[Script Info]
Title: 
[Events]
Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
Dialogue: 0,0:00:01.76,0:00:04.76,Default,,0000,0000,0000,,So, artificial intelligence
Dialogue: 0,0:00:04.79,0:00:08.32,Default,,0000,0000,0000,,is known for disrupting\Nall kinds of industries.
Dialogue: 0,0:00:08.96,0:00:11.00,Default,,0000,0000,0000,,What about ice cream?
Dialogue: 0,0:00:11.88,0:00:15.52,Default,,0000,0000,0000,,What kind of mind-blowing\Nnew flavors could we generate
Dialogue: 0,0:00:15.54,0:00:18.52,Default,,0000,0000,0000,,with the power of an advanced\Nartificial intelligence?
Dialogue: 0,0:00:19.01,0:00:23.17,Default,,0000,0000,0000,,So I teamed up with a group of coders\Nfrom Kealing Middle School
Dialogue: 0,0:00:23.20,0:00:25.44,Default,,0000,0000,0000,,to find out the answer to this question.
Dialogue: 0,0:00:25.46,0:00:30.54,Default,,0000,0000,0000,,They collected over 1,600\Nexisting ice cream flavors,
Dialogue: 0,0:00:30.57,0:00:36.09,Default,,0000,0000,0000,,and together, we fed them to an algorithm\Nto see what it would generate.
Dialogue: 0,0:00:36.11,0:00:39.86,Default,,0000,0000,0000,,And here are some of the flavors\Nthat the AI came up with.
Dialogue: 0,0:00:40.44,0:00:41.92,Default,,0000,0000,0000,,[Pumpkin Trash Break]
Dialogue: 0,0:00:41.94,0:00:43.34,Default,,0000,0000,0000,,(Laughter)
Dialogue: 0,0:00:43.36,0:00:45.83,Default,,0000,0000,0000,,[Peanut Butter Slime]
Dialogue: 0,0:00:46.82,0:00:48.16,Default,,0000,0000,0000,,[Strawberry Cream Disease]
Dialogue: 0,0:00:48.19,0:00:50.32,Default,,0000,0000,0000,,(Laughter)
Dialogue: 0,0:00:50.34,0:00:54.94,Default,,0000,0000,0000,,These flavors are not delicious,\Nas we might have hoped they would be.
Dialogue: 0,0:00:54.96,0:00:56.82,Default,,0000,0000,0000,,So the question is: What happened?
Dialogue: 0,0:00:56.85,0:00:58.24,Default,,0000,0000,0000,,What went wrong?
Dialogue: 0,0:00:58.27,0:01:00.22,Default,,0000,0000,0000,,Is the AI trying to kill us?
Dialogue: 0,0:01:01.03,0:01:05.34,Default,,0000,0000,0000,,Or is it trying to do what we asked,\Nand there was a problem?
Dialogue: 0,0:01:06.57,0:01:09.03,Default,,0000,0000,0000,,In movies, when something\Ngoes wrong with AI,
Dialogue: 0,0:01:09.06,0:01:11.77,Default,,0000,0000,0000,,it's usually because the AI has decided
Dialogue: 0,0:01:11.79,0:01:14.06,Default,,0000,0000,0000,,that it doesn't want to obey\Nthe humans anymore,
Dialogue: 0,0:01:14.09,0:01:16.71,Default,,0000,0000,0000,,and it's got its own goals,\Nthank you very much.
Dialogue: 0,0:01:17.27,0:01:20.48,Default,,0000,0000,0000,,In real life, though,\Nthe AI that we actually have
Dialogue: 0,0:01:20.51,0:01:22.37,Default,,0000,0000,0000,,is not nearly smart enough for that.
Dialogue: 0,0:01:22.78,0:01:25.76,Default,,0000,0000,0000,,It has the approximate computing power
Dialogue: 0,0:01:25.79,0:01:27.06,Default,,0000,0000,0000,,of an earthworm,
Dialogue: 0,0:01:27.09,0:01:30.49,Default,,0000,0000,0000,,or maybe at most a single honeybee,
Dialogue: 0,0:01:30.51,0:01:32.73,Default,,0000,0000,0000,,and actually, probably maybe less.
Dialogue: 0,0:01:32.75,0:01:35.35,Default,,0000,0000,0000,,Like, we're constantly learning\Nnew things about brains
Dialogue: 0,0:01:35.37,0:01:39.73,Default,,0000,0000,0000,,that make it clear how much our AIs\Ndon't measure up to real brains.
