AI for Good? No. AI for War.
Competing paths for the future
Any given technology is value-neutral. It has no inherent identity as good or as evil. As with anything, context matters.
This is as true for AI technologies as it is true for any other.
I think that it is possible to look at AI from a place that understands, for example, the morally questionable ways many (most? all?) LLMs were developed through the theft of copyrighted works, without necessarily implicating the underlying technology.
Similarly, it is not the technology itself that is responsible for regrettable outcomes, say in the example of the use of ChatGPT by the Tumbler Ridge shooter. It is the very human guiderails and processes that matter and that shape AI’s use.
It is so important to separate the businesses and business models from the technologies themselves. Likewise, we need to distinguish between any individual business and the wider economic and political system in which it operates.
Looking at that wider context can help reveal where power resides, who is shaping the development paths of technologies, and who benefits and who suffers from them.
We can see this at play in two different discourses and applications of AI.
On the one hand, you have very commendable initiatives that seek to position AI for the good of people. Take this new CIGI policy brief by Cornelia C. Walther, who sets out a case for “Prosocial AI” — “systems tailored, trained, tested and targeted to advance human well-being and planetary health.” Walther maps out for middle-income countries a “fourth path” that differs from the approaches of Silicon Valley, Brussels, and Beijing and would instead shape AI development in ways that actually benefit humanity.
There is a lot to commend this brief, and as I’ve explored before, it is important to imagine and map paths to different futures. However, the vision work like this sets out is particularly jarring given the very real ways AI is being developed and deployed right now.
As reported in The Guardian, AI is being used by the US and Israel in their strikes against Iran as it “shortens the kill chain,” reducing the planning time needed to approach and launch a strike. This use follows on from Israel’s own copious use of AI in Gaza. One Israeli intelligence source that used Lavender, Israel’s AI targeting system, described exactly how that kill chain was shortened: “I would invest 20 seconds for each target at this stage, and do dozens of them every day. I had zero added-value as a human, apart from being a stamp of approval. It saved a lot of time.”
International and humanitarian law requires weapons to be able to distinguish between military and civilian targets. However, academics quoted in a Nature piece are clear that “LLM-powered, fully autonomous weapons without human oversight are not currently reliable and do not comply with international laws.” Even where there is oversight, such as in the Lavendar case above, it is often nominal. Indeed, “there is no evidence that AI lowers civilian deaths or wrongful targeting decisions and it may be that the opposite is true.”
There is nothing inherent about AI that makes its use for raining down death already a fait accompli, while its use for “prosocial” purposes remains a matter of academic debate. That one use receives hundreds of millions in funding, and the other, if it occurs to policymakers at all, is an afterthought, is a reflection not of technology but of the human political and economic structure we have built and the outcomes they incentivize.
I’m not as naive as to think that there should be no defence applications of advanced technologies. I’m a diplomatic historian by training who has lectured on centuries of bloody, war-filled history. Threats exist, and people will exploit weakness. However, that same background also makes me keenly aware that technological arms races and concerns about security can create the very threats they were meant to ward off. Moreover, it makes me very conscious of how pursuing security above all else can cost us our very humanity.
At a time when Canada’s new Defence Industrial Strategy signifies a largely uncontested and unprecedented peacetime militarization of the Canadian economy, these decisions about the structures we are establishing and the incentives we are creating matter. What path are we actually laying down for the future? What type of application of these “dual-use” technologies, such as AI, are we actually incentivizing? And, who, ultimately, will pay the price?