WEBVTT - AI: Friend or Foe? 0:00:04.440 --> 0:00:12.520 Technology with tech Stuff from dot com. Hey there, and 0:00:12.600 --> 0:00:16.480 welcome to tech Stuff. I'm your host Jonathan Strickland. I'm 0:00:16.480 --> 0:00:19.600 a senior writer for how Stuff Works dot com. Where 0:00:20.200 --> 0:00:23.400 do you know what we do? We explain the universe. 0:00:24.400 --> 0:00:28.960 And this topic today is all about a little exchange 0:00:29.040 --> 0:00:34.080 that happened online over the course of several days. Actually, 0:00:34.080 --> 0:00:38.159 it started with Elon Musk and he was addressing a 0:00:38.200 --> 0:00:43.120 governmental body and talking about his view that artificial intelligence 0:00:43.240 --> 0:00:47.240 needs to have strict regulations attached to it in order 0:00:47.280 --> 0:00:52.440 to prevent some sort of catastrophic future, possibly sky Net related, 0:00:52.840 --> 0:00:57.560 where the robots and other artificial intelligent constructs rise up 0:00:57.600 --> 0:01:01.320 against their human masters and crush us under their metaphorical 0:01:01.680 --> 0:01:07.199 or perhaps literal boots. And then you had Mark Zuckerberg 0:01:07.319 --> 0:01:11.720 of Facebook on a live broadcast on Facebook Live from 0:01:11.760 --> 0:01:14.960 his backyard during a barbecue. He was asked a question 0:01:15.160 --> 0:01:19.120 about this sort of thing, and he specifically said that 0:01:19.160 --> 0:01:22.640 he thought this was a very pessimistic view of the 0:01:22.680 --> 0:01:25.600 future of artificial intelligence, and that within five or ten years, 0:01:25.680 --> 0:01:29.679 artificial intelligence would be transforming our lives in ways that 0:01:29.720 --> 0:01:32.120 we can't even imagine, and they would all be awesome 0:01:32.280 --> 0:01:35.520 and fantastic and magical and we should love that. And 0:01:35.560 --> 0:01:38.880 then Musk struck back on Twitter and said that he 0:01:38.920 --> 0:01:41.880 had talked to Zuckerberg about this process before, but frankly, 0:01:42.360 --> 0:01:45.720 Zuckerberg just is out of his depth with artificial intelligence. 0:01:45.720 --> 0:01:48.080 It's not something that he's an expert at, and he's 0:01:48.120 --> 0:01:54.480 really speaking from inexperience. I find this exchange amusing, as 0:01:54.520 --> 0:01:59.200 does a lot of the journalists area of technology. I mean, 0:01:59.240 --> 0:02:01.320 we've got a lot of people who are commenting on this, 0:02:01.920 --> 0:02:07.200 but personally I also find it a little confounding because, uh, 0:02:07.400 --> 0:02:12.920 Elon Musk has said some stuff that contradicts his own 0:02:13.080 --> 0:02:18.280 companies policies if you look at it carefully. Um, he 0:02:18.600 --> 0:02:24.920 has specifically resisted the concept of regulations for self driving cars, 0:02:24.919 --> 0:02:32.679 but you could argue very uh realistically and convincingly, I 0:02:32.720 --> 0:02:36.960 would say that self driving cars are an implementation of 0:02:37.040 --> 0:02:42.280 artificial intelligence. So we're gonna dive into this. We're gonna 0:02:42.320 --> 0:02:45.880 look at the different opinions about artificial intelligence, kind of 0:02:45.919 --> 0:02:49.040 explore the concept of artificial intelligence in general, see where 0:02:49.080 --> 0:02:52.240 it came from, and what does it really mean and 0:02:52.560 --> 0:02:55.320 who's right or as I put it in my notes, 0:02:55.440 --> 0:02:58.960 Musk's position is AI without regulation is going to totally 0:02:59.040 --> 0:03:02.160 kill us, dude. And Zuckerberg's position is a is going 0:03:02.200 --> 0:03:05.680 to improve our lives in countless ways. Bra So who's 0:03:05.760 --> 0:03:10.440 right or neither of them right? Well, to start off with, 0:03:10.520 --> 0:03:14.800 let's talk about the birth of the term artificial intelligence. 0:03:15.200 --> 0:03:18.440 It was coined by John McCarthy, who passed away in 0:03:18.440 --> 0:03:20.800 two thousand eleven at the age of eighty four. He 0:03:20.919 --> 0:03:24.639 worked at Stanford as a professor emeritus of computer science, 0:03:25.120 --> 0:03:28.400 and he also co founded the Artificial Intelligence Project at 0:03:28.480 --> 0:03:32.000 m I T as well as the Stanford Artificial Intelligence Labs. 0:03:32.000 --> 0:03:38.480 So somebody who certainly has had a long and storied 0:03:38.520 --> 0:03:42.040 past in the development of artificial intelligence. He first used 0:03:42.080 --> 0:03:44.760 the term artificial intelligence in a proposal for a summer 0:03:44.840 --> 0:03:48.400 research conference at Dartmouth in nineteen fifty five. It was 0:03:48.440 --> 0:03:52.040 the first time the term ever appeared in a printed publication. 