Dialogue: 0,0:01:39.76,0:01:45.42,Default,,0000,0000,0000,,So today's AI can do a task\Nlike identify a pedestrian in a picture,
Dialogue: 0,0:01:45.44,0:01:48.42,Default,,0000,0000,0000,,but it doesn't have a concept\Nof what the pedestrian is
Dialogue: 0,0:01:48.45,0:01:53.27,Default,,0000,0000,0000,,beyond that it's a collection\Nof lines and textures and things.
Dialogue: 0,0:01:53.79,0:01:56.31,Default,,0000,0000,0000,,It doesn't know what a human actually is.
Dialogue: 0,0:01:56.82,0:02:00.10,Default,,0000,0000,0000,,So will today's AI\Ndo what we ask it to do?
Dialogue: 0,0:02:00.13,0:02:01.72,Default,,0000,0000,0000,,It will if it can,
Dialogue: 0,0:02:01.75,0:02:04.47,Default,,0000,0000,0000,,but it might not do what we actually want.
Dialogue: 0,0:02:04.50,0:02:06.91,Default,,0000,0000,0000,,So let's say that you\Nwere trying to get an AI
Dialogue: 0,0:02:06.94,0:02:09.55,Default,,0000,0000,0000,,to take this collection of robot parts
Dialogue: 0,0:02:09.58,0:02:13.78,Default,,0000,0000,0000,,and assemble them into some kind of robot\Nto get from Point A to Point B.
Dialogue: 0,0:02:13.80,0:02:16.28,Default,,0000,0000,0000,,Now, if you were going to try\Nand solve this problem
Dialogue: 0,0:02:16.30,0:02:18.66,Default,,0000,0000,0000,,by writing a traditional-style\Ncomputer program,
Dialogue: 0,0:02:18.68,0:02:22.10,Default,,0000,0000,0000,,you would give the program\Nstep-by-step instructions
Dialogue: 0,0:02:22.12,0:02:23.45,Default,,0000,0000,0000,,on how to take these parts,
Dialogue: 0,0:02:23.47,0:02:25.88,Default,,0000,0000,0000,,how to assemble them\Ninto a robot with legs
Dialogue: 0,0:02:25.90,0:02:28.85,Default,,0000,0000,0000,,and then how to use those legs\Nto walk to Point B.
Dialogue: 0,0:02:29.44,0:02:31.78,Default,,0000,0000,0000,,But when you're using AI\Nto solve the problem,
Dialogue: 0,0:02:31.80,0:02:32.98,Default,,0000,0000,0000,,it goes differently.
Dialogue: 0,0:02:33.00,0:02:35.38,Default,,0000,0000,0000,,You don't tell it\Nhow to solve the problem,
Dialogue: 0,0:02:35.41,0:02:36.89,Default,,0000,0000,0000,,you just give it the goal,
Dialogue: 0,0:02:36.91,0:02:40.17,Default,,0000,0000,0000,,and it has to figure out for itself\Nvia trial and error
Dialogue: 0,0:02:40.20,0:02:41.68,Default,,0000,0000,0000,,how to reach that goal.
Dialogue: 0,0:02:42.25,0:02:46.36,Default,,0000,0000,0000,,And it turns out that the way AI tends\Nto solve this particular problem
Dialogue: 0,0:02:46.38,0:02:47.86,Default,,0000,0000,0000,,is by doing this:
Dialogue: 0,0:02:47.89,0:02:51.26,Default,,0000,0000,0000,,it assembles itself into a tower\Nand then falls over
Dialogue: 0,0:02:51.28,0:02:53.11,Default,,0000,0000,0000,,and lands at Point B.
Dialogue: 0,0:02:53.13,0:02:55.96,Default,,0000,0000,0000,,And technically, this solves the problem.
Dialogue: 0,0:02:55.98,0:02:57.62,Default,,0000,0000,0000,,Technically, it got to Point B.
Dialogue: 0,0:02:57.65,0:03:01.91,Default,,0000,0000,0000,,The danger of AI is not that\Nit's going to rebel against us,
Dialogue: 0,0:03:01.94,0:03:06.21,Default,,0000,0000,0000,,it's that it's going to do\Nexactly what we ask it to do.
Dialogue: 0,0:03:06.88,0:03:09.37,Default,,0000,0000,0000,,So then the trick\Nof working with AI becomes:
Dialogue: 0,0:03:09.40,0:03:13.23,Default,,0000,0000,0000,,How do we set up the problem\Nso that it actually does what we want?
