Moral Relativism

I’ve mentioned WEIRD on this blog before. For those who haven’t heard, the basic idea is that college students in North America are very unlike most people on Earth, yet psychology usually considers them type specimens for our entire species.[1] This calls into question a lot of “universals” proposed in psychology papers.

You might think morality would be a clear exception to that. Young people are fitter, old people have already contributed most of what they will to society; if one of each group is put in danger, we should try to save the former first before the latter. Right?

We are entering an age in which machines are tasked not only to promote well-being and minimize harm, but also to distribute the well-being they create, and the harm they cannot eliminate. Distribution of well-being and harm inevitably creates tradeoffs, whose resolution falls in the moral domain. Think of an autonomous vehicle that is about to crash, and cannot find a trajectory that would save everyone. Should it swerve onto one jaywalking teenager to spare its three elderly passengers? Even in the more common instances in which harm is not inevitable, but just possible, autonomous vehicles will need to decide how to divide up the risk of harm between the different stakeholders on the road. […]

… we designed the Moral Machine, a multilingual online ‘serious game’ for collecting large-scale data on how citizens would want autonomous vehicles to solve moral dilemmas in the context of unavoidable accidents. The Moral Machine attracted worldwide attention, and allowed us to collect 39.61 million decisions from 233 countries, dependencies, or territories.

Awad, Edmond, Sohan Dsouza, Richard Kim, Jonathan Schulz, Joseph Henrich, Azim Shariff, Jean-François Bonnefon, and Iyad Rahwan. “The Moral Machine Experiment.” Nature 563, no. 7729 (November 2018): 59. https://doi.org/10.1038/s41586-018-0637-6.

Well, the data is in. I could do an entire blog post on just their summary, but for now merely note the benevolent sexism,[2] focus on punishment, classism, deontology, and cat hatred. That left bar chart is confusing; the bar between the elderly and the young isn’t indicating that both would be spared equally often, but that children would be spared 49 percentage points more often.

Figure 2 (global preferences) from Edmond et. al (2018).

Sure enough, there’s a clear preference for sparing the young over the elderly. But hold on here; this was an online survey, and the map of people playing the “game” shows a definite skew towards North America and Europe. This summary is “global” in that it aggregates all the data together, but not in the sense that it represents the globe’s preferences. We would do better to break down the responses into countries and analyze that.

First, we observe systematic differences between individualistic cultures and collectivistic cultures. Participants from individualistic cultures, which emphasize the distinctive value of each individual, show a stronger preference for sparing the greater number of characters (…). Furthermore, participants from collectivistic cultures, which emphasize the respect that is due to older members of the community, show a weaker preference for sparing younger characters (…). Because the preference for sparing the many and the preference for sparing the young are arguably the most important for policymakers to consider, this split between individualistic and collectivistic cultures may prove an important obstacle for universal machine ethics. …

We observe that prosperity (as indexed by GDP per capita) and the quality of rules and institutions (as indexed by the Rule of Law) correlate with a greater preference against pedestrians who cross illegally (…). In other words, participants from countries that are poorer and suffer from weaker institutions are more tolerant of pedestrians who cross illegally, presumably because of their experience of lower rule compliance and weaker punishment of rule deviation. This observation limits the generalizability of the recent German ethics guideline, for example, which state that “parties involved in the generation of mobility risks must not sacrifice non-involved parties.” …

… we observe that higher country-level economic inequality (as indexed by the country’s Gini coefficient) corresponds to how unequally characters of different social status are treated. Those from countries with less economic equality between the rich and poor also treat the rich and poor less equally in the Moral Machine. … the differential treatment of male and female characters in the Moral Machine corresponded to the country-level gender gap in health and survival (a composite in which higher scores indicated higher ratios of female to male life expectancy and sex ratio at birth—a marker of female infanticide and anti-female sex-selective abortion). In nearly all countries, participants showed a preference for female characters; however, this preference was stronger in nations with better health and survival prospects for women. In other words, in places where there is less devaluation of women’s lives in health and at birth, males are seen as more expendable in Moral Machine decision-making.[1]

Just consider the consequences of all this: do we have to change the moral calculus of a self-driving car if the owner sells it to someone in another country, or if they merely drive into one? If we tweak the calculus to remove all benevolent sexism, people will feel these cars are unfairly harming women; either we need to pair driver-less cars with a global education campaign to eliminate sexism, or there’ll be a mass movement to bake sexism into our cars. At the same time, self-driving cars will save quite a few lives no matter what moral system they follow; should we sweep all this variation under the rug, and focus on the greater good?

Our moral code depends strongly on where we live and how well we’re living, so how could we all agree to a universal moral code, let alone follow it? Non-normative moral relativism, contrary to the name, is the human norm, and imposing a universal moral code on us will cause all sorts of havoc.

Except when it comes to cats.

[HJH 2018-12-05: Huh, where did that graphic go? I’ve popped it back into place.]


