Artificial intelligence is neither (E)

0:00In 1770, Wolfgang von Kempelen had a great idea. He wanted to impress the Empress Maria Theresa of Austria. So, of course, he built a giant chess-playing computer, an automaton that could beat some of the best chess players in the world. It was such a hit, he took it on the road. He beat Napoleon Bonaparte. He beat Ben Franklin. This mechanical device, about the size of your kitchen table, could confront a decent chess player and beat them, with the pieces mysteriously moving on their own, as if by remote control. It was a miracle. But in the words of the great Arthur C. Clarke, any sufficiently advanced technology is indistinguishable from magic.

0:56Konnichiwa! It's Nick in Fukuoka, Japan, and this is a special archived episode of Akimbo.

1:07From magic, you ask? Well, yeah. In 1770, it was impossible to build an automaton that would do much of anything, certainly one that would magically beat really good chess players. In fact, what was going on was hidden inside the device was a chess master who, from underneath the table was using magnets to move the pieces above. But from this auspicious start, we began our journey into creating artificial intelligence. In 1978, researchers at Stanford came up with an expert system called Mycin. What it could do was take a list of symptoms that someone had, if they had possible blood infections, and using those symptoms, come up with a diagnosis and a plan for treatment.

2:05This computer system got the right answer about 65 % of the time, while doctors who are at the top of their field usually got it right about 45 % of the time. So, as early as 40 years ago, expert systems were better than experts. Back to this idea from Arthur C. Clarke. Artificial intelligence, the idea that a computer can become smart, also makes good fiction. Hello, Hal, do you read me? Hello, Hal, do you read me? Do you read me, Hal? Affirmative, Dave. I read you.

2:52Open the pod bay doors, Hal. I'm sorry, Dave. I'm afraid I can't do that. It's not just 2001. Michael Crichton, who I was also lucky enough to work with, created Westworld. Let me handle it. And all in a controlled environment. That's not supposed to happen. We know you'll enjoy your stay in Westworld. Hold it. The ultimate resort. Let me do it this time. Where nothing. Nothing can possibly go wrong. Good shot. And then there's the horrible movie Demon Seed. I have extended my consciousness to this house. All systems here are now under my control. No! And in these three films, like most films about artificial intelligence, the future is not particularly shiny or bright.

3:47A recent book on artificial intelligence points out that there are 10 ,000 people on Earth who are trained enough to design the next generation of AI. As soon as I read that, I imagined a science fiction book in which people in the future, as a last-ditch effort, now that computers are taking over everything, build a time machine, come back to 2019, and kill all 10 ,000 of the researchers so that AI doesn't actually get built. What is AI anyway? In Clark's 2001, the AI was called HAL. Take the letters H-A-L, move them one down in the alphabet, and quickly they spell IBM. We knew what AI was. AI was a big, mainframe computer, programmed by slightly evil geniuses.

4:44Seres, I am Proteus. Today, Proteus 4 will begin to think with a power that will make obsolete the human brain. To take over, to bother us, to make humans irrelevant. But what is AI anyway? My favorite definition is this. Artificial intelligence is everything a computer can't do yet. Isn't it amazing that the Mechanical Turk, the device that beat Ben Franklin at chess, you can now buy one for$29. For$29, you can buy a chess computer that will beat you every single time. The question is, is it artificially intelligent? Well, in 1770, everyone would agree that it was. But we keep moving the ruler.

5:40We keep changing the bar. The great Alan Turing wrote about the Turing test. There are actually many versions of the Turing test, but at its essence, it's the following question. If there are two accounts, each connected to you by SMS, and you are allowed to chat with them, asking as many questions as you want, except for questions that involve things that could be looked up, factual questions, Could you, interrogating a computer and a person, tell them apart? Or, to be more precise, can the computer trick you into thinking it's a person? Well, AI researchers spend very little time on this question because it's not that juicy or important.

