The Last 6 Decades of AI — and What Comes Next | Ray Kurzweil | TED (YouTube Video Transcript)

(00:00:00) Your YouTube transcript will appear here (00:00:04) so we've heard a lot about artificial (00:00:07) intelligence uh I've actually been (00:00:09) involved with AI for 61 years which is a (00:00:14) record um and we've heard a lot about (00:00:17) what people think about AI today uh (00:00:21) so tried to figure out what did we think (00:00:25) about artificial intelligence 61 years (00:00:28) ago so first of all people ask well what (00:00:31) are you into I'd say artificial (00:00:32) intelligence and they say what's that so (00:00:36) no one was really aware of it um the PE (00:00:41) I joined in (00:00:43) 1962 1956 was the conference where (00:00:46) artificial intelligence got its name so (00:00:49) the the views were quite (00:00:52) different uh people who were in computer (00:00:55) science had heard of artificial (00:00:57) intelligence most people were quite (00:00:59) skeptical they thought it would never (00:01:01) happen or if they thought it would (00:01:02) happen maybe it would happen in a (00:01:04) century or several centuries uh but the (00:01:07) people that actually came to that (00:01:09) Dartmouth conference in (00:01:11) 1956 they were quite optimistic some of (00:01:14) them including Minsky thought it would (00:01:16) take like one semester to (00:01:21) reach uh to reach uh the level of (00:01:24) intelligence that humans had (00:01:28) um and in fact that that led to our (00:01:31) first argument he was my mentor for 50 (00:01:33) years uh but we argued about that cuz I (00:01:36) I thought it would take decades but we (00:01:38) would see it within our (00:01:40) lifetime so we we are the only species (00:01:43) that actually creates tools that (00:01:45) enhances our intelligence I mean I'll (00:01:48) bet almost everybody has one of these (00:01:51) that makes us more intelligent this (00:01:53) connects to the cloud gets more (00:01:55) intelligent every year uh basically the (00:01:58) singularity is going to bring that into (00:01:59) our minds we're going to become smarter (00:02:02) and there's two different things we have (00:02:04) in our anatomy that enable us to do that (00:02:07) one is our brain but we're not the only (00:02:09) species that has a brain or even a (00:02:11) comparable brain elephants and whales (00:02:14) actually have a brain that's larger than (00:02:16) us but there's another aspect of their (00:02:18) Anatomy that they don't have and that no (00:02:21) one else has aside from humans which is (00:02:25) our (00:02:27) thumb so I can look at a tree and I can (00:02:29) imagine GE I could take those poles and (00:02:31) create a tool and then I can actually do (00:02:34) it now monkeys if you look at them and (00:02:37) have a thumb but it doesn't really work (00:02:38) very well it's actually an inch down (00:02:41) they they can grab things very (00:02:44) uh without much (00:02:47) force uh they can create a first (00:02:49) generation of tools uh but they can't (00:02:52) use that tool to create another set of (00:02:54) tools (00:02:56) so uh they really can't create a whole (00:02:59) set of tools that uh will enhance their (00:03:02) intelligence we're the only species that (00:03:05) does that and that's what artificial (00:03:06) intelligence is doing from the very (00:03:09) first homonid that uh created a very (00:03:12) primitive tool to Gemini and gp4 today (00:03:17) uh we create tools that make it (00:03:20) smarter (00:03:22) and so I've been actually monitoring uh (00:03:27) the growth of (00:03:29) computation uh uh which is right here I (00:03:32) spent like 45 years on (00:03:35) this uh and as you go up the chart it (00:03:38) represents exponential growth uh you (00:03:41) might think that someone was in charge (00:03:42) of this gee we've done this much it's in (00:03:45) a straight line let's get our next (00:03:48) computer to be right here but no one was (00:03:51) aware of it no one even knew that this (00:03:52) was happening for the first 40 years uh (00:03:55) I discovered this 45 years (00:03:57) ago uh I had various reasons to feel it (00:04:00) would continue at this pace in (00:04:04) 1939 uh that represents (00:04:11) 0.007 calculations per second per (00:04:14) constant dollar at the upper right