AI and machine learning professor at Gonzaga University Graham Morehead joins WIRED to answer the internet's burning questions about artificial intelligence. What are the origins of AI? What's the difference between AI, AGI, and ASI? What will the implications be if China achieves artificial super intelligence first or the United States does? What does the next 10 years of AI development look like? Will AI take all human jobs? Answers to these questions and many more await on AI Support.
Director: Jackie Phillips
Director of Photography: AJ Young
Editor: Richard Trammell
Expert: Graham Morehead
Line Producer: Joseph Buscemi
Associate Producer: Paul Gulyas
Production Manager: Peter Brunette
Production Coordinator: Rhyan Lark
Casting Producer: Nick Sawyer
Camera Operator: Oliver Lukacs
Sound Mixer: Lila Rowel
Production Assistant: Abigayle Devine
Post Production Supervisor: Christian Olguin
Post Production Coordinator: Rachel Kim
Additional Editor: Jason Malizia; Shane Boissiere
Assistant Editor: Andy Morell
Director: Jackie Phillips
Director of Photography: AJ Young
Editor: Richard Trammell
Expert: Graham Morehead
Line Producer: Joseph Buscemi
Associate Producer: Paul Gulyas
Production Manager: Peter Brunette
Production Coordinator: Rhyan Lark
Casting Producer: Nick Sawyer
Camera Operator: Oliver Lukacs
Sound Mixer: Lila Rowel
Production Assistant: Abigayle Devine
Post Production Supervisor: Christian Olguin
Post Production Coordinator: Rachel Kim
Additional Editor: Jason Malizia; Shane Boissiere
Assistant Editor: Andy Morell
Category
🤖
TechTranscript
00:00Hello, I'm Graham Moorhead. I teach AI machine learning at Gonzaga University. Let's answer some questions from the internet. This is AI support
00:13At Justin dart 82 asks, what are the different types of artificial intelligence broadly speaking?
00:19There's type 1 and type 2 type 1 is like your instinct that part of the brain that can take in lots of information
00:26But reacts very quickly with the emotion type 2 is more like a methodical series of steps
00:33Like when someone asks you to do a math problem neural networks are type 1
00:38They can take in many inputs and they give a quick output
00:41But that output sometimes is not exact type 2 AI is logic and multiple steps
00:48Jsura asks, when did AI start you may have heard that there was a conference it was in
00:551956 it was the first conference where they used the term artificial intelligence. It was at Dartmouth in New Hampshire
01:02But AI existed before that
01:04We've all heard of the imitation game
01:06Alan Turing cracked the Enigma machine to save us in World War two
01:11That was a kind of AI but also he wrote papers in the 1930s and 40s that we still read today that are
01:18Important for understanding AI. At BFB morning show asks, what's up with grok?
01:24Grok is a newer entry into the AI
01:29Ecosphere we had Google come out with a paper in 2017 called attention is all you need and that was the BERT model
01:36But no one really cared until open AI came out with chat GPT
01:41Then we had Facebook or meta come out with llama and a French company came out with mistral
01:46Of course the Chinese company came out with deep seek Elon Musk didn't want to be left out of this
01:52So he came up with grok and it's the same structure now grok 3 the newest one was trained on more
02:00data more compute than any other of these models had ever been trained on before and
02:08It's good
02:09The big benefit comes from seeing other companies make mistakes
02:13The newer models get the benefit from those past mistakes and not make them
02:18Eugene asks is AI taking our jobs or not? The answer is yes, but don't worry
02:23Think back to the invention of the ATM people thought it was going to get rid of bank tellers
02:28But we have just as many bank tellers now as we did back
02:31Then AI is going to take a lot of jobs who just have to think differently about what jobs we do
02:37I would encourage you to think about how can you disrupt yourself?
02:41How could you take part of your own job and have a I do it for you
02:45Then you'll be more productive at daily AI tweets asks
02:49Why is a I dumb because it doesn't have a brain just a lot of circuits and algorithms
02:53The shape of thought is what's different when you think about?
02:58This thing is moving and it's going over here. You think about an object in space and you think about
03:05Consequence of actions for AI it's just numbers
03:10Matrices and vectors and all these numbers it is not shaped
03:15Internally like your thoughts are shaped Jabbar JD asks
03:19What's the difference between AI and AGI and the so-called super intelligent ASI?
03:26ASI is artificial super intelligence. Well AI of course was coined at that conference in
03:331956 they thought they were 20 years away from having something equivalent to a human
03:38But obviously it took a lot longer than that AGI
03:42Artificial general intelligence what that means is that you have an AI that can think adaptively
03:48Think about all the things that you can do even if you weren't trained to do them
03:52You can adapt to the situation AI can't do that very well yet an AGI
03:58would be able to
03:59ASI is gonna come very soon after that because once you have one computer that can do as much as a human you can just
04:07Turn on a second computer and now it can double user
04:10Okay, toe 69 69 asks how far are we from an AI partner like in the film her from the outside?
