Rise of the machines? The energy implications

Figure 1: Will the machine’s rise up to overthrow us?

There’s been quite a lot of discussion on the internet recently on the topic of AI (Artificial Intelligence), prompted by several movies out on the topic (the latest Terminator one, Ex-Machina, Chappie, Interstellar, etc.), as well as the British TV series Humans (actually a remake of a Swedish TV series). We’ve also had dire warnings from Stephen Hawking and Bill Gates about how AI could spell the end for humanity (and this from the guy who came up with Windows?). Others worry that AI’s might steal all of our jobs. Anyway, I thought it would be useful to bring a engineering prospective to this topic and try to separate the plausible from the implausible.

Figure 2: The trouble with robots….some strings attached!

For starters if you ever look at those walking robot brought out by the likes of Honda or other firms, you’ll often notice some cables sticking out of the back. They are there because robots do have a number of pretty hefty power requirements. Typically robotic arms are powered by electric servo motors, which can result in a high instantaneous power demand as the robot moves (which is why some run off three phase supplies, particularly if the job involves heavy lifting), which means a good sized power supply. Alternatively either hydraulic or pneumatic actuators could be used. But both of these need either a compressed air or compressed fluids supply or an on board compressor, which can be quite energy hungry.

Thus any future robot or android would have to have an energy source capable of powering everything. And I’m told by experts in robotic’s that you’d be looking at something with a power consumption in the kW’s range to do that (assuming you want you’re machine to have roughly human level strength). Chuck in all the other systems such a machine will need, some gyroscopes (for balance), sensors (camera’s, microphones, radar) a pretty sophisticated computer system, etc., and you’ve got quite a hefty power demand. My guess is that a future android would probably need somewhere between 2 kW’s to 4 kW’s to run and operate. Assuming a 12-24 hr operating cycle (the rest of they time, we’ll assume its powered down or recharging) that leaves us with an energy demand of up to 24 kWh’s per day, perhaps even as high as 72 kWh’s per day, about the ball park range of daily energy consumption between that of a small car and an SUV.

And if that if that level of energy consumption came from fossil fuels, the result would be to give any future android a carbon footprint so high it would turn James Hansen into an cyborg hunting, gun totting Kyle Resse. And we’ve already got major issues to worry about with electronics waste as it is! It also raises the question, how would you supply such power to an independently moving robot or android? The Termintor film won’t have been quite as scary if every couple of minutes Arnie has to break off his pursuit of Sarah Connor to raid a hardware store and swap over about a hundred D-cell batteries 🙂

Keep in mind the sort of power requirements we’re talking about here would require a fairly heavy battery and it seems doubtful you’d squeeze that into a human sized body (the 24 kWh battery of a Nissan Leaf for example weights 214 kg’s and fills the whole bottom of the car). A smaller Li-ion battery could do the trick (assuming future advances in battery tech), with regular recharging. However, keep in mind that this would mean many hours of down time, i.e. our android can’t work 24/7, yet we can hire someone at minimum wage to do the same job, with the same level of “downtime” without the capital costs of a robot (nor the costs of all of that electricity!). Plus our human worker can pull a double shift if necessary, if you pay him enough money for overtime.

Wireless charging of an android is a possibility, as was the premise in the film Ex-Machina (although they sort of forgot about that towards the end!…when she wandered off grid!). While such power transmission levels are a bit beyond current technology, its a plausible future scenario. However, this would constrain our android to a fixed route and location. And needless to say, the machines will have difficulty taking over if they can’t walk more than a few metres outside the building before running out of juice!

Figure 3: RTG’s are a compact power source, but still less than ideal…..

Most people will at this point say “nuclear power” is the answer. Ya, and do you really want a walking, talking, Chernobyl wandering around in close proximity to people? What could possibly go wrong! The Terminator in this scenario won’t have needed a gun to get Sarah Connor. All he’d need to do was sit down near her for ten minutes reading the paper until she’d received a fatal dose of radiation!

Furthermore RTG’s (NASA’s term for small portable nuclear batteries) aren’t really up to the job. They are designed to supply modest amounts of power (generally under a kW) to deep space probes over very long time periods. Once you account for the additional “packaging” (shielding, heat sinks, energy conversion systems, etc.) no RTG currently available could power any of the hypothetical androids from science fiction, as most RTG’s are simply too big and too bulky.

