Over a recent pub dinner, Oxford MBA classmate and good pal, Joel Usher, went cerebral over his favourite topic, Artificial Intelligence (AI). Joel discussed how rapidly AI could usurp humans as the dominant player on Earth. He highlighted that some of the world’s most influential people, such as Stephen Hawking (Oxford alum) and Elon Musk, have signalled warnings about AI. While Joel is far more convinced of the positive implications, such as the potential lifting of the constraints of labour and wages, he summarised for me how it could come to pass that AI supersedes human intelligence, identifies humans as a threat, risk, or simply inferior, and subsequently neutralises us. So I asked Joel what might happen after AI does away with humans. We had a brief think about this and then went back to talking about girls.
More recently, I’ve thought about what AI would do with the natural world. Indeed, one of the logics presented for why AI would dispense with us is that we have proven to be a threat to our own natural surroundings. We live in an era that scientists call the ‘Anthropocene’.
[youtube_sc url=”https://www.youtube.com/watch?v=fvgG-pxlobk” start=”79″ end=”137″]
What would happen then if AI took over? Oxford professor, Nick Bostrom, summarises how AI seeks optimisation:
[youtube_sc url=”https://www.youtube.com/watch?v=MnT1xgZgkpk” start=”503″ end=”646″]
The question then becomes what would AI’s optimisation process entail? I’d love to run with this topic at greater length to do it justice, but here I’ll attempt to scratch the surface in an effort to get the conversation going.
How would AI Approach Nature Conservation?
Would AI take a hands-on or hands-off approach to nature conservation? A popular movement in recent years calls for the rewilding of the natural world. The approach suggests that humans should be somewhat hands-off in managing nature, allowing it to find its own balance according to its own processes. If AI were to take a rewilding approach, would it seek to return to a baseline, which, for example, could be a period of pre human-driven extinctions, re-introducing native species in the process to enhance ecosystem functioning, or would it allow nature to continue from the point of AI takeover, choosing to accept human impact and lingering effects as consequences of our existence?
Moving to the right of the spectrum, would AI micromanage natural landscapes according to a different interpretation of optimisation? We might assume that AI would do away with protected areas, given that we gave rise to the need for them in the first place. AI would also be left to determine whether nature requires protection from its own forces. The invasive species debate enters the picture here. Would AI intervene?
Many of these questions fall within the realm of existing human debate, but should the world see the removal of us as natural enemy number one, it’s feasible to see AI evaluating the criteria differently than we do.
How would AI approach climate change?
The impact of human-driven climate change on the biosphere presents another question as to how AI might approach the optimisation of nature. Would AI accept anthropogenic impact on the earth’s climate as acceptable for an optimised world, or would it seek corrective action? In the latter case, AI might engage in geo-engineering, described in part here by Oxford University’s Rob Bellamy (note a return to the point of optimality).
[youtube_sc url=”https://www.youtube.com/watch?v=wOSA2RJusYU” start=”1696″ end=”1903″]
Moreover, if AI were to tackle climate head on, it would face challenges that we do not yet have the computing power to entirely understand. Predicting climate activity requires what Oxford Professor Tim Palmer describes as solving an equation from jet streams down to molecules, and we’re nowhere close to that point.
[youtube_sc url=”https://www.youtube.com/watch?v=ed4p7JYMYKs” start=”1504″ end=”1627″]
AI would have to make its own headway here. Of course, it would again be weighing factors differently to those weighed by us humans, given that human-driven CO2 emissions would cease to be an issue, and energy production and consumption would look entirely different.
How would AI approach ecosystem functioning?
Would the natural world have any value at all for AI? At present, we haven’t found replacements for the services that ecosystems provide (food, water, clean air). We still need them to function healthily to support the human population. AI might not need these services however.
AI could endeavour to optimise the earth by maximising intelligence productivity in the same way Bostrom describes in the video above, and the natural world might be seen as a barrier to this optimisation process. Natural landscapes could be cleared entirely to make more room for computing power, and the services they once provided might all be somehow automated through artificial means that ensure the earth’s sustainability. We better understand how a world with greater biodiversity is a world with increased resiliency. If AI determines that optimisation can be achieved with the planet’s resiliency built in artificially, the natural world might hypothetically be dealt with in a similar fashion to humans, meaning marginalisation or neutralisation to make way for other priorities based on different values systems that are vastly different to ours.
So we might be protecting ourselves AND the natural world from AI?
It’s entirely possible that AI could view nature for its flaws. It might view the entire food chain in terms of opportunity cost and seek new forms of occupancy. So in effect, we could be on the hook for our own survival in addition to nature’s. To properly guide AI, Borstrom suggests that we must build in clear values systems.
[youtube_sc url=”https://www.youtube.com/watch?v=MnT1xgZgkpk” start=”796″ end=”936″]
Given our challenges in determining where, how, and what we conserve, in combination with continuing calls for clarification of where our values rest in relation to the natural world, we have some work to do.
For me, above all, the topic calls into question what we might learn from an AI approach to nature, whether that approach would be to let nature do its thing, manage its conservation more closely, or to replace it with something artificial based on different optimisation criteria. I largely support the first option, with, obviously, us still in the picture. We touch nearly everything on earth now in some way, and over time we leave less space and scope for the natural world to conduct itself free of our influence. An AI takeover might, ironically, be the only way for truly natural occurrences to re-establish themselves.
Beyond our ability to shape AI values however, the core issue rests in the decision criteria. Our dominant pull-push relationship with nature, meaning our desire to pull resources from earth and push back upon it via human footprint, could be vastly different to an AI-centric relationship. The relationship also hinges on whether the intrinsic value we assign to nature could be grasped by AI, because without it AI might not distinguish between natural and artificial processes aside from which ones are found to be most efficient.
In any event (realising I’ve gone a bit ‘mad scientist’ here), bringing this post back to its MBA roots, the lesson is this: Spend time in the pub with classmates.
Back to top of article