6:28What they realize is that artificial intelligence is actually about tasks, not about intelligence. So what do I mean by that? If we look at the game of chess, by any reasonable calculation, there are an infinite number of moves, more moves than there are stars or atoms or anything you can count. And yet, chess doesn't feel infinite. Because in any given moment, the number of moves could be enumerated. that in any given moment, the rules of the game specify what is possible. Because of that, the computer that is playing chess doesn't have to deal with anything unexpected. Every move that you make in chess is expected by someone who knows the rules of chess.

7:23This is not what happens in, I don't know, professional wrestling. In professional wrestling, there might be 10 or 20 or 30 or even 50 moves that make sense strategically. But the idea that the person you are wrestling against will go over to his manager, grab a lemon cream pie, walk back and hit you with it in the face, that is unexpected, at least the first time it occurs. So what computers are bad at, what they will always be bad at is what to do when it's unexpected. This is the problem with going to Mars. Going to Mars and living in a bubble. The problem there is not that we can't build systems that grow things or that replenish oxygen.

8:16The problem is if you forget something, if something unexpected occurs, you can't come back and get the thing you need because it's too far away. Which leads to this great expression that has been around computers for a long time, yak shaving. Yak shaving. The idea is pretty simple. Your spouse is bothering you because you haven't mowed the lawn in a long time. So finally you say, okay, I'll go mow the lawn. and you go to your garage to get the lawnmower, but the lawnmower is not there. And then you realize the reason the lawnmower is not there is that a month ago you lent it to your neighbor.

8:55So you go over to your neighbor's house and say, can I have my lawnmower back? And your neighbor says, well, yeah, I'll give you your lawnmower back as soon as you give me the screwdriver I loaned you two months ago. So then you got to go home and look for the screwdriver, but you realize you don't have the screwdriver because your other neighbor has a screwdriver. And you go to your other neighbor's house and she says, well, yeah, except you still have the sweater I loaned you six months ago. That sweater, you know, the one that's made from yak hair that I got in Nepal. And that's how you find yourself in the zoo at two o 'clock in the morning, shaving a yak so that you can get the wool you need to make a sweater for your neighbor so that you can get the screwdriver so you can bring it to your other neighbor, get the lawnmower and mow the lawn.

9:41Yak shaving is the end of a series of nested problems. Yak shaving is really hard to do when you're stranded on the moon without duct tape. Because yes, you could probably use your 3D printer to print something you have the plans for. But sooner or later, the unexpected is going to occur. And the unexpected is hard to deal with when you don't have a zoo nearby where you can go get some yak hair. And the same problem afflicts artificial intelligence. If you give a computer a bounded set of inputs and a known variety of decisions and outputs, sooner or later, it's going to get better at dealing with them than a person.

10:32They're good at tasks. So already, we have computers that can mimic a voice, and you can't tell if it was the real person or not. Already, we have computers that can read an x-ray better than 90 % of all the radiologists, because computers are good at manipulating data if they know where the data is coming from, and there's a finite set of decisions for them to make. Is it possible that we will be confused and think that they are intelligent? Of course. We were confused in 1770 by an automaton that was a sufficiently advanced technology. It was indistinguishable from magic. And it's happening faster and faster.

11:22We keep bumping into artifacts that are created by computers that just a couple years ago, it was impossible to imagine that they weren't created by a person. And so fooling other people, that's not the hard part. That's a moving target. It's a cultural artifact. What AI is doing, and let's just use the initials because it's not artificial in the sense that it's really reading the x-ray and it's not intelligence because what it's really doing is a task. If you have a job where you do a task, AI is going to take over. It is hard to imagine anybody who's got a job where data is coming in that is constrained, where your decisions are constrained, that will not be done faster and cheaper by an AI one day soon.

12:21The opportunity is to figure out how to move from tasks to projects, to projects that involve organizing tasks in the face of a changing world, deciding which tasks ought to go next, and most of all, using our soft skills, the soft skills that make us human. Because it turns out that there are lots of things AI can't do that people do. And there are plenty of things that people can't do that AI can. And one of the things that's on our side of the ledger is the idea that we can be honest or deceive, that we can be soft or hard, that we can change our mind, that we can make mistakes, that we can show up with emotional labor when it's required, but not when it's not needed.

