Oxford’s School of Geography and the Environment was addressed by visiting professor Peter Dauvergne on the subject of ‘Artificial Intelligence and the Future of Sustainability’.
Dauvergne is a Professor of International Relations at the University of British Columbia, Canada, and a Fellow of the Royal Society of Canada. He began by explaining the title of his new book, Genius Squared. He was deeply impressed by the chess showdown that occurred in December 2017, when DeepMind’s AlphaZero was pitched against Stockfish 8, both software programs designed to play chess.
DeepMind, an AI startup acquired by Google in 2014, had developed ‘a new kind of chess program’. Its deployed machine learned to self-play approximately 20 million games of chess in four hours, using enormous computing power. It then demolished Stockfish 8, already considered to be far superior to any human player. Over 100 games, AlphaZero drew in 72, won in 28 and lost on no occasion. Moreover, noted Dauvergne, ‘it played aggressively, intuitively…’ It upended the tendency of chess software programmes to play defensively. It was this bravura performance, and not just the raw results, that pointed to a dramatic increase in artificial intelligence, hence Genius Squared.
Relating this initial insight to sustainability led Dauvergne to first highlight six ‘bird’s eye view’ trends in AI, an area of research and progress so expansive that he admitted it was difficult to track:
1. Machine learning is transforming many aspects of AI.
2. Machine interaction with humans (Siri, Alexa) has been compared to ‘the new electricity of the 21st century’. That may prove to be a boast, but machine interaction is highly significant and has only just begun.
3. AI is rocketing in political and military significance, with commensurate (and terrifying) implications given that the two dominant players are the US and China.
4. AI empowers big business, with extensive implications for consumption and capitalism.
5. Some claims of AI evangelists will prove to be hyperbole.
6. There are powerful incentives to conceal or deny risk and ‘downsides,’ including algorithm bias, privacy issues, transparency issues, inequality, unemployment threats, increased resource extraction, repression of populations by AI-empowered state actors, and weaponisation.
Dauvergne apologised for not getting to the specific subject of sustainability sooner in his talk, but juxtaposed this with the prediction that AI is so significant a change in human affairs that it will completely transform global environmental politics in the next decade.
For example, he noted, every single smart device brings in its wake new plastics. Between 2005 and 2018, more plastic was created globally than in all recorded history up to 2005.
Currently the estimate is that ocean plastic waste will double in the period 2010-25, with the evidence of its accumulation partly explaining why we have suddenly become so aware of it in a very short period of time. He added that between 5-10% of marine plastic pollution is from car tyres, a recent insight, while some estimates put the total at up to 28% if you include otherwise invisible tyre ‘crumb’ and not just whole tyres.
Dauvergne, previously author of Will Big Business Destroy our Planet?, painted an almost dystopian (if nominally exciting) future in which incredibly clever micro-targeting of advertising in a digital, consumer context will supercharge consumption and empower large corporations. Aided by autonomous, AI-enhanced mining equipment and other forms of agricultural/fisheries and civilian transport and machinery, this process might clean out the planet, leaving it far more depleted than it already is, with climate change ignored and most megafauna extinct.
Equally a concern is weaponisation. Dauvergne noted that China is believed to be working intensively on the military application of all forms of AI, while publicly only discussing civilian applications, while the Pentagon recently tendered a large contract in this area. Asked who is leading this unofficial and mostly invisible ‘arms race’, Dauvergne recommended Kai-Fu Lee’s book AI Super-Powers, adding that the US is ahead in raw discoveries, but that China was ‘racing’ to catch up and possessed a large advantage in procuring data (from surveillance, partly), tending to ignore many of the privacy issues that are normative in the west.
To impart a sense of the rapidity of these changes, Dauvergne cited one commercial prediction, that ‘derived business value’ from AI was $700m in 2017, but expected to be $4 trillion by 2022, an expansion amounting to an explosion.
Turning to the widely heralded advantages of AI for conservation and sustainability, Dauvergne showed a slide of a submersible piece of autonomous equipment called a ‘ranger bot’ that scours coral reefs, autonomously killing one species of starfish that is considered to be invasive and destructive. One can imagine drones to map biodiversity or fight poachers and illegal logging; satellite surveillance to track fishery fleets; and autonomous cars that reduce road traffic accidents. Crop diseases will be more easily identified by apps like Plantix, while medical diagnostics will undergo a revolution, putting a ‘doctor’ in reach of everyone via their smart device. In all such examples Dauvergne emphasised that humans stood potentially to advance their lot, with sustainability enhanced in some cases.
The difficulty with all these examples, he said, was that the same technologies will likely be deployed by illegal fishers, loggers and poachers, (and legitimate resource extracting industries and states), while the ranger bot example is one of many cute headlines that will represent a ‘good news story’ that is ‘incremental…but does nothing to solve the underlying problems [such as coral reef destruction resulting from climate change]’.
Floor discussion following the talk elucidated the broad insight that AI is poised to become, in numerous ways, a ‘non-human actor,’ replacing humans in numerous roles and thus increasing unemployment. But Dauvergne was adamant that it cannot be ‘put back in its box’: ‘we can’t wish it away,’ he said. What we can do is be perceptive to its true risks and honest about its true (or desirable) limits and applications.
Climate change was raised only by one member of the audience at the very end, creating the sense that AI is so enormous that it needs to be mapped and tracked on its own terms. In its myriad applications, argued Dauvergne, AI is neither ‘good’ nor ‘bad’ for anything including the planet and sustainability. Rather, it greatly amplifies what humans would otherwise have done within their own human limits. We have to hope that the genius squared will be a responsible actor.