Justice and Artificial Intelligence

 

Historians of technology are familiar with the problem of technological advances that outstrip the legal system, leading to situations that are clearly unfair, but leave some people with no legal recourse.  In a recent article in The Dispatch, artificial intelligence (AI) expert Matthew Mittelstaedt calls for a case-by-case approach to the problem of AI advancing beyond the borders of the law.

           

In the article, reporter Alec Dent cited the case of someone who recently filed for a copyright on an AI-generated piece of artwork.  The U. S. Copyright Office rejected the application, which was filed on behalf of the AI program itself, because it lacked the "human authorship" needed to create a valid copyright claim.  We don't know what would have happened if the programmer or software developer would have filed for a copyright on his or her own behalf, saying that the AI program was just a tool like an artist's brush.  But as AI systems take over more and more formerly creative jobs done by humans, the distinction may become harder to make.

 

ChatGPT, the powerful AI chatbot developed by the OpenAI firm, has attracted a lot of attention in academia for its ability to come up with plausible-sounding paragraphs of English text on virtually any topic, including those typically assigned as essay questions in the humanities.  It's not exactly like students weren't cheating before ChatGPT—there are numerous online stores where essays and even completed lab reports and exams can be purchased.  But ChatGPT creates work that, if not exactly original, hasn't existed in exactly the same form before, and thus falls between the stool of previously existing work that is merely copied, and the stool of truly original work that a human mind has originated. 

 

The original intent of copyright law, and anti-cheating rules at universities for that matter, was to allocate the rewards (or penalties) due to a piece of original work fairly.  If Mr. A wrote an essay on the downfall of the Soviet Union, whether it was for his history class or for The Atlantic, Mr. A should get the credit (or blame) for it, not some word-processing software or AI program. 

 

One big problem I foresee will arise from the defective anthropology that prevails in our modern culture.  Human beings are different from other animals and from machines due to a difference in kind, not merely in degree.  In the rush to embrace ever more impressive applications of AI, a lot of people have lost sight of this fact, if they ever believed it in the first place. 

 

The people at the U. S. Copyright Office seem to believe it, at least so far, but the person who applied for a copyright in the name of an AI program seems to think that there's no essential difference between human intelligence and AI.  Unfortunately, that view is very popular in high places—much of academia, the Silicon Valley world, and even in parts of government. 

 

A degraded view of humanity results when one assumes that there is no essential distinction between humans and advanced AI programs.  Of course there is a practical distinction, so far—AI systems still make stupid mistakes that a normally intelligent five-year-old wouldn't make.  But the AI optimists see this as merely a temporary condition that will disappear with further advances in the field.

 

Saying that human intelligence and AI is basically the same drags humans down to the level of machines.  And all the consequences of treating humans as machines will result from that attitude.  No AI expert wants to be treated like a machine.  But allowing AI programs to hold copyrights or be in charge of decisions that formerly required human input does exactly that to the subjects or patients of the AI system in question.

 

Justice is something that happens among human beings, and human beings are not machines.  People advocating for robot rights and similar policies that attempt to attribute human-like characteristics to AI programs are not exalting robots—they are degrading humans in a subtle and indirect way, a way that selectively degrades some humans more than others. 

 

We have already heard of cases in which AI-informed medical or legal decisions turned out to be highly discriminatory against certain minority groups.  The AI developers are rightly exercised about such problems, but it takes human beings to recognize that other human beings are being treated unfairly.  The whole body of the law can be regarded as a huge algorithm for executing justice among people—after all, what are laws but a set of rules and procedures for carrying them out? 

 

But as of yet, we have not handed over the execution of the laws to machines.  Lawyers, judges, and juries do that.  The institution of the jury trial, as rare as it's getting to be in criminal justice, is a common-law recognition that ordinary people deserve to be judged by other ordinary people, not just experts, whether the experts are human beings or computers. 

 

Turning over works of creativity and judgment to AI systems may be efficient.  It may even be fairer in a human sense than leaving such actions in the fallible hands of human beings.  But beyond a certain point—that point to be judged by humans, not by machines—it becomes a dereliction of duty, just as a student typing in his history assignment to ChatGPT and handing in the AI program's answer is a dereliction of his duty to think for himself.

