What is the singularity?

Before I can make any assertions about an event, I have to define the event beyond just a buzzword label. The technological singularity is a hypothetical event whereby technological advancements compound and snowball. By this definition, you could easily assert that we’re already in the singularity. Computers, for instance, accelerate business, research, and technological deployment. This, in return, accelerates the advancement of computers. It’s a positive feedback loop.

So what sets apart The Singularity from today?

My personal definition is: When my life drastically changes.

Sophisticated, I know. But there’s some value in evaluating global technological shifts from the subjective perspective. The singularity is fundamentally about disruption. Disruption of economics and commerce, disruption of lifestyles and life trajectories.

Key disruptors

Deep learning

The internet is presently losing its composure over GPT-3. An old college buddy of mine works at OpenAI and, almost two years ago, he told me that they were getting to the point of fundamentally understanding the nature of knowledge and intelligence. I didn’t really believe him at the time, but I also didn’t understand the groundbreaking Transformer architecture upon which GPT-3 is based.

I first got my hands on a Transformer in the form of Google’s Universal Sentence Encoder. This technology alone will disrupt the world once it gets fine-tuned and deployed globally. How? Why? That’s a post for another time. Suffice to say, I now agree with my buddy - we are starting to grasp the fundamental nature of intelligence.

I have been listening to podcasts and YouTubers talk about GPT-3 and even a lot of tech savvy folks can’t seem to make heads or tails of it. They make assertions like oh, this is just some sort of compression. Is it actually processing and thinking? Or is it just storage?

It’s actually both.

I first started tinkering with neural networks way back in 2009 and 2010, before the current wave of deep learning. Back then, I was experimenting in C++, manually creating what are now known RNNs and such. I had no idea what I was doing and was way in over my head, but it’s nice to know that I was just a couple years ahead of the curve. During that time, I had some fundamental realizations about neurons (biological and artificial). Neurons (or nodes) are responsible for both memory and processing.

For conventional computing, it’s difficult to grok the idea that a unit does both storage and computing. But that’s just how it is. There is no hard drive in your brain, no CPU. You have gray matter (the “thinking” neurons) and white matter (the wiring) and a few regions that tend to specialize. None of those regions specialize in either processing or knowledge, though some do specialize with forming cohesive memories.

But I digress. The point is that deep learning can and will disrupt pretty much every domain of human expertise.

Medicine

Diagnosing patients is fundamentally a pattern matching problem. Deep learning is the most sophisticated pattern-matching tool we have, regularly outperforming humans once you have the right training data. I suspect that we will see early disruption in rural and poor regions of the world where need for medical care trumps the status quo. With the launch of Starlink, everywhere in the world will have access to quality, reliable internet. This will create a huge market for cloud-based medical diagnostic platforms.

A traveling physician with the most basic training in the middle of the Sahara will be able to upload pictures and patient history to such medical portals via Starlink and rapidly get back a list of most likely diagnoses as well as follow-up tests that can nail down the diagnosis. Think about Star Trek medical tricorders. That kinda thing. Wave a bluetooth smartphone attachment over someone and, within moments, get a detailed and sophisticated state of their health.

Will this technology cost people jobs? I think so. Physicians and specialists are expensive. I suspect that hospitals will find that they can have fewer physicians on staff and go with cheaper labor, such as nurses and physician’s assistants who can leverage these cloud-based services to create better, more consistent care for cheaper.

Transportation

I was getting an Uber back from the car dealership a few weeks back and was quietly contemplating how my driver would be out of a job within 5 years. Uber, Tesla, Lyft, Ford, and pretty much every other major manufacturer are working tirelessly to bring full self-driving vehicles to the market. These vehicles will ultimately be safer and cheaper than anything on the road today, by a long shot.

Economic forces will mandate that these technologies be adopted, though the entrenched industries will resist it. Still, there are something like 7 million drivers in America alone who will be out of work within a decade. When combined with electric vehicles, we are going to see some huge knock-on effects, such as a collapse in demand for automotive mechanics and parts, since EVs are far more reliable and cheaper to maintain. Though, this lower cost could be offset by unment demand for transportation. Cheaper transport could result in vastly more passenger-miles per year.

Transportation costs will drop drastically, which will probably cause people to travel more, but will also reduce the cost of many goods, including materials for homes.

Science

Embedding true knowledge and intelligence into tireless machines is already starting to revolutionize science. Deep learning is already aiding in drug discovery as well as fundamental physics at places like CERN and the LHC. As these tools become more commercialized, we will see every researcher benefitting from deep learning. If nothing else, it will help consolidate the state of the industry, reading every paper in existence and summarizing it, allowing researchers to focus on the gaps and also to keep up to date much quicker.

As deep learning proliferates, science will accelerate in most areas, creating compounding returns for better computers, better AI, and the snowball will continue. We’re already seeing deep learning being used to streamline data acquisition, used to create better models. This can be see in concepts like AutoML.

Quantum computing

IBM, Google, and D-Wave are all currently locked in an arms race for Quantum Suppremacy (which sounds cyberpunk AF). I don’t think I can overstate how disruptive the proliferation of quantum computing will be. You can already sign up and use quantum computers as a cloud service.

