The Hypocrisy of American Commitments to AI Safety
The US is much more committed to robot domination than international law
You don’t have to be a legal expert to know that the last few years have not exactly been a high point for countries adhering to their international commitments. Leading the retreat is Trump’s America, breaking commitments with enthusiastic abandon from Iran to trade.
Unfortunately, it’s time to add something else to that list: AI safety, and the importance of human intervention as an AI safeguard. Consider the following pair of recent developments in the nascent world of drones and artificial intelligence.
First, news from the US navy. Their engineers have built a new kind of AI system, called ‘Heterogeneous Collaborative Unmanned Systems’, or HCUS. Heterogeneous, if you need to Google that (I certainly did), means diverse in character. These sophisticated AI systems are housed in a variety of different robot shells, with the ability to operate in teams. They can ‘talk’ to each other, co-ordinate on assigned tasks and enable the delegation of increasingly complicated military operations.
Like a child learning to find their own feet, with this new capacity for independent problem-solving HCUS need less direction from their human overlords compared to other drones. Unlike with children, America’s navy is desperately keen to deploy HCUS as soon as possible into battlefields and high-risk geopolitical arenas to conduct risky intelligence-gathering and interception work.
The second recent development in AI comes from the arcane world of international law, where developed states are increasingly pushing for greater co-operation and rule-setting around artificial intelligence. In particular, the OECD (a group of developed countries) and the G20 (the biggest developed countries) have agreed on principles around AI safety, under the moniker ‘human-centred artificial intelligence’. Forty-two countries, including the United States, agreed to endorse this sentiment:
AI systems should be designed in a way that respects the rule of law, human rights, democratic values and diversity, and they should include appropriate safeguards — for example, enabling human intervention where necessary — to ensure a fair and just society.
This principle reflects concerns by a wide range of actors that so-called ‘lethal autonomous weapons’ - drones with the capacity to kill without human input - are a serious risk to humankind. These advocates argue that some level of human input is needed morally, to evaluate proportionality and to avoid a ‘accountability gap’ where no human can be held responsible for an AI mistake.
It is this quandary that the OECD/G20 is looking to address by asking states to commit to a safeguard system based on human intervention. At the same time, the United States (as well as China and Russia) is working to address quite the opposite challenge, deep in their military labs: building AI and drones with the greatest degree of autonomous capability possible. This aligns with their military incentives: future battles between drones are likely to be decided by which technology is the most autonomous. But these two developments, one towards AI safety and one towards AI independence, do not sit particularly easily with each other.
The (Lost) Art of the Deal
It would be one thing if America signed on to the OECD/G20 principles that warn against autonomous AI while working secretly to develop the world’s strongest autonomous military. But the US took it to another level by broadcasting their latest step towards robot domination - the new HCUS - to the world at large, less than two weeks after the G20 principles were announced. They promise a week-long demonstration of the drones, no less.
None of this has inspired much critique (or attention) in the media, or from the OECD/G20 states (as far as we are aware). This might have something to do with the fact that HCUS themselves are typically small, and appear not-threatening. But that does not change the fact that their existence, and public showcase, flies in the face of the commitments America has made to AI safety. The fundamental utility (and technological breakthrough) of HCUS is that they do not require human involvement, or at least they require much less human involvement than previous iterations of drone technology.
And one would be wrong to assume that HCUS are relatively harmless because they are not large and bristling with missiles (as, say, a Predator drone is). This is partly because some HCUS will indeed be equipped with weapons, including warheads, to autonomously target enemy vehicles or infrastructure - further in breach of the spirit of the OECD/G20 agreement. The US is hardly hiding the offensive capabilities of its new drones: the Economist quotes that the demonstration of HCUS will “culminate in a remote operator on-demand offensive attack on a simulated target”.
However, the most important reason to be wary of HCUS is because the technological potential of the breakthrough far exceeds the first applications for the new AI. We are talking about advanced military systems that have the ability to work with each other in sophisticated ways, including re-orientating tactics in response to new threats and co-ordinating tracking and targeting across large distances and many drones. The military potential for ‘drone swarms’ is well-documented, and provides just one (admittedly scary) example of what HCUS technology is capable of.
Maybe the hypocritical development and flaunting of HCUS should not have surprised me, given the track record of the world and America specifically failing to live up to international commitments. But the utterly brazen nature of this particular example still rankles. America publicly committed to the principle of AI retaining human control in the same month as the unveiling of a new class of militarised drone, the chief technological distinction of which is the lack of human input required to complete military missions.
It’s true that none of the commitments at the OECD or G20 were “legally binding” for the countries that signed up. But it’s unquestionable that the international order relies on states living up to commitments they make to each other. America’s hypocrisy continues to erode the integrity of this all-important principle.