0:03:53.800 --> 0:03:56.600 So what is artificial intelligence? I mean, it's such a 0:03:56.680 --> 0:03:59.120 huge term, has been used by so many people that 0:03:59.320 --> 0:04:02.160 it's lost a lot of its meaning. Also, I should 0:04:02.160 --> 0:04:05.280 point out that John McCarthy I mentioned earlier as a 0:04:06.240 --> 0:04:09.520 one of the creators of Lisp, the programming language that 0:04:09.600 --> 0:04:12.920 was used in artificial intelligence. So if the name sounds familiar. 0:04:12.920 --> 0:04:14.760 It means that you listen to the History of Programming 0:04:14.800 --> 0:04:18.320 Languages episodes, or that you're just familiar with John McCarthy's work. 0:04:19.360 --> 0:04:21.760 But yeah, artificial intelligence is one of those terms that, 0:04:21.960 --> 0:04:25.120 since its introduction, has been used to describe so many 0:04:25.160 --> 0:04:28.040 different things, and used in such a vague way so 0:04:28.160 --> 0:04:31.320 many times that for many people it seems like a 0:04:31.360 --> 0:04:36.120 meaningless term. It's almost like it's just a general catch 0:04:36.160 --> 0:04:40.520 all for the scary possibilities of technology that gets away 0:04:40.560 --> 0:04:43.520 from us. It reminds me of Humpty Dumpty and through 0:04:43.560 --> 0:04:47.599 the Looking Glass. He says that that words mean exactly 0:04:47.640 --> 0:04:50.520 what he wants them to mean, neither more nor less. 0:04:51.320 --> 0:04:52.880 He says, you know who is to be the master? 0:04:53.040 --> 0:04:55.800 That is the only thing that's important. I'm more in 0:04:55.920 --> 0:04:58.000 charge than the words are. So if I use a 0:04:58.040 --> 0:05:00.680 word to mean something, that's what it means. That's what 0:05:00.760 --> 0:05:03.440 I feel artificial intelligence has become for a lot of people. 0:05:04.080 --> 0:05:05.719 And I also think that that's what leads to a 0:05:05.720 --> 0:05:10.560 lot of disagreements. That some people have one idea of 0:05:10.600 --> 0:05:14.240 what artificial intelligence is and other people have a totally 0:05:14.279 --> 0:05:17.440 different idea of what artificial intelligence is. But because they're 0:05:17.480 --> 0:05:20.719 both using the phrase artificial intelligence. It seems like they're 0:05:20.720 --> 0:05:23.880 talking about the same thing, and that's why they're hitting 0:05:24.160 --> 0:05:27.200 some massive disagreements, or at least one of the reasons 0:05:27.480 --> 0:05:31.040 why they are disagreeing. It's really because, if you dig 0:05:31.120 --> 0:05:36.920 down further, they're talking about two different things frequently. Now, 0:05:37.000 --> 0:05:40.480 John McCarthy's use of artificial intelligence was given this context 0:05:40.600 --> 0:05:44.280 in his proposal. Quote the study is to proceed on 0:05:44.360 --> 0:05:48.440 the basis of the conjecture that every aspect of learning 0:05:48.720 --> 0:05:52.719 or any other feature of intelligence can, in principle, be 0:05:52.920 --> 0:05:56.560 so precisely described that a machine can be made to 0:05:56.720 --> 0:06:01.560 simulate it. End quote. Here's the thing about that definition. 0:06:02.480 --> 0:06:06.159 It already is vague because it's talking about the every 0:06:06.200 --> 0:06:09.960 aspect of learning or any other feature of intelligence. We 0:06:10.080 --> 0:06:14.360 have not fully defined what intelligence is within the human experience. 0:06:15.120 --> 0:06:21.080 There are aspects of intelligence that are very vague and fuzzy, 0:06:21.160 --> 0:06:25.640 and we only have kind of a partial understanding of 0:06:25.640 --> 0:06:29.679 what it actually is. An example I might give as consciousness. 