In AI race, ethics collateral damage?

Should AI be allowed to power killer robots? Anthropic, which has developed software that can be used to target weapons autonomously, thinks it should not. Pentagon, which has contracted the San Francisco-headquartered AI company to work with classified data, thinks it should. Trump has sent Anthropic packing, threatening to place it in the same national security risk bucket as Chinese tech major Huawei. His argument is that law, not company morals, will dictate how the US prosecutes its wars. Pentagon used Anthropic's code during the latest Iran strike, even as US Department of Defense negotiated more acceptable terms with Anthropic's rival OpenAI. Anthropic says it will sue.

There is a troubling ethical dimension to the development. Typically, lawmakers get to decide the limits of technology development, or its dispersal. But the law is still some way behind AI's development. The ethical burden, thus, must be shared by producers and consumers until voters decide on the matter. This was how human cloning for reproductive purposes was halted long before it was widely banned. Tech companies share reservations over AI safeguards. But these may not withstand the competitive intensity of strong investor interest. So, there may be a case for allowing the law to catch up with AI.

The US approach to regulating AI has been through a conversation with technology creators. Any ethical concerns emerging from the other side of the table acquire special emphasis in dialogue-based rule-setting. Unlike earlier technology developments, the state has a diminished role in the evolution of AI. Tech creators must be comfortable with the products they are bringing to the market. They need to be sure automation can fulfil customer expectations. Hopefully, Anthropic will find a resolution to its moral dilemma in court. The incident is, of course, a chilling reminder of the intensity of the AI race, where ethics can very easily become collateral damage.