04:18It looks like we're not that far. There are people who have had online
04:22Weddings where they invited their friends and had an online
04:25Ceremony where they became betrothed to their AI loves so for some people we're already there
04:31But there is something that the AI in her had that no AI has yet
04:37And that's a will a desire the AI in the movie
04:43Wanted to do things on her own and by the end of the movie moves on to have her own life AI as it currently exists
04:51Doesn't want anything. It just responds with words when you give it words user
04:57Incognito to to asks. Hello
04:59I don't understand what tokens are and how they work with chat GPT when open AI or Google or all these other companies
05:07Designed their AI the first choice they have to make is how big is the vocabulary?
05:12How many words will it know about and the bigger you make it?
05:16The harder it is to train your model
05:18Sometimes their words other times their word parts when you say a word like gin like gin the drink that's in there somewhere
05:26that's already a
05:27Token, but the word ginormous is not in there ginormous gets split into three tokens
05:34gin or
05:35Muss and that first part of ginormous is a token and that token is the same token as gin the drink
05:42It thinks they're exactly the same at the beginning which should give you a hint about why AI is stupid
05:48Sometimes at Nicole Heather are chat GPT can't even get answers, right? I hate chemistry
05:55How does AI get wrong? I'm irritated chemistry is very difficult for AI when AI
06:01Operates on words that you give it it thinks about the world in a sequence of these words or tokens
06:08Which we just learned about chemistry is often three-dimensional structures
06:12it's difficult for AI to conceptualize that three-dimensional world if you think about a
06:18Glass of wine it can render a beautiful photograph of a glass of wine
06:23But it cannot render a full glass of wine because that's what it's seen
06:28AI is great at giving you something it has seen at
06:32Lamarck D asks, how is deep seek different to chat GPT? Is it better?
06:37They're very similar chat GPT version 4o came out in May of 2024 and it was amazing
06:45It was the best thing we'd seen and they have a reasoning model called Oh one that was built on that
06:51Deep seek comes along and gives us a model this just as good maybe a little better on some things
06:57Anytime somebody does something first someone else can come along and do the same thing cheaper
07:02Maybe even a little bit better at Rose Manoan asks
07:06How can AI be biased AI is learned from the Internet every bias across the Internet is somewhere
07:13Represented in AI it's learning from you. It's learning from me learning from everybody also remember
07:19Every book every blog post ever written on
07:23Flat earth theory is in there somewhere take it with a grain of salt at AI and design asks
07:29What will happen if the US achieves ASI first? What if it's China?
07:34Will there be a difference both the US and China have amazing AI capabilities?
07:39There are so many great papers coming out of China
07:42And I'm glad they published them in English because we learned a lot from them and they learn a lot from us
07:47They have more AI researchers than we do they have more STEM students than we do AI has been deployed in China
07:55Largely as a part of what we call their surveillance state in the US
07:58We typically use AI to empower people to be more creative and do more things right now. There is somewhat of a spirit of
08:06Competition, but also cooperation. I hope that's maintained
08:09I want you to think about ASI as being a virtual Einstein once we have that we can tell Einstein Einstein
08:17Why don't you please go do?
08:1910,000 years of research figure out time travel figure out
08:23Antigravity and then come back and tell us what you've learned ten seconds later. It comes back
08:28I just did ten thousand years of research and here's the secret the time travel in
08:33Antigravity think about what that virtual Einstein could do whoever gets to ASI first
08:38There's no points for second place at Carrie Cassidy 18 asks
08:42How is AI powered don't you have to charge it at a car charging station?
08:46AI uses a lot of power the big facility called Colossus
08:51they just put in Memphis, Tennessee uses about
08:5350 megawatts and uses a lot of power and a lot of water to cool the machines and we might be bringing on so
09:00many new AI training centers in the next 10 years that just AI will use the same amount of power as the whole country uses
09:08Now at luscious gem asks AI needs water
09:12How why the facility I just mentioned in Memphis uses swimming pools full of water
09:18maybe every minute it goes into these little tubes and each little tube goes up to one of the
09:24GPUs these are graphical processing units the actual
09:28computational engine of AI and every one of them has a little loop for the water to go by it and
09:34Take the heat and then the hot water is ejected out of the facility at Timbo cop asks
09:39I have questions about well-intended calls for educators to teach their students AI
09:45Literacy are we AI literate ourselves? Do we know what AI literacy involves who will teach us? It's important to read books
09:53I suggest you read books
09:55It's also important to get a vague sense of how these things work
10:00Most people are never gonna understand the inner mechanics of it
10:03Just like you don't understand how your car works, but you still drive it
10:07You got to think about AI as being your expert friend who knows a lot but has some bad
10:15Misconceptions so you don't trust everything if I were you I would use it a lot
10:20And trust it very little k0 crop asks is AI generated misinformation
10:27Going to ruin history. What are its potential implications for future?