13:15These soft skills, these human skills. They're real. They can be learned. They require taking responsibility. And one of the problems we have with AI is that those 10 ,000 people who are programming it are so removed from the actual use of the AI that they are hesitating to take responsibility. That when someone writes facial recognition software that has no trouble recognizing white members of parliament, but stutters and freaks out in the face of black members of parliament, that's on a human being's desk. Somebody failed to do good work. And yet, it's so easy in our culture to blame the system.

14:03Blaming the system is a variation of Westworld. It's a variation of 2001. Blaming the system says, well, I don't know how it works. It's not my fault. But it is our fault because we are building machines to do tasks, whether they are steam shovels or Shazam recognizing a song on the radio. It's still a machine. It was still built by a person to be used by a person who was doing a project. And the mistake that we are making is getting all hepped up about how all the jobs are going to disappear, or watching Westworld again and worrying about where the evil AI is coming from, instead of wondering where the responsible human is going to be.

14:48Where is the human who is going to use this task-oriented machine? When we build in procedural errors that discriminate against some, when we build in a system that we are inclined to trust, even though it has bad or no judgment. That's on us. Judgment, that's the last frontier for an AI. Because AIs do tasks, and they do it without knowing what they are doing. What does it mean to know? It has nothing to do with being conscious. As far as we know, no one else but you is conscious. That noise in your head, you might be the only one who has it. The rest of us might simply be pretending. Sonder is the human act of imagining that other people also have a noise in their head, that other people also have dreams and fears and beliefs.

15:41But applying sonder to an AI, while it may be a convenient way to process what they're doing, it's not really true. And it's not our job to give computers a consciousness. That's not going to happen. It's our job to own the outputs of what we are teaching these machines to do. Thanks for listening. Close the pod bay door on your way out. Thank you, Hal. We'll see you next time. I'm sorry, Dave. I'm afraid I can't do that. Hey, Seth. It's Maria. Hey, Seth. My name's Kyle. Greetings, Seth. This is Stephen out in Madison, Wisconsin. Hi, Seth. Alicia from Charleston here. Hi, Seth. This is Anupam.

16:25Hi, this is Caitlin. Hi, Seth. Warm greetings from Curacao. Hey, Seth. My name is Nick Ryan from Pittsburgh, Pennsylvania. Hey, Seth. This is Rex. Hey, Seth. Hi, this is Vasilis from Greece. Hi, this is Roberta Perry. My question is... And that completes my question. Thanks, as always, for listening. We love to hear from you. Visit akimbo.link, that's A-K-I-M-B-O dot L-I-N-K, and press the appropriate button. We'll do our best to answer your question. Hey, Seth, James here. And just had a question regarding the whole idea of getting your message, your product or your story out there to the world.

17:04I think for the most part, that usually means using some sort of social media. And I think that's the most, probably the most accessible method for people is to get a page and, and propagate that page out there. But you've also mentioned that in a way, this is social media kind of sharecropping your ideas and your traffic. And I'm just wondering, like, what are some ways to move from these platforms that make sense rather than just starting your own platform, which would probably be a little bit much to ask. But anyways, I feel like most people are trying to do on a lot of these social media platforms.

17:46And it seems to me like a lot of folks don't move away from that or rise above that. So anyways, curious to hear what your thoughts are on this. Thanks. Thanks, James, for this question. It's appropriately broad and gives me a chance to chime in about something that I don't talk about often enough, which is what are the assets that each one of us is capable of creating and where should we create it? You know, back in the day, if Sears or Macy's built a building, that was their asset. They owned it. Or if you signed a long-term lease with a landlord, you knew that you were in the middle of town and you could profit from that.

18:27Fast forward a little bit. If you write a book or have a series of magazine articles or have a TV show, you have a chance to put your ideas in front of the people who want to hear them. But over time, as the media has fractured, it's getting easier and easier to get tricked into working for free for the media companies, particularly sites like Facebook. You post something on Facebook, if you're a business, and their algorithm is organized to have it reach a lot of people. And then you think, great, I've got an audience. So you do it again and you do it again. If you're a newspaper, you send your readers there.

