The coming Wave by Mustafa Suleyman

A “… wave is a set of technologies coming together around the same time, powered by one or several new general-purpose technologies with profound societal implications …”.

Proliferation is their default behaviour, and it is given a boost by demand (as people realise the value of the tech) and cost reduction (due to people tinkering).

The most powerful of these waves has been computing which “… transformed society faster … and proliferated faster than any invention in human history …”.

The reason technology keeps on progressing is because people like money (and other selfish rewards) and do not like hardship; and technology is a way to lessen hardship while making money.

General purpose technologies are those that have wide application.

They can be made more specific through combination with other general purpose technologies.

They quickly become invisible to us as they permeate our lives.

Twenty four is the supposed number of general purpose technologies that have emerged in the entire history of humankind.

On the foundation of these were built all our specific technologies.

GPTs become waves when they diffuse widely. Without an epic, near-uncontrolled global diffusion, it’s not a wave, it’s history.

GPTs have such wide impact that “… its makers quickly lose control over the path their inventions take once introduced to the world …”.

The author gives a few examples of how well meaning inventors’ had their inventions appropriated for less than noble uses.

Even though history has many who tried to stop the spread of technology they have been unsuccessful, and in fact the time they spend trying to limit the spread of tech. eventually comes to bite them back as missed opportunity for societal progress.

The author cites examples through history.

They also mention that nuclear weapons are an exception where the spread of technology has been contained because of mutually assured destruction (and luck).

Being a fringe technology from the 1950s to 2012 and then accelerating with breakthroughs such as AlexNet and Deepmind DQN with huge investment by companies such as Microsoft and Google, catalysed through 5G, availability of large training data and increasingly democratised through Meta’s PyTorch or OpenAIs APIs – “… mass-scale AI rollout is already well underway …”.

The author says that “AI really isn’t “emerging” anymore. It’s in products, services, and devices you use every day …” and “Within a few years AIs will be able to talk about, reason over, and even act in the same world that we do.“.

This blows my mind. Unprecedented is not the right word, I’d call it alien.

I am truly in awe of the world we are getting into. And not only is AI getting more powerful, it’s also getting more efficient in terms of model size and cost.

For example, the author’s current company, Infection AI, “… can reach GPT-3-level language model performance with a system just one twenty-fifth the size …”.

It is normal for technology to get efficient with time, but when I think that GPT-3 was just a year old when the author wrote this book … I’m impressed at the speed of progress.

But when I close my laptop and come back to real life and see that not much is different.

But this, the author reminds me, is because our prehistoric brains are bad at dealing with diffused development like this, he compares it to climate change, another big threat facing us, where too we are looking in the other direction.

Suleyman implores that we stop worrying about AI sentience and focus on what it can do right now

The author calls discussions on AI sentience, singularity and other such sensationalist topics, red herrings.

Is a term coined by the author that represents the middle step between the AI we have today and the AGI we’re promised tomorrow.

Simple (or “narrow”) AI today is good at working at a specific task. ACI will be good at that AND connecting the results of that task to more broader ends.

The author gives the example (calling it the “modern Turing test” to check if an AI is ACI) of an AI being given the task of “Go make $1 million on Amazon in a few months with just a $100,000 investment”.

Such abilities are outside of today’s AI but the author says they’re coming “… thousands of these models, and they will be used by the majority of the world’s population.“.

The human race is about to get a bunch of help at a lot of tasks, for both good and bad.

We’re making real progress in being able to safely and intelligently modify/edit DNA (human and others).

DNA sequencing has seen a million-fold fall in price over the last two decades. CRISPR is here and helping people with erstwhile incurable diseases like sickle cell.

And you can also buy literal DNA printers today and start making new strands at home. If this is not playing God, then I don’t know what is.

Personalised medicine does not seem far fetched and neither does literally turning back ageing.

Robotics

They are already in farms, warehouses, and operating rooms.

What is to come is when large number of robots start to coordinate (aided/controlled by AI) and start doing still like “… build dwellings in a matter of days for a fraction of what traditional construction might have cost“.

Robots can be considered the physical manifestation of AI.

Clean, abundant energy

According to the author “Renewable energy will become the largest single source of electricity generation by 2027“.

While solar and hydro are the popular choices today, the author brings out nuclear fusion as the energy source of the future.

A few decades into the future there is hope of decentralisation of energy grids and energy independence.

Nanotechnology

That too at the atomic level (a typical atom is about 0.1 to 0.5 nanometers wide) where atoms are manipulated individually to assemble into almost anything.