10:32Historiography. Yes, it might ruin history, but it's not a new problem whenever
10:37There are victors in a war and they get to write the history those books may or may not be true whatever
10:43We get from our AI you got to remember it might not be true
10:47The only way not to ruin history is to get to first-hand sources and cross-reference with evidence at higher
10:55AI asks what's next for AI? Where do you see AI in the next decade now?
11:00There's a lot of very popular things that AI will do it's going to cure many diseases
11:05Alpha fold is a model that came out of Google
11:08Demiscivious was the head of this project and he used methods that had been applied to game playing and he applied them to figuring out
11:15When a protein is made, how does it fold? What's its actual shape in the old days? It might take us six months and
11:24$100,000 to determine how one protein folds now
11:27We know how they're structured this came out of AI
11:30That's probably gonna lead to curing many cancers and many other diseases launch the train asks
11:36Am I crazy or is AI therapy helping me? The first time AI therapy helped someone was around
11:431965 the AI was called Eliza all caps and a certain scientist made it and
11:49Started to show it to people his secretary would ask him to leave the room so she could talk to Eliza
11:55Privately I think that when you're talking to an AI therapist
11:59You're doing the internal work on yourself and the AI merely provides a framework or almost an excuse
12:06It's doing next token prediction. Just the next word
12:09It doesn't know what it means and it cannot conceptualize how it would feel to a human to hear those words
12:17Sad meat asks, how soon do you realistically think that AI will gain full sentience?
12:24Sentience is like consciousness. It means it's like something to be you. It's not like that for AI yet
12:30AI has no will no
12:33Feelings it's not sad. It's not happy. Even if it says to you the phrase I'm happy to the AI
12:40Those are just tokens when we figure out what consciousness is
12:44Which we haven't yet
12:46Then maybe we can begin to program or begin to build hardware that can support it
12:53Giulio Tononi and Sir Roger Penrose
12:56They're studying consciousness and they believe it could be something that involves quantum states and it's not something
13:04Computational AI that you use right now and maybe for at least 10 20 years will not be sentient at Silly Fox girl
13:13NYA asks, how do I avoid AI? Well, you can go camping
13:18It's gonna be harder and harder to avoid AI AI is everywhere when you talk to your phone and it types out your words
13:24That's AI when you walk past
13:28Maybe a government building and there's a security camera pointing at you and it's recognizing people or things
13:35That's AI now. You don't have to use AI you can live your life and
13:40Write your own essays do your own homework and I encourage you to do that
13:45Unpopular Dave asks with the advancements of AI video
13:49How will we be able to differentiate between what's real and what's not?
13:53This is a very challenging problem, and I know some researchers who are working on
13:59embedding keys in
14:01Video and in text later on when you look at that data, you can make sure it was not changed
14:07Another way is when AI changes something there are little
14:12watermarks in the video or the audio and we can develop another AI that can see those and identify this
14:19Text came from a large language model or this video came from a deep fake when you watch an AI video
14:26You may see a person shake their hair and the hair changes color or things that weren't there
14:32Come into view all of a sudden
14:35AI doesn't know about object permanence. It doesn't know about
14:40physics or the world and that's what I would look for if I were looking to see if a I was deepfaked at
14:46atmospheric with a zero asks
14:48How does AI learn to distinguish between truth and popular opinion in the material?
14:53It gets to learn from it doesn't AI has no clue
14:57All AI is trying to do is predict the next word and all it's trying to predict is this word
15:02Likely it doesn't care or know if it's true careful fig 8 4 8 2 asks
15:08What is the difference between AI and ML AI is artificial intelligence a broad umbrella?
15:15And there's many things under it. Typically you think if a human does something that takes intelligence
15:21Can I write a program that does that for them? That's artificial intelligence
15:25Now within that one of the methods we use is called machine learning. You have a bunch of
15:32Examples and you don't program every little line. You just tell the AI look at these examples and teach yourself
15:40That's machine learning at Albert
15:42Underscore JJ JJ is AI that can write its own code and a GI
15:48Not necessarily because we already have this and we don't have a GI yet
15:52There are a number of AI agents that can write code
15:56You can ask an AI agent to write code and then run that code and see if you can ask it
16:02to write code to improve itself and then run that code and very quickly it
16:08Decays like images. We've all seen image generators
16:12You give it a phrase and you imagine what it's going to show you and it shows you something beautiful
16:16Well, we did that and then took a bunch of these and trained another AI and it made images that were not quite as good
16:23but we use those and trained another level and did this again and again and what you find is that
16:28It becomes like a copy of a copy of a copy. It's
16:31terrible mode collapse is what happens and that's going to happen right now when AI writes itself and
16:39Changes itself over time because these agents and these AI's don't have
16:45Consciousness at code with the Dahlia asks is tech evolving too fast for our own good
16:51Today's AI can compose music write essays and code, but are we losing the human touch?