 

The law will eventually catch up to today's AI innovations, as it always does.  Of course, by then we will have new advances, and so for a time at least we will see a kind of legal-AI arms race with AI leading and the law lagging behind.  But we will all be losers if legislators and AI developers forget that human beings are different in kind, not just degree, from AI programs.  If we forget that critical fact, we may deserve what happens if we do.

 

Sources:  Alan Dent's article "The Gaps Between the Law and Artificial Intelligence" was published on Mar. 8, 2023 at https://thedispatch.com/article/the-gaps-between-the-law-and-artificial-intelligence/.  I also referred to Wikipedia articles on ChatGPT and OpenAI.

 

NOTE:  There used to be an RSS feed that readers could subscribe to who wished to be notified of new blog articles, which are issued every Monday morning.  A reader recently pointed out to me that this feature was no longer functioning.  I am currently trying to repair the problems and add an automatic email notification system, but it may take a while, so I ask readers interested in these features to be patient.

Sentiment, Calculation, and Prudence

 

Some engineers eventually become managers, and managers not only of engineering projects but of entire companies or even public organizations.  The COVID-19 pandemic has thrown a spotlight on the question of how those in charge should allocate scarce resources (including technical resources) in the face of life-threatening situations.  And so I would like to bring you a brief summary sketch of three ways to do that:  two that are widely applied but fundamentally flawed, and one that is not so well known but can actually be applied successfully by ordinary mortals like ourselves.

 

None of this is original to me, nor to Robert Koons, the philosopher who describes them in a recent issue of First Things.  But originality is not usually a virtue in ethical reasoning, and in what follows, I'll try to show why.

 

In the 1700s, the Enlightenment thinkers Adam Smith and David Hume devised what Koons calls a "difference-making" way of coming up with moral decisions.  The way this process works is best described by an example.  To properly assess an action or even the lack of an action, you must figure out the net difference it makes to the entire world.  Koons uses the example of a homicidal maniac who, if left to himself, is bound to go out and kill three people.  Suppose you know about this maniac: you can either do nothing, or choose to kill him.  If you do nothing, three people die; if you kill him, only one person dies.  Other things being equal, the world is a better place if fewer people die, so the logic of difference-making says you must kill him.

 

That's an extreme example, but it vividly illustrates the rational basis of two popular ways of making moral decisions involving public health.  Let's start with the commonly-heard statement that every human life is of infinite value.  Few would dare to argue openly with that contention, yet if you try to use it as a guide for practical action, you run into a dilemma.  Even something as simple as your driving a car to the grocery store exposes other people to some low but nonzero chance of being killed by your vehicle.  If you take the infinite value of human life seriously, you will never drive anywhere, because infinity times (whatever small chance there is of running over someone fatally) is still infinity.

 

Koons calls one way of dealing with this dilemma "sentimentalism."  He's not talking about people who watch mushy movies, but the fact that the sentimentalist, in the meaning he uses, abandons logic for emotion and settles for life more or less as it is, but feels bad whenever anybody dies.  Such people exist in a constant state of deploring the world's failure to live up to the ideal that every human life is of infinite worth, but otherwise derive little moral guidance from that principle in practice.

 

The more hard-headed among us say, "look, we can't act on infinities, so if we put a finite but large value on human life, at least we can get somewhere. " Applying the difference-making idea to human lives valued at, say, a million dollars, allows you to make calculations and cost-benefit tradeoffs.  Engineers are familiar with technical tradeoffs, so many engineers find this method of moral decision-making quite attractive.  But one of many problems with this approach is that it requires one to take a "view from nowhere":  there are no boundaries to the differences a given decision makes, other than the world itself.  Again, if we try to be truly logically consistent, calculating all the differences a given life-or-death decision makes is practically impossible.

 

At this point Koons calls Aristotle and St. Thomas Aquinas to the rescue.  Operating under the umbrella of the classical virtue called prudence, Koons asks a given person in a given specific set of circumstances to judge the worthiness of a particular choice facing him or her.  He sets out four things that make a human act of choice worthy:  (1)  whether the human is applying rational thinking to the act, rather than random chance or instinct; (2) what the essential nature of the act is; (3)  what the purpose or end of the act is; and (4) what circumstances are relevant to the act. 