0:06:31.040 --> 0:06:35.520 Defining consciousness is a particularly troublesome and difficult thing to 0:06:35.640 --> 0:06:39.680 do in human beings, let alone in machines. So we 0:06:39.839 --> 0:06:45.120 know that consciousness is a manifestation of the brain, the 0:06:45.200 --> 0:06:47.560 idea of the mind being a manifestation of the brain. 0:06:47.640 --> 0:06:52.000 All of this is dependent upon actual physical matter, the 0:06:52.040 --> 0:06:54.599 gray matter in our heads. And we know this because 0:06:55.120 --> 0:07:01.800 various UH diseases, disorders, injuries that affect act consciousness are 0:07:01.800 --> 0:07:04.920 the ones that are affecting the brain. It's if you 0:07:05.600 --> 0:07:09.440 UH suffer an injury to the brain that is in 0:07:09.520 --> 0:07:13.000 one of these areas that define consciousness, your sense of 0:07:13.040 --> 0:07:17.480 consciousness is likewise affected. That tells us that there is 0:07:17.480 --> 0:07:22.520 this physical connection, that there's not this metaphysical mind necessarily 0:07:22.640 --> 0:07:27.000 that is a layer on top of our physical brains. 0:07:27.080 --> 0:07:32.920 But beyond that, defining consciousness is really tricky, and there 0:07:32.960 --> 0:07:37.960 are plenty of psychologists, neurologists, philosophers who have debated the 0:07:38.080 --> 0:07:42.360 nature of consciousness for ages, and we're no closer to 0:07:42.480 --> 0:07:46.160 really defining it than we were back then. Sometimes we 0:07:46.880 --> 0:07:51.160 narrow it down by saying, what isn't consciousness? So we're 0:07:51.160 --> 0:07:54.360 whittling it away, kind of the way the sculptor whittles 0:07:54.400 --> 0:07:58.240 away everything that isn't a statue when they start with 0:07:58.240 --> 0:08:01.200 a block of marble. You know, you just say, are 0:08:01.280 --> 0:08:03.880 you in a car of an elephant? Here's your block 0:08:03.880 --> 0:08:06.160 of marvel. Cut all the stuff that doesn't look like 0:08:06.200 --> 0:08:08.760 an elephant away, and what you're left with is the 0:08:08.800 --> 0:08:12.360 elephant that might be what consciousness is for us. We 0:08:12.440 --> 0:08:15.720 remove all of the concepts that are not consciousness, and 0:08:15.720 --> 0:08:19.200 whatever is left over that becomes our definition. It's not 0:08:19.320 --> 0:08:24.760 exactly satisfying at the moment. Well McCarthy expounded on artificial 0:08:24.800 --> 0:08:29.000 intelligence in nineteen sixty in a paper titled Programs with 0:08:29.160 --> 0:08:33.760 Common Sense, which kind of gives you another perspective of 0:08:33.800 --> 0:08:38.200 what artificial intelligence could be. And this leads us to 0:08:38.960 --> 0:08:42.960 ask other questions like what exactly kind of what? What 0:08:43.000 --> 0:08:46.760 are the implementations of artificial intelligence? What what types are there? 0:08:47.679 --> 0:08:50.800 And again, the number of types of AI depends upon 0:08:50.840 --> 0:08:54.400 whom you ask and how they frame the answer. They're 0:08:54.440 --> 0:08:56.839 simple answers where some people will say, oh, there's two 0:08:56.840 --> 0:09:00.120 types of AI. They're strong AI and there's weak a I. 0:09:00.960 --> 0:09:04.520 Or some people will say general AI and narrow AI. 0:09:05.440 --> 0:09:07.880 Others will say no, no, no, there's like thirty three 0:09:08.040 --> 0:09:12.080 types of artificial intelligence. Or they might say there's three 0:09:12.160 --> 0:09:15.720 or four types of artificial intelligence. With those two types, 0:09:15.760 --> 0:09:19.360 those are the easiest to kind of explain in broad categorizations. 0:09:20.200 --> 0:09:24.200 Narrow or weak AI is the artificial intelligence we would 0:09:24.240 --> 0:09:29.000 create that's dedicated to a narrow task or series of tasks. 0:09:29.040 --> 0:09:32.560 Strong AI is generally understood to mean a machine that 0:09:32.640 --> 0:09:38.760 has sentience, consciousness, and mind, those qualities that we associate 0:09:38.920 --> 0:09:43.480 with human intelligence. But again, we cannot even fully describe 0:09:43.960 --> 0:09:49.640 those concepts within the context of humans, so trying to 0:09:49.679 --> 0:09:54.040 figure out how to imbue machines with those elements is 0:09:54.120 --> 0:09:59.239 even more complicated. Now there's also the concept of general intelligence. 0:09:59.320 --> 0:10:04.040 This is not reliant upon consciousness or mind or sentience. 