💡E02: Three Underrated Technologies from the Future: Brain Recording; Deep Geothermal; IVF + Updates BCI; Clean Meat; LEO Internet++

Your guide to deep tech written by a human. Please excuse poor content, grammar and typos

May 31, 2023

gm as the kids say. as I always say: you get into crypto in your 20s; and rare earth mining in your 30s. If I’ve learnt anything, it’s that technology is eventually about digging stuff out the ground.

This week you get some post from the mailbag; the three most underrated technologies I came across on my travels: brain recoding, deep geothermal energy, and assisted reproductive technologies (IVF, etc). Interviews with Julius Kümmerle, CEO of SafeAD on the State of Computer Vision for Autonomous Driving and James Dborin and Meryem Arik of TitanML on the State of Neural Network Compression. And news on brain-computer interfaces, clean meat, and LEO satellite internet. Stick around next week for the three most *overrated* technologies. (Hint: fusion and bitcoin, sorry not sorry).

🚁— Lawrence

❌ Mailbag

As you know, last week I launched the State of the Future site, which probably should have went out months ago but I wanted it to be as accurate as possible. I shouldn't have worried. The Internet has been kind enough to tell me *exactly* what I got wrong. Keep it coming, it makes me stronger like Learning From Human Feedback (RLHF). The main comments were about DAOs, fusion and eVTOLs.

DAOs

Despite my Bitcoin scepticism and the short-term adoption path for Web3, I’m still in camp “DAOs as global resource co-ordination structures”. A bit abstract yes, but it’s accurate. Last year I wrote (and yes, I am going to quote myself in this Substack, if Ben Thompson can do it, so can I):

“So look, TheDAO was an experiment, and in the last 6 years has proven that it’s viable. We have a new digital-only structure to coordinate capital and labour. Today they are only usable by crypto people. Samudai is building a platform to bring DAOs to the masses and by doing so ushering in a new era of configurable labour legos. And look I’m not saying we’re reinventing capitalism out here, but I mean, what else are we doing if not changing the balance of power between capital and labour?”

The last year, especially the issues around liability has changed my view on timeline, but not on the overall impact. Anyway Navin from Samudai says:

“I’m seeing a lot of new serious research oriented and community focused DAOs (eg ValleyDAO, ResearchHub DAO) and validates the DAO potential in scientific research as well as academic research funding”

I sort of half knew this because of VitaDAO (Book recommendation: Jellyfish Age Backwards) but I think this is an interesting pathway for DAO adoption that actually adds value rather than is a massive crypto hype train. More of this please.

Nuclear Fusion

Yeah I know man. Look I was super bullish on fusion, unless I spoke to people in the industry. Many of these interviews I’ll publish on here soon. But suffice to say, there is a pretty strong disagreement about the timeline on fusion. Startups and VCs are talking early 2030s and experts/people working at ITER/others say 2040s. From the perspective of my children, both are good outcomes. For a VC one outcome is money, the other is no money. So people are pretty opinionated. It’s another one of those binary polarised debates like bitcoin or AGI. It’s either vapourware/fraud/snakeoil or it’s the future of money/humanity/energy. I’m honestly not overly opinionated, but as of Wednesday 31 May, I am weighting the assessments of fusion experts over startups/investors. Anyway more specifically on this next week, but one useful datapoint this week was from the Journal of Fusion Energy:

“To be precise, we should now say “fusion was said to be 19.3 years away 30 years ago; it was 28.3 years away 20 years ago; 27.8 years away 10 years ago.” And now, scientists believe fusion energy is only 17.8 years away.”

Andrew Holland, CEO of the Fusion Industry Association told me on Twitter:

“Well. 17.8 years away is the middle of the consensus - with key decision makers (including, notably the investors and the U.S. government) saying its much faster.”

Adding:

Any timelines after five years can be extended or shortened with more resources.

I’m not convinced, if someone can tell me how we solve the first wall problem in less than 10 years then I’m listening…

eVTOLs

Maciej Wolski on Linkedin:

"No eVTOLs in market today...", 2030 prediction: "By 2030 no customers have paid for an eVTOL journey" - have you heard about Jetson? They have already sold many of their machines. Even if this is just the beginning - until 2030, I would expect many eVTOL journeys completed." Currently, the price is rather high to justify normal traveling and the time of flight is also relatively short - so it is time for early adopters. But as you can see on their website - the entire production is sold out. And this is just the beginning, I believe by 2030 we will have such flying machines for more than one person and with longer flight ranges.”