[1] Henrich, Joseph, Steven J. Heine, and Ara Norenzayan. “Beyond WEIRD: Towards a Broad-Based Behavioral Science.” Behavioral and Brain Sciences 33, no. 2–3 (June 2010): 111–35. doi:10.1017/S0140525X10000725.

[2] Glick, Peter, and Susan T. Fiske. “An Ambivalent Alliance: Hostile and Benevolent Sexism as Complementary Justifications for Gender Inequality.” American Psychologist 56, no. 2 (February 1, 2001): 109–18.

Judicial Math

Whew, quite a week of news, eh? The Manafort verdict has stuck with me, if only for this detail.

One of the jurors from the recently-concluded trial of Paul Manafort has described herself as a strong supporter of President Trump. She said she drove every day to the Alexandria courthouse where Mr. Trump’s former campaign chairman was being tried with her “Make America Great Again” cap in the back seat, and that she planned to vote again for Mr. Trump if he runs for reelection in 2020. She said she thought prosecutors had targeted Mr. Manafort as a way to get dirt on Mr. Trump, and that she didn’t want Mr. Manafort to be guilty. Nonetheless, she voted to convict him because the evidence of his guilt “was overwhelming.” […]

The jury couldn’t come to unanimous agreement on 10 other counts and a mistrial on those charges was declared. Ms. Duncan revealed that there was just one juror who held out on conviction on those counts, citing reasonable doubt. The other eleven jurors were convinced of Mr. Manafort’s guilt.

I don’t know why that juror held out, so let’s instead consider a hypothetical. Earlier, I argued that Democratic and Republican voters were more polarized than first appeared because roughly 10-20% of the population can be convinced of nearly anything. The first juror in the Manafort trial to out themselves bought pretty heavily into some of Trump’s conspiracy theories, so they must have some grip on the general public.

What if this 10-20% of the populace was so deep into these theories that they’d never find one of Trump’s associates guilty? That would be a huge problem if they were on a jury. What are the odds of such an event occurring?

We can calculate this ourselves, via the Binomial distribution.

The expected number of jurors that'll never convict on a 12-jury panel. I'm cheating a bit and using a Beta, to create more visual distance between the 10% and 20% cases; for the latter, only use those values which correspond to an integer along the X axis.Assuming a 12-person jury, if 10% of the population would refuse to convict under any circumstance, then there’s about a 72% chance of at least one such person being a juror; if 20%, then there’s a whopping 93% chance. Since the US Federal courts require unanimity to reach a verdict, those are also the minimum odds of a mistrial on one count!

There’s an obvious workaround, drop unanimity and permit eleven people to reach a verdict. The minimum odds of a mistrial drop to 34%, if 10% of all people would refuse to convict, or 72.5% in the 20% case. Is that acceptable to you, or would you like those values to be lower? We can use math and computers to determine the ideal quorum of jurors needed to satisfy your threshold. Let’s define t as the minimum odds of a mistrial, n as the number of jurors, k as the minimum number of guilty votes needed to achieve a conviction, and q as the proportion of people guaranteed to refuse to convict. For any given combination of those, the minimum odds are

t = sum from p=(n-k+1) to n (n, p) q^p (1-q)^(n-p)

The good news: you can drive t to be as low as you wish. The bad: you accomplish that by inflating the size of the jury pool while keeping the quorum low, which means the weight of the evidence necessary to convict drops. Avoiding partisan bias means more false convictions, and vice-versa, so we have to calculate our preferred trade-off.

A chart of the minimum odds of mistrial, for a given jury size and quorum necessary to convict.

This math is par for the course. Every judicial system puts numbers to these questions:

  1. How many guilty people should be allowed to walk free?
  2. How many people should be convicted of a crime they didn’t commit?
  3. How much should we invest into those who have been convicted of a crime, and how should we spend those funds?
  4. How much should we spend on crime prevention, and which programs are the most effective?

For instance, its been estimated that at least 4.1% of all convicts given a death sentence in the US were falsely convicted; is that rate of killing innocent civilians acceptable, or should it be lowered? Of the hundred thirty-seven prisoners freed from US jails in 2017, their average time behind bars was 10.7 years; is putting an innocent person behind bars for that length of time something we can tolerate as a society, or should it be lowered? If it should be lowered, are we going to do that by doing more aggressive post-conviction audits, better training for police and prosecutors, both, or are there more effective tactics out there?

Working out this math also changes our judicial philosophy. If we build our system so that it punishes the guilty, then our false conviction rate had better be low. If instead we build our system so that it makes them better citizens, then putting an already-good citizen in there isn’t a big loss and we can instead tune other variables.

The only real choice here is if we consciously put those numbers in place ourselves, receive a nasty shock when we later calculate them, or pretend those questions don’t exist. Currently, we’re doing a lot of the last two in Canada and the US.