What is the bottom line?

Quantum computing excels at difficult problems. I won’t explain how it works but you can check out this great video from IBM for a deeper dive. Quantum computing is hundreds of millions times faster at some problems than conventional computing. I suspect, as qubits increase, we will see quantum computing become billions or trillions of times faster.

Optimization

Maybe you’ve heard the “traveling salesman” problem. Maybe you haven’t. What you might not have heard, though, is that training AI is all about optimization. OpenAI allegedly spent about $7 million just to train GPT-3 in cloud computing. That is prohibitively expensive, obviously. What if the training process of advanced AI models dropped precipitously? What if the most process-intense parts of optimizing a deep neural network became millions of times more efficient? Billions? Within 10 years, I think that training the equivalent of GPT-3 will be trivial, costing pennies rather than millions of dollars.

What happens when literally everyone can train something as powerful as GPT-3 in an afternoon?

It’s hard to even wrap my head around this possibility. Anything that you have enough data for, or a good enough simulation environment, can be automated to super-human levels in an afternoon. The economic value here, let alone the disruption to everyday life and work, is incomprehensible. This is why the biggest companies in the world are locked in this arms race.

Simulation

Right now, the largest networks of computers, such as those for Folding@Home and those used to decode LHC output, are slow and expensive. Ludicrously expensive. These kinds of networks have been used to help develop the COVID-19 vaccines.

What if the entire backlog of F@H could be done in a few days? What if new vaccine and drug candidates can be tested in massive batches? Thousands of candidates tested against tens of thousands reactions? What if you can then simulate the entire genome of a patient and their drug interactions during an office visit?

This is not even entirely hypothetical, quantum computing is already being used in material science to help create the next generation of lithium batteries.

What life might be like in 2030

Where we live

Between Starlink, remote work, and job destruction, I suspect the current exodus from cities will explode. I think we will see more and more communities popping up in the cheapest places to live, such as the mostly empty central states. Self-driving electric vehicles are expected to reduce the cost of passenger-miles to 5 cents or less. You’ll be able to summon a car to take you from your home to anywhere for just a few dollars. I expect a lot of people will also switch to a more rural, homestead style of living without their conventional jobs. People still need occupations and UBI will create a safety net.

Beyond that, I suspect economic pressures will force more people to cohabitate. I anticipate a rise in intentional communities and cohousing projects. We may very well find ourselves living in eco-villages before too long.

Where we work

Honestly, I think that unemployment will skyrocket by 2030. Simply put, many people will be unemployable. This is not due to personal failing or laziness. It will simply be due to the fact that people cannot compete with the kinds of AI and automation I’ve outlined above. Only those who are capable of gaining domain expertise will be able to find jobs conventional jobs.

I do suspect the gig economy will continue, though. Homestead life will see more people switching back to gardening, crafts, and other more personal services as a way to make some extra money. Think of things like childcare, pet grooming, and other domestic service. A good friend of mine is a massage therapist - I suspect she will be immune to job losses due to the singularity.

If I do lose my job to the singularity, I plan to focus on writing.

Health and medicine

With a mass exodus from cities, the introduction of UBI, and hopefully universal healthcare, we will probably be under far less stress as a society. One can only hope! Beyond that, I suspect that the combination of quantum computing and deep learning will result in massive breakthroughs in regenerative medicine, de-aging medicine, and remediations for chronic conditions. We’re already seeing the beginnings of some of this with nanotech based medicine.

Taken all together, I think we will be far happier and healthier in 10 years, and we will probably be approaching indefinite lifespan. Indefinite lifespan is the idea that all common causes of death will be “solved”. Infectious disease will be a thing of the past as well as chronic disease.

Back around 2010 is when I started paying attention to stem cell therapy and regenerative medicine in general. I predicted then that we would see the first major breakthroughs by 2020. Today, there are a handful of therapies going through clinical trials. The market for regenerative medicine is huge. The first to market will make a bloody fortune, so you can bet your biscuit that someone is working on it!

A series of implants and/or external sensors will be able to evaluate your health on an ongoing basis. Heart attack, stroke, and aneurysms are gone for good. Early detection of and outright prevention of all cancer will be in place.

Entertainment

GPT-3 can procedurally generate just about any text, this technology will only get better with time. Imagine you fire up your Nook or Kindle and say you want to read a brand new high fantasy story. I want to read something like Game of Thrones, you say to your device. It goes and thinks for a moment and generates the first page of a story. You start reading and, as you do, biometric sensors track your response to what you’re reading. When you finish each page, the next one is generated on the fly based upon your reaction to the previous page. You read and entire story that is custom-tailored to your exact preference and reading level.

Take that one step further.

Let’s imagine that the fourth or fifth generation of GPT technology can generate movie and TV scripts in real-time. Then you have some other generative deep learning models that can translate the screenplay directly into audio and video. Yes, I’m saying that you’ll get highly customized TV shows and movies generated in near real-time just for you!

Some commentators on Reddit suspect that Microsoft licensing GPT-3 means that videogames will have procedurally generated dialog before too long. Take that to a few more evolutions and you’ll get entire games that are procedurally generated.