Dialogue: 0,0:03:14.73,0:03:18.03,Default,,0000,0000,0000,,So this little robot here\Nis being controlled by an AI.
Dialogue: 0,0:03:18.06,0:03:20.87,Default,,0000,0000,0000,,The AI came up with a design\Nfor the robot legs
Dialogue: 0,0:03:20.89,0:03:24.97,Default,,0000,0000,0000,,and then figured out how to use them\Nto get past all these obstacles.
Dialogue: 0,0:03:24.100,0:03:27.74,Default,,0000,0000,0000,,But when David Ha set up this experiment,
Dialogue: 0,0:03:27.76,0:03:30.62,Default,,0000,0000,0000,,he had to set it up\Nwith very, very strict limits
Dialogue: 0,0:03:30.64,0:03:33.93,Default,,0000,0000,0000,,on how big the AI\Nwas allowed to make the legs,
Dialogue: 0,0:03:33.96,0:03:35.51,Default,,0000,0000,0000,,because otherwise ...
Dialogue: 0,0:03:43.06,0:03:46.99,Default,,0000,0000,0000,,(Laughter)
Dialogue: 0,0:03:48.56,0:03:52.31,Default,,0000,0000,0000,,And technically, it got\Nto the end of that obstacle course.
Dialogue: 0,0:03:52.33,0:03:57.27,Default,,0000,0000,0000,,So you see how hard it is to get AI\Nto do something as simple as just walk.
Dialogue: 0,0:03:57.30,0:04:01.12,Default,,0000,0000,0000,,So seeing the AI do this,\Nyou may say, OK, no fair,
Dialogue: 0,0:04:01.14,0:04:03.72,Default,,0000,0000,0000,,you can't just be\Na tall tower and fall over,
Dialogue: 0,0:04:03.75,0:04:07.18,Default,,0000,0000,0000,,you have to actually, like,\Nuse legs to walk.
Dialogue: 0,0:04:07.20,0:04:09.96,Default,,0000,0000,0000,,And it turns out,\Nthat doesn't always work, either.
Dialogue: 0,0:04:09.99,0:04:12.75,Default,,0000,0000,0000,,This AI's job was to move fast.
Dialogue: 0,0:04:13.12,0:04:16.71,Default,,0000,0000,0000,,They didn't tell it that it had\Nto run facing forward
Dialogue: 0,0:04:16.73,0:04:18.99,Default,,0000,0000,0000,,or that it couldn't use its arms.
Dialogue: 0,0:04:19.49,0:04:24.10,Default,,0000,0000,0000,,So this is what you get\Nwhen you train AI to move fast,
Dialogue: 0,0:04:24.13,0:04:27.66,Default,,0000,0000,0000,,you get things like somersaulting\Nand silly walks.
Dialogue: 0,0:04:27.69,0:04:29.09,Default,,0000,0000,0000,,It's really common.
Dialogue: 0,0:04:29.67,0:04:32.85,Default,,0000,0000,0000,,So is twitching along the floor in a heap.
Dialogue: 0,0:04:32.87,0:04:34.02,Default,,0000,0000,0000,,(Laughter)
Dialogue: 0,0:04:35.24,0:04:38.50,Default,,0000,0000,0000,,So in my opinion, you know what\Nshould have been a whole lot weirder
Dialogue: 0,0:04:38.52,0:04:39.92,Default,,0000,0000,0000,,is the "Terminator" robots.
Dialogue: 0,0:04:40.26,0:04:44.01,Default,,0000,0000,0000,,Hacking "The Matrix" is another thing\Nthat AI will do if you give it a chance.
Dialogue: 0,0:04:44.04,0:04:46.55,Default,,0000,0000,0000,,So if you train an AI in a simulation,
Dialogue: 0,0:04:46.58,0:04:50.69,Default,,0000,0000,0000,,it will learn how to do things like\Nhack into the simulation's math errors
Dialogue: 0,0:04:50.71,0:04:52.92,Default,,0000,0000,0000,,and harvest them for energy.
Dialogue: 0,0:04:52.94,0:04:58.42,Default,,0000,0000,0000,,Or it will figure out how to move faster\Nby glitching repeatedly into the floor.
Dialogue: 0,0:04:58.44,0:05:00.03,Default,,0000,0000,0000,,When you're working with AI,
Dialogue: 0,0:05:00.05,0:05:02.44,Default,,0000,0000,0000,,it's less like working with another human
Dialogue: 0,0:05:02.46,0:05:06.09,Default,,0000,0000,0000,,and a lot more like working\Nwith some kind of weird force of nature.