0:10:04.480 --> 0:10:08.920 With general intelligence, the idea is that a machine would 0:10:08.920 --> 0:10:12.960 be able to apply intelligence to any problem rather than 0:10:13.000 --> 0:10:16.160 just a specific or narrow band of problems. So, in 0:10:16.160 --> 0:10:21.480 other words, a generally intelligent machine could be used to 0:10:21.559 --> 0:10:27.559 solve problems of various degrees and various contexts. So you 0:10:27.640 --> 0:10:32.000 might have a general intelligence robot, and the general intelligence 0:10:32.080 --> 0:10:34.920 robot can do things like figure out how to manipulate 0:10:35.400 --> 0:10:41.040 physical objects, how to maneuver around within an environment, UH, 0:10:41.080 --> 0:10:43.920 and a few other elements as well. It's having a 0:10:43.920 --> 0:10:47.960 more general approach to problem solving as opposed to something 0:10:48.000 --> 0:10:54.119 that was made specifically to handle a particular task UH. 0:10:54.160 --> 0:10:58.360 But general intelligence. True general intelligence would be capable of 0:10:58.480 --> 0:11:02.679 applying an intelligen an approach to any kind of problem, 0:11:02.679 --> 0:11:08.160 not just a related family of problems. In an article 0:11:08.200 --> 0:11:11.640 for Government Technology, there was a writer named Arrend Hints 0:11:11.640 --> 0:11:14.760 of Michigan State University who laid out four broad categories 0:11:14.760 --> 0:11:20.760 of artificially intelligent machines. And these would go well into 0:11:21.120 --> 0:11:25.040 pretty sophisticated artificial intelligence. From the very get go um 0:11:25.240 --> 0:11:29.480 I argue that artificial intelligence is composed of lots of 0:11:29.520 --> 0:11:34.800 different facets, and you can find elements of artificial intelligence 0:11:34.800 --> 0:11:38.320 and many different programs that exist today, none of which 0:11:39.320 --> 0:11:44.239 are approaching this general intelligence model, and certainly not approaching 0:11:44.280 --> 0:11:49.640 strong AI. But here's how Hence breaks it down. He says, 0:11:49.640 --> 0:11:53.959 Type one is a reactive kind of intelligence. These are 0:11:54.000 --> 0:11:59.800 machines that take action in response to some state, and 0:12:00.040 --> 0:12:03.080 they don't form memories, they don't use past experience to 0:12:03.160 --> 0:12:08.000 inform current decisions. And he argues Deep Blue IBM's Deep 0:12:08.000 --> 0:12:12.439 Blue was that kind of a machine, and Deep Blue 0:12:12.880 --> 0:12:17.200 was the computer that defeated Gary Kasparov in a series 0:12:17.400 --> 0:12:20.800 of chess matches in the nineties. Actually, the first series 0:12:21.000 --> 0:12:23.880 ended in a draw between the two. The second series 0:12:23.920 --> 0:12:28.320 Deep Blue one, and then IBM quickly um ended up 0:12:29.360 --> 0:12:34.200 retiring Deep Blue from that point forward. But Deep Blue 0:12:34.360 --> 0:12:37.360 would just look at the state of the chessboard at 0:12:37.400 --> 0:12:40.280 any given moment and then make a decision based upon 0:12:40.280 --> 0:12:43.200 that state. When it was Deep Blues turn, it didn't 0:12:43.280 --> 0:12:46.760 build up a series of decisions, didn't track what was 0:12:46.800 --> 0:12:52.840 happening turn overturn. Uh, So it didn't evolve in any way. 0:12:52.840 --> 0:12:56.000 It had no internal representation of the world. It just 0:12:56.080 --> 0:12:59.199 would look at what was happening right now and make 0:12:59.240 --> 0:13:03.000 a decision. Now, there was a researcher named Rodney Brooks 0:13:03.080 --> 0:13:05.400 at AI researcher who said, these type one machines are 0:13:05.440 --> 0:13:09.240 the only ones we should ever try to make, because 0:13:09.920 --> 0:13:13.720 to make a machine more intelligent, one that contains a 0:13:13.840 --> 0:13:19.080 virtual representation of the world, would be impossible. That we 0:13:19.440 --> 0:13:23.840 as humans would be incapable of building a virtual representation 0:13:24.559 --> 0:13:27.840 that is accurate enough for such a machine to make 0:13:28.080 --> 0:13:32.240 good decisions. It would have a faulty representation of the world, 0:13:32.320 --> 0:13:35.520 and therefore any decision it would make would not be 0:13:36.040 --> 0:13:41.640 ideal and it would