 

Unlike the difference-making approach, which imposes the impossible burden of near-omniscience on the decider, judging the worthiness of an action doesn't ask the person making the decision to know everything.  You simply take what you know about yourself, the kind of act you're contemplating, what you're trying to accomplish, and any other relevant facts, and then make up your mind.

 

In this process, some decisions are easy.  Should I kill an innocent person, a child, say?  Item (2) says no, killing innocent people is always wrong. 

 

Here's another situation Koons uses, but with an example drawn from my personal experience.  You walk outside your building past a bicycle owned by a person you really hate (call him Mr. SOB) and would like to see out of the way.  You notice that someone who hates Mr. SOB even more than you do has quietly disconnected the bike's brake cables, so that unless Mr. SOB checks his brakes before he gets on his bike, he will ride out into the street with no brakes and quite possibly get killed.  If you decide to say or do nothing, you have not committed any explicit act; you have simply refrained from doing anything.  But item (3) says your intentions in refraining were evil ones:  you hope the guy will get killed on his bike.  In this case, not doing anything is a morally wrong act, and you are obliged to warn him of the danger. 

 

And in less extreme cases, such as when public officials decide how to trade off lockdown restrictions versus spending money on vaccines or public assistance, the same four principles can guide even politicians (!) to make decisions that do not require them to be all-knowing, but do ask them to apply generally accepted moral principles in a practical and judicial way.

 

Of course, judiciousness and prudence are not evenly distributed virtues, and some people will be better at moral decision-making than others.  But when we look into the fundamental assumptions behind the decision-making process, we see that the difference-making approach has fatal flaws, while the traditional virtue-based approach using prudential judgment can be applied successfully by any individual with a good will and enough intelligence to use it.

 

Sources:  A much better  explanation of these approaches to moral reasoning can be found in Robert C. Koons's original article "Prudence in the Pandemic" which appeared on pp. 39-45 of the October issue of First Things, and is also accessible online at https://www.firstthings.com/article/2020/10/prudence-in-the-pandemic.

The Value of Personal Data

In some countries, all mineral rights are owned by the government.  In these countries, your family may have owned a plot of ground for generations.  But if the government thinks there's oil under it, they can come in, drill a well in your back yard, and make millions off the oil they find—and not give you a cent.  And it's all legal.

Other countries with different traditions regarding property rights view this situation as unjust.  In the U. S., for example, mineral rights usually vest in the property owner, which is how many otherwise dirt-poor Texans got rich when oil was discovered on their previously worthless land.

What goes for land that you buy should also go for things that you do.  If your actions lead to the creation of something that is of value, it would seem only fair that you should receive the properly evaluated equivalent for that value—in money or other valuable and exchangeable form. 

In Don't Be Evil, journalist Rana Foroohar describes how large tech companies such as Facebook, Amazon, and Google, as well as many smaller ones, collect data from us that is estimated to be worth nearly $200 billion.  When you click on a link, or lately even talk about certain things in the hearing of your personal assistant or your mobile phone, that information is noted, logged, and used to sell advertising and other things that the giant tech companies get real money for.  And as the Internet of Things grows with its ability to track our movements and other actions, this data stream will only get bigger. 

What do you get in exchange for providing the lifeblood of commerce for these firms?  They would say that you receive lots of free stuff in return—free web searches, a free personal Facebook page, free ads for things you may want to buy, and so on.  And this is true.  But it is far from the ideal free-market exchange of value, in which both parties come on a more or less even footing to an agreement after sharing essential information and comparing the potential exchange with any others that they might make with other parties.  

To put this situation in perspective, Faroohar points out that $200 billion is more than the total value of the annual U. S. agricultural output.  In other words, it's as if the large food companies (ConAgra, Tyson Foods, etc.) took everything that U. S. farmers grow, but paid them nothing for it.  Nobody would put up with that, and nobody would keep farming for very long either.

