Tej Kohli: A tax on AI could help to reduce inequality

Tej Kohli is the founder of the not-for-profit Tej Kohli Foundation whose ‘Rebuilding You’ philosophy supports the development of scientific and technological solutions to major global health challenges whilst also making interventions to rebuild people and communities around the world. Tej Kohli is also an impact investor who backs growth-stage artificial intelligence and robotics ventures through the Kohli Ventures investment vehicle. Tej Kohli’s blog is #TejTalks and he is the author of Rebuilding You: The Philanthropy Handbook.

Twitter @MrTejKohli.

As thousands of workers commence working from home due to Coronavirus, the Internet is awash with memes about the tempting distractions of YouTube. The artificial intelligence used by YouTube to continually serve relevant distractions is a modern shoulder devil for the home worker.

Baby Shark has nearly 4.8 billion views on YouTube. In 2011 Google revealed that streaming 1-minute of video on YouTube consumes 0.0002 KwH of energy. That means that so far, people watching the 136-second-long Baby Shark video have collectively consumed about 2,112,000 KwH of energy. To give that context, in 2019 the average UK home consumed 3,100 KwH of electricity. Based on YouTube’s estimate of energy used per video, Baby Shark has so far consumed as much energy as 681 average British households do in a single year.

This yardstick calculation doesn’t account for any of the energy consumed searching for the video (a single search activates servers in six to eight data centres) or the energy consumed in the last leg of data transmission, which is high for wireless devices, or the energy used by the screen that Baby Shark was watched on. So, in truth the energy collectively used to take Baby Shark to nearly 5 billion viewers is likely many multiples higher than two million KwH.

What is certain is that this ‘invisible’ energy usage is a fact of modern life. What is also certain is that it is not good for the environment. The energy consumed to power huge data centres is low per user but is collectively astronomical. And the advent of artificial intelligence as an omnipresent part of modern life means it will get worse. Researchers at the University of Massachusetts Amherst estimate that the carbon footprint of training a single AI model is 284 tonnes — five times the lifetime emissions of an average American car.

We are now arriving into a future where 1 trillion sensors are connected to the Internet of Things by 5G via the Cloud. This new reality will prove a tipping point in the enablement of artificial intelligence in every aspect of our daily lives. This artificial intelligence will require unprecedented amounts of data. And all of this new data will need to be processed, often in real time, through a huge increase in the amount of processing taking place in data centres around the world. Data usage will be multiples that of the Internet as we know it.

The impact of this inevitable AI revolution on society is likely to be short term pain but with long-term gain. AI is likely to exacerbate inequality. Without democratisation of AI, AI will bring a monopolistic tendency to dozens of industries, eroding the competitive mechanisms of markets in the process. The wealthy would be the first to gain from AI. It will be the low paid and unskilled workers who may see their jobs replaced by AI. And it will be small businesses who cannot afford to invest in AI that will be left unable to compete.

Which brings me to my core point. If AI is going to be doing more of our work for us, making more decisions for us and even taking our jobs, then it should also bear its fair share of the tax burden too. If humans are to be masters and artificial intelligence is to be our servant, then AI should provide for us too. We can achieve this by taxing data centres.

A tax on data centres would be easy to implement. It is no more possible to hide the acreage of a data centre than it is to hide acres of land. The tax levied on data centres could equate to their energy use to reward those processing data more efficiently with the lowest carbon footprint. Data centres would pass on the tax burden to their clients within their pricing, and the burden would trickle down through the supply chain to the end consumer.

And unlike regressive forms of taxation that hit the lowest paid the hardest, a tax on data centres would fall more heavily on the wealthy, who tend to consume more processed data. A cleaner uses less data than a financial analyst. The driver of a brand-new self-driving car will be using a lot more data centre processing than the driver of a manual car. An older person living on a small pension is unlikely to spend their days attached to a smartphone.

The owners and investors in the new wave of technology businesses people like me — who are profiting handsomely from AI, will also be paying more, through taxes on the data centres that will power the enterprises upon which the future AI economy is built. At present it is the big tech companies with digital products and services who are most famously adept at avoiding taxation in the countries where their customers reside. A tax on data centres would be one way to ensure that they pay their fair share — albeit indirectly.

Those responsible for more Co2 emissions through their heavy use of data centres would also indirectly pay more tax. This would be no more outlandish than the UK’s annual car tax, which correlates the amount of tax payable with engine size and emissions. In short, if the future AI economy is powered by data centres rather than by the labour of people, why shouldn’t those data centres also fund the public services that all people rely on?

A tax on data centres could also help to govern artificial intelligence. There are those who worry that left ungoverned, AI will cause too much inequality. There is even a risk that AI could become society’s master and that we will all become its servant. But a tax on data centres would place economic controls on the most outlandish and energy-intensive uses of AI. It would incentivise the most efficient use of AI and bring into equilibrium the minimum data processing required in order to derive the maximum benefit for people and society.

You would still be able to distract yourself with Baby Shark when working from home. But by doing it you could also be helping to fund vital public services.