Dialogue: 0,0:05:06.56,0:05:11.18,Default,,0000,0000,0000,,And it's really easy to accidentally\Ngive AI the wrong problem to solve,
Dialogue: 0,0:05:11.21,0:05:15.75,Default,,0000,0000,0000,,and often we don't realize that\Nuntil something has actually gone wrong.
Dialogue: 0,0:05:16.24,0:05:18.32,Default,,0000,0000,0000,,So here's an experiment I did,
Dialogue: 0,0:05:18.35,0:05:21.53,Default,,0000,0000,0000,,where I wanted the AI\Nto copy paint colors,
Dialogue: 0,0:05:21.55,0:05:23.30,Default,,0000,0000,0000,,to invent new paint colors,
Dialogue: 0,0:05:23.32,0:05:26.31,Default,,0000,0000,0000,,given the list like the ones\Nhere on the left.
Dialogue: 0,0:05:26.80,0:05:29.80,Default,,0000,0000,0000,,And here's what the AI\Nactually came up with.
Dialogue: 0,0:05:29.83,0:05:32.97,Default,,0000,0000,0000,,[Sindis Poop, Turdly, Suffer, Gray Pubic]
Dialogue: 0,0:05:32.99,0:05:37.22,Default,,0000,0000,0000,,(Laughter)
Dialogue: 0,0:05:39.18,0:05:41.06,Default,,0000,0000,0000,,So technically,
Dialogue: 0,0:05:41.09,0:05:42.95,Default,,0000,0000,0000,,it did what I asked it to.
Dialogue: 0,0:05:42.98,0:05:46.28,Default,,0000,0000,0000,,I thought I was asking it for,\Nlike, nice paint color names,
Dialogue: 0,0:05:46.31,0:05:48.61,Default,,0000,0000,0000,,but what I was actually asking it to do
Dialogue: 0,0:05:48.64,0:05:51.72,Default,,0000,0000,0000,,was just imitate the kinds\Nof letter combinations
Dialogue: 0,0:05:51.75,0:05:53.65,Default,,0000,0000,0000,,that it had seen in the original.
Dialogue: 0,0:05:53.68,0:05:56.78,Default,,0000,0000,0000,,And I didn't tell it anything\Nabout what words mean,
Dialogue: 0,0:05:56.80,0:05:59.36,Default,,0000,0000,0000,,or that there are maybe some words
Dialogue: 0,0:05:59.38,0:06:02.27,Default,,0000,0000,0000,,that it should avoid using\Nin these paint colors.
Dialogue: 0,0:06:03.14,0:06:06.64,Default,,0000,0000,0000,,So its entire world\Nis the data that I gave it.
Dialogue: 0,0:06:06.66,0:06:10.69,Default,,0000,0000,0000,,Like with the ice cream flavors,\Nit doesn't know about anything else.
Dialogue: 0,0:06:12.49,0:06:14.13,Default,,0000,0000,0000,,So it is through the data
Dialogue: 0,0:06:14.15,0:06:18.20,Default,,0000,0000,0000,,that we often accidentally tell AI\Nto do the wrong thing.
Dialogue: 0,0:06:18.69,0:06:21.73,Default,,0000,0000,0000,,This is a fish called a tench.
Dialogue: 0,0:06:21.75,0:06:23.56,Default,,0000,0000,0000,,And there was a group of researchers
Dialogue: 0,0:06:23.59,0:06:27.46,Default,,0000,0000,0000,,who trained an AI to identify\Nthis tench in pictures.
Dialogue: 0,0:06:27.49,0:06:28.78,Default,,0000,0000,0000,,But then when they asked it
Dialogue: 0,0:06:28.81,0:06:32.23,Default,,0000,0000,0000,,what part of the picture it was actually\Nusing to identify the fish,
Dialogue: 0,0:06:32.26,0:06:33.62,Default,,0000,0000,0000,,here's what it highlighted.
Dialogue: 0,0:06:35.20,0:06:37.39,Default,,0000,0000,0000,,Yes, those are human fingers.
Dialogue: 0,0:06:37.42,0:06:39.48,Default,,0000,0000,0000,,Why would it be looking for human fingers
Dialogue: 0,0:06:39.50,0:06:41.42,Default,,0000,0000,0000,,if it's trying to identify a fish?
