In the second case, how is this different from e.g. Kelley Blue Book for car prices?
---
The problem with the first case is that we do not have existing case law about AIs, and we also have no laws on the books that clarify what to do in AI cases, so this is making the lawsuits and investigations very messy and time-consuming, since judges have to waffle about to figure out what the correct, legal thing to do is.
Pretty much all Western law systems judge the severity of action using intent, or mens rea; it doesn't absolve you of crime if you didn't mean to do it, but your penalties will generally be lesser. The people using the algorithm can wash their hands of intent by saying "we signed a contract with a third party for their algorithm and didn't reasonably expect biased outcomes." Then you make the third party a defendant, and then the third party claims "trade secrets" because we have no laws about what is and isn't protected trade secrets with AI, etc. and that becomes a whole legal case on its own that takes even more time for the legal system.
---
The difference in the second case is that the corporate landlords in question are automatically pegging their rates to the algorithm's recommendation and not giving property managers discretion. The algorithm's customers combined also control over half of supply of apartments in the Seattle area.
---
But the point of legal clarity, is to reduce the amount of questions, which both 1) heads off potential issues and 2) makes the issues that come into the legal system faster and clearer to process. It also potentially actually makes things easier for companies who make AIs too, since legal defense spending isn't exactly cheap.
>I don't get it. For the first example, the system is only a tool used and the responsibility is still with whoever does the sentencing/traffic stops etc.
Sometimes though the rules of society must take into account the actual reality of humanity, with all our foibles, and work anyway. Pointing the finger at "individual responsibility" doesn't always cut it. Certain tools simply push too far to the edge of well known human mental failure points. If the overall system for activities involving serious life/safety/security/liberty human failure modes tends to push towards these failure modes and lean too heavily on an expectation of human perfection, and then imperfect humans fail resulting in loss, it can be very worth considering whether the problem is human imperfection or if it's just a bad system. The incredible safety record of the modern airline industry for example comes heavily from treating accidents as system failures by default. Outside of a few edge cases that are constantly working to get shrunk, no accident should ever result from just a single problem including human problems, from a single person getting tired or making a mistake. There are checklists and formalized procedures. There are layers and layers of redundancy, of everyone checking each other. We don't just take incredibly complex tools like aircraft and then if one crashes say "well the pilots didn't have the right stuff!"
Or for a current AI-related ongoing change, consider self-driving technology. There have been two overall broad approaches, one "working its way up" from driver-assist tech, and the other aiming to start at full self-driving immediately even if that means tight geographic restrictions and vehicle tech requirements. In the first, the idea is that "the driver is still always in charge" and "it's only a tool" until full self driving is achieved. And perhaps in principle either could get there in the end without wildly different issues.
But in practice, that's not how humans work at scale. For many people if you give them something that works 99.5% of the time and then fails catastrophically 0.5% of the time and tell them that they need to constantly act as if the tool wasn't there and might not be there at any instant, well, they just won't do that consistently. They'll come to depend on it. Attention will wander, distractions will set in, and they won't maintain mental state the same as if they were driving themselves. If there is some warning they may need to take over sure they'll be able to execute a state change and do so, but if it's a matter of seconds, they won't. Humans are creatures of shortcuts and habits, that's just how it is. So maybe L3/L4 driver assist just isn't ok, and saying that it's "only a tool" doesn't cut it.
>In the second case, how is this different from e.g. Kelley Blue Book for car prices?
The issue is if it's all being done automatically, not merely as a recommendation, and at scale.
In general an area the law has yet to really effectively grapple with is that of emergent effects, where individually something would be fine, but with enough scale has an effect that mimics something we already aren't ok with. Here, the effect of fully automated algorithmic pricing of a sufficiently high percentage of the market of nominally independent actors could create the effect of formal collusion to change market prices.
Another example would be AI, networking and storage with sufficient cameras. Putting a security camera out on your property is perfectly legal. So is many cameras. So is saving video, and humans looking at it. There is no "reasonable expectation of privacy" in public per se. Individually there'd be no issue. But if there are a sufficient number, networked, with sufficient storage/memory/computer and AI thrown at it, the effect suddenly changes and becomes as if someone was being persistently trailed/tracked the same as if a GPS tracker had been stuck on them or the like. And worse it's for everyone simultaneously. The sum is greater then the individual parts. Such a system could effectively not just trace everyone's movement but also interactions and start to build up their social network graphs and no doubt all sorts of other personal information.
How to grapple with things that are fine and even incredibly helpful by themselves but become dangerous at scale may be one of the great challenges of this century.
And if you try, you end up writing something vague that doesn’t actually fix the problem and instead bans something else.
Case in point is the EU having to write something vague and it turns out they had no idea what was coming.
Maybe just wait until there is a problem before trying to write vague laws IMO.
I.e. "We don't approve loans to {ethnic majority} on purpose. We only use our models, which only rely on historical data." -> No proof, no case, no problem
The point of any AI law should be to put the onus on the model executor to explain how their model doesn't violate existing law.
You can use any model you want, but you can't point to its unexplainability as a free pass for ignoring protected classes.
And if you can't explain its results? Well, then you're opening yourself up to lawsuits.
Requiring the claimant have some evidence might not be a perfect legal regime, but the converse is a de-facto ban on innovation.
Can a human _prove_ that their decisions are made without bias? No. Are we going to hold them to the same standard proposed here for an AI? Also no. In practice, rules like the one proposed here just enshrine legacy decision systems which are already understood to biased and ineffective.
The most mysterious black box we have is the human brain.
What would you say of requiring training data and hyperparameter archiving for subsequent legal review?
> The regulation, proposed by the Commission in 2021, was designed to ban some AI applications like social scoring, manipulation and some instances of facial recognition. It would also designate some specific uses of AI as “high-risk,” binding developers to stricter requirements of transparency, safety and human oversight.
The issue is that as written it would still not cover general AI and that would still be legally grey.
Some of those uses in the US have already been problematic, like Madison Square Garden using computer vision to ban any lawyers from firms suing them from their vast properties, which is actually against various laws like their liquor license.
Every technology, at some point, will inevitably bring questions of legal liability and culpability, and AI has a large tendency to do so because of how applicable it is. It’s not the worst idea to try and get ahead of the problem by defining a legal framework.
Examples of how a free-for-all has resulted in legal questions;
* police and the judiciary in the US have come under fire several times for using proprietary AI to determine things like sentencing or traffic stops
* there is currently a class action lawsuit in Washington state about whether or not price recommendations by a third party’s algorithm constitutes collusion on price-fixing if enough corporate landlords use it