This is the most far out though and maybe only will be seen by the end of this century.

Quantum computing

This is still some time away but promises multiple orders of magnitude improvement over the computers we have today.

And with such power we can perform extremely high fidelity simulations that have a ton of real world applications like modelling chemical reactions, protein folding, rush hour traffic to arrive at real time local optima. Bad things can also happen like Q-Day.

I’d personally call this “democratisation and decentralisation”.

The coming wave of technologies is going to wrest power and control away from central operators like we’re already seeing examples of with the military ($3M patriot missiles vs $100 DIY combat drones) and publishing (newspapers vs tweets).

The wave will give immense power to motivated individuals.

The author explains this concept well with a simple example: “… no matter how wealthy you are, you simply cannot buy a more powerful smartphone than is available to billions of people … same billions will soon have broadly equal access to the best lawyer, doctor, strategist, designer, coach, executive assistant, negotiator, and so on … If everyone has access … also includes those who wish to cause harm“.

Computing speed and its impact on software capability have been rapidly developing since the 80s, but development in other fields which may seem to have slowed down is going to get supercharged by the coming wave.

Objects in the physical world derive their value from the sophistication of their design more than the raw materials they are made from (the author gives the example of a Ferrari).

With AI helping the creation of new materials and powering things like 3D printers we will be seeing increasingly sophisticated design here on out. Already there are examples like some car parts designed with help from AI.

We may think of technologies as having dual uses such as a kitchen knife that can be used to chop vegetables and also stab someone.

But this definition is too black-and-white when it comes to AI.

Instead the author prefers “omni-use” which is more general and its benefits or drawbacks runs across a spectrum.

For e.g., should we use an AI that helps discover new compounds for better drugs but that can also help design toxic molecules if used by nefarious actors.

The previous three points are applicable to all technology that we’ve seen through history. Autonomy, however, is new to the coming wave.

Being autonomous means that technology does not need to ask humans before “taking action”. “… a world where constant intervention and oversight are increasingly unnecessary. What’s more, with every interaction we are teaching machines to be successfully autonomous“.

The pros and cons of this are easy to imagine.

And so the author asks a more central question around containment: “Will new inventions be beyond our grasp?“. Since autonomous technology can build new things on its own (like an AI improving its own code), for the first time in our history there will come a time where no human (or a group of humans) is able to explain how something works and why it did what it did.

This is, of course, the main reason why AI and the coming wave cannot be contained; we’ll basically be too dumb to contain it once it starts improving itself.

“Ultimately, in its most dramatic forms, the coming wave could mean humanity will no longer be at the top of the food chain. Homo technologicus may end up being threatened by its own creation“.

AI advancements over the last decade have not gone unnoticed by nations and many are putting resources into technologies of the coming wave because “… choosing to limit technological development when perceived adversaries pile forward is, in the logic of an arms race, choosing to lose“.

There are four quarters where growth is coming from: US, EU, China and India.

China is a notable example for having surpassed US and other countries in AI R&D and other fields like EVs, batteries, robotics, 6G, quantum computing and bioscience. It’s the country’s explicit goal to become the global leader in AI by 2030.

The thinking that China is only good at imitating and making inferior copies is outdated.

UK is regarded for its strength in AI, Japan, Germany and South Korea for robotics, Israel for cybersecurity.

Basically, the challenges that face humans today are too big for them to solve alone.

Also, technology has percolated so thoroughly through every part of modern day living that it is nigh impossible to avoid it or give up its fruits.

And incentives are such that they nudge more of it to be around. So techno-denialism is naive. The author introduces the “great dilemma” where we both need coming wave technologies to maintain and improve standard of living, and are also threatened by its disruptive capability.

Damned if you do, damned if you don’t.

The author calls the social contract by another name, “the grand bargain“. He says that the coming wave is going to threaten this, that too in just the next 2 decades (really? that soon?).

If I consider the four pillars of the wave outlined above

• Asymmetric impact and omni-use means the state will have a hard time controlling bad actors.
• Fast evolution and autonomy will keep the state perennially two (or ten) steps behind.

The nation state relies on shared imagination of control lying with a central overseer (which then transforms into real control as the overseer takes action) and since the coming wave can potentially empower bad actors to wrest that control away, yes, I can understand how the nation state can be undermined by the coming wave if left unchecked (like crypto-currency and its relationship with central banks).

Allowing this to happen will, according to the author, be a really bad idea.

“An influential minority in the tech industry not only believes that new technologies pose a threat to our ordered world of nation-states; this group actively welcomes its demise. These critics believe that the state is mostly in the way. They argue it’s best jettisoned, already so troubled it is beyond rescue.“.

