Slashdot Mirror
Can We Build Ethics Into Automated Decision-Making? (oreilly.com)
"Machines will need to make ethical decisions, and we will be responsible for those decisions," argues Mike Loukides, O'Reilly Media's vice president of content strategy:
We are surrounded by systems that make ethical decisions: systems approving loans, trading stocks, forwarding news articles, recommending jail sentences, and much more. They act for us or against us, but almost always without our consent or even our knowledge. In recent articles, I've suggested the ethics of artificial intelligence itself needs to be automated. But my suggestion ignores the reality that ethics has already been automated... The sheer number of decisions that need to be made means that we can't expect humans to make those decisions. Every time data moves from one site to another, from one context to another, from one intent to another, there is an action that requires some kind of ethical decision...
Ethical problems arise when a company's interest in profit comes before the interests of the users. We see this all the time: in recommendations designed to maximize ad revenue via "engagement"; in recommendations that steer customers to Amazon's own products, rather than other products on their platform. The customer's interest must always come before the company's. That applies to recommendations in a news feed or on a shopping site, but also how the customer's data is used and where it's shipped. Facebook believes deeply that "bringing the world closer together" is a social good but, as Mary Gray said on Twitter, when we say that something is a "social good," we need to ask: "good for whom?" Good for advertisers? Stockholders? Or for the people who are being brought together? The answers aren't all the same, and depend deeply on who's connected and how....
It's time to start building the systems that will truly assist us to manage our data.
The article argues that spam filters provide a surprisingly good set of first design principles. They work in the background without interfering with users, but always allow users to revoke their decisions, and proactively seek out user input in ambiguous or unclear situations.
But in the real world beyond our inboxes, "machines are already making ethical decisions, and often doing so badly. Spam detection is the exception, not the rule."
Ethical problems arise when a company's interest in profit comes before the interests of the users. We see this all the time: in recommendations designed to maximize ad revenue via "engagement"; in recommendations that steer customers to Amazon's own products, rather than other products on their platform. The customer's interest must always come before the company's. That applies to recommendations in a news feed or on a shopping site, but also how the customer's data is used and where it's shipped. Facebook believes deeply that "bringing the world closer together" is a social good but, as Mary Gray said on Twitter, when we say that something is a "social good," we need to ask: "good for whom?" Good for advertisers? Stockholders? Or for the people who are being brought together? The answers aren't all the same, and depend deeply on who's connected and how....
It's time to start building the systems that will truly assist us to manage our data.
The article argues that spam filters provide a surprisingly good set of first design principles. They work in the background without interfering with users, but always allow users to revoke their decisions, and proactively seek out user input in ambiguous or unclear situations.
But in the real world beyond our inboxes, "machines are already making ethical decisions, and often doing so badly. Spam detection is the exception, not the rule."
I'm not sure I consider the Amazon directing you to Amazon products as a very good example of "Automation", since that has a giant bias plugged into the engine by Amazon. You are trying to ascribe ethics to a system where humans are obviously in firm and direct control over results.
To me considering ethics and automation is more of a general concern where the automation is making derived choices that are pretty far removed from human directive. I think you can build in ethics to try and be kind to people, it's not impossible - but even the choice to try and include some kind of ethical directive, is still really at the mercy of humans and how much time and effort they are willing to put into such things...
Perhaps the most effective solution is for some company to come up with a really kick-ass ethical choice helper for automation, that becomes so popular that companies are clamoring to include it. Otherwise it will get placed in the asme leaky lifeboat that Accessibility is always placed in.
"There is more worth loving than we have strength to love." - Brian Jay Stanley
I suppose it's likely on the order of Monday not being the best day of our next week... ethical decisions made by artificial intelligence will not be above reproach.
Though, perhaps, like the standard realists have for automated vehicular piloting, all AI ethical decisions have to do to pass muster is exceed the effectiveness of decisions that would've been made by their biological inventors.
Fortunately for the future of the robot overlords, we haven't set this bar that high.
Happiness in intelligent people is the rarest thing I know.
Ernest Hemingway
I wrote that subject line as a lead-in:
Like 'mercy', 'ethics' requires understanding of human beings and human-related matters.
Since the poor, weak excuse for 'AI' they keep slinging around lately cannot 'think', and therefore is entirely incapable of understanding humans, they are also incapable of being 'ethical'.