hand (00:04:16) corner you've got uh a Google computer (00:04:21) which was uh 130 billion calculations (00:04:25) per second and recently Nvidia just came (00:04:29) out with a chip which is half a trillion (00:04:30) calculations per second so this little (00:04:33) chart (00:04:34) represents a growth of 75 quadrillion (00:04:39) fold increase that's why we didn't have (00:04:41) large language models in (00:04:44) 1939 or even three years ago we did have (00:04:47) something called large language models (00:04:49) they didn't work very well 3 years ago (00:04:52) began to work two years ago we've seen a (00:04:54) tremendous progress that's happened in (00:04:56) the last two (00:04:58) years uh in 1999 (00:05:01) I was asked to make a prediction of when (00:05:03) would we see AGI artificial general (00:05:07) intelligence and so I I figured that (00:05:10) this chart would continue which it has (00:05:14) and I figured we'd need about a trillion (00:05:15) calculations per second uh to do AGI so (00:05:20) I estimated (00:05:22) 2029 um (00:05:25) that was (00:05:27) met uh with a lot of skepticism M uh (00:05:31) Stanford had actually been monitoring my (00:05:33) predictions they called an International (00:05:36) Conference to talk about my prediction (00:05:38) and hundreds of AI scientists came from (00:05:41) around the (00:05:42) world um and they agreed that that it (00:05:46) would happen we would achieve AGI but (00:05:48) not within 30 years the estimate was 100 (00:05:52) years and I've talked actually some of (00:05:54) the people who were there who said 100 (00:05:56) years then and they're basically (00:05:58) agreeing it's going to happen very soon (00:06:00) musk says it's going to happen in 2 (00:06:02) years it's not an unreasonable position (00:06:05) other people saying three or four years (00:06:07) I'm sticking with five years but uh it (00:06:11) could happen soon but it's everybody (00:06:12) agrees now uh AGI very (00:06:16) soon so I have another book coming out (00:06:20) the singularity is (00:06:25) nearer and I've got about 50 graphs in (00:06:28) there uh (00:06:31) I can't explain it right now but if you (00:06:33) talk to me later I can explain these (00:06:35) charts but it basically shows uh that a (00:06:40) that artificial intelligence is going to (00:06:41) take over (00:06:43) everything um the the amount (00:06:47) of the amount of money that we make (00:06:50) right now is 10 times greater in (00:06:52) constant dollars than it was 100 years (00:06:54) ago we were very very poor 100 years ago (00:06:57) there was no government programs uh so (00:07:00) we're much richer than we were then (00:07:04) and the movement not only computation (00:07:07) but every single (00:07:09) technology uh is done by creating taking (00:07:13) the latest thing we've created and (00:07:16) making the next one we take the latest (00:07:18) chip and we use that to create the next (00:07:21) one (00:07:22) um we have greater wealth as I said that (00:07:26) leads to better education leads to (00:07:28) better doctors leads leads to healthy (00:07:30) people leads to more Global wealth all (00:07:33) of these things work together AI (00:07:36) supercharges (00:07:38) everything so I could talk about each (00:07:42) thing is being actually revolutionized I (00:07:44) think the most interesting thing is (00:07:46) actually medicine there are a lot of (00:07:48) people who are experts in AI who are (00:07:51) against what's happening and they're (00:07:52) very nervous about it and they think (00:07:54) it's going to wipe us out um but people (00:07:57) tend to get diseases which will which (00:08:00) are threatening to them uh and what's (00:08:04) going to happen people are going to get (00:08:06) diseases and AI is going to come up with (00:08:08) a cure uh very soon uh which will lead (00:08:12) to a great deal of (00:08:14) appreciation (00:08:16) um people say that AI is not creative (00:08:19) it's very creative you can actually put (00:08:22) together possibilities that might work (00:08:25) for example madna was trying to create (00:08:28) their covid vaccine they actually put (00:08:30) together a list of different mRNA (00:08:32) sequences now what would we do in the (00:08:34) past someone would come in and say well (00:08:36) okay there's several billion let's try (00:08:38) this one or maybe they pick three you (00:08:41) can't (00:08:43) uh do a clinical test on billions of (00:08:46) different possibilities but that's (00:08:48) exactly what they did by simulating the (00:08:51) reaction and that took two days so in (00:08:53) two days I created the medona (00:08:56) vaccine uh and that is still on the (00:08:59) market it it's been the best vaccine it (00:09:02) was done in two two days and we're going (00:09:05) to do that with every other thing (00:09:07) there's some very promising cancer cures (00:09:09) that are out there which AI produced uh (00:09:13) they're looking very promising uh the (00:09:15) next years is going to be remarkable for (00:09:18) medicine we had 190,000 proteins done by (00:09:21) people uh in (00:09:23) 2022 2023 Alpha fold 2 did 200 million (00:09:29) basically Ally every protein and how (00:09:31) they (00:09:32) fold uh every Protein that's used in (00:09:36) humans and and every other species on (00:09:38) Earth uh done in a few months uh we're (00:09:42) going to be able to go through uh cures (00:09:45) for diseases at the same (00:09:49) rate (00:09:50) so we're going to simulate trials (00:09:54) digitally uh it'll be much safer it'll (00:09:56) be a million times (00:09:58) faster um (00:10:00) and by the end of this decade as we go (00:10:02) into the 2030s we're going to achieve a (00:10:05) new (00:10:06) Milestone it's called Longevity escape (00:10:10) velocity let me say that again because (00:10:13) you're going to be hearing a lot about (00:10:14) that longevity escape velocity right now (00:10:18) you go through a year and you use up a (00:10:21) year of your (00:10:22) longevity however scientific progress is (00:10:25) also progressing which is actually (00:10:27) bringing us back it's giving us cures (00:10:29) for diseases new forms of treatment so (00:10:33) right now you're getting back about four (00:10:34) months so you lose a year you get back (00:10:37) four months or you're losing eight (00:10:39) months however the scientific progress (00:10:41) is on and exponential it's going to get (00:10:44) faster and faster and as we get to the (00:10:46) early 2030s let say between 2029 and (00:10:50) 2035 depending on how diligent you are (00:10:54) uh you're going to get back a full year (00:10:56) so you lose a year you get back a year (00:10:59) as we actually go past that point you'll (00:11:01) actually get back uh more than a year (00:11:03) and you'll go backwards in time uh which (00:11:06) would be cool (00:11:11) um now some people are concerned we're (00:11:14) going to run out of (00:11:16) resources uh and actually if we just (00:11:18) went ahead and didn't make any new (00:11:21) resources we would run out of resources (00:11:23) like energy for example uh but it's not (00:11:26) happening in a (00:11:28) vacuum uh AI is revolutionizing (00:11:31) everything for example we only have to (00:11:34) connect um one part in (00:11:37) 10,000 of the sunlight that falls on the (00:11:40) earth to meet all of our energy needs (00:11:42) it's plenty of Headroom uh and that's (00:11:44) growing exponentially and we'll achieve (00:11:47) that within 10 years and that's also (00:11:49) growing (00:11:51) exponentially (00:11:53) so we will have plenty of (00:11:56) resources (00:11:58) um and (00:12:00) when we get to the 2030s Nanobots will (00:12:03) connect our brains to the cloud uh just (00:12:06) the way your phone does it'll expand (00:12:09) intelligence a millionfold by (00:12:13) 2045 that is the (00:12:16) singularity we will be (00:12:20) funnier uh (00:12:23) sexier smarter more creative free from (00:12:26) biological limitations will be able to (00:12:29) choose (00:12:30) our (00:12:31) appearance we'll be able to do things we (00:12:33) can't do today like visualize objects in (00:12:35) 11 Dimensions we can speak all languages (00:12:39) we'll be able to expand Consciousness in (00:12:40) ways we can barely imagine uh we will (00:12:45) experience richer culture with our extra (00:12:48) years so I've recently become a (00:12:51) grandfather I'm very much looking (00:12:53) forward to that spending more time with (00:12:56) family friends loving and being being (00:12:59) loved all enhanced by AI I believe this (00:13:03) gives life its greatest meaning thank (00:13:06) you very much