But just living an ordinary life these days means that you constantly do things that produce little bits of valuable data for the likes of Facebook, Google, and Amazon, whether you really mean to or not.  And in a technical sense, it is perfectly legal.  The cadres of tech-company lawyers who write the incomprehensible boilerplate on every software agreement that you lie about reading before being able to use the software make sure of that. 

One of the good outcomes of the otherwise horrific Nuremberg Trials of Nazi war criminals was the development of the Nuremberg Code, which has since been adopted to govern experiments involving human subjects.  One of the core principles of the code is that participants must give informed voluntary consent to being experimented on.  In other words, they must clearly understand the possible consequences of participating in the experiment and be able to say yes or no freely after making an informed decision.

If we regard the entire data-mining exploits of the big tech companies as a large-scale long-term experiment on the public, it is easy to see that we as individuals are at a vast disadvantage compared to the firms that profit from the data we generate.  Withholding our data would be difficult or impossible, especially when we don't even know that we're providing it (e. g. when Alexa or your mobile phone eavesdrops on your conversations).  And we have no idea what consequences will result from our actions.  And I include among these consequences the fact that the rich monopolies represented by the above-named firms get even richer, while in exchange I receive certain services that are convenient, true, but have value that I would be hard put to estimate in dollar terms.  Even if I did, it's doubtful that the value I perceive as getting from these firms would come anywhere close to the money they make off me by mining it.

The fact that I have to go through mental contortions even to think this way shows how deeply disguised the process is.  As an engineer, I'm trained to think of worst-case scenarios, and if I let my imagination wander in that direction with regard to the situation of data mining, I might come up with something like this:  The U. S. economy becomes even more two-tiered, with a very small number of very wealthy people working for or associated with the largest monopolistic tech firms, and everybody else on some kind of government-paid dole to keep them from starving, because most other jobs have vanished.  Research and development dries up here and moves to China, where most future technology developments happen under the firm control of the government there. 

I could go on, but I think I have made sufficiently clear the point that every day, with every click on a site associated with the largest tech firms, we allow them to obtain data that we make, but that they profit from. 

I do not pretend to have a good solution to this problem.  When similar situations arose in the past, such as during the "robber-baron" period of the 1800s when railroads monopolized essential transportation and commodities, the government had to intervene with countervailing forces embodied in things like the Interstate Commerce Commission and antitrust laws.  If the economy, the job market, and society in general is not to be further hollowed out by the activities of the large online tech firms, which are now indisputably having a negative effect on the political viability of our democracy, something needs to be done.  But I'm not sure what. 

Sources:  Rana Faroohar's book Don't Be Evil was published by Random House in 2019. The statistic about the value of data mined from the public being worth an estimated $197.7 billion by 2022 is on p. 25

The Robot Says You Flunked: Algorithms versus Judgment

Harvard and MIT have teamed to develop an artificial-intelligence system that grades essay questions on exams.  The way it works is this.  First, a human grader manually grades a hundred essays, and feeds the essays and the grades to the computer.  Then the computer allegedly learns to imitate the grader, and goes on to grade the rest of the essays a lot faster than any manual grader could—so fast, in fact, that often the system provides students nearly instant feedback on their essays, and a chance to improve their grade by rewriting the essay before the final grade is assigned.  So we have finally gotten to the point of grading essays by algorithms, which is all computers can do.

Joshua Schulz, a philosophy professor at DeSales University, doesn't think much of using machines to grade essays.  His criticisms appeared in the latest issue of The New Atlantis, a quarterly on technology and society, and he accuses the software developers of "functionalism."  Functionalism is a theory of the mind that says, basically, the mind is nothing more than what the mind does.  So if you have a human being who can grade essays and a computer that can grade the same essays just as well, why, then, with regard to grading essays, there is no essential difference between the two. 

With all due respect to Prof. Schulz, I think he is speculating, at least when he supposes that the essay-grading-software developers espouse a particular theory of the mind, or for that matter, any theory of the mind whatsoever.  The head of the consortium that developed the software is an electrical engineer, not a philosopher.  Engineers as a group are famously impatient with theorizing, and simply use whatever tools fall to hand to get the job done.  And that's what apparently happened here.  Problem:  tons and tons of essay questions and not enough skilled graders to grade them.  Solution: an  automated essay grader whose output can't be distinguished from the work of skilled human graders.  So where is the beef?

