All right well welcome back thank you so
much to those of you who tuned into my last video it was just amazing to see
some of the reactions from all around the world. So the reactions
really broke into two parts the first was oh my god this is amazing look at
what this technology can do. And the other reaction was really: oh my
god, my job! What's going to happen? In this video we're going to take a look
at artificial intelligence and why you really don't have to worry about your
job... in the short term! And what some of the lackings of GPT3
are while it's amazing in terms of what it
can do there's a lot that it can't do i'm going to show you some more demos of
what gpt3 is able to do in just the last week and lastly, I'm
going to talk about why, while you don't have to worry about your jobs there are
things that we do have to worry about as a society,
in terms of artificial intelligence in terms of biases
for racism, sexism and more let's get into it. So let's jump back to the computer and
see some of the demos that have been released in the last week. So this is an
example of a cool new dictionary that i saw released this week using
GPT3 and the idea of the dictionary is it's able to use
context to give you the definition of a word most of the time a dictionary just
gives you definitions for different words that you look
up but here you can actually see that you can give it the context of the words:
"The bark of a tree." and then it'll give you the definition
based on the context of that word. So this is an example of an
image recognizer it's using a combination
of image recognition as well as gpt3 so it's recognizing the ingredients on a
package and then able to tell you whether those
ingredients are good for you bad for you or what they actually come
from so here we see it's identifying the ingredients. It's able to find those
ingredients, it's actually looking up emojis for each of those ingredients as
well. And able to give you a sense if those
ingredients are good for you or bad for you.
So here's another example of gpt3 working in a spreadsheet this time in
google docs similar to a demo i showed last week
it's able to complete and autocomplete formulas and equations based on what
it's seeing in the surrounding area so it has a simple function called gpt3
which populates the data from what it sees around it and this is a great
example of a tool that can really be powerful in
the hands of folks who use spreadsheets. Again lots of people were
concerned that this can eliminate jobs but i actually think it
makes it powerful to get better information and use tools
more effectively. Here's another example of GPT3 doing
layout for design and here the designer is asking for a button
that looks like a watermelon and it's generating the CSS for creating
that button, with the text inside of it. A button for
every color of the rainbow, and it automatically produces a design
that satisfies that constraint. A table of the richest countries in the
world with column names and gdp... So a little bit of editing and polish,
let's see if that fixes it. It was close, but the autocomplete isn't
perfect and again we really need to be thinking of this
as a very sophisticated autocomplete. So it's not going to replace designers or
developers it's going to produce the beginnings of something that's
incredibly useful but it doesn't really have testing or quality associated with
it it's still up to skilled developers and
designers to make sure that they use this as a starting point
to produce something useful and to produce something that is well tested.
So here's another one that i really like. This is a quote engine,
and you can just hit refresh and it will give you a new quote every time.
"A bad question will beat a good answer every time."
"Tradition is by definition progress against the odds."
"The environment shapes evolution and the economy shapes society. Ultimately
evolution shapes everything. So let's do it right." I couldn't agree more,
good job AI. And here's another design example of
using text to UI within figma. This is a variation of one
of the things we saw last week So in this example called magicify he's
able to produce the beginnings of a webpage by producing
sections of that web page one at a time. So he's able to give testimonials or
logos, placeholder graphics and the UI is able
to generate the beginnings of those designs.
So for the developers out there, here's another technical one where you're able
to type in a sentence and GPT3 is able to produce an
artificial intelligence-based SQL query that actually works within a
SQL database. So really cool example of a technology. So
it's able to produce and autocomplete amazing things like html and css and
code and sql statements but the reason i say jobs aren't really
at jeopardy is because it really does take a human to test these things.
You need to make sure that these statements are correct the designs are
valid they actually work with humans that people understand the text and the
copy that it's being produced and that it's producing content that is
useful and valid. So there are four types of artificial intelligence that are
important to talk about. The first is reactive machines and so a reactive
machine is basically a computer an algorithm and
it's reacting to the environment if you think about certain types of
artificial intelligence it's really looking at the board of what
the environment looks like such as a chess board or text that you
give it and reacts to that board. The second type of
artificial intelligence has memory and so it's using
what it learns over time getting smarter and smarter in terms of memory it's able
to learn and adapt from the things that it's doing.
Self-driving cars are a good example of an artificial intelligence that's
learning from its environment and getting better and better.
as it learns. So the third type of artificial intelligence is
theory of mind and that's where you can really put yourself in someone else's
shoes and imagine what they're thinking and really solve abstract problems. So
for example if i tell you you should really click the like button.