I’m open to being pushed on this one, I am not as confident on the prediction as Fusion, but I really don’t see the economics of this working out. It’s plausible there will be a few early adopters that currently use helicopters. Maybe some tourist flights. But in less than 7 years I can’t imagine how the price will be competitive with all the different mobility options. I am more closely aligned to Horace Dediu’s vision of micro mobility than flying taxi’s at least for the next decade.

Finally, I was speaking to reader and renaissance man, Dimi De Jonghe about creativity in the age of LLMs. If “content” (text, images, code, music, soon video: See Ben Mildenhall on NeRFs) is about to increase 10-100x then what are the implications for human creators? As a “writer”, I think it’s about being more human. Having a style, a tone, a voice. So I am throwing off the shackles of Grammerly and ChatGPT and going rogue.

“Written exclusively by me. Please excuse poor content, grammar and typos’

✍️ Three Undervalued Technologies from The Future

As much at State of the Future is supposed to be a methodical approach to comparing technologies, sometimes things don’t add up. I actually think the interesting stuff that comes from the work is the stuff that doesn’t quite fit in. It’s the stuff that takes longer to write about. Or I stay awake thinking about. You sort of have to do the work to know what that *feels* like. It's like an itch. This intuition is probably a function of experience or simply data.

Morris Clay

and I have spoken about this a lot as he build the VC Tech Stack. There probably is a way to formalise this into a prompt and automate it in some way. But until then. Here are the technologies I think at undervalued (not in a quantitive price/valuation/media coverage sense, but more in the Conversations with Tyler sense).

Decoding speech from brain waves with 75% accuracy

Until recently, any brain-recording technique relied on surgical interventions like stereotactic electroencephalography (EEG) or electrocorticography (ECoG). The skull is pretty effective at blocking signals from the brain, so sensors must be embedded on top of the brain to get a clear signal. The holy grail for decades has been to get accurate signals without surgery. This would mean we could run more experiments collecting vastly more data. Last year, in a relatively underreported story, using non-invasive electroencephalography (EEG)and magnetoencephalography (MEG), Meta decoded speech from brain activity with up to 73 percent top-10 accuracy from a vocabulary of 793 words from just three seconds of brain activity.

Meta and others are now working to extend the work into speech production, enabling true brain-computer interfaces. I think we have Crossed the Rubicon. Caeser would soon become a Dictator. Combine this with MetaAI and Oculus, and I dunno, Zuck certainly knows his Roman history…

Anyway, Meta achieved these results with data from fewer than 200 people. With faster and cheaper experiments, EEG and ECoG devices and brain-decoding software will get cheaper. More data fed into more powerful machine learning models will enable us to decode a broad range of neural activity. Neuroscience has been pulled firmly into the realm of AI. We have been surprised by the capabilities of large language models when trained on text and images. I can’t help but think there is something quite profound brewing with brainwaves-to-output models. The key thing for investors to think about here is the interface between invasive BCI-as-medical-device and non-invasive BCI-for-commercial-applications. In the short-term <3 years, the medical device pathway is the only route to market. This is costly and has a limited market potential. The bigger opportunity is for commercial applications like VR but there is no go-to-market for that yet. Timing risk.

Go Deeper » Brain-computer interfaces

0.1% of Earth’s heat could supply our energy needs for two million years

Our energy needs can be pretty much solved if we can just dig a little deeper into the earth. We already tap the Earth’s crust for energy, although mainly where it breaches the earth, in the form of geothermal energy. 27 countries already generate about 15.4 GW of electricity. But recently engineers have begun to explore the possibility of digging deeper. The decay of naturally occurring radioactive elements continuously replenishes the heat at a flow rate of roughly 30 terawatts. This is almost double all human energy consumption. And this process will continue for billions of years. And more than that, it could be the cheapest of any renewable. We already drill to a depth of around 200 meters and generate about 15GW of electricity annually through geothermal power plants. However, that is limited to areas where hydrothermal resources are readily available near the surface. Accessing heat from deeper into the earth’s crust is the realm of deep geothermal energy.

There are various types of deep geothermal technologies: enhanced geothermal systems (EGS), super-hot EGS, and advanced geothermal systems, all of which extract heat from resources with varying temperature, depth, well permeability, and reservoir permeability. But they face the same problem, drills don’t work at temperatures deep into the earth. Pulse-power and other technologies are being developed to replace diamond-based drilling equipment. Although there are numerous engineering challenges to overcome and a societal aversion to fracking, it’s remarkable that we are only a few engineering challenges away from an abundant renewable baseload resource requiring no new science. The Oil & Gas industry has millions of people and trillions of dollars to solve the challenge, and they are doing so as a side effect to chasing fracking tech.