Someone will now attempt to argue that 'ethics' is just a set of rules to follow -- or perhaps I should say 'laws' -- and there are always exceptions to rules and laws where there are humans and human lives to consider. Therefore: machines should not be involved in making decisions requiring 'ethics', they are entirely unqualified to do so by their very nature.
Furthermore: all so-called 'AIs' should be supervised by humans at all times; no 'autonomy'. There always needs to be at least one human being there to allow or disallow what any of these machines does.
The programmer is ethical.
Browsing at +1 - no ACs, I ignore their posts. So refreshing!
Unethical animals programming ethics into a dumb machine?? Dog help us all!
Automated systems that apply 'known' and accepted rules equally, is maths. Those formulation of those rules is ethics and is fine as long as the majority are aware of the rules and approve of them. The more impactful the rules, the more it could favour one against the other, the greater majority required to apply that rule but the starting point should always be more that 50% of the eligible citizenry and the upper limit, depends which you favour mathematically fractions or decimal places, it make a difference, as in 2 out of 3, probably the upper limit of foolish majority restraint, (decimals is more a choice between 60% and 70% for whole numbers sake but doesn't read as well as 2 out of 3 or 2 to 1).
The difference between AI and straight maths, of course corrupted AI which makes unethical decisions to favour it's programmers, versus a simple spreadsheet, which applies the rules, that everyone is aware of and the majority have agreed to.
The silly bullshit waffling around about mob rule, what a crock of shit, who complains about mob rule, the 1% who consider the entirely of the 99% the mob, the opposite of mob rule (majority rule, which is just emptily slandered by calling it a mob), is entirely corrupt elite rule, who inevitable govern to suit themselves at the expense of the mob and fear the mob, the majority, will hole the elite, the extreme minority accountable for their corruption, avarice and very venal and abusive natures.
Yeah, I want automated decision making, fuck AI and fuck the cunts who propose it, you arseholes are just totally full of shit (AI as a layer of bullshit to hide decision making that favours a tiny minority at the expense of the majority, a new layer of bullshit added to the old elite lies). I want those maths rules to be open and clear and up for debate and affirmation or rejection by the majority, maybe a super majority in some cases 2 out of 3 or 2 for and 1 against.
Chaos - everything, everywhere, everywhen
Can we build ethics into Human decision-making? Only once we have done this do we have any hope of building it into AI.
Specialist Mac support for creative pros, Melbourne
Seriously, how do you define "ethics" so that it would be an acceptable definition to, well, everyone?
Because it won't be accepted as "ethical" unless its decisions agree with you (for all values of "you", including "me").
"I do not agree with what you say, but I will defend to the death your right to say it"
Asimov was NOT saying here are 3 simple laws you can program your robots with and everything will be hunky dory.
He was saying the Exact Opposite!
If you actually read and understood any of his robot stories, the theme was consistent: robots are no better than we program them to be and they can never be as good/smart/ethical as we are, period.
The 3 laws robot short stories are all well written (Asimov, after all) and tell compelling and educational stories about robotic potential when given decision making ability outside a well defined prescribed rules (such as in manufacturing where there are no real decisions to be made, just a script blindly followed).
I know none of you will actually read them and you will keep referring to them and getting it entirely backwards, but at least I tried.
and then you can start thinking about how to code it.
Things that were ethical when I was a kid would mortify today's people - bringing guns to school, designated smoking areas for students, nude swim class, bullying as a way life... no, ethics is a variable based on time, race, age, and income.... Build "ethics" into any algorithm and the masses will find it unethical.
Currently the various expert systems and automation have zero morals and ethics. Their only criteria are maximize profit, minimize risk. If someone ends up dying in screaming agony, meh.
This is just another extension of the principle that the only people you can get on the phone are people with no power to say yes. Their job isn't to make an ethical decision, their job is to make sure the people with authority to make a decision don't have to personally feel the consequences of tossing ethics out the window.
Same with the software. The programmers are just following orders, it's not like they're using the software on actual people. The people using it are just following orders, it's not like they wrote the software to make those decisions. The people at the top just specified software and ordered it's use. It's not actual people, just boring statistical data on a quarterly report.
Of course, in reality the software is an extension of those who give the orders. They just want people to blame "the computer" for as long as possible, just like in the '70s when, according to the CSR on the phone, the computer was infallible.