potentially do more harm than good. Uh. 0:13:41.880 --> 0:13:43.960 These sorts of machines are always going to make the 0:13:44.040 --> 0:13:48.000 same decision given a certain set of criteria. So let's 0:13:48.000 --> 0:13:51.000 go back with Deep Blue. Let's say the Deep Blue 0:13:51.040 --> 0:13:53.040 is looking at the board, the chess board. It is 0:13:53.080 --> 0:13:55.920 Deep Blues turn. It's been maybe you know, a half 0:13:55.960 --> 0:13:59.000 dozen turns in the chess game, and it makes a 0:13:59.040 --> 0:14:02.400 decision base upon all the positions of the pieces that 0:14:02.440 --> 0:14:06.640 are in play at that time. This decision is based 0:14:06.720 --> 0:14:12.120 on the probabilities of other moves that the opponent might make, 0:14:12.480 --> 0:14:15.319 the strength of any given move against what the current 0:14:15.360 --> 0:14:17.559 conditions are. There are a lot of factors that go 0:14:17.640 --> 0:14:22.240 into that one decision, But the argument goes that type 0:14:22.240 --> 0:14:26.360 one machines will always come to the same conclusion given 0:14:26.400 --> 0:14:29.160 that same set of criteria. So, in other words, if 0:14:29.200 --> 0:14:34.680 in game one, Deep Blue is given this arrangement of pieces, 0:14:35.360 --> 0:14:39.440 it will make a specific decision by weighing all those 0:14:39.720 --> 0:14:43.600 options out and going with the best one. In game two, 0:14:44.040 --> 0:14:48.280 if that exact same configuration of pieces were to be 0:14:48.400 --> 0:14:51.480 presented to Deep Blue and it's Deep Blues turn, it 0:14:51.520 --> 0:14:54.160 would make the same decision. In that case, It's not 0:14:54.200 --> 0:14:56.920 gonna improvise, it's not going to change, it's not going 0:14:57.000 --> 0:15:01.200 to learn from its past experience, is it is just 0:15:01.360 --> 0:15:04.840 going to make a decision based upon the parameters that 0:15:04.920 --> 0:15:08.600 are in front of it at that very moment. Now, 0:15:08.680 --> 0:15:12.120 this is not necessarily a bad thing. You may want 0:15:12.200 --> 0:15:17.560 certain artificially intelligent machines to be very predictable in their response. 0:15:18.040 --> 0:15:20.800 If you have a smart thermostat and it's been trained 0:15:21.200 --> 0:15:24.680 to learn what you prefer over time, and it's learned 0:15:24.680 --> 0:15:27.280 that you like it cool in the mornings, so you 0:15:27.360 --> 0:15:31.080 like a nice maybe seventy degrees for your thermostat fahrenheit, 0:15:31.240 --> 0:15:34.600 that is obviously, Uh, it's not going to improvise one 0:15:34.640 --> 0:15:36.640 day and say, you know what, I'm just gonna try 0:15:36.720 --> 0:15:38.840 something here. I bet he's really gonna like it. If 0:15:38.840 --> 0:15:41.880 I said eighty eight degrees fahrenheit on a humid Georgia 0:15:41.960 --> 0:15:46.160 July morning, Let's see what happens next. What happens next 0:15:46.240 --> 0:15:51.640 is Jonathan just sweats like crazy. So there are implementations 0:15:51.640 --> 0:15:55.040 where you would want to type one AI machine and 0:15:55.120 --> 0:15:59.640 nothing more advanced than that. Then Type two intelligent machines 0:15:59.680 --> 0:16:02.520 would be ones that have at least some version of 0:16:02.520 --> 0:16:06.680 a memory. They can track changing variables over time and 0:16:06.720 --> 0:16:11.040 analyze past behavior before making a decision on a course 0:16:11.160 --> 0:16:15.200 of action. Now, this memory doesn't go into a lexicon 0:16:15.400 --> 0:16:18.320 of memories. It's like short term memory that never gets 0:16:18.360 --> 0:16:22.040 converted into long term memory. It's stored temporarily, kind of 0:16:22.040 --> 0:16:25.480 like random axis memory is in computers, and then after 0:16:25.520 --> 0:16:30.120 that short while it can be overwritten. These machines would 0:16:30.120 --> 0:16:32.200 be able to do a little bit more than what 0:16:32.280 --> 0:16:34.520 Deep Blue could do. It wouldn't just be making a 0:16:34.520 --> 0:16:37.880 decision based upon the current state of the game. It 0:16:37.880 --> 0:16:41.600 would also remember how the last few moves went and 0:16:41.680 --> 0:16:44.120 what brought it there, and how the other player had 0:16:44.160 --> 0:16:47.600 been playing, and might be able to take that information 0:16:47.920 --> 0:16:52.200 and incorporate that in its decisions for the following moves, 0:16:52.960 --> 0:16:57.080 which means it could potentially adapt its play style. So 0:16:57.640 --> 0:17:02.400 this is a slightly more sophisticated version of the intelligent machine. 