Not that hoi-polloi are in favour of government either. The author says “… trust in government, particularly in America, has collapsed“.

And cites Pew Research results, growing inequality gap, Democracy Perception Index poll and other factoids.

So it would appear that even without the coming wave, nations of the world have a lot of trust issues to deal with.

Think about the parallels between what tech. firms offer and what the government does.

Both provide a range of services, both don’t charge a lot of money for those services (vs the value we derive from them) and both deeply impact lives of billions of people.

Companies like Apple, Google, Microsoft and Amazon have cash reserves the size of countries.

This is going to exacerbate with the coming wave as the “superstar effect” makes these large companies even larger.

Another thing that governments do is focus more on services than goods, the author expects this too to be the behaviour of future tech.-govt. entities, already we see the subscription business model that many tech. companies choose to focus on today instead of offering one-time purchases.

The same technologies that can potentially topple governments can also be used by them to tighten control if intercepted at the right time.

China comes to mind as a modern surveillance state that can enhance control further using the coming wave but honestly a lot of countries are going to do this.

Surveillance is lucrative business and not only for the state wanting to monitor its subjects, but also for the enterprise wanting to know what its employees are up to.

AI massively aids in surveillance by removing the human effort required to parse through the video (and other kinds of data) streams to identify “entities of interest”.

We already see this happening (sometimes without the subject knowing) in places like China and the United States.

A few times the AI does get things wrong like this time and this time.

But as algorithms improve it will get harder to plead innocence and claim that the machine got it wrong.

Almost everything these days is running on software and everything is connected to the Internet – a situation ripe for a cyberattack, but not a basic run-of-the-mill cyberattack.

But rather, one that is powered by AI, learns as it attacks systems, patches up its vulnerabilities and actively stops attempts at shutting it down.

Robotics (incl. UAVs) when powered with AI, can be used by actors to cause harm and/or surgically disable key state actors. Example Mohsin Fakhrizadeh.

The goal need not even be to convince (that the other side is right) but to confuse so that decision making is paralysed. This will be a real challenge for states in the decades ahead. Even private firms need to watch out. And regular people too.

Well meaning research has gone bad from time to time; like the Russian Flu of 1977. The risk is increased if AI and DNA printing get in the wrong hands and are used to synthesise molecules with devastating effects.

AI led automation will lead to mass unemployment and unlike the past where older jobs were replaced by new ones, this time it is not going to happen because AI will take up those new jobs as well.

This will lead to social unrest and volatility, and governments will lose some portion of tax revenue which means more public services will not be available leading to additional unhappiness with the state.

The immense democratisation of power can lead to groups of people saying “we don’t need the government any more”, and not in a rebellious way but matter-of-factly and literally.

The state may unwillingly (or willingly) abdicate its role in our lives as we hire erstwhile state provided services like health, security, transport ad hoc from technology firms. This is the off-grid movement supercharged.

As AI, 3D printing, solar PV systems, robotics, satellite internet, VR and such enable people to be increasingly self-reliant and nudge them to intelligently wonder “seriously, what does the government do for me again?”.

As I wrote before, the author thinks that stopping the coming wave is not possible.

Plus, given the benefits technology we may not even want to stop it. So, given that this is coming, how should we prepare? The answer is containment.

It means carefully orchestrating the development, spread and adoption of new technologies so that we only get the good parts and do not get the bad parts.

Because “Technology evolves week by week. Drafting and passing legislation takes years“, because “Even technologists and researchers in areas like AI struggle with the pace of change. What chance, then, do regulators have, with fewer resources?“.

We need Public private partnership!

Yeah, what were you expecting? A miracle cure?

The author says that even powerful groups of governments cannot effectively contain the wave.

Instead what is needed is partnering with the very companies that are creating the wave.

And being smart about it

The author says that there’s not a blanket containment strategy for each kind of technology, instead the approach needs to be tailored for each.

For this he gives a list of nine questions that policy makers can ask to make a better containment plan per technology.

The author also gives ten steps that humanity can take towards effective containment starting from the most basic step of making sure that the AI is trained on representative, its biases weeded out and a reliable kill-switch (a very important thing, as I have learnt from the game Horizon).

Additionally, the following are suggested:

1. Having governments employ in-house AI experts paid at the same level of their corporate counterparts.
2. Requiring license(s) to create specific AI agents.
3. Compensating those who lose their jobs because of AI.
4. Global treaties around AI.
5. Greater public participation.
6. Outright apocalypse preparation in case things still go sideways.