I predict that the conversation around deep geothermal will change as the cost of the energy transition becomes starker, and over the next few years, it becomes clear we are falling further behind on the 1.5%, and maybe the 2% target (See the latest IAE report). The conversation around energy moonshots will accelerate. Yes, solar, wind, and battery prices continue to plummet (seriously, the decline is unbelievable).

Credit:

Azeem Azhar

But we need to go faster. Deep geothermal balances capital requirements, probability of success, and capacity compared to alternative moonshots like geoengineering, space-based solar power, and nuclear fusion. And aside from some engineering work on drill bits (diamonds break at the depths we are talking about), it’s really a strategic decision from the O&G majors.

Go Deeper » Deep Geothermal

2% of all babies born worldwide are via IVF

People sort of already know the global fertility rate is falling, and that it’s a bad thing. Generally people are aware that ‘demography is destiny’ but I don’t think it’s really understood how much of public policy is downstream of fertility. Fiction does a better job of telling that story — Children of Men and The Handmaid’s Tale spring to mind.

So with that dystopian background, know this:

the global fertility rate was 2.5 children per woman in 2019, down from 4.7 in 1950.
It is forecast to fall below 1.7 by 2100.
One in seven couples has trouble conceiving, and 48 million couples and 186 million individuals live with infertility globally.
These numbers are increasing yearly, with male reproductive problems increasing by about one per cent annually in Western countries.

This is a slow-moving crisis. The simple truth is that we can’t allow the fertility rate to fall to 1.7 without having to rethink how our societies are organized entirely. And we can’t allow reproductive problems to increase by one per cent annually. Not only is this a human tragedy, but a falling population shrinks the economy and can cause societal instability. I think the shrinking fertility rate is as much of a threat to human thriving as climate change, and one that gets far less attention.

We are already applying technologies to address the problem. In 2018, eight million children were born worldwide using IVF and other assisted reproduction techniques. Over 2.5 million cycles are performed yearly, resulting in around 2% of all babies. And IVF is getting better with live birth rates by cycle five, increasing from 76% in 2005 to 80% in 2010. Further improvements are combining advanced computer vision and genomic analysis, known as pre-implantation genetic diagnosis (PGD), to improve embryo selection. Synthetic embryos, where stem cells are used instead of eggs or sperm, are controversial, but progress continues. This line of research could one day help reduce failed pregnancies.

Despite the scale of the consequences, progress on the technological side is slow. And even pro-natalist policies are less effective than in the past. Despite the Chinese state shifting from a one-child policy to a two-child policy in 2016 and a Three-Child Policy in 2021, fertility rates remain stubbornly below two.

There are few immediate catalysts suggesting we can halt the declining fertility rate. The best levers are tax and government spending to make it cheaper to have children. Longer-term, we need to understand better why infertility is increasing and fix the biological, chemical, and societal problems. One *speculative* saving grace may be artificial intelligence as a replacement for unborn human workers. This may make the dependency ratio redundant, fixing the economy but profoundly affecting society.

Go Deeper » Assisted Reproductive Technologies

Share State of the Future

☎️ Interviews

Julius Kümmerle, CEO of SafeAD on the State of Computer Vision for Autonomous Driving [Computer Vision]

I liked this line: “We figured that autonomous driving today is not so much about that big dream of a fully autonomous car driving through Phoenix, where Waymo showed that cars could drive autonomously under very specific conditions. In our view, for now, it’s more about making existing cars safer through driver-assistant systems and achieving partial autonomy, like Mercedes did for level 3 autonomous driving.” Pragmatism after years of hype.

James Dborin and Meryem Arik of TitanML on the State of Neural Network Compression [LLM]

A particularly relevant interview as we think more about the future value chain of LLMs as “The One LLM to Rule Them All/General Models” versus “Small models, fine-tuned locally”. Since the leak of LLaMA there has been so much work on running the model on laptops and smartphones. It sort of feels like the Internet vs propriety models, and betting against the Internet is a bad strategy. Anyway making models smaller with compression is one pathway to reducing operating cost of running these models. We’re also interested at Lunar in new semiconductor designs that effectively do the same but being more energy efficient. ASICs yes, but also analog and neuromorphic designs too.

🗞️ News

Neuralink: Elon Musk's brain chip firm says US approval won for human study [Brain-Computer interface], [Neuroprosthetics]

Elon Musk's brain-chip company, Neuralink, has announced that it has received approval from the US Food and Drugs Administration (FDA) to conduct its first tests on humans. The company aims to restore vision and mobility by connecting brains with computers. Neuralink's microchips, tested in monkeys, are designed to interpret brain signals and relay information to devices via Bluetooth. However, the company does not have immediate plans to start recruiting participants. An earlier attempt by Neuralink to win FDA approval was reportedly rejected on safety grounds. Other companies in the Brain-Computer Interface (BCI) space include Kernel, a startup developing non-invasive brain recording technology, and Paradromics, a company working on high-data-rate neural interfaces. Synchron, another key player, has developed a stentrode, a neural interface that can be implanted without open brain surgery, and has received FDA approval for clinical trials.