Furthermore an RTG could get our android a little hot under the collar. RTG’s generate lots and lots of heat, about 90% of their energy output is typically heat in fact. And this is exactly why space agencies use them, as this heat protects spacecraft and their instruments from freezing up in the cold of deep space (or in a dust storm on Mars). However for an android it would create a massive heat dissipation problem. Arnie’s Terminator might have difficulty blending it when he’s got two massive engine exhausts stick out of his neck, spewing out hot exhaust’s like a coal rolling hot rod!

Figure 4: ….not least because they could leave our AI a little hot under the collar!

One plausible energy source for a future android of the future is a fuel cell. These are becoming ever more compact and reliable. Low temperature fuel cells (such as PEM’s) operate at 45-80ºC, with an efficiency of 40-60% well within the range of heat that can be dissipated. Also new fuel sources in the form of ceramic pellet’s (which soak up hydrogen like a sponge, eliminating the need for pressurised tanks) are under development. There’s also DMFC’s (Direct Methanol Fuel Cells), which can run on alcohol based fuels. While they run at slightly higher temperatures, the fuel they use already exists and in theory if you’re android is starting to run low on fuel, take it to a bar and give a few shots of vodka….of course I can’t help but notice that this is a little close to the supposed power source for the robot “Bender” from Futurama. Its worrying when the most technically plausible sci-fi show happens to be the one that goes out of its way to be as unscientific as possible!

Figure 5: Will robots of the future run on alcohol?….same as the Irish!

Although on a serious note, a fuel cell powered android would also require an air supply to operate, so that would rule out certain sci-fi scenario’s where they function in the vacuum of space without difficulty.

However even with the correct power source, could Androids replace people in many job roles? as portrayed in films such as “I-robot”. I suspect that as AI improves we’ll see more and more use of them and ultimately jobs going. The services industry could see many of its more tedious jobs replaced by AI (thought not androids, probably just banks of computer servers) within the next few decades. But equally, there will be a need for someone to supervise and maintain these machines, so you’ll end up hiring more programmers and engineers. Even so, there is a limit to how much of an economy can be “automated”.

Robots cost a lot of money, so its often not worth one’s while buying them unless the fixed costs (buying & maintaining it) can outweight the long term costs of paying some person to do the same job. Robotic cleaners, for example, have been around since the 1990’s yet you’ll still see plenty of people mopping floors in any subway station or office complex. Similarly, there have been various attempts to develop agricultural robots, yet you’ll still see lots of people working in fields come harvest time.

And any future android, or human like robot, ain’t going to be cheap. My guess, assuming we could develop the tech to build one, is a price tag in the order of hundreds of thousands, if not millions of dollars a pop. And as I discussed in a prior post (with regard to nuclear reactors) while mass production of a product can bring down the unit costs, there’s a limit to how far it can do so. We’ve been using industrial robots in factories for several decades now and they still cost thousands (or much more!) to install.

Indeed, back during the 70’s and 80’s many manufacturers, notably car companies, when gaga for robots. They built factories full to the brim of robots doing every conceivable task, so much so that the few human employee’s left dreaded going to the toilet least the find a robot in the cubicle waiting to wipe their ass. However, corporations soon realised that this was merely pushing up their capital costs and not increasing profitability. Those robots needed maintenance, which meant they were replace lots of low wage workers with a smaller number of better paid engineers. And there was the issue of spare parts and other associated costs. Plus there were quality control issues. When machines screw up they tend to do it in a big way, potentially leading to entire production runs having to be scrapped or hours of down time on the production line (very costly in these days of Just In Time inventory management).

So many corporations went back to drawing board in the 90’s and began putting more people on the production line again. Robots were restricted to jobs which were either difficult or too dangerous for people or required a high level concentration and skill. Welding car bodies being a good example, lifting and placing heavy items being another. Jobs which required patience and care (e.g. lining up two parts for assembly, inspection of critical weld joints) were reserved for people, not least because this left a good few mark one eyeballs watching the line ready to call in any issues before they became a major problem.