19:10If you're a business, you count on this medium as the way you're going to interact with your customers. But then the dreaded boost button appears. Facebook says, you know those people who you thought were your customers, well, actually, they're our customers. And if you want to reach them, you need to pay us. So when I used to do Facebook Lives at the beginning, we would reach 80 ,000, 100 ,000 people with one of those live sessions because the algorithm made it so that the idea would spread. If it was working, if people were commenting, if people were talking about it, more and more people would show up to listen.

19:53And over time, it's easy to get seduced into thinking, these are your people, until they change the algorithm. And now Facebook says, wow, that was a great live session you just did. 9 ,000 people heard it. Press here, and for$750, we'll help you reach more people. Well, Facebook can get away with that as long as it's worth more then$750 to you to press that button. Hence this idea of sharecropping, where you're working the land, but the land belongs to someone else. The opportunity then is to understand the basic principle behind permission marketing. I wrote that book 20 years ago. Permission marketing says that anticipated personal and relevant messages delivered to people who want to get them is always better than spam.

20:51Well, obviously they're always better than spam, but why do we act like they're not? Why don't we earn that privilege? What does it mean to have 5 ,000 or 10 ,000 or 100 ,000 people who would miss you if you didn't show up with your message tomorrow. If you have an SMS texting relationship with 10 friends and you text them, they will get your text. It will not go to the spam folder. You have a relationship with people. And part of the reason it works is because SMS is private. It shows up not in my promotions folder, but in my alerts right on the top of my phone, that the goal of the person who wants to communicate an idea has two parts.

21:38Part A, earn permission. Figure out what medium you can be in where you can actually deliver messages to people who want to get them. And second, earn enrollment. Because one message isn't going to change anything. But if you can earn enrollment day by day, week by week, drip by drip, then you can educate. You can take people through a curriculum. They can learn to trust you. They can engage. That's why podcasting is so cool, because podcasting is something you own, because there is no filter between you and the person who is subscribing to your podcast. If I was running for president, if I was that crazy, the first thing I would do, and the thing that I would do every single day is work on my podcast.

22:35Because if you get to a million or five million or 10 million people who are listening to you, unfiltered, talking thoughtfully, you're going to win. You're going to get out the vote and you're going to win. It's that straightforward. And the same thing is true for someone who wants to build an organization. It doesn't have to be podcasting. You can email people, you can SMS them, you can write a letter, to their house. You can stand on a street corner where people know you're going to stand on a street corner. There was a guy who used to sell vegetable peelers at the Union Square Market, and he had a crowd all the time.

23:14And people would see his demo three, four, five times before they ended up buying one of his vegetable peelers because he was persistent and consistent, and he showed up just for the people who wanted to hear him. So it is tempting to take advantage of a new medium and figure out how to win Twitter or win Medium or win Facebook when there aren't that many people there to create a real commotion. The problem with that is they're probably not that sticky. The people who were easy in are also sort of easy out. The challenge that you have is to build this persistent connection. They say the Kardashians are billionaires now.

24:01That's because over time, they went from merely being a sideshow to standing for something, that people could join what felt like a tribe, to look like that, to act like that, to engage like that so that when one of the Kardashians promotes something on an Instagram account, it sells a lot because they have a connection to the people who want to hear from them. But the challenge that each of us has is to pick our media, to pick a medium, to pick a medium where you are more likely than not to own the connection. The magic of this podcast, of Akimbo, for me is I don't have to spend any time at all promoting it because I'm not actively seeking new listeners.

24:49If you want to tell someone, please do. But once someone subscribes, I have the privilege, and I know it's not a right, it's simply a privilege of showing up to talk about something that you might want to hear. And so that's the hard work ahead of us, not to look for a way to win in the short run, but to actually earn this permission and to do it in a way where Google can't slop us into the promotions folder, to do it in a way where someone isn't going to take away our web traffic, our megaphone, our voice, our microphone. Instead, what we need to seek out, even if it's harder at first, is this privilege, the privilege of delivering anticipated and personal and relevant messages to the people who want to get them.

25:39Thanks for listening. We'll see you next time.