The thing I think about most here is the pathway from medical device to consumer device. Any invasive BCI is obviously a medical device and will be amazing as a neuroprosthetic for treatments of neurological disorders and injuries. But I imagine Elon's and most investor interest is for the commercial side. The vision of a BCI enabling thought-controlled interfaces. Obviously all of the work on a medical device pushes forward the field in terms of the neuroscience but has a relatively impact. I am personally more interested in Meta's work on Using AI to decode speech from brain activity and other projects collecting EEG data.

Brain implants help paralysed man to walk again [Brain-Computer interface], [Neuroprosthetics]

More on BCI this week, in this case, a40-year-old Dutch man, Gert-Jan Oskam, who was paralysed in a cycling accident 12 years ago, has been able to walk again due to electronic brain implants. This medical breakthrough, still in its experimental stage, wirelessly transmits his thoughts to his legs and feet via a second implant on his spine. This represents a significant advancement in the field, building on previous research involving spinal cord stimulators and brain-computer interfaces, which have been used to restore some motor function or control external devices. The development was led by Swiss researchers, with Prof Jocelyne Bloch of Lausanne University carrying out the delicate surgery to insert the implants. After a few weeks of training, Oskam could stand and walk with the aid of a walker. The system is still at a basic research stage, but the aim is to get it out of the lab and into the clinic as soon as possible.

This work is more of a significant step in BCI development. This serves as a robust proof of concept, showing that it's possible to decode brain signals into complex motor actions like walking. This success could accelerate BCI development by attracting more funding and interest. More importantly, this development could pave the way for regulatory approval and more widespread use of BCIs. However, the technology is still experimental and bulky, and can only be used for limited periods.

New Study Is Extremely Embarrassing for Lab-Grown Meat [Clean Meat]

A study by researchers at UC Davis suggests that lab-grown meat's environmental impact could be significantly higher than that of retail beef, based on current production methods. The researchers found that the global warming potential of lab-grown meat, measured in kilograms of CO2 emissions, is between four and 25 times greater than the average for beef products sold in stores.

One of the main reasons for this is the need for highly-refined growth media, which allow cells to multiply in a lab setting. However, the researchers suggest that there could be effective ways to reduce this carbon footprint significantly in the long run. If lab-grown meat companies were to switch to food-grade ingredients, the global warming potential of cultured meat could end up being anywhere between 80 percent lower to 26 percent higher than conventional beef production.

I'm already pretty bearish on clean meat: "By 2030 clean meat disappoints, contributing <0.1% of protein intake", so this work potentially validates the assessment. This study highlights the issue of pharmaceutical-grade media which is currently required for optimal cell growth and safety requirements for regulatory approvals. An interesting opportunity in the space is enabling this shift from highly-refined to food-grade media which would reduce environmental impact and cost.

NASA’s Laser Link Boasts Record-Breaking 200-Gb/s Speed [LEO Satellite Internet]

Researchers from NASA, MIT, and other institutions have achieved a record-breaking space-to-ground laser-communication link with data rates of 200 gigabits per second. This rate allows a satellite to transmit over 2 terabytes of data in a single 5-minute pass over a ground station. The TeraByte InfraRed Delivery (TBIRD) system, which orbits about 530 kilometers above Earth, made this possible. TBIRD achieved downlink rates of up to 100 Gb/s with a ground-based receiver in California, 100 times faster than the quickest internet speeds in most cities. Despite the challenges of using terrestrial components in space and the distortion of laser beams from atmospheric effects, the researchers developed solutions to ensure successful data transmission. The team is now exploring where to apply this technology in upcoming missions and how to extend it to different scenarios, such as geostationary orbit and future missions to the moon. The technology may also find use in high-speed atmospheric data links on the ground.

I am sceptical of the commercial viability and market opportunity of LEO Satellite Internet. It just seems difficult to compete with terrestrial communication, simply because space is further than the earth and so will always be more expensive. That said, a 200 Gbps laser link would certainly improve satellite internet services, offering faster download speeds and lower latency. I still think precise pointing and atmospheric disruptions are hard engineering challenges, but people said the same before beamforming. The immediate applications will be for a space-based data economy in which faster speeds will allow for more complex computing and sensor capture. I do wonder about inter-satellite links too for things like data centres in space.

Right, that’s your lot. Leave a comment and give me some RLHF like one of your French models.

State of the Future