So quite frankly if you need some “help around the house” (why is it that the only job sci-fi can envisage androids of the future doing is folding bed sheets?) have you considered Kids mark 1? They can be easily produced via a simple (but fun) process and they are considerably cheaper to build and maintain than any cyborg or android will ever be. Although you might be tempted to try returning them under warranty, particularly when they hit their teenage years! 😉

But the reality is that while AI’s (although not androids) might gradually start to take over certain roles and jobs, but there are going to be practical limits to what they can do or are allowed to do. Take the driver-less car. Many I know in transport research seem to assume its a slam dunk that once AI technology reaches a sufficient level of maturity they will take over the roads. I’m not so sure, as this would sort of defeat the purpose of individually owned cars. And one has to question if its going to be possible to solve many of the obstacles to their use, most notably public acceptance. Note that we’ve long had the technology to make aircraft or indeed even cargo ships fully autonomous, but we still have crews on board, largely because the public demand such things.

Figure 6: Driverless vehicle technology is advancing rapidly, but what about public acceptance?

And in part this boils down to a matter of responsibility, i.e. who is to blame if things go wrong? Its also why I suspect that if we ever did develop advanced concious AI’s (and that’s a big if), it would be necessary to give them some sort of “rights” almost as soon as they were smart enough to understand what those rights meant. Because with rights comes responsibilities, meaning they can be made liable if they screw up, just like the rest of us.

But regardless of how smart computers get, there ability to learn is going to be constrained by the information they have access too. As computer science expert Mark Bishop points out, any AI will be constrained by its programming and what information its sensors can access. Like the prisoners in the Shadow’s in the cave analogy, an AI say, running a car, is going to be limited in its knowledge of the outside world by programming devoted entirely to vehicle control and sensors designed purely to detect potential road hazards. There would be no reason for it to talk (like KITT in Knight Rider) or develop any form of social skills, it would be more like a sort of robot version of Top Gear’s “The Stig”.

And we’d have to be pretty stupid to make the decisions that would allow AI’s to gain control of critical systems. Take the premise behind the Terminator series, a super intelligent computer is put in charge of the USA’s nuclear weapons (a rip off I might add of the plot of the Forbin project film of the 1970’s). Why would you do this? Any monkey can push a button (again costs, isn’t it going to be cheaper just to pay a few grunt’s to sit in a bunker under a field in Kansas?). A computer that smart in control would constitute massive over-design. Also this creates a single point of failure in a system (all the sov’s need to do is take out one PC…by upgrading it to Windows 10 perhaps!….and they’ve disabled all of the US defences). Indeed I can think of many good reasons, not to use such a computer. Not because it might start WW3, but because it might conclude that nuclear war would be a pointless act of stupidity that would merely result in its own destruction and the destruction of the human race and hence the resources it depended on for its survival. It might decide its best course of action was to sabotage its arsenal, so less “judgement day” and more “saviour day”.

There are certainly legitimate concerns here. Mark Bishop’s main worry isn’t super intelligent cyborg’s taking over the world, its computers slightly smarter than the ones we already have being turned into autonomous killing machines. The trouble with this is, history tells us that if you make wars easier to wage and remove yourself more from the messy business of killing, politicians are more likely to go on a bombing spree. Just look at the issue of drone strikes. But this is more a social and moral issue than a technical one.

So AI’s are unlikely to ever run outside of our control, simply because they will be constrained by the same laws of physics we are. Furthermore, they will face a variety of social and economic constraints that will restrict their use to certain fields and disciplines. And if, as I suspect, they are even more energy and resource dependant than us, that will place a massive restriction on their proliferation, as well as giving them a very strong incentive to keep our civilisation going. The AI’s will in short, not take over, no so long as the engineer’s and the bean counters are the ones making the decisions. And if they were to get into a position of taking over (because we stopped listening to the engineer’s and the bean counters), the first act of our AI rulers would presumably be to note the error of their predecessors and appoint some engineers and accountants to an advisory role.

Which is why I for one welcome our future AI overlords. And I’d like to let them know I can be useful in rounding up others to work in their giant underground methanol distilleries 😉