COMMENTARY: By Tracy Campbell
There is something grotesque about taking one of the greatest intellectual achievements in human history — a technology capable of translating languages, modelling proteins, accelerating scientific discovery, assisting doctors, educating children and placing an astonishing repository of human knowledge at an ordinary person’s fingertips — and harnessing it to the ancient government business of deciding who may be killed.
States hardly needed technological assistance with that.
For as long as states have made war, they have assassinated, disappeared, tortured and executed.
- READ MORE: Lavender’: The AI machine directing Israel’s bombing spree in Gaza
- The controversial documentary NAZA: 10 reasons to defend this whistleblower war on Gaza film
- Other AI killing in Gaza
They have acted on rotten intelligence, mistaken civilians for enemies, bombed the wrong houses and buried the innocent alongside the intended target.
Humanity has accumulated centuries of law in an attempt to restrain that capacity for violence.
Now we have given it an algorithm.
A +972 Magazine investigation into Israel’s “Lavender” system used in the Gaza genocide describes an AI-assisted process that marked tens of thousands of Palestinians as suspected militants for potential assassination. Six Israeli intelligence sources described enormous target lists, scant human scrutiny and predetermined tolerances for civilian deaths.
Significance of Lavender
The significance of Lavender reaches far beyond one system, one military, or one war.
Artificial intelligence does not hate anyone. It has no territory to conquer, no frightened electorate to appease, no flag to salute and no enemy it wants dead.
It cannot decide that one nationality, ethnicity or political movement is hateful. It cannot lust for revenge.
Humans supply the malice
And AI offers governments a hideous new possibility: not merely to make lethal decisions, but to industrialise them.
A human life can become data, the data a probability, the probability a classification and the classification a target. Somewhere along that progression, the unique, precious universe of a living person — somebody’s child, parent, lover, neighbour — disappears into a machine-generated assessment which can be processed thousands of times faster than any human being could properly investigate the people concerned.
That isn’t intelligence replacing human cruelty — it is human cruelty acquiring computational power.
Recreated in machinery
Alan Turing wondered whether machines could think. John McCarthy gave artificial intelligence its name. Claude Shannon helped establish the mathematics on which our information age was built. Marvin Minsky devoted much of his life to understanding whether intelligence itself could be recreated in machinery.
They helped begin an intellectual adventure of almost preposterous ambition: to understand intelligence sufficiently well that we might reproduce some part of it.
Today Demis Hassabis speaks of developing AI to benefit humanity. Dario Amodei leads Anthropic, founded around the problem of making immensely powerful AI safe.
Sam Altman heads OpenAI, whose stated mission is to ensure artificial general intelligence benefits all humanity.
Surely this cannot be what benefiting humanity was supposed to mean.
AI may help us understand disease, proteins, climate, mathematics, language and perhaps eventually the nature of intelligence itself. Yet the same extraordinary capacity to detect patterns across quantities of information beyond the reach of a human mind can be turned towards finding patterns in a population and deciding which people resemble the people a state wants dead.
And therein lies another danger.
Human judgement, evidence
When a human intelligence officer says, “I believe this person is a combatant,” we understand that a human being has made a judgment and can demand the evidence behind it.
When a machine assigns a probability, the mathematics can lend the conclusion an entirely undeserved aura of objectivity. But an algorithm cannot cleanse dubious intelligence, discriminatory assumptions or corrupted data merely by processing them at enormous speed.
Nor can it assume responsibility.
The analyst can point to the algorithm. The commander can point to the analyst. The politician can invoke the demands of war. The engineer can say the machine was only an advisory tool.
The machine, magnificently innocent of everything, points nowhere.
That is why human responsibility must remain absolute. I would go so far as to say that any use of AI in warfare should be prohibited.
Artificial intelligence must never become moral laundering for state violence. A computer-generated confidence score does not acquire the moral weight of evidence merely because mathematics produced it.
10,000 suspicions
Processing 10,000 suspicions in the time a human being could properly investigate 10 does not make those suspicions more reliable. It merely makes killing scalable.
Turing did not ask whether machines could think so that governments could think less.
Human beings have finally built machines capable of doing things our ancestors could scarcely have imagined.
If one of the achievements of artificial intelligence is that it enables us to kill people faster, we will not have taught machines something new.
We will simply have taught them something terribly old.
Tracy-Moana Campbell is an employment relations advocate based in Masterton and an activist on Palestinian human rights. She had a commercial orange grove in Lebanon 10km from the Israeli border in 2005-2011. Israel bombed her oranges in 2006. It is a war crime to destroy civilian food sources. This commentary was originally a social media post but has been republished with the author’s permission.