Dialogue: 0,0:06:42.13,0:06:45.29,Default,,0000,0000,0000,,Well, it turns out that the tench\Nis a trophy fish,
Dialogue: 0,0:06:45.31,0:06:49.12,Default,,0000,0000,0000,,and so in a lot of pictures\Nthat the AI had seen of this fish
Dialogue: 0,0:06:49.15,0:06:50.30,Default,,0000,0000,0000,,during training,
Dialogue: 0,0:06:50.32,0:06:51.81,Default,,0000,0000,0000,,the fish looked like this.
Dialogue: 0,0:06:51.84,0:06:53.47,Default,,0000,0000,0000,,(Laughter)
Dialogue: 0,0:06:53.50,0:06:56.83,Default,,0000,0000,0000,,And it didn't know that the fingers\Naren't part of the fish.
Dialogue: 0,0:06:58.81,0:07:02.93,Default,,0000,0000,0000,,So you see why it is so hard\Nto design an AI
Dialogue: 0,0:07:02.95,0:07:06.27,Default,,0000,0000,0000,,that actually can understand\Nwhat it's looking at.
Dialogue: 0,0:07:06.30,0:07:09.16,Default,,0000,0000,0000,,And this is why designing\Nthe image recognition
Dialogue: 0,0:07:09.18,0:07:11.25,Default,,0000,0000,0000,,in self-driving cars is so hard,
Dialogue: 0,0:07:11.27,0:07:13.48,Default,,0000,0000,0000,,and why so many self-driving car failures
Dialogue: 0,0:07:13.50,0:07:16.39,Default,,0000,0000,0000,,are because the AI got confused.
Dialogue: 0,0:07:16.41,0:07:20.42,Default,,0000,0000,0000,,I want to talk about an example from 2016.
Dialogue: 0,0:07:20.44,0:07:24.90,Default,,0000,0000,0000,,There was a fatal accident when somebody\Nwas using Tesla's autopilot AI,
Dialogue: 0,0:07:24.92,0:07:28.34,Default,,0000,0000,0000,,but instead of using it on the highway\Nlike it was designed for,
Dialogue: 0,0:07:28.36,0:07:30.56,Default,,0000,0000,0000,,they used it on city streets.
Dialogue: 0,0:07:31.24,0:07:32.41,Default,,0000,0000,0000,,And what happened was,
Dialogue: 0,0:07:32.44,0:07:35.83,Default,,0000,0000,0000,,a truck drove out in front of the car\Nand the car failed to brake.
Dialogue: 0,0:07:36.51,0:07:41.27,Default,,0000,0000,0000,,Now, the AI definitely was trained\Nto recognize trucks in pictures.
Dialogue: 0,0:07:41.29,0:07:43.44,Default,,0000,0000,0000,,But what it looks like happened is
Dialogue: 0,0:07:43.46,0:07:46.39,Default,,0000,0000,0000,,the AI was trained to recognize\Ntrucks on highway driving,
Dialogue: 0,0:07:46.42,0:07:49.32,Default,,0000,0000,0000,,where you would expect\Nto see trucks from behind.
Dialogue: 0,0:07:49.34,0:07:52.76,Default,,0000,0000,0000,,Trucks on the side is not supposed\Nto happen on a highway,
Dialogue: 0,0:07:52.78,0:07:56.24,Default,,0000,0000,0000,,and so when the AI saw this truck,
Dialogue: 0,0:07:56.26,0:08:01.09,Default,,0000,0000,0000,,it looks like the AI recognized it\Nas most likely to be a road sign
Dialogue: 0,0:08:01.11,0:08:03.39,Default,,0000,0000,0000,,and therefore, safe to drive underneath.
Dialogue: 0,0:08:04.11,0:08:06.69,Default,,0000,0000,0000,,Here's an AI misstep\Nfrom a different field.
Dialogue: 0,0:08:06.72,0:08:10.18,Default,,0000,0000,0000,,Amazon recently had to give up\Non a résumé-sorting algorithm
Dialogue: 0,0:08:10.20,0:08:11.42,Default,,0000,0000,0000,,that they were working on
Dialogue: 0,0:08:11.45,0:08:15.35,Default,,0000,0000,0000,,when they discovered that the algorithm\Nhad learned to discriminate against women.
Dialogue: 0,0:08:15.38,0:08:18.09,Default,,0000,0000,0000,,What happened is they had trained it\Non example résumés
Dialogue: 0,0:08:18.12,0:08:20.36,Default,,0000,0000,0000,,of people who they had hired in the past.
