AI world domination? Not so fast...
Imagine all AIs shutting down – from your social media feed to Google Search – just because companies can’t process any more data. Well, this is more likely than AI taking over the world. Why?
Category
🗞
NewsTranscript
00:00 AI is everywhere, behind your social media feed, your email assistant or Google search.
00:05 Now imagine all of this comes to a halt.
00:08 Okay, that might be an exaggeration, but running out of resources to train and operate our
00:14 AIs is way more likely than AI taking over the world.
00:19 AI computing is happening in data centers, essentially buildings that house computers
00:24 and servers.
00:25 If we continue to scale up AI development, we'll simply exceed their limits.
00:30 That leaves us with two options.
00:33 Build more of them or make computing more efficient.
00:35 Well, there are problems with either one of these solutions.
00:39 Oh, and while we're at it, we may also run out of new data to train our AIs.
00:45 But more on that later.
00:47 Why we can't just build more data centers.
00:50 Simply put, we're about to run out of resources to operate them.
00:54 For reference, one data center alone can consume as much electricity and water as 50,000 homes.
01:01 In times of climate change and declining resources, it's very unlikely we can keep this kind
01:06 of consumption up.
01:08 So how about making the computing inside the data centers more efficient then?
01:14 Moore's Law.
01:17 So let's get to the technical part.
01:19 Stay with me here.
01:21 Every piece of tech nowadays runs on microchips.
01:24 In those chips, there are transistors.
01:26 The more transistors on a chip, the more computing power.
01:30 Still with me?
01:31 Then let me introduce you to Moore's Law.
01:33 It's not a scientific law or anything.
01:35 It's a term formulated by Intel co-founder Gordon Moore back in the 70s.
01:40 He predicted that the number of transistors that would fit onto a chip would double about
01:45 every two years.
01:46 And until now, his prediction has held true, leading to ever more powerful chips.
01:52 However, experts think that this will reach a physical limit somewhere in the 2020s.
01:59 This is because we don't have the tech to make transistors even smaller or to cool them
02:03 down anymore.
02:04 Aside from these physical constraints, we might also run into a problem with new training
02:09 data for AIs.
02:11 AI feeding itself.
02:13 As AI is more widely being used, there's a growing concern that it'll erase real innovation
02:20 and new ideas.
02:21 Why?
02:22 Even though it seems that programs like ChatGPT or MidJourney are creative, they really aren't.
02:28 Artificial intelligence draws on existing data.
02:31 So if AI is incorporated basically everywhere, there's a chance that no new original data
02:37 will be produced.
02:39 Essentially, AIs would feed themselves, cancelling out everything that's new and groundbreaking,
02:45 making creativity stagnate.
02:47 If you think now, come on, that's never going to happen.
02:50 You might be right, but some experts estimate that by 2026, up to 90% of content on the
02:56 internet could be AI-generated.
02:58 Alright, let's take a breather.
03:02 How to deal with all of this?
03:05 Data centers could be built underground in already existing mines or caves.
03:10 This would save space and resources.
03:13 Also, natural water flows may help with cooling them and reduce water consumption.
03:19 However, this is all very dependent on their location and hard to imagine on a global scale.
03:25 As for more efficient microchips, we need to rethink the tech itself.
03:30 Classic computer chips are not designed for AI-specific tasks, leading to a waste of energy
03:36 and suboptimal performance.
03:38 With chips tailored to AI, computers can do much more with the same resources.
03:43 And as for the AI that feeds itself, we need to take a holistic approach.
03:49 Trust in human inventiveness and creativity, as well as benefiting from AI's abilities
03:54 while doing that.
03:56 I think we drastically need to rethink how we're approaching this technical revolution.
04:01 Using up all of our resources for our short-term growth now and hindering developments in the
04:06 future is the wrong way to go about it, in my opinion.
04:10 What do you think?