"The customer's interest must always come before the company's."
Which customer? Is the company thriving, expanding, and being able to take advantage of economies of scale to provide better and cheaper product to future customers considered in customer's interest or not? Or are we talking about customers who can't afford the product, so we should just give them the product for free and bankrupt the company, since customer interest comes first? Or are the workers at the company also customers, or is it ethical to exploit them just to provide cheaper products?
The above are hyperbole's, but herein lies the problem, if you want to put "customer" interest above the company, you must specify "which customers".
I think you can build in ethics to try and be kind to people
Ethics is not about being kind to people it is about doing the right thing. For example, a system to spot cheating on an exam is not going to be particularly kind to the people it catches and it would be highly unethical for it to be kind by ignoring the cheating. Since doing the "right thing" is subjective and extremely contextual any ethics in automated decision making is going to have to be directed by a human and, since people may vary on what they believe is ethical, very hard to get right.
Even something very basic like not killing people is not going to be easy to implement e.g. should an automated car prioritize the lives of the occupants over others or vice versa? It's made even harder by the fact that computer algorithms do not comprehend the ethical consequences of their choices: all the programmer does is tweak the parameters to make it behave in a way that they believe is ethical which ultimately means the "ethics" will be determined by large corporations or, if regulated, governments which is frankly rather a depressing thought givengovernments' and companies' past records on making ethical choices.
You know what, I think I actually will read them. I have been needing some heavy duty sci fi to occupy my time, but wasn't sure what. His stuff should be easy to find at a used book store?
No.
Because there are powerful vested interested with a desire to prevent such a thing.
General Relativity: Space-time tells matter where to go; Matter tells space-time what shape to be.
True for the first part, but...
If you actually read and understood any of his robot stories, the theme was consistent: robots are no better than we program them to be and they can never be as good/smart/ethical as we are, period.
Read Robots and Empire.
"Ethical problems arise when a company's interest in profit comes before the interests of the users"
Unfortunately that's the law in the US. Any company that neglects opportunities for profit is subject to lawsuits from shareholders. A corporation's sole responsibility is to look out for the interests of shareholders.
Neglecting customers/users might ultimately reduce profits, thus must be considered. But ethics? Where in the hierarchy of concerns is ethics? For each management organization that will differ.
There is a new type of corporation called a "Public-benefit corporation" in the US and going by similar names in other jurisdictions. It allows for public benefit to be a charter purpose in addition to the traditional corporate goal of maximizing profit for shareholders. Rules vary by location but it's an idea that may spread. Ethics can be an important part of such organizations.
...omphaloskepsis often...
Basically we need to include the goal in any automated decision making process, probably including result measurements after the fact (hopefully in a check/fix loop). And to do that we need to define the result we want.
The issue is when optimizing for one area makes another break in "interesting" ways. Like the old Chinese curse of "May you live in interesting times." Those only exist because of relationships we don't expect or measure through optimization.
I mean why would you measure the people you can't help? That list is long and boring right? Until it's not...like when you lack diversity (hiring only white, male, cis, etc).
In fact, I wonder if tools used for finding diversity issues could help pinpoint issues like this? Or are we stuck waiting for people to scream about something still? I have no idea...not my field.
But basically, "cost reduction" as a required goal makes this an unlikely result for anything created by private companies, in my opinion. See cases of "cheaper to pay off the hurt people than fix the problem" even after you realize there is a problem. And that's assuming we can even realize the issue.
An imperfect creator cannot create a perfect creation. Any AI created by humans will have imperfections equivalent to human imperfections.
"A person is smart. People are dumb, panicky dangerous animals and you know it." - K
See Turns Out Algorithms Are Racist. And don't forget that time Twitter taught Microsoft’s AI chatbot to be a racist asshole in less than a day. If we can do the opposite and make sure our algorithms aren't racist unintentionally and that our machine learning assistants don't become racist, you'll have won half the battle of making them ethical. This is going to be a hard problem to solve. It's not enough for programmers to be ethical they need to have outside observers helping them, especially people not like them.
"You'll get nothing, and you'll like it!"
I did In effect he was saying if your robot brain is vastly superior to a humans then it will be able to interpret the laws in a human way. However, that robot was unique in all the universe and was essentially a new species. The core point stands. Robots can never be better than us.
