Resources used to create this essay.
Mustafa Suleyman and Seth Rosenberg: https://www.youtube.com/watch?v=cNzRviY4Ei8
The economy and national security after AGI | Carl Shulman: https://www.youtube.com/watch?v=wTci0CdOPIc&list=PL-BRtcBm4Yj4aKn72p4PjyqHh0ZQFdI1A&index=1
Sam Altman & Brad Lightcap 20VC conversation: https://youtu.be/G8T1O81W96Y?si=IF6LfghiRr7dhIRK
How to survive the looming AI take over.
“Heavier-than-air flying machines are impossible.”
- Mathematician and physicist, Lord Kelvin (1895)
“There is not the slightest indication that [nuclear energy] will ever be obtainable. It would mean that the atom would have to be shattered at will.”
- Albert Einstein (1934)
“I see little commercial potential for the internet for at least ten years”
- Bill Gates (1994)
“The truth is no online database will replace your daily newspaper, no CD-ROM can take the place of a competent teacher, and no computer network will change the way government works”
- Computer Scientist, Clifford Stoll (1995)
Over the coming decade, there is a high chance that AI is going to take most human jobs.
AI and robotics will be able to perform most, if not all, tasks currently done by humans (cognitive work and manual labor). There will be very few jobs where humans will be able to compete with AI systems – they will be more efficient, and far cheaper.
Most economists and mainstream predictions about AI are not pricing in the speed of this change.
Major job displacement and rapid economic growth within a matter of years, not decades.
Artificial Intelligence is fundamentally different from other technologies – different to airflight, nuclear energy and the internet.
General intelligence won’t just replace a segment of the economy; it’s going to flip the entire economy upside down and then send it into orbit.
If we’re lucky, rapid economic growth will result in a wider distribution of wealth.
If we’re unlucky, wealth will be concentrated in the hands of the few.
So where does this leave us?
The obvious optimistic long term hope, is that massive economic prosperity will allow us to comfortably(ish) transition to more uniquely human domains – creativity, novelty, nostalgia.
But there is going to be an awkward and uncomfortable in-between.
As Government and National Security, workplaces, institutions and the wider population slowly and then suddenly awaken to the magnitude of what is coming, there will be panic, pain and opportunity.
The Opportunity
I’ve spoken heaps about the pain, so let’s talk about the opportunity.
We can expect AI to become increasingly, generally intelligent – able to do most cognitive work within a matter of years, with robotics and manual labour soon after.
This isn’t just data collection or a glorified chat bot.
We’re talking about highly cognitive and nuanced work requiring multiple steps over weeks or months or years – ‘Here’s my credit card, go and find a niche product and market, develop and design a product, build me an online website and shop, and make me rich.’
Of course there will be bottle-necks, but it’s coming. And it will come for the highest dollar value opportunities first – AI researchers, managerial work, software development, lawyers.
So assuming AI (and new LLM model releases) become increasingly and rapidly, Generally Intelligent, where is the opportunity?
Here’s three.
- Collection and curation of high quality data
- Post-Training refinement
- Ambiguity-friendly domains
Collection and curation of high quality data
As Artificial Intelligence system becomes increasingly general, data and quality of data is the golden ticket.
The quality of data you can collect, curate and feed back into an increasingly general AI system will become increasingly valuable. Both for yourself, for your work, for a customer or client.
Creating User Experiences, products or processes which naturally and natively collect high quality data is the big opportunity.
Post-Training
When it comes to training Large Language Models, which are the current and most popular incarnations of Artificial Intelligence (ChatGPT and Claude for example) the terminology can be confusing.
In very simple terms, you can think of a pre-training and then a post-training phase.
In the pre-training phase, AI is trained using vast amounts of compute and internet data.
There will be opportunity to contribute to the post-training phase with bespoke experiences on top of increasingly intelligent base models, If you have collected and curated high quality data.
This could look like a bespoke postrained LM interface built on top of an existing open source model with custom data. Or, like a custom Integration or bot plugged into, or on top of, an existing proprietary model like ChatGPT.
Ambiguity-friendly domains
On the road to beyond human level general intelligence (super-intelligence), AI will first become most valuable in “problem domains that make a virtue out of imprecision”.
This will create a small window of big opportunity.
What does that mean?
An ambiguity-friendly domain simply means a space or problem where there isn’t always a single, clear answer – think precise medical diagnosis (non-ambiguous) versus individually curated treatment plans.
AI will eventually do both, so ‘wrapper companies’ will generally not be a good long term play.
But some domains are inherently ambiguous and benefit from multiple perspectives. So regardless of increasing AI capabilities, there will likely always be some room for creative and interpersonal collaboration – though how this looks long term is cloudy at best.
To make this all easier to envision, let’s use an example.
Imagine you’re a travel agent.
At some point in the not so distant future, a personal AI system will know all my previous travel history, my preferences, my calendar and availability. It will be able to fully plan, coordinate and book my holiday from start to finish, better than myself, and better than you (the travel agent).
When AI systems start booking flights and hotels (coming very soon), there will still be opportunities where you can help curate bespoke travel experiences by working with these increasingly intelligent AI systems.
There will even be opportunity to build ‘wrapper’ companies, or products and services on top of openly available AI systems, or with agent plug-ins, which facilitate more bespoke ‘post-training’ enhancements.
Hidden gems, off-the-beaten-path experiences, and tailored dining options based on unique likes and preferences.
But as AI becomes increasingly intelligent and integrated into our lives, we will require less and less human interaction.
Once we’ve reached the level of integrated General Intelligence required to fully plan and book a holiday from start to finish, we can assume most other areas of the economy and work have also been automated away, including most manual labor.
At this point, the cost difference between AI and human labour will be so extreme, that you’d have more chance hiring Elon Musk as your travel agent, or Justin Beiber to fix your leaking pipes, than an average human – in ten years time from now (maybe less).
This is the result of the potential unprecedented economic growth.
In the videos above, you saw the two CEOs of two of the biggest companies in the world who will be responsible for creating the increasingly general intelligence that is probably going to take your job.
Sam Altman and Mustafa Suleyman from OpenAI and Microsoft AI.
From my understanding, Altman believes the window of opportunity is much smaller than Suleyman.
But both ultimately land in the same camp – we’re going to have increasingly intelligent AI systems that will manage most of our lives and transform most of the economy in a relatively short period of time.