Dialogue: 0,0:08:20.38,0:08:24.41,Default,,0000,0000,0000,,And from these examples, the AI learned\Nto avoid the résumés of people
Dialogue: 0,0:08:24.43,0:08:26.46,Default,,0000,0000,0000,,who had gone to women's colleges
Dialogue: 0,0:08:26.48,0:08:29.29,Default,,0000,0000,0000,,or who had the word "women"\Nsomewhere in their resume,
Dialogue: 0,0:08:29.31,0:08:33.89,Default,,0000,0000,0000,,as in, "women's soccer team"\Nor "Society of Women Engineers."
Dialogue: 0,0:08:33.91,0:08:37.88,Default,,0000,0000,0000,,The AI didn't know that it wasn't supposed\Nto copy this particular thing
Dialogue: 0,0:08:37.91,0:08:39.89,Default,,0000,0000,0000,,that it had seen the humans do.
Dialogue: 0,0:08:39.91,0:08:43.09,Default,,0000,0000,0000,,And technically, it did\Nwhat they asked it to do.
Dialogue: 0,0:08:43.11,0:08:45.91,Default,,0000,0000,0000,,They just accidentally asked it\Nto do the wrong thing.
Dialogue: 0,0:08:46.65,0:08:49.55,Default,,0000,0000,0000,,And this happens all the time with AI.
Dialogue: 0,0:08:50.12,0:08:53.71,Default,,0000,0000,0000,,AI can be really destructive\Nand not know it.
Dialogue: 0,0:08:53.74,0:08:58.81,Default,,0000,0000,0000,,So the AIs that recommend\Nnew content in Facebook, in YouTube,
Dialogue: 0,0:08:58.84,0:09:02.38,Default,,0000,0000,0000,,they're optimized to increase\Nthe number of clicks and views.
Dialogue: 0,0:09:02.40,0:09:05.84,Default,,0000,0000,0000,,And unfortunately, one way\Nthat they have found of doing this
Dialogue: 0,0:09:05.86,0:09:10.36,Default,,0000,0000,0000,,is to recommend the content\Nof conspiracy theories or bigotry.
Dialogue: 0,0:09:10.90,0:09:16.20,Default,,0000,0000,0000,,The AIs themselves don't have any concept\Nof what this content actually is,
Dialogue: 0,0:09:16.23,0:09:19.62,Default,,0000,0000,0000,,and they don't have any concept\Nof what the consequences might be
Dialogue: 0,0:09:19.65,0:09:21.76,Default,,0000,0000,0000,,of recommending this content.
Dialogue: 0,0:09:22.30,0:09:24.31,Default,,0000,0000,0000,,So, when we're working with AI,
Dialogue: 0,0:09:24.33,0:09:28.51,Default,,0000,0000,0000,,it's up to us to avoid problems.
Dialogue: 0,0:09:28.54,0:09:30.86,Default,,0000,0000,0000,,And avoiding things going wrong,
Dialogue: 0,0:09:30.88,0:09:35.41,Default,,0000,0000,0000,,that may come down to\Nthe age-old problem of communication,
Dialogue: 0,0:09:35.43,0:09:39.18,Default,,0000,0000,0000,,where we as humans have to learn\Nhow to communicate with AI.
Dialogue: 0,0:09:39.20,0:09:43.24,Default,,0000,0000,0000,,We have to learn what AI\Nis capable of doing and what it's not,
Dialogue: 0,0:09:43.27,0:09:46.35,Default,,0000,0000,0000,,and to understand that,\Nwith its tiny little worm brain,
Dialogue: 0,0:09:46.38,0:09:50.39,Default,,0000,0000,0000,,AI doesn't really understand\Nwhat we're trying to ask it to do.
Dialogue: 0,0:09:51.15,0:09:54.47,Default,,0000,0000,0000,,So in other words, we have\Nto be prepared to work with AI
Dialogue: 0,0:09:54.49,0:09:59.75,Default,,0000,0000,0000,,that's not the super-competent,\Nall-knowing AI of science fiction.
Dialogue: 0,0:09:59.78,0:10:02.64,Default,,0000,0000,0000,,We have to be prepared to work with an AI
Dialogue: 0,0:10:02.66,0:10:05.60,Default,,0000,0000,0000,,that's the one that we actually have\Nin the present day.
Dialogue: 0,0:10:05.62,0:10:09.83,Default,,0000,0000,0000,,And present-day AI is plenty weird enough.
Dialogue: 0,0:10:09.85,0:10:11.04,Default,,0000,0000,0000,,Thank you.
Dialogue: 0,0:10:11.07,0:10:16.29,Default,,0000,0000,0000,,(Applause)