Will the rise of AI turn off the lights?

The price of artificial intelligence may be too high in pure costs of electricity.
2 January 2019 | 9652 Shares

The true cost of AI may be the power which it draws. Source: Shutterstock

We’re poised on the brink of Life 3.0. That’s the message expounded by Max Tegmark in his excellent publication of the same title. But whether or not we’ll exist to see it remains an interesting point of debate.

Although it’s difficult to summarize an entire volume in just a few sentences (some might say foolhardy), one of the book’s themes is the way which artificial intelligence may take us human beings, as a race.

Machine learning and its rather more complex cousin, deep learning (think many layers of machine learning), both require significant amounts of computing power. Happily, for the technologically-inclined, computing is becoming cheaper, cycle for cycle: there has been a doubling of processor speed every couple of years since the year dot, and there’s no real sign that the trend will end.

Moore’s law. Source: Wikicommons

The opening of Life 3.0 contains an all-too-believable flight of fancy which involves a group of gifted individuals using public cloud computing to stage a self-learning machine, which gradually accrues increasingly large revenues. The revenues are plowed back into buying more compute, which increases the AI’s abilities, which eventually leads to…well, you’ll have to buy the book to find out!

In addition to significant computing muscle, deep learning algorithms require lots of data. So much so that we can happily leverage the buzzphrase from the early noughties: big data. AI in all its forms loves big data, the more the better. The greater the quantity of information going into AI code, the better the learning ‘experience’, the better the eventual results.

Further happy coincidences are fuelling the fire of machine learning, the first being our everything-digital lifestyles, which feeds every online interaction into a database, somewhere.

Of potentially more significance is the emergence of small, networked, cheap computing, control & monitoring platforms, better known as the Internet of Things (IoT) devices. IoT is pouring a stream of binary into the data lakes of the world, and within the next dozen years or so, the pouring will become a torrent.

All the ingredients are either already in place, or will be soon, therefore, for artificial intelligence to take us into the great unknown. Falling computer hardware costs plus big (and bigger) data equals AI with the computing power and material it needs.

But no discussion of computing power is complete without giving serious consideration to that other form of power – electrical energy.

Some areas of the US, where the geography’s right, are having something of an economic renaissance, hosting the vast server farms which run the services on which we rely every day. And those are the server farms which, increasingly, power AI compute and storage.

A Utah copper mine – one of the tech industry’s by-products. Source: Wikicommons

According to a Japanese study, by 2030 the tech-derived demand for power in that country will outstrip its current generating capacity. And technology’s carbon output may well be 20 percent of the world’s total in the next ten years.

Amazon Web Services, for example, is run on gargantuan batteries of servers, complete with cooling systems and supporting hardware. But Amazon’s carbon footprint for its XaaS provision is something about which the company is remarkably tight-lipped.

The power required to service the data lakes, and their processing by the next generation of artificial intelligence models is outstripping the world’s provision of renewable energy. Tech companies like Google and Apple, aware of the precariousness of their situation, state that all of their power requirements produce zero net emissions. Facebook is set to follow suit, it has said.

Claims of this sort are always subject to accusations of green-washing (how much carbon offsetting contributes to the net figures, for example). But even if the Apples of this world are setting examples, it’s not one necessarily followed by companies like Tencent, Alibaba, and Baidu, the Chinese giants whose carbon footprints can only be a matter of conjecture.

Artificial intelligence then comes with an overhead of its inherent hunger for power. It is a long game of the highest order who bets that AI can solve its own energy crisis. Which will come first, a computer which can devise a power generation revolution or a machine-driven ecological catastrophe?