Brett.

    Any new entrant Super AGI or broader super AI will need to displace any market entrenched Narrow Super AI. The owner of a Narrow Super AI can broaden or improve its Narrow Super AI. Big Tech companies can also try to buyout or use mergers or partnerships with entrenched Narrow AI.

  29. Medical advances & globalisation have enabled humans to approach max lifespan of ~130 years BUT not 1 sec has been added to "extend" human lifespan.
    My 2c worth, genetic engineering may eventually enable longer lived humans & humans will become 2 distinct species: " naturals"- those staying organic & " synthetic"- those embracing AGI – HG Wells had the right idea but not in the way he envisioned.

  30. That's the weird/great thing about new technologies: new avenues for people to engage in trade that are unexpected. Humans will find a way to keep themselves busy by trading for new services/experiences/wants.

    The only things necessary for survival are clean air, clean water, food, shelter, and clothing. Most of the work people engage in are not directly related to those things. We have expanded our opportunities to trade with each new technology.

    Mass communications (radio, tv, internet, telephones) opened up so many new jobs in entertainment, finance, information sharing, etc. Powerful farm equipment opened up jobs in food snacks, restaurants, hotels, sports venues, bars, etc. Printing presses enabled newspapers, educational books, schools/colleges, religious vocation/instruction, magazines, journalism, etc.

    As Brian reports, technologies will enable advances in medicine, which has already extended human lifespans significantly.

    We will find a way to keep ourselves busy and advance our capabilities. It's what humans do…

  31. That's a good question. And who knows- it may turn out that a starship is the perfect platform from which to seed the upper atmosphere with sulphur dioxide to control heating of the planet?

  32. The biggest problem will be finding the human capital grokking that immense mountain of skills and data, the ones coming as result and required to have all those nifty technologies into market.

    And also filling up the rest of the economic roles required to maintain a functional society. From cleaners to cooks, to scientists and technologists, the world has never been so hungry for skilled people. Funnily, at a time when people seem strongly against breeding more people, because that's bad for mother Earth or something.

    Yeah, AI and robots will come to the rescue. Nevertheless I can't help but notice that every new technology and tool out there (included AI and robots), still requires a small army of trained humans to develop them and maintain them.

  33. I know thousands of rocket launches doesn't compare to the millions of flights that happen each year. Still I wonder: does the ultra high-altitude Starship emissions have any specific bad impact on the higher levels of the atmosphere?

  34. I don't disagree with any of this. It's pretty much in line with my own predictions for the past 20 years or so, when I realized that "AI" was way too broad a term with far too many meanings. There are probably at least a dozen definitions, and it causes confusion in any discussion of the subject where terms are not rigorously defined at the outset (which is rare).

    I do have some mild concerns that learning the nature of consciousness might be rather disturbing when it comes to learning about ourselves, but you can't swerve away from gaining knowledge.

    For example, there is some evidence that almost all of our decisions are made before we are consciously aware of them, and that our consciousness just gives us a rationalization for why we made them and, possibly, looks for lessons learned. Another being that perhaps we actually do die at the end of every Planck time and are reformed by a propagating wave, almost a soliton, through space-time, that gives us the illusion of unbroken existence, much like a celluloid movie would have us looking at a blank screen most of the time.

    Even so, I wouldn't mind a few centuries to try and figure it all out with my strong AI friends to help (and my weak AI devices to take care of everything in life that doesn't require an IQ of above 150 or so).

  35. Does a pocket calculator count as Artificial Narrow Intelligence? What about a slide rule?
    Either way most calculations are done much faster than human mental or pencil & paper calculation.

  36. There is a weird glitch on the previous post about Starlink.
    When I click on show comments I just see it telling me there are 13 comments before they disappear.

  37. So many of these things will work only in micr0g, or far better in micr0g. Plus a lot of stuff we have not thot of yet as we are only now starting to experiment with the advantages of micr0g, rather than the physiological disadvantages of humans unnecessarily in micrf0g, on a purported trip to Mars.

    "It can generate folded structure predictions that can then be used to solve experimental structures by crystallography" is a great example. Micr0g crystal growth is the first big thing along these lines, I believe. This would include alloys, broadly defined. Many of these things will likely need ISM and thus ISMRU to be done at all, in Space, at scale. Not on Mars or the Moon, or Earth. "Is the surface of a planet the right place to do micr0g?" This is a major tech of the 20s, already happening on the ISS. The astronauts doing the lab assistant work are very excited! We should be too. The *gold* in this gold rush IS the micr0g.

    edit: another example, the ColdQuanta tech seems like would be highly advantaged by not having to support the atoms in g, with the lasers. A very gentle laser would hold the atoms in place, cold and fast.

  38. "Any Super AGI or broader super AI will need to displace Narrow Super AIs in every valuable domain or market."

    Not necessarily. An AGI can simply *use* narrow AIs. Your own brain doesn't solve everything with general intelligence, after all. It solves very little with general intelligence, as a matter of fact; It relies on a series of domain specific intelligences tied together with a bit of general intelligence glue.

    Imagine if you had to reason out how to take each step as you walked, or how to construct a sentence as you talked. It would be nightmarish.

    What general AI needs is the ability to generate domain specific AIs, as it needs them.

Comments are closed.