While I understand how extrapolating causes projections such as this, I have to say I believe that this level of energy consumption to mine Bitcoin is very unlikely to actually occur.
I can think of a number of reasons that would slow the amount of energy used to mine Bitcoin. For example, if Bitcoin were consuming that much energy, the cost of energy would increase because the supply would become limited for all other purposes. At some point, mining then becomes unprofitable, so effort would decrease, increasing the supply of electricity, etc. Supply and demand dynamics. Also, mining effort is likely to diversify further into other cryptocurrencies. ASIC's cause many (and certainly nearly all serious Bitcoin miners) from having an easy ability to mine another currency at scale, as they are not able to be repurposed to other algos, unlike GPU mining.
Other factors come into play, such as the next halving of the block reward as well as difficulty decreases if mining effort causes the average time between blocks to lengthen past the target.
I just don't see any way possible that by 2020 all of the world's energy (as is used today) will be devoted to Bitcoin. We don't grow electrical generation capacity fast enough to keep up. So something will give. I don't see the electrical generation side giving unless Bitcoin miners are going to start building power plants. I mean large, industrial-scale power plants. (For all I know they already have done this somewhere.)
I guess we shall see. Certainly, energy consumption of mining is a concern to the extent that it provides a negative talking point about cryptocurrency. It is also a legitimate concern with respect to the use of energy. However, Proof of Work is a useful concept to slow the rate of coin emission. I see coins that implement "proof of useful work", distributed computing work units, as a worthy way to mitigate this problem to some extent. However, at this point at least, I would never recommend that this is implemented in Bitcoin. The type of distributed computing work units are too diverse and more work should be done in this area (other technical and practical problems to confront and resolve) before it would ever be something I'd advocate for in a currency as important as Bitcoin. This is an area that interests me and that I hope to spend some R&D effort on.
Best regards,
Ben
Edited to add: Currently, the incentives to mine Bitcoin cause a feedback loop where more mining effort is added to the network (energy usage increases), difficulty increases (due to more effort finding blocks faster). Over the long term, this feedback loop will confront the realities of global energy supply being limited and/or cost of energy being priced out of profitability. So this is a short-term phenomenon in my view. If it became necessary, there are ways that software changes could change the economic balance, such as artificial difficulty increases or other changes to how difficulty was calculated, adjustment of the 10 minute/block target, etc. I don't think anyone involved with Bitcoin wants it to consume all of the planet's energy. And as I stated before, even if they did want that, I don't think the world would let that happen.