My New Obsession is AI Regulation
There’s starting to be A LOT of talk about regulating AI, and rightfully so.
This open letter to halt AI development has been signed by 1124 people at the time I opened it, and includes some people you may have heard of like Elon Musk, Steve Wozniak, Yuval Noah Harari, Andrew Yang, and one of the godfathers of modern AI Yoshua Bengio. We can debate until the cows come home the real motivations behind some of those people choosing to sign, but I completely agree with the sentiment.
Realistically, though, it’s all talk - you can’t really put the AI genie back in the bottle at this point, nor will all the governments just start mass-banning it (though there has been some talk of that in Europe). The reason you won’t see outright bans is the same reason why lawyers have jobs: regulating behaviour is really hard and no one really follows rules.
People in power know that regardless of the rules they pass, enforcement is a gigantic kettle-of-fish mess in any context, and the stakes are just too high to halt development and hope everyone else is doing the same. Kinda like nukes, except way more lucrative for the average-joe capitalist.
All of this regulation reading has got me reflecting on why regulation is so hard, so I thought I’d reflect a bit on a recent experience I had.
All regulation is ultimately human regulation
All regulation is ultimately human regulation. All rules (except for mathematical or scientific rules, I suppose) are made by humans, for humans, and about things humans care about.
The weak link there, of course, is humans. Ultimately, we live in a world full of self-interested people forced to try to co-exist peacefully while fulfilling their own personal desires and agendas. That is not meant to sound cold or cynical, it’s just kind of the truth.
Anyone who has worked in or near policy-making knows concretely what the real issues are. People have personalities, agendas, thoughts, and feelings, and everyone is struggling to make meaning out of their lives and to have some sort of impact on the world that gives them that meaning. But to some extent, what happens in this world can often be a zero-sum game, which means not everyone gets to pursue their agendas, or make the meaning they’re hoping to make.
By definition:
- everyone is a different person and has a different set of thoughts, feelings and goals,
- policy is about creating rules and procedures meant to govern groups of people, and
- no policy can possibly encapsulate and make room for everyone’s different thoughts, feelings and goals
That is what makes politics so hard. You can’t satisfy everyone all of the time, so ultimately some choices will have to be made.
Who makes the choices?
That’s the question, isn’t it? Politics and policy is, like everything else, about power in the end. Who gets the power, how they wield it, and what constraints are put on them, is a never-ending subject. Different groups, organizations, countries, etc. will all have different ways of answering these questions, and there isn’t a clear answer to any of them.
What is clear is that the games played in the policy-making arena are very human games, which means they are subject to human dynamics. Hero-worshipping, emotional decision-making, economic constraints, public perception - the list goes on, and on, and on, and is so daunting that it seems most people throw their hands up and say:
Well it’s all broken, but I don’t know how to fix it, so I’m just going to stop paying attention.
But these are all generalities. Let’s talk specifics.
My recent brush with policy-making
For the past few years, the big fire-drill in policy-making circles has revolved around the Covid-19 pandemic. The world shut down in an unprecedented way, and there’s almost no human institution that was left untouched by its influence.
I work in a very small corner of the world - the Ontario criminal justice system. During the pandemic, I happened to be serving as a volunteer board member of my professional bar’s legal association. When the pandemic hit, that meant I was quickly forced onto a regional working committee with judges, crown attorneys (Canadian prosecutors), law enforcement, court employee union representatives, and others. It was our job to meet twice daily for months, and play our part in modernizing and technologizing our little corner of the Canadian justice system overnight, while helping to keep it from totally collapsing.
What I learned during that experience, I’m told by others more experienced in the policy-arena than me, is a set of lessons broadly applicable to policy-making writ large:
- only the thoughts and feelings of people with a seat at the table matter
- there’s usually a pecking order in place for the different representatives from the various interested stakeholders, and decisions strongly reflect that pecking order
- the representatives with a seat at the table are just people, who bring their own personalities and flaws to every contribution they make
- you never actually have anywhere near the quantity or quality of information necessary to make good decisions
- multi-stakeholder committees are mostly about jockeying for influence, resources and risk/liability-minimization to protect their own agendas
- those agendas usually involve long-term plans with set-in-stone budgets and action items that are in year 3 of 5
- making new policy is extremely detail-oriented and hard work, and
- even when you seem to have done an objectively good job, people aren’t happy anyways
It was an eye-opening experience, to say the least, and gave me a lot more respect and sympathy for anyone doing that kind of job for a living.
