As the Bitcoin bubble neared its zenith in 2017, the internet filled up with dire predictions that “crypto-mining” would melt the world’s power grids.
Whole countries’ energy consumption levels became yardsticks for Bitcoin’s thirst for power. Ireland, Hungary, New Zealand – take your pick. Call it the Ohms Race: so-called miners frantically pumped electricity into processors in the hope of flushing out that elusive $100,000 cryptic coin. Or $10 cryptic coin, depending on which side of the peak they were on when it turned up.
And although Bitcoin brownout alarmism quickly died down, the episode should ring alarm bells in the autonomous vehicle community. Because the amount of power needed to mine virtual coins doesn’t even represent an amuse-bouche in relation to the stonking torrents of amps required to make self-driving cars work.
This was explained in the FT yesterday, with all the insouciance that the house journal of neo-liberal techno-grandiosity could manage. (FT – Driverless cars: mapping the trouble ahead)
You see, for AVs the map is the territory. For the tens of thousands of dollars-worth of on-board sensing gubbins to work, it has to be fed a massively-detailed 3D map of the vehicle’s surroundings. Everything that doesn’t move, from trees to signs to sandwich boards to rubbish bins, needs to be there for the vehicle to compare with what it ‘sees’ at any given microsecond.
Data-wise, these maps are colossal. Gigabytes to describe a modest stretch of uninterrupted blacktop in the middle of nowhere. God-knows-what-abytes for urban streetscapes. There’s no way of streaming so much information to the vehicle fast enough when it’s travelling so it must be loaded aboard on hard drives and carried around everywhere the vehicle goes.
There’s brave talk of perfecting AVs’ artificial intelligence so they can interpret their way along streets using simpler data. But for the foreseeable future they’ll need these 3D mega-maps. And presumably they’ll be restricted to operating within the area of mega-map they can practically carry around.
3D electronic maps may be new but the territory we’re moving into with them isn’t. Joseph Tainter covers it in The Collapse of Complex Societies: he shows that societies collapse when their investments in social complexity and their energy subsidies reach a point of diminishing marginal returns.
Theoretically, AVs are the answer to several problems Namely, ordinary cars’ huge redundancy (idle 95% of their lives), congestion, and the cost of accidents caused by human mistakes that machine-driven cars supposedly wouldn’t make. The most happy-clappy proponents of AVs enthuse over their potential to completely eliminate jams and accidents while slicing 60% off the cost of motoring by replacing inefficient car ownership with 24/7 ‘ridership’.
If AVs were simply ordinary cars but with added sensors and a clever brain to do the driving, these pipe-dreams might be valid. But they’re not. It turns out that for AVs to work as intended we’re going to have to build nothing less than a 3D virtual copy of most of the surface of the planet within five or 10 yards of a road.
Creating, storing, updating and transmitting that data will require a huge network of servers – plus fall-backs up the wazoo to ensure 100% uptime. Between the servers and the vehicles will be a layer of energy-gobbling supercomputers performing the gazillions of calculations per second required to coordinate traffic and feed individual vehicles the data they need to make their own decisions.
AV developers and their would-be customers simply assume the necessary power will be there. And this at a time when the cost and complexity of building a single power station is proving a stretch even for an advanced economy like the UK.
There’s every reason to think that Level 5 vehicle automation will be as near-impossible to achieve as commercial nuclear fusion. The torrents of capital and talent going into AV development are pouring into, if not a dead end, a neither-here-nor-there scenario of partial automation and fiendishly complex lash-ups aimed at allowing semi-AVs and conventional cars to coexist.
It’s a question of whether the game is worth the candle. That originally referred to a game of cards where the stakes were less than the price of the candle burned to light the play.
Will the outcome of the billions poured into self-driving cars be worth it? No. Not when you find out you’ll have to build and maintain a massively expensive 3D facsimile of the world before any AV can even think of operating in the real world like an ordinary car.
Should we stop and do something more useful, or at least redefine our goals for AVs? Of course.
But we won’t. Humans simply aren’t smart enough.
QuadRanting 