A small part of ethics is in the form of rules that we can express and follow. Even ignoring that these rules constantly change and adapt, they are only a small part of the whole.
Most of ethics happens with at most very general, unspecific rules. Basically "don't be an asshole". Good luck expressing that in a programming language. Most of this requires you to be and feel like a human and to use empathy - by putting ourselves into another persons position in our imagination, we can deduce which behaviour we would find acceptable and which not if the roles were reversed. We are light years away from such a thing in AI.
Assorted stuff I do sometimes: Lemuria.org
I'd read the Foundation Trilogy first, or at least the first book, before reading the robot novels. The robot novels are mysteries, but the Foundation Series is a must-read about the cyclic rise and fall of civilization
Article makes it sound like Humans have ethics all worked out.... hilarious! How about when robots start asking about CEO pay, outsourcing, leveraged buyouts, wiretapping, alcohol legality etc etc etc!
Who gets to decide who's ethics to follow.
The only way to build ethics into anything, including human decision making, is to start with facts. These days, stupid people have a loud voice, amplified by social media, and use it to claim "alternative facts" which aren't facts at all, but they are stupid and don't understand what a fact actually is.
Are you going to base ethical decisions on "alternative facts"? I don't want to be around for that, but here I am.
How are you going to eliminate politics from the programming process that will train these AI devices? Whose ethics will they base their decisions on?
... we will also build in Prejudice, Racism, Sexism and any number of other "unwanted" elements.
The creation of an AI instruction set that manages to be without prejudice will be nearly impossible, given that whoever makes the list will be skewing the process with their own priorities:
Human Life vs Overall Health of the Planet vs Quality of Life vs Sustainability, etc
Governments will want their say, while special interest groups (like the Rich) will also want to build in their own influence to further their continued existence over other, "lesser" interests...
Next will come those who "Game the System" so it will work in their favour no matter what. A "Back Door" if you will.
You can already see the HAL 900-like issues that will arise.
It's a shame that the 3-Laws system is so flawed.
It''s also a shame that "Do No Harm" is too broad and untenable as a foundation, either.
It's simple. as ethics are always in the eye of the beholder, you cannot build in real ethics as it's always the view of one person/group.
stop asking.
Also the humans in the stories seemed to be naive, believing that it was impossible to have the rules broken.
If you create a "perfect" creation, then it wouldn't have free will. The moment you insert free will, then you'll create the thing with an imperfection, the ability to choose wrong (or right). Unless you value free will above being perfect, then choosing wrongly is something that can be included in a perfect design.
There are a number of various stories about how choosing perfection leads to less than optimal results.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
You'd need a consolidated moral center.
Allahu Akbar!
Have gnu, will travel.
Nope. Too much concentration of power. Won't work. Next plan.
Communism was a terrible idea, but it is only one terrible idea among many. All 'command economies' suck, as they require power to be concentrated, where it will corrupt.
Can't be fixed. Suggest something self organizing, like capitalism.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Can We Build Ethics Into Automated Decision-Making?
Yes, if the algorithm is transparent, publicly published, and more or less straight-forward. Which is to say, if they're NOT ethical, people will bitch up a storm until it's fixed. Because democracy works. The exact same thing happens with, say, police department's policies on arresting people. There's a host of honestly vague and confusing algorithms set in law about when it's legal to arrest someone. We as the public have an equally vague sense of what those laws/rules/algorithms are and if a case comes forward where the law seems wrong, people complain, it gets political, and laws change. Case in point: Racial profiling is now illegal.
Same damn thing happens with or without the automation part. The only difference is if we expect a human to follow policy or a computer to follow programming. There will be deviations and blind-spots in the programmed algorithm just like there's humans that don't follow policy. Corruption will likely continue in some form or fashion. This is a boring and moot question from philosophers trying their damned hardest to forcefully insert a philosophical debate into an engineering problem. Stop that.
https://cdn.lolwot.com/wp-cont...
But that's no true scotsman report!
Every decision any program makes is based on an underlying ethical code. Retrieving data without corrupting it, for example. Faithfully reproducing and transmitting what you type. Retrieving the information that your requested. Making robocalls. These all have ethical underpinnings--either for good or bad.
The operation of software is an expression of the ethics of its programmer. You can't leave out ethics, good or bad, it's baked into the fabric of the code by the programmer.
Liar! ;)