16:56Where should we draw the line in the silicon sand? I think every person should draw their own line. Think about it like
17:04Exercise I did not come here on foot. I came in a car the car made it easier and faster
17:09But does that make me lazy? I
17:12Find time to still exercise even though I have machines that can bring me places very quickly the same for the mind
17:20Even though you have a machine that can write an essay for you. You should still write essays
17:24You should still be creative. You should still express yourself. AI just lets you get rid of the boring stuff
17:31Original confusion asks, what is the difference between predictive AI and generative AI?
17:37they're just different ways of looking sometimes at the same AI sometimes differently a
17:42Generative AI learns from a distribution
17:45think about all the words on the internet being a
17:49distribution of words or all the pictures on the internet being a distribution of images a generative AI
17:56can make an image that looks like it was taken from the internet or
18:02Generate words that looks like they were generated by a human predictive AI is
18:09Just a different way to think about it
18:10maybe you want to predict who's gonna win a big basketball game or the Super Bowl or what the weather will be tomorrow at
18:18Even sir wave and asks I'm lost here. How can AI create jobs?
18:22Isn't it supposed to replace humans you sent that on a computer the computer was going to take away jobs
18:29But somehow it made many new jobs the internet took away jobs
18:33But it also made new jobs a hundred years ago. Would you have been able to explain to someone?
18:39What a youtuber is we all need to become AI?
18:43Managers by the end of this year, there will be millions if not billions of AI
18:48Agents, these are little AI's that could work for you the jobs of the future. A lot of them will be
18:56management
18:57Managing the work of AI's at the omni liberal asks
19:01Please stop using dog AI for legal documents either go to someone that has learned to read legal documents or do some homework yourself
19:08I would say a similar answer use AI for the easy stuff easy doesn't mean quick
19:14there are things that are easy to do that take forever like reading through a document and
19:19Finding that one section that talks about a certain thing then maybe you can't search for it because you don't know what words were used
19:25But AI can find it for you at any good name left asks when they get good enough
19:31How will they be proven to be deep fakes deep fakes again are AI?
19:37Generated videos that look like someone, you know AI as it stands now
19:42Doesn't understand the three-dimensional world
19:44So we still have that way to look at these deep fakes and say something's wrong
19:50Either the dynamics in the video or something
19:54Disappeared or came into existence
19:56but at some point that won't be true at some point AI will operate in a three-dimensional world and it won't be
20:04Possible for maybe even the best of their AI to determine that it was deep faked
20:09Therefore we have to go to the other strategy
20:11You have to have trusted people who when they're taking video or when they're taking a picture or audio
20:18They're baking into it a
20:20Signature that makes it impossible to cheat. That's the only way in the future. We'll know that something is not deep-faked at
20:27Discivenant asks, what sector do you think will be most impacted by AI in the next five years?
20:33I'm not sure I could tell you even for one year
20:35I do think that everything that involves words is going to be very much affected then imagery
20:42Then video there is so much coming that we don't know where to start
20:47Think about the proteins that just came out of alpha fold somewhere in there
20:52There's a few needles in the haystack that can cure this cancer in that cancer
20:56There are things that can help just general metabolic health
21:00Maybe we'll all live longer at Kaylee author asks how dangerous is AI you got to think about AI as a tool?
21:08There are dangerous people out there that might use that tool
21:11There are bad actors who can use it in a bad way
21:13But AI on its own has no will or desire at a jinky almond are asks
21:20Considering the potential of artificial intelligence reaching human like intelligence. Do you believe it should have rights?
21:26No, for reasons I said before I believe AI does not have consciousness. So I do not believe we should give it rights
21:34Anonymity underscore anonymous asks what things is it important not to tell chat GPT and why what if you tell your deepest?
21:42Darkest secrets to chat TPT then it's a chance that that's in its memory somewhere and it used that
21:49To train its next word prediction
21:52So the basics are you don't tell chat TPT things you wouldn't want to put on a website that just faces the world
21:59I'd hypnotize D asks with all these companies talking about AI
22:03How long do you think it will be until we get Skynet?
22:07Skynet is a choice and we should choose not to do it now
22:12There are limited cases where you have to have an AI in control of a weapon
22:17Think about a Patriot missile when you have some kind of missile coming in to kill you
22:23You need a missile to take down that missile
22:26But it's too fast for a human to sit there and aim it correctly
22:31You need an AI to take over for those limited things
22:34We as humans need to make policy choices to say there will always be a
22:40Responsible human who makes the top choice
22:44Okay, those are all the questions. Thanks for watching AI support