0:17:03.680 --> 0:17:07.120 Now we've got two more types of intelligent machines to cover. 0:17:07.800 --> 0:17:10.480 But before I jump into type three in type four 0:17:10.640 --> 0:17:14.320 and then go further into this discussion about artificial intelligence, 0:17:14.840 --> 0:17:24.840 let's take a quick break to thank our sponsor. Now, 0:17:24.880 --> 0:17:29.240 a type three intelligent machine incorporates what he says is 0:17:29.280 --> 0:17:32.600 a theory of mind. These machines would have an internal 0:17:32.680 --> 0:17:35.760 concept of the world as well as the beings that 0:17:35.840 --> 0:17:40.000 actually inhabit that world, and an understanding that those beings 0:17:40.280 --> 0:17:45.560 also possess intelligence that guide their behaviors. In a way, 0:17:45.600 --> 0:17:48.280 you could think of this as awareness of others. So 0:17:48.960 --> 0:17:54.639 these machines would know that people aren't just bags of 0:17:54.680 --> 0:17:58.040 meat that do things, that we have intelligence, and that 0:17:58.040 --> 0:18:03.080 that in fact guides our behavior and understanding that then 0:18:03.119 --> 0:18:07.720 affects the decisions that the AI makes. But this is 0:18:07.720 --> 0:18:11.119 still not quite at that level of strong AI that 0:18:11.160 --> 0:18:14.080 I was mentioning earlier. To get there, you have to 0:18:14.160 --> 0:18:18.000 hit type four machine intelligence, and that is when you 0:18:18.080 --> 0:18:21.000 hit self awareness, where the machine is not just aware 0:18:21.280 --> 0:18:26.879 that other beings possess the quality of intelligence, but it 0:18:27.000 --> 0:18:30.439 is aware of its own self and its own state 0:18:30.800 --> 0:18:37.080 and its own being in relation with everything else. It 0:18:37.240 --> 0:18:41.399 is this sort of machine that could, in theory, start 0:18:41.480 --> 0:18:44.800 to design improvements to itself so it could be recursive 0:18:44.880 --> 0:18:50.439 and that it is able to make improvements. And then 0:18:50.560 --> 0:18:54.000 you get into the situation that some futurists think of 0:18:54.040 --> 0:18:59.200 as the singularity, where you have self improving, artificially intelligent 0:18:59.280 --> 0:19:04.680 machines that are able to evolve. It's such a remarkable 0:19:04.760 --> 0:19:10.679 rate where every generation of improvements is a huge leap 0:19:10.720 --> 0:19:13.320 from the last one, and it's taking less and less 0:19:13.359 --> 0:19:16.959 time between generations for things to change that it becomes 0:19:16.960 --> 0:19:20.800 impossible to describe what the present set of circumstances are 0:19:21.080 --> 0:19:24.199 because the present would be changing so quickly that it 0:19:24.280 --> 0:19:29.679 becomes a meaningless concept. This is the singularity. That is 0:19:29.800 --> 0:19:35.760 one potential outcome of this instance, if it were in 0:19:35.800 --> 0:19:38.840 fact possible, which we don't know if it is possible yet. 0:19:39.560 --> 0:19:42.159 It's some people treated as a foregone conclusion that we 0:19:42.200 --> 0:19:45.760 will eventually have machines that will be able to attain 0:19:45.840 --> 0:19:51.120 self awareness and potentially self improvement. And once you get 0:19:51.119 --> 0:19:56.639 to that stage, how do you avoid this singularity? People 0:19:56.640 --> 0:19:59.159 would argue that's impossible to avoid, But there are a 0:19:59.200 --> 0:20:01.840 lot of people who say, we have no reason to 0:20:01.880 --> 0:20:04.199 believe that this is something that is going to happen, 0:20:05.040 --> 0:20:09.199 or that's even possible from a technological perspective, or maybe 0:20:09.240 --> 0:20:13.360 possible from a technological perspective, But we're talking decades, if 0:20:13.400 --> 0:20:16.160 not a century or more out in front of us, 0:20:16.560 --> 0:20:20.840 based upon our limited understanding of intelligence and our limited 0:20:20.840 --> 0:20:23.359 amount of processing