“I mean, it’s not that complicated”
Not to minimize the work we did, but my little multi-stakeholder committee wasn’t curing cancer or regulating artificial intelligence. We were figuring out things like how to disseminate public links to Zoom-based courtrooms to justice professionals, how much lead-time we should require for email requests that were sent between different stakeholders, and how many times a day courtrooms should be cleaned. It wasn’t easy, but it wasn’t rocket science.
Before sitting on that committee, I would have said all of that seemed simple. I could have given you some common-sense answers that seemed to accommodate most peoples’ interests, and then buttoned it with a smug kicker like “I mean, it’s not that complicated”. But of course, I would have been wrong.
I would have been wrong because I didn’t know about the personalities, or the lack of information, or the action items that were in year 3 of a 5 year funded budget. I certainly didn’t understand negotiating with union representatives, or the dangers of issuing new internal policies on a daily basis that no one could possibly keep up with.
Turns out, things that seem “not that complicated” are in fact very complicated.
The human element
In my opinion, we seemed to do a lot of policy-making based largely on what a powerful person at that table had for breakfast that morning, or by the time of day rendering everyone too lazy to shoot down a new and idiotic idea, or with undue influence from the fresh energy of a stakeholder’s recent replacement representative not yet beaten down by the grind of doing the job for months on end, but with zero understanding of the actual dynamics at play.
Then I started to think about bigger multi-stakeholder tables. Tables with media coverage, millions of dollars on the line, and reputations at stake for very senior, public and powerful people. Nobody at our table was really dealing with any of that, and we still made many sub-optimal decisions influenced by the wrong things. But we were trying our best.
And then there’s enforcement
Metallica was right - Nothing Else Matters
We have enough trouble agreeing on rules. The truth is, rules are such a small part of regulating human behaviour that we spend way too much time on them, and not nearly enough on the realities of enforcement.
Unless rules have reliable teeth, they’re essentially useless
I see examples of this everyday as somebody working in the justice system, and can give you as many examples as you can handle of the ’enforcement gap’ between how laws say people should behave, how they actually behave, and how difficult it is to (1) catch rule-breakers, and (2) reliably enforce commensurate consequences.
There are good reasons for all of this, but the result remains the same: humans are hard-wired to push boundaries, and the only boundaries are reliable mechanisms of enforcement.
That is, for example, why international law doesn’t work - international laws can be passed with the harshest language and best of intentions, but there isn’t actually an effective international law enforcement agency to carry them out.
Even domestically, what does enforcement actually look like? Criminal laws and the justice system, which work but are riddled with problems, and then the civil/regulatory sphere, which is no better (and maybe much worse).
With AI (and tech in general, including any behaviour taking place online) the technical challenges of just finding the conduct are immeasurably more complicated - I found this guy’s twitter thread about his paper on the subject really interesting.
The point?
All of this is to say, we’re going to have enough trouble globally even agreeing on what the laws regulating AI should look like, and the conflicting standards across physical boundaries (that the online world doesn’t care about or respect) are going to wreak havoc in every sphere. And once laws are in place, the actual hard work begins of learning how to enforce those rules that most won’t understand, and probably won’t make much sense in their first iterations.
So we’re doomed?
I don’t know. Maybe. Hopefully not. I do know one thing from having watched a justice system respond in realtime to a crisis (Covid-19): when things get scary, rights and feelings go out the window in favour of crushing every ant-sized problem with blunt, industrial jack-hammer sized solutions, so it’s not like we’re never good at enforcement. The question is more: what will it look like, and how scary will things have to get first?
It’s going to be interesting to watch.