Yeah i don't know if i should click the like button.
But if you click the like button google will recommend my video and
it'll make it easier for me to make better videos for you in the future.
Well i do like better videos so future me would really appreciate that. So the
idea of theory of mind is you can put yourself in someone else's shoes and
this is a real leap in terms of artificial intelligence
this is sometimes called general artificial intelligence or AGI
it has all that ability of memory it also has all that ability of a reactive
machine and can really think and solve complex problems. The
fourth type of artificial intelligence is really self-aware. That's not only
where the artificial intelligence can think and put itself in the shoes of
other people but it realizes that it itself is aware of its own existence and
this is the science fiction of AI and we're really nowhere close to number
three number or four. So when you think about gpt-3 it really doesn't have a lot
of memory it's mostly a type 1 machine, a reactive machine. It's
reacting to the text that you give it and it's spitting out an output and so
it doesn't have memory from session to session it doesn't learn
from one behavior to another. You give it examples and it reacts to those examples
and even in the name GPT the PT stands for pre-training you pre-train
GPT it's pre-trained on billions of pieces of information
but that training doesn't continue to happen as you're running the algorithm.
So that's a really important distinction in terms of artificial
intelligence you can really think of it as the world's most sophisticated
autocomplete. So similar to google it's filling in the
result of a search query something that you're giving it and then it's
predicting what the output's going to be. Now because of the sophistication of
gpt3 it's able to complete html, it's able to complete code,
poetry, other languages, and text and it makes it look amazingly sophisticated.
But it is using an extension of that autocomplete
philosophy, with a lot of data and a lot of AI
behind it. Now GPT3 is an amazing piece of technology and i can totally
understand why it would cause people to be concerned for their jobs
in fact being concerned for employment isn't anything new, and people have been
concerned for their employment with advances
in technology since Aristotle and the ancient greeks.
Farmers have been concerned about plowing technology, scribes have been
concerned about the printing press, typist, mathematicians, and secretaries
have been concerned about computers and what that would do to their jobs.
In fact there's a term for this it's called technological unemployment and
it's been going on for decades and centuries and centuries and
the reason i think it's not of great concern is while technology creates
advances and may shift or eliminate some jobs it often
creates brand new opportunities and brand new
jobs. There are a lot more people employed today than they were employed
hundreds of years ago and so it's important that we as people look at this
technology as a brand new tool to allow us to build amazing technology.
The thing that people should be worried about when it comes to artificial
intelligence is bias. Here's a video from Chukwuemeka
about some of the ways that biases and artificial intelligence can be
problematic. okay now, we trying then... come to your hand too black too black,
yeah. come again, try. let's start napkin and napkin again... Napkin again What's the color of your hand? Ha ha ha. So it's a simple example of bias but
there are lots of biases in artificial intelligence and as we're building
technology either for auto-complete, auto-suggest, image
recognition or artificial intelligence it's important that we think about
biases and populate the databases that we're building
with diverse information, such that we really represent a good cross-section of
all of society. Here's a video of Joy from MIT talking
about some of the biases that are introduced in facial recognition as well
Hi I'm Joy and I research how computers detect
recognize and classify people's faces in my TED featured talk, i spoke about my
experience with the coded gaze, my term for algorithmic bias.
The system i was using worked well on my lighter skinned friend's face
but when it came to detecting my face it didn't do so well
until i put on a white mask. Joy has an amazing TEDx talk that i've linked up
here in the corner so if you want to check that out, give it a listen.
As developers build tools for artificial intelligence it's important that they
think about biases, bad jokes things that are offensive
historical inaccuracies and more in order to build better AI and AI that
is more inclusive and holistic in terms of society.
This is something that OpenAI is actively thinking about. In fact here's a
tweet from Sam Altman, one of the founders of OpenAI
about how they're really thinking about bias and making sure
that these applications that you're seeing demos of
are properly vetted before they're released. That's one of the reasons why
these application demos are available on
twitter but they're not publicly available, these aren't apps that you can
download yet because they haven't gone through that
vetting process which they're still working out.
It's great to see OpenAI taking bias seriously and i encourage other
developers who are thinking about artificial intelligence or GPT3 to think
about bias and how they can make their applications more inclusive.
I want to thank everyone for joining me on this video, i'm excited to announce i
crossed a thousand subscribers. Thank you so much. I love talking about
entrepreneurship, design and technology, and how these things can
really change the world for the better. If you like that sort of thing, you know
what to do. If you have questions or want to check me out on twitter,
post other demos that you think I should highlight, or other technology i'd love
to hear from you. i'm Greg Raiz until the next one.
