Automatic conversion of text to video is fascinating, but insanely energy-intensive. This is yet another reason why data centers should be subject to licensing requirements, as some politicians are now proposing.
Imagine that you have to fully charge your mobile phone as it runs out. And that 1800 others do like you. About as many as there are people living in Vardø.
At the same time, a data center at the giant Open AI in the US is starting to transform the phrase “man walks dog” into a 60-second video using artificial intelligence (AI).
Who do you think uses the most electricity?
What we know is that a data center can use the same amount of power to let ChatGTP create a high resolution image like what you use to fully charge your mobile phone.
But what happens to the energy consumption when the text you enter will become a video, and not just a still image? Open AI, which is behind ChatGPT, recently launched the world first Sora, which takes this very leap.
As artificial intelligence researchers, we see that public information is needed in the field of AI and energy use.
Exactly how much electricity The AI system Sora uses for his calculations is still unknown. But if we stick to the comparison that one AI-generated image requires as much electricity as one phone charge, then a 60-second video from Sora contains 1,800 such images.
Thus, it is not unreasonable to estimate that Sora uses smallest as much electricity to make his 60-second video as you and the 1,799 other mobile subscribers put together spent on charging.
Or if you want: as much electricity as it takes to leave 20 panel ovens running at 1000 watts each for one hour.
In addition, enormous amounts of computing power must be used to exercise the models behind the films. But that is not included in this calculation.
Currently, Sora is not available for the general public. If it does, the users will in total draw enormous amounts of power from energy- and nature-sapping data centers around the world. This increases the timeliness of a proposal SV has now joined: that the state should introduce a licensing system for data centres, to ensure control.
One of the party’s arguments is that licenses can be used to reject data centers that will use power for cryptocurrency mining.
Being able to say no to data centers that plan to gobble up electricity for toy-like video production is, in our eyes, another very good reason to introduce a licensing scheme.
The backdrop for SV’s proposal is that huge data centers are now on their way into the power grid and nature here at home. The AI systems of OpenAI “live” in giant centers of the type that Google will build in Skien.
In parallel, data centers are planned both in Troms and Nordland, in Hamar and in Oslo.
Artificial intelligence is not one thing, but many. Some KI models require a lot of power, others much less. We can divide the AI tools into two general categories:
- Specialized AI models to solve clearly defined tasks. Like looking for broken bones in an X-ray or improving water purification processes. Such systems do not necessarily require that much computing power.
- General AI models. Like ChatGPT and Sora. These should give us answers to most things, consist of extremely large data sets and thus require a lot of computing power. They use a lot of electricity to create images. Each pixel – each tiny dot – in a color image consists of three numbers that determine the color. A lot of calculations have to be done in order for each pixel to get the right color.
Since the energy consumption of different AI systems is so different, it is important that many people understand what this entails. Especially in light of statements from the technology leaders who speak the loudest and have the most people’s ears in AI development worldwide.
From the outside, it may look like a race is on to have the biggest and most powerful AI models. And that the goal is to achieve the greatest possible influence in our daily lives.
OpenAI CEO Sam Altman is one of these visionary technology leaders. He makes no secret of the fact that the development he wants requires revolutionary breakthroughs in the energy field.
What is a possible and desirable development in the AI field will be one of the big questions in the years ahead. It will be absolutely necessary to see this in the context of the other major societal challenges, such as the climate crisis.
Hopefully, AI can contribute to solving the climate challenges. Among other things, by streamlining energy-intensive processes and finding new technological solutions.
For that we need small, specialized models. Not the big, general models.
Norwegian politicians should not sponsor the ongoing AI race by simply giving the technology companies access to the energy they claim they need.
Concessions should be introduced and distributed to data centers according to the possible social benefit they have.
Tags: electricity Speech
-