power. When you compare it to something 0:20:23.400 --> 0:20:26.440 like the human mind, keeping in mind that the human 0:20:26.480 --> 0:20:30.800 brain has got billions of neurons in it, and we 0:20:30.920 --> 0:20:37.240 have artificial neuron networks, but they are dwarfed by the 0:20:37.280 --> 0:20:41.680 connections that you find in the human brain. So there's 0:20:41.720 --> 0:20:46.320 a lot of heated debate in the artificial intelligent world 0:20:46.880 --> 0:20:51.080 about whether or not this is something we should even 0:20:51.119 --> 0:20:55.840 concern ourselves with. But some people say that self awareness 0:20:55.920 --> 0:20:59.520 could arise from a system that has a given amount 0:20:59.520 --> 0:21:03.679 of complex city without us having any deeper understanding of 0:21:03.720 --> 0:21:07.119 what consciousness is. In other words, if you were to 0:21:07.160 --> 0:21:12.440 make a machine that was complex enough, consciousness could be 0:21:12.480 --> 0:21:18.000 an emergent behavior, something that naturally occurs once you reach 0:21:18.200 --> 0:21:23.199 a system of significant complexity. That's a little difficult to 0:21:23.200 --> 0:21:26.400 wrap your head around, but keep in mind we humans 0:21:26.440 --> 0:21:29.639 have been harnessing and creating stuff without having a full 0:21:29.800 --> 0:21:35.200 understanding of it. For ages. We were using electricity well 0:21:35.280 --> 0:21:39.920 before we understood the actual physics of electricity, So it's 0:21:39.960 --> 0:21:42.280 a little different. I mean, you can't compare electricity to 0:21:42.320 --> 0:21:46.240 consciousness directly, it doesn't make any sense. But just to 0:21:46.280 --> 0:21:50.280 say there is precedence in human beings creating something that 0:21:50.320 --> 0:21:55.040 they do not fully understand whether or not that's actually possible, However, 0:21:55.440 --> 0:21:58.359 is it we don't know. There's no way for us 0:21:58.400 --> 0:22:04.400 to answer. There is no construct, no machine complex enough 0:22:04.920 --> 0:22:08.120 for us to run an experiment and see if consciousness arises. 0:22:08.680 --> 0:22:11.399 And if it does arise, how do we recognize it? 0:22:12.119 --> 0:22:17.280 How do we know that a machine actually possesses that 0:22:17.280 --> 0:22:20.040 that feature? How do we know a machine truly has 0:22:20.119 --> 0:22:24.480 become self aware and conscious? We'll talk about that a 0:22:24.520 --> 0:22:27.119 little bit more later on, because, of course, there are 0:22:27.160 --> 0:22:29.200 a lot of people who have come up with ideas 0:22:29.720 --> 0:22:32.639 on how we would judge whether or not a machine 0:22:32.680 --> 0:22:36.000 had achieved consciousness, and some of them are more serious 0:22:36.000 --> 0:22:41.840 than others. There's also some serious objections. Well, those are 0:22:41.920 --> 0:22:45.119 the four types that were laid out by this one 0:22:45.600 --> 0:22:50.520 person from Michigan State University. But then, uh, John Spacey 0:22:50.680 --> 0:22:53.879 over on Simplicable has an article where he talks about 0:22:53.920 --> 0:22:59.679 thirty three types of artificial intelligence thirty three. Now, I'm 0:22:59.720 --> 0:23:01.959 not going to go through all thirty three, and I 0:23:02.040 --> 0:23:04.440 also want to point out that the thirty three types 0:23:04.480 --> 0:23:07.240 he points out are really more like thirty three facets 0:23:07.280 --> 0:23:11.360 of artificial intelligence. It's not thirty three degrees of artificial 0:23:11.400 --> 0:23:14.880 intelligence where we start with dumb machine and we end 0:23:14.920 --> 0:23:21.520 with super smart robo master. It's more like different aspects 0:23:21.560 --> 0:23:26.080 of intelligence that are in various stages of development and 0:23:26.160 --> 0:23:32.960 research in the artificial intelligence field. So it's not really 0:23:33.000 --> 0:23:37.280 separate categories. It's more like specific implementations of intelligence. An 0:23:37.320 --> 0:23:42.320 example would be effective computing, effective being a F F E, C, T, 0:23:42.720 --> 0:23:49.160 I V E as into affect something. Effective computing tries 0:23:49.240 --> 0:23:52.800 to suss out the emotions people are experiencing and to 0:23:52.880 --> 