human compatible ai and the problem of control

In proposing a set of Principles for Beneficial Machines, the author plots a potential strategy – but is clear about just how complicated the path ahead will be. AI has huge potential in areas such as healthcare diagnostics, and it would clearly be unwise not to pursue this, but domains that encompass themes such as autonomous weapons and massive surveillance grids require more fundamental consideration across society. Offering a new way forward for AI development, Russell opts for a well-argued middle ground. Isaac Asimov’s Laws of Robotics, some of the backstory from Frank Herbert’s “Dune” universe and a number of recent blockbuster films all contribute to the wide-ranging discussion, which also embraces a slew of moral and economic issues. Naturally, science fiction has dealt with this set of challenges numerous times and, while the vision he provides is far from a road map, Russell uses some classic fictional themes as markers.

His core analysis hinges on whether we are building an intelligence that will support the ambitions of humanity or whether such a creation will inevitably focus on reinforcing its own goals – and so ultimately out-evolve us and render human endeavour irrelevant. In this timely, vivid book, Stuart Russell sets out his views on where AI should (and should not) be taking us, and provides a new approach to helping us get there. In a largely unregulated space, huge amounts of venture capital and other private funds are being ploughed into developments that may fundamentally change human existence – changes over which the general population has little control.

Yet fundamental questions, both moral and practical, remain about what our ultimate targets for AI should be, while research and investment press forward at a rapid pace. It is frequently presented by commentators as being the great new hope of humanity – one that suggests a leap forward into a golden age. Today, the term “artificial intelligence” has emerged from its fringe existence and is scarcely out of the headlines. The discussion that emerged was fascinating but disquieting – as it soon became clear that their views on the emerging potential of AI were based on wildly different experiences, assumptions and expectations. A decade ago, I sat in a seminar room in Silicon Valley and listened to talks by two leading AI protagonists: one an elderly East Coast professor, the other a billionaire tech entrepreneur. The story of artificial intelligence development to date has been complex and often frustrating, with many false starts.

Human compatible ai and the problem of control