68 Science and technology The EconomistJanuary 20th 2018
Computers and criminal justiceAlgorithm’s dilemma
I
N AMERICA, computers have been
used to assist bail and sentencing deci-
sions for many years. Their proponents
argue that the rigorous logic of an algo-
rithm, trained with a vast amount of
data, can make judgments about wheth-
er a convict will reoffend that are un-
clouded by human bias. Two researchers
have now put one such program, COM-
PAS, to the test. According to their study,
published in Science Advances, COMPAS
did neither better nor worse than people
with no special expertise.
Julia Dressel and Hany Farid of Dart-
mouth College in New Hampshire select-
ed 1,000 defendants at random from a
database of 7,214 people arrested in Bro-
ward County, Florida between 2013 and
2014, who had been subject to COMPAS
analysis. They split their sample into 20
groups of 50. For each defendant they
created a short description that included
sex, age and prior convictions, as well as
the criminal charge faced.
They then turned to Amazon Mechan-
ical Turk, a website which recruits volun-
teers to carry out small tasks in exchange
for cash. They asked 400 such volunteers
to predict, on the basisof the descrip-
tions, whether a particular defendant
would be arrested for another crime
within two years of his arraignment
(excluding any jail time he might have
served)—a fact now known because of
the passage of time. Each volunteer saw
only one group of 50 people, and each
group was seen by 20 volunteers. When
Ms Dressel and Dr Farid crunched the
numbers, they found that the volunteerscorrectly predicted whether someone
had been rearrested 62.1% of the time.
When the judgments of the 20 who
examined a particular defendant’s case
were pooled, this rose to 67%. COMPAS
had scored 65.2%—essentially the same as
the human volunteers.
To see whether mention of a person’s
race (a thornyissue in the American
criminal-justice system) would affect
such judgments, Ms Dressel and Dr Farid
recruited 400 more volunteers and re-
peated their experiment, this time adding
each defendant’s race to the description.
It made no difference. Participants identi-
fied those rearrested with 66.5% accuracy.
All thissuggests thatCOMPAS, though
not perfect, is indeed as good as human
common sense at parsing pertinent facts
to predict who will and will not come to
the law’s attention again. That is encour-
aging. Whether it is good value, though, is
a different question, for Ms Dressel and
Dr Farid have devised an algorithm of
their own that was as accurate asCOM-
PASin predicting rearrest when fed the
Broward County data, but which in-
volves only two inputs—the defendant’s
age and number of prior convictions.
As Tim Brennan, chief scientist at
Equivant, which makesCOMPAS, points
out, the researchers’ algorithm, having
been trained and tested on data from one
and the same place, might prove less
accurate if faced with records from else-
where. But so long as the algorithm be-
hind COMPASitself remains proprietary,
a detailed comparison of the virtues of
the two isnot possible.Are programs better than people at predicting recidivism?T
HE exhausting chore of raising young is
one a few birds manage to avoid. By
laying their eggs in the nests of others, they
dupe those others into feeding their nest-
lings. Such brood parasitism has arisen in-
dependently at least three times, in the
groups known as cuckoos, cowbirds and
honeyguides. That gives biologists a tool
with which to explore the phenomenon of
convergent evolution, in which unrelated
lines with similar ways of life evolve simi-
lar adaptations that help them to thrive.
One feature shared by cuckoos, cow-
birds and honeyguides is that the shells of
their eggs are all thicker than those of the
birds they parasitise—sometimes by as
much as 30%. This lookslike a classic case
of convergent evolution, but no one has
been able to prove the point by demon-
strating a benefit derived from it that is con-
nected directly with brood parasitism. Li-
ang Wei, of Hainan Normal University, in
China, thinks he has now done so. His
work, just published in the Science of Na-
ture, suggests that the greaterthickness of
brood-parasites’ eggshells provides insula-
tion, which speeds up the eggs’ incubation.
This ensures they hatch before their hosts’
eggs do, thus granting the parasitic hatch-
lings time to dispose of their incipient ri-
vals by puncturing the eggs containing
them orpushing those eggsout of the nest.
To test this idea, Dr Liang and his col-
leagues decamped from their home in Chi-
na’s southernmost province to Heilong-
jiang, its northernmost. Their destination
was Zhalong National Nature Reserve,
where reed warblers are parasitised by
common cuckoos. They picked this pairing
of host and parasite because the two spe-
cies’ eggs are, by chance, of similar sizes.
That made it easier to compare the rates atwhich the two sorts of egg lost heat, since
the ratio of surface area to volume is a cru-
cial variable in matters thermodynamic.
The team searched the reserve for war-
bler nests. When they found one that also
had a cuckoo egg, they removed that egg,
together with a warbler egg, and brought
the pair to their laboratory, where they
placed them in an incubator at 37.5°C—
their natural incubation temperature.
On the first day of this process, and on
three further occasions, spaced three days
apart, each of the 15 pairs of eggs the team
had collected was taken out of the incuba-
tor for 20 minutes—the average amount of
time clutchesare left unattended by war-
bler mothers (the fathers having long de-
parted) when they go foraging. As soon as
the eggs came out of the incubator, the
team took thermal images of them, permit-
ting their temperatures to be estimated.Then, justbefore they put the eggs back,
they took a second set of images.
The findings were clear. During the 20-
minute periods outside the incubator the 15
warbler eggs lost 4.42°C on average, where-
as the 15 cuckoo eggs lost an average of
4.15°C. This may not seem a huge differ-
ence, but experience of incubating bird’s
eggs artificially shows that actually it is.
Given these findings, Dr Liang argues
that the thicker eggshells do indeed give
the embryos within the developmental
edge they need to hatch first. As to why
warblers and other victims of brood para-
sitism fail to retaliate bythemselves evolv-
ing thicker eggshells, and thus faster-hatch-
ing young, that is unclear. Presumably the
extra cost of doing so is not worthwhile in a
world where a minority of nests are paras-
itised. Proving this, though, would require
a whole new research project. 7EvolutionShell game
An intriguing example of convergent
evolution is explained