0:23:57.560 behave appropriately according to the parameters of the programming, not 0:23:57.600 --> 0:24:02.560 according to social rules necessarily, So these are machines that 0:24:02.560 --> 0:24:06.040 would be able to recognize emotions and respond in a 0:24:06.080 --> 0:24:11.360 way that was appropriate compared to their actual programming. Another 0:24:12.720 --> 0:24:16.679 type that Space lists is computer vision, which is the 0:24:16.720 --> 0:24:20.359 area of computer science focused on analyzing and understanding visual 0:24:20.400 --> 0:24:24.359 information computationally. So I've talked about this in episodes of 0:24:24.400 --> 0:24:27.520 tech Stuff in the past. You've heard about various projects 0:24:27.560 --> 0:24:31.240 to train computers to recognize images like pictures of cats 0:24:31.840 --> 0:24:36.440 versus other stuff like pictures of Deloreans. So this, as 0:24:36.480 --> 0:24:39.600 it turns out, it's really hard to do. This is 0:24:39.640 --> 0:24:42.600 one of those gaps we see between human intelligence and 0:24:42.640 --> 0:24:48.360 machine intelligence. Even with deep learning and artificial neural networks. 0:24:49.440 --> 0:24:53.320 This is really tricky stuff. It doesn't take a long 0:24:53.400 --> 0:24:57.800 time to teach your typical human concepts that they can 0:24:57.840 --> 0:25:04.040 then apply across a broad spectrum of examples. So I 0:25:04.200 --> 0:25:07.320 like to use the example specific example of a coffee mug, 0:25:07.440 --> 0:25:13.080 or just a mug a mug. So imagine a mug. Now, 0:25:13.119 --> 0:25:15.800 the mug you are imagining probably looks a certain way. 0:25:16.040 --> 0:25:18.359 But if I were to show you a totally different mug, 0:25:18.440 --> 0:25:20.679 you would recognize that as a mug, even if it 0:25:20.760 --> 0:25:25.080 was a different size, different color, different shape of the handle, 0:25:25.600 --> 0:25:29.040 different shape of the container itself, As long as it 0:25:29.280 --> 0:25:34.919 adhered to the general parameters that we associate with the 0:25:35.040 --> 0:25:38.679 concept of a mug, you know what I was talking about, 0:25:39.119 --> 0:25:43.120 You would know that that was a mug. Computers not 0:25:43.240 --> 0:25:47.080 that good at this, right, Like it might require you 0:25:47.160 --> 0:25:51.320 to train a computer by showing it thousands or tens 0:25:51.359 --> 0:25:55.080 of thousands of images of mugs so that it builds 0:25:55.200 --> 0:26:01.800 up the various elements that define mug nous, so that 0:26:01.880 --> 0:26:04.200 if it were to look at a brand new image 0:26:04.680 --> 0:26:07.919 of a mug that is unlike all the others that 0:26:08.080 --> 0:26:11.520 have preceded it, it would still be able to identify 0:26:11.560 --> 0:26:15.680 that as Yes, that is a mug. This is hard 0:26:15.720 --> 0:26:20.679 to do. We've seen some advances in this field, but 0:26:21.280 --> 0:26:27.159 it demonstrates the huge gap in that specific part of 0:26:27.200 --> 0:26:31.879 intelligence between machines and humans. That isn't to say that 0:26:31.920 --> 0:26:36.280 it's not getting better with machines. It is, but that's 0:26:36.320 --> 0:26:39.280 just one example that I wanted to give, and it 0:26:39.520 --> 0:26:43.960 kind of hammers home this idea that general intelligence with 0:26:44.080 --> 0:26:48.000 machines is a long way off. There's just so many 0:26:48.000 --> 0:26:55.600 different aspects of it. Um Beyond this Spacey continues to 0:26:55.680 --> 0:26:59.200 define various terms within AI, and I don't necessarily think 0:26:59.200 --> 0:27:02.160 of them as types of artificial intelligence, but again aspects 0:27:02.359 --> 0:27:06.680 of artificial intelligence. Not every AI implementation will need all 0:27:06.760 --> 0:27:09.600 of these aspects. Some of them are going to be 0:27:09.720 --> 0:27:15.200 much better with a very narrow range of artificial intelligent features. 0:27:15.920 --> 0:27:18.560 For example, your rumba probably does not need to be 0:27:18.600 --> 0:27:20.720 able to pick up on what your mood is, whether 0:27:20.800 --> 0:27:24.040 or not you're sad, or anything along those lines. But 0:27:24.080 --> 0:27:26.399 if you are talking about a machine that needs to 0:27:26.400 --> 0:27:30.680