What a Day To Be Wrong

I don’t to come across as a predictive genius. Just today, in fact, I guessed that Manafort’s verdict would land today, and he’d be found guilty on most of the eighteen counts.

Former Trump campaign manager Paul Manafort has been convicted on 8 of 18 counts by a federal jury in Virginia, and a mistrial was declared on the remaining 10 counts when the jury deadlocked.

See? I was one short of “most.” That note from the jury threw me off, I thought it indicated more consensus than there was. Anyway, the guilty convictions were for multiple years of tax fraud and a touch of bank fraud. It’s a bit disappointing, as he’s likely guilty of a lot more, but those guilty verdicts alone could carry a maximum of 80 years. And “deadlocked” isn’t the same as “not guilty,” the rules allow prosecutors to try again. While they must be feeling a bit sore about the judge they got, between the guilty verdicts they got and Manafort’s second trial in a few weeks those prosecutors are probably more soothed than sore and content to let bygones be bygones. The deadline’s a week from now, so we won’t have to wait long to learn how that prediction turned out.

As for Michael Cohen, I was telling everyone to beware of all the spin in the air and wait for more concrete steps.

Cohen pleaded guilty to eight counts: five of tax evasion, one of making a false statement to a financial institution, two related to illegal campaign contributions.

Looks like the early reports were more accurate than I gave them credit for. I definitely recommend giving the plea agreement charging document a read, it’s a trip.

1. From in or about 2007 through in or about January 2017, MICHAEL COHEN, the defendant, was an attorney and employee of a Manhattan-based real estate company (the “Company”). COHEN held the title of “Executive Vice President” and “Special Counsel” to the owner of the Company (“Individual-1”).
2. In or about January 2017, COHEN left the Company and began holding himself out as the “personal attorney” to Individual-1, who at that point had become the President of the United States. […]

27. In or about August 2015, the Chairman and Chief Executive of Corporation-1 ( “Chairman-1”), in coordination with MICHAEL COHEN, the defendant, and one or more members of the campaign, offered to help deal with negative stories about Individual-l’s relationships with women by, among other things, assisting the campaign in identifying such stories so they could be purchased and their publication avoided. Chairman-1 agreed to keep COHEN apprised of any such negative stories.
28. Consistent with the agreement described above, Corporation-1 advised MICHAEL COHEN, the defendant, of negative stories during the course of the campaign, and COHEN, with the assistance of Corporation-1, was able to arrange for the purchase of two stories so as to suppress them and prevent them from influencing the election. […]

37. In or about January 2017, MICHAEL COHEN, the defendant, in seeking reimbursement for election-related expenses, presented executives of the Company with a copy of a bank statement from the Essential Consultants bank account, which reflected the $130,000 payment COHEN had made to the bank account of Attorney-1 in order to keep Woman-2 silent in advance of the election, plus a $35 wire fee, adding, in handwriting, an additional “$50,000.” The $50,000 represented a claimed payment for “tech services,” which in fact related to work COHEN had solicited from a technology company during and in connection with the campaign. COHEN added these amounts to a sum of $180,035. After receiving this document, executives of the Company “grossed up” for tax purposes COHEN’ s requested reimbursement of $180,000 to $360,000, and then added a bonus of $60,000 so that COHEN would be paid $420,000 in total.

That’s a violation of campaign finance laws aimed at influencing the 2016 election, in consultation with a presidential candidate, his team, and said candidate’s business. As if that wasn’t bad enough, Cohen’s lawyer followed up with a one-two punch.

Today he stood up and testified under oath that Donald Trump directed him to commit a crime by making payments to two women for the principal purpose of influencing an election. If those payments were a crime for Michael Cohen, then why wouldn’t they be a crime for Donald Trump?

“I can tell you that Mr. Cohen has knowledge on certain subjects that should be of interest to the Special Counsel,” [Lanny] Davis told [Rachael] Maddow. “The obvious possibility of a conspiracy to collude and corrupt the American democracy system in the 2016 election,which the Trump Tower meeting was all about. But also knowledge about the computer crime of hacking and whether or not Mr. Trump knew ahead of time about that crime and whether he cheered it on. We know that he publicly cheered it on, but did he also have private information?”

Before the Lanny Davis news, though, I’d predicted that Trump’s West Virginia rally would be off the hook.

Trump has concluded after about an hour and 15 minutes, longer than usual. No mention of Cohen, Manafort, or Mueller, and less time spent on the Russia probe – just a couple sentences – than at most events.

Phooey, and the pre-show was so promising. Both the Manafort and Cohen verdicts were handed out within minutes of each other, and a mere two hours before Trump’s rally, so the silence may simply be because he didn’t have time to absorb what was happening? Maybe instinct kicked in, and Trump realized anger would only make him look more guilty? Or maybe visions of pardons are dancing through his head? It’s still too early to be sure, but nonetheless that’s another flubbed prediction.

Ah well, back to that post on probability theory.