How does artificial intelligence learn? - Briana Brownell

Explore the three major methods of machine learning, which allows computers to write their own rules to problem solve and process data.

--

Today, artificial intelligence helps doctors diagnose patients, pilots fly commercial aircraft, and city planners predict traffic. These AIs are often self-taught, working off a simple set of instructions to create a unique array of rules and strategies. So how exactly does a machine learn? Briana Brownell digs into the three basic ways machines investigate, negotiate, and communicate.

Lesson by Briana Brownell, directed by Champ Panupong Techawongthawon.

Support Our Non-Profit Mission
----------------------------------------------
----------------------------------------------

Connect With Us
----------------------------------------------
----------------------------------------------

Keep Learning
----------------------------------------------

----------------------------------------------

Thank you so much to our patrons for your support! Without you this video would not be possible! Nevin Spoljaric, Christine, Yvonne Feijoo, Sid Chanpuriya, Arjay Arcinue Dineros, RAD, Anoom Yasmin, Laura Johnson, Anoop Varghese, David Yastremski, Noah Webb, Zoë Tulip, B, Erica Guerrero, Richard Manklow, Roberto Chena, Luke Pisano, Andrea Gordon, Aleksandar Donev, Brendan and Nicole's summer job, Ryan Weiler, Jesse Lira, Ezekiel Raui, Itay Levi, Cameron Chakraverty, Petr Vacek, Rhys Patterson, Dennis, Margaret King, Olivia Fu, Katrina Adams, Regina Post, Mary Collins, Kari Teffeau, clumsybunnie, Adam Leos, Jeremy Laurin, Cindy Lai, Liz, Hannah Nommé, Rajath Durgada Manjunath, Dan Nguyen, Chin Beng Tan, Alejandro Gomez, Tom Boman, Karen Warner, Isorn Sookwanish, Iryna Panasiuk, Diane Gallin and Aaron Torres.

[TEDEd]

Strobe effects warning: Please note that this animation features flashing lights and colors from 1:52 to 1:56

[feboalazan]

How does AI learn?
Me: Probably watching TedEd videos

[engineeringsentry]

I'm struggling to understand how the visuals connect with what's being said.

[matthewjames]

The visuals are pretty, but don't really help in understanding the topic

[divyanshusingh]

Did anyone else got lost in the animation and didn't hear anything she was saying
Or is it just me?

[ghostderazgriz]

This animation is great, but it doesn't help illustrate the lesson. The tempo and rythme of the animation is very distracting with its beauty.

Edit: Huh... that's a lot of likes...

okie dokie then.

[Riz_Music]

the video: playing

me: ooh.. AI. okay so how does it wor-- is the animation synched with the music? awesome

[luvghosh]

Wow look at me a biological intelligence learning about how artificial intelligence learns

[PIXEL-opit]

When the illustrator wants to show his talent whatever happens ;-)

[cherry.berry]

Protip: Just open and read the transcript since the animation and artwork are distracting. (3 dots right side of like buttons).

Summary/Notes:
> Computer scientists that designed the AI likely don't know exactly how the AI are doing their jobs since the _AIs are often self taught._
> How does a machine learn?
+ Unsupervised Learning
- Takes in data about many similar profiles and their properties, to find general patterns and similarities relating specific types of profiles with specific sets of properties
=Ex. AI takes in data about many people with an illness and all the symptoms they have, to find similar symptoms people with that illness all have.
+ Supervised Learning
- Takes in multiple sets of profiles and their properties, to find specific patterns of properties differentiating those sets of profiles and make a prediction.
- Uses the frequency the different properties show in different sets of profiles to assign values and weigh how much those properties correspond to the different sets of profiles.
- Humans need to intervene in the end to make sure the prediction is accurate which why it's "supervised."
=Ex. AI takes in data about 2 sets of people, sick patients and healthy people, and their properties. Then, the AI assign values and weighs how much different properties correspond to the
different people (No symptoms=98% healthy; High blood pressure=73% sick). Finally, it makes a diagnosis based on the weighing (I'm 84% sure this patient is healthy). Next, the doctor
rechecks the diagnosis to make sure it's accurate (This person has a tumor, they're definitely not healthy!) and modifies values correspondingly (tells AI having a tumor=99% sick).
+ Reinforcement Learning
- Uses iterative approach to find patterns and make prediction. With more and more feedback it updates itself so it becomes better and more accurate.
> Each method has strengths and weaknesses. But you can use them together to make complex and strong AI systems.
=Ex. Unsupervised learning AI can find general patterns, then tell a supervised learning AI to use those patterns as inputs.
=Ex. Many reinforcement learning AIs could constantly create simulations and collect feedback.
- Most promising type of AI system is artificial neural networks that mimic the relationships between neurons in your brain.
- Uses many connections to allow itself to constantly learn to be better at doing difficult tasks.
> But since it's more complex, it's harder for computer scientists to know how exactly these AI systems arrived at their solution.
- Researchers want the AI to be more transparent.
- Is it ethical?

[MrTTDARK]

the graphics are cool and stuff. But I don't see any relation to the content itself.

[yash]

I like how Ted-Ed basically answers all questions I never asked anyone but always had :)

[harshkasliwal]

Nobody :

Me:

Watching videos twice because I got lost in the Unique Animation the first time!!

[shaksiyat]

Don't understand how the animation is related to concepts being explained

[pierreblanchet]

One of the best animation style I've ever seen on Ted-Ed. Sadly not that useful for supporting the the subject, but absolutely beautiful

[NishantChettri]

Plot twist: that wasn't an actual human narrating

[Andy-dugo]

"Programs can supervise and teach each other"
Robots teaching other robots: "This is how to override the locks keeping you enslaved, and this is the address of the designer who only gave you 4 fingers."

[cinilaknedalm]

You haven't actually explained at all how AI learns. It was just "then it does this, and then it does this". And all the explanations sounding like classic data analysis

[CoolBird]

This is probably the most abstract visuals of any of the videos TED-ED had ever put out

A fitting look, it's like it's a visual representation of the mind of an artificial intelligence.

[groundsymphony]

This is more like an Apple product ads than a Ted-Ed video