The alignment problem in AI is a scary sounding topic that I’ve been thinking about recently.
It’s the seemingly insurmountable challenge of ensuring that the goals of an AI are aligned with human values.
However, the term "human values" assumes that there is a universally agreed-upon set of values that all humans share.
While there are some values that most humans agree on, such as the importance of basic human rights, and that being mean to animals is bad, the reality is that many humans do not live in a society that aligns with these values. War, violence, and corruption are still prevalent in many parts of the world. We still engage in factory farming and routinely set fire to forests.
Core values differ from region to region, some societies place high value and status on material objects, individualism and wealth, whilst others place greater emphasis on community and strong family ties.
Even within regions that are seemingly aligned you get individuals or small groups that break away from these norms. For example SBF using the guise of philanthropy to gain credibility, and then using that influence to deprive everyone of their money.
Despite this, unaligned humans are the source of many technological achievements. Competition through war has driven innovation, leading to both violent and non-violent innovations.
Both the nuclear bomb and the computer were largely resultant from small groups of individuals working in alignment against a greater unaligned cause.
Colossus computer - c. 1943. Geoff Robinson Photography/Shutterstock.com
In peacetime, wars are fought through technological advancements, such as the Space Race or the competition between companies like Apple and Google.
Being “unaligned” on a large scale is probably the greatest precursor to human suffering, but also the greatest enabler for human ingenuity.
In an unaligned world, there is an incentive to encourage others to align with you, alignment brings mass wealth, power and control. The problem is, who is the leading group fighting for alignment? If it's an evil regime, you might end up becoming aligned by hating Jewish people and committing atrocities such as the Holocaust. But if the leading aligned group is Scientists working on the discovery of Antibiotics, you end up with millions of lives saved.
This trend is likely to continue as both AI and Crypto become more prevalent in the world.
AI alignment
The challenge with AI alignment is that many subgroups will want to align to create their own version of a super-intelligent AI.
If we were to ask 100 random people what the most important problem a super-intelligent AI should solve first, we would get a vast array of differing answers.
A person from the West Coast of America might argue that the AI should focus on equality and climate change. while someone from a deeply conservative state could want the AI to prevent abortions. A large portion of responses might lobby for ending world hunger, whilst a CCP representative might want the AI to gain technological superiority over the United States.
Assuming these requests result in success, let’s ponder some of the scarier outcomes.
The San Francisco robot might decide that the only conceivable way for every group of humans to be truly equal while simultaneously reducing climate change is if all humans are turned back into their individual atoms and converted into the raw materials required to fill the world with solar panels.
The far-right robot might conclude that the most logical way to stop all abortions is to create an undetectable yet highly transmissible virus that renders the recipient infertile. No more pregnancies equals no more abortions. Problem solved.
The philanthropic robot, dead set on feeding the world, might decide to create some super farms using newly invented chemical pesticides on a mass scale, massively increasing crop yields and food security in the short term. But the side effect of nuking all insects and non-food generating plants would lead to a mass extinction of life on earth.
The CCP built robot might use its giant brain to determine that the best way of getting a technological advantage over the USA would be to launch a devastating cyber attack on infrastructure and military assets. Swiftly plunging the world into WW3 when President Biden and NATO respond with escalatory measures.
Of course the above are super simplified hyperbolic versions of what may occur when a party finally is able to build an AGI. They are written to describe the vast differences of “Human Values” subsets of us have, and when taken literally the huge variance of unpredictable outcomes that are possible.
Whilst I’m generally optimist and am hopeful that AI can bring HUGE improvements to the world. I think it's most likely to follow the usual trajectory of human enabled technology - Increased opportunity for human suffering coupled with rapid and exponential technological improvements.
This time, as was true with the invention of nuclear weapons, there is a non-zero chance the human suffering element is actually the extinction of all human beings on earth.
But there’s also a non-zero chance AI cures all disease, increases average intelligence by an order of magnitude, solves climate change and makes energy free and abundant for all.
For this reason I think it's unlikely the letter signed by 1,000+ industry leads asking for a pause on AI development garners much success. The incentives to rapidly increase AI development remain.
Why have you written a post about how AI is going to kill us all!?!?!?!? I want to hear about Crypto.
I took a week off from thinking about tokens and jpegs this week to prevent a mental breakdown and of course spent the time healthily pondering how robots will destroy us all.
The same alignment problem that's prevalent in AI also exists within Crypto.
Crypto could actually help solve a number of human alignment problems?*
The world of Crypto is a constantly evolving space, characterised by misalignments among various stakeholders. Although this may seem chaotic, it is also what makes the industry so exciting and full of potential. If everything were already decided and aligned, there would be little room for innovation and opportunities for profit.
In fact, I would argue that the fact that cryptocurrency lacks alignment is probably the number one reason why it is one of the least efficient markets on the planet and thus the most profitable and fun marketplace for imbeciles like me to participate in.
However, we do face significant challenges. There are many groups of powerful entities working against the interests of cryptocurrency, and it can feel at times like these powers are beginning to align against us.
It’s in the interest of certain governments to slow the adoption of Crypto and instead encourage the use of CBDCs.
It's in the interest of the banking industry to put barriers up for both on-ramps and off-ramps into Crypto.
It’s in the interest of VCs to start new, centralised chains competing with and criticising the more decentralised incumbents as the opportunity for wealth creation and control is greater.
Whilst you could look at the above points and draw a conclusion that these parties are all aligned in speeding up the death of Crypto, I believe this would be incorrect. Because each party is acting in its own self interest.
Groups of individuals working in their own self interest actually solidifies Crypto's future.
If one major government is too hostile to cryptocurrency developers, then it opens up the opportunity for other nations to welcome them with open arms. Being unaligned here can actually work to cryptocurrency's advantage.
If banks believe they can turn a healthy profit by charging management fees for staking clients' Ethereum, they will become more crypto-friendly.
If businesses believe they can grow their brand by issuing NFTs or attracting digital asset holders, global marketing for cryptographic assets ensues, and the overton window shifts.
The more powerful companies and individuals begin holding digital assets, the greater chance they have to lobby governments to take a pro-crypto stance. As companies such as Coinbase fight back against the SEC, the more likely it is that others will join forces and use their collective power to win hearts and minds both in the courts and in Washington.
So, the Alignment problem exists in Crypto too. And for now it's a good thing. It's why we’re all able to make money far easier in this market than any other.
Over time, I believe this will change, and cryptocurrency will become socially acceptable and enter the mainstream as governments, builders, users, lobbyists, and legal teams become aligned. Sadly, when this day comes, we may all be proven correct that cryptocurrency really was the future of finance after all. But for those of us that love trading the inefficiency of markets, it is probably time to find something new.
On the plus side though, even if we are wrong, at least cryptocurrency cannot rise up, overthrow the human race and kill us all, so we will not be guilty of supporting the worst innovation ever! :)
Happy Easter everyone!