The thing that bothers Prof. Schulz is that the use of automated essay-grading tends to blur the distinction between the human mind and everything else.  And here he touches on a genuine concern:  the tendency of large bureaucracies to turn matters of judgment into automatic procedures that a machine can perform. 

Going to extremes can make a point clearer, so let's try that here.  Suppose you are unjustly accused of murder.  By some unlikely coincidence, you were driving a car of a similar make to the car driven by a bank robber who shot and killed three people and escaped in a car whose license plate number matches yours except for the last two digits, which the eyewitness to the crime didn't remember.  The detectives on the case didn't find the real bank robber, but they did find you.  You are arrested, and in due time you enter the courtroom to find seated at the judge's bench, not a black-robed judge, but a computer terminal at which a data-entry clerk has entered all the relevant data.  The computer determines that statistically, the chances of your being guilty are greater than the chances that you're innocent, and the computer has the final word.  Welcome to Justice 2.0. 

Most people would object to such a delicate thing as a murder trial being turned over to a machine.  But nobody has a problem with lawyers who use word processors or PowerPoints in their courtroom presentations.  The difference is that when computers and technology are used as tools by humans exercising that rather mysterious trait called judgment, no one being judged can blame the machines for an unjust judgment, because the persons running the machines are clearly in charge. 

But when a grade comes out of a computer untouched by human hands (or unseen by human eyes until the student gets the grade), you can question whether the grader who set the example for the machine is really in charge or not.  Presumably, there is still an appeals process in which a student could protest a machine-assigned grade to a human grader, and perhaps this type of system will become more popular and cease to excite critical comments.  If it does, we will have moved another step along the road that further systematizes and automates interactions that used to be purely person-to-person.

Something similar has happened in a very different field:  banking.  My father was a loan officer for many years at a small, independent bank.  He never finished college, but that didn't keep him from developing a finely honed gut feel for the credit-worthiness of prospective borrowers.  He wouldn't have known an algorithm if it walked up and introduced itself, but he got to know his customers well, and his personal interactions with them was what he based his judgment on.  He would guess wrong once in a great while, but usually because he allowed some extraneous factor to sway his judgment.  For example, once my mother asked him to loan money to a work colleague of hers, and it didn't work out.  But if he stuck to only the things he knew he should pay attention to, he did pretty well.

Recently I had the occasion to borrow some money from one of the largest national banks in the U. S., and it was not a pleasant experience.  I will summarize the process by saying it was based about 85% on a bunch of numbers that came out of computer algorithms that worked from objective data.  At the very last step in the process, there were a few humans who intervened, but only after I had jumped through a long series of obligatory hoops that allowed the bankers to check off "must-do" boxes.  If even one of those boxes had been left blank, no judgment would have been required—the machine would say no, and that would have been the end of it.  I got the strong impression that the people were there mainly to serve the machines, and not the other way around.

The issue boils down to whether you think there is a genuine essential difference between humans and machines.  If you do, as most people of faith do, then no non-human should judge a human about anything important, whether it's for borrowing money, assigning a grade, or going to jail.  If you don't think there's a difference, there's no reason at all why computers can't judge people, except for purely performance-based factors such as the machines not being good enough yet.  Let's just hope that the people who think there's no difference between machines and people don't end up running all the machines.  Because there's a good chance that soon afterwards, the machines will be running the people instead.

Sources:  The Winter 2014 issue of The New Atlantis carried Joshua Schulz's article Machine Grading and Moral Learning on pp. 109-119.  The New York Times article from which Prof. Schulz learned about the AI-based essay grading system is available at http://www.nytimes.com/2013/04/05/science/new-test-for-computers-grading-essays-at-college-level.html.  The Harvard-MIT consortium's name is edX.

Note to Readers:  In my blog of June 16, 2014, I asked for readers to comment on the question of monetizing this blog.  Of the three or four responses received, all but one were mostly positive.  I have decided to attempt it at some level, always subject to reversal if I think it's going badly.  So in the coming weeks, you may see some changes in the blog format, and eventually some ads (I hope, tasteful ones) may appear.  But I will try to preserve the basic format as it stands today as much as possible.