#3: What AI Can and Can't Do
Welcome back to another episode of The Junction,
a podcast that explores the intersection of people,
process and technologies for SMBs and nonprofits.
So today we want to talk about
what AI can and can't do.
According to a recent report by McKinsey, about half
the activities carried out by workers could be automated.
So, Chase, let's just jump into it.
We've talked about it in the first couple of episodes.
People are worried about job displacement and
what does this mean for my role?
How do I function outside of right?
Like, AI is here, it's here to stay.
Is it going to take my job?
I think we should debunk the things
around what it can and can't do.
It's that fear of the unknown.
This is kind of the unique thing, because
if you've been following along in the news,
it seems like AI can seemingly do everything.
There's stable diffusion, right?
It can make graphics, you can give it input, it
can make logos, it can come up with content.
What's really interesting that maybe some people don't
pick up on is if you ask stable
diffusion, which is an image generation, right?
Generate an image of the beach with the
ship and the sun setting, it'll do that.
But if you ask it to have some text, any kind of
text, and write the name of the ship is Pirates R, it
knows that it needs to put text on the ship.
But then when the text shows up,
it's nothing like an actual letter.
It just knows that text kind of looks like this.
And that's a really good analogy to
AI can do a lot of things.
But if the person can't do it well, or if
it is something that requires an excessive amount of finesse
or creativity or expertise, then the AI is going to
naturally be challenged and it will be difficult for it
to do those things, just as you said that.
So I'm thinking through you're
talking about it generating content.
Does it run on prompts?
Like, can you say now for the next three
months, go create a blog about this thing.
What's really interesting about this whole thing is that
a lot of people are just thinking about prompts.
So the answer is yes, right?
Like, let's ask it some questions.
Let's give it some context.
Let's have it do something for us.
Generate an image, generate content, like you
said, answer these questions for me.
There is this idea of AGI it's, in a sense.
Like the robot thinks for itself, right?
And what some people are doing with there's
something in the world called auto GPT, right?
Like, let me give it a go and
let the robot try to figure it out.
What's really interesting here is that you
now effectively take the human out after
the beginning of the prompt.
And now the AI is doing everything.
So the thought right, is that now you've got
this AI model, like, basically doing all the work.
And what's really interesting is not that what it comes up with
is any good, it's that it can do it at all.
So going back to the question,
what can and can't it do?
If it can't do it right now, then
it's probably a decent indicator that it won't
be able to do it in the future.
Decent, like, sure, with enough time, money,
and code, we can do anything, right.
If it can do it now, well, maybe it isn't
any good, but the fact that it can do it
at all means that it's only going to get better.
Yeah, I think I keep going back to this human element.
Right.
I look at anything that any use
case that we're exploring will certainly in
the beginning require a ton of checking. Right.
Checking the work, but kind of
going back to the prompts.
So I've been seeing a lot more like webinars out
there about just around this, like, chat GPT prompts.
What to ask?
In fact, the other day I downloaded
a prompt cheat sheet on LinkedIn. I saw that in slack.
I know.
I dropped it in slack.
I pulled it up.
Actually, it's right in front of me.
And it's got different categories
for writing creativity, content creation,
spreadsheets, programming, data, science.
So it's kind of walking you through how to and then
it gives you the plug, this action in here and then
in brackets, what program and language do you want it in?
Yeah.
Now I've heard this time and time
again here it's writing code, obviously.
I think we've experimented a little bit with that
just to see what it can do in some
use cases is not right, you're saying?
Oh, yeah, it does keep in mind, right, it's been
trained at least in the GBT 3.5 and 4.0, it's
been trained on basically the entire internet, right.
Like digest the entire Internet.
And now it has all this data at its hands.
Well, it doesn't have data that answers your exact
question, but it has a really good idea and
it's predicting what it thinks it should be.
The way that I think about these prompts, like
the sheet that you're talking about, go back to
one of the previous episodes where we're asking individuals,
we're asking this idea of the model, hey, what
do you think about this?
Well, replace the model with that intern. Right.
Like, you would ask the same question.
There's just a slight difference in the way
that you ask those things because the model
is expecting you to give it more context. Right, right.
Hey, Mel, write a blog about automation.
Well, your next question is like, okay, I can do
that, but what do you like, give me more context.
And what we're effectively doing in these prompts is
providing more context than we normally would out of
the gate to somebody now to this bot.
So the bot doesn't generate something that's useless.
Right.
You could say, hey, OpenAI
write a blog about automation.
It'll do it, but it's not going to be
exactly what you're wanting because in your mind, you're
like, well, I really wanted it to be about
salesforce, but you didn't specify that.
So with all of these prompts, right, this prompt
engineering idea, the whole idea is to provide as
much context in the initial prompt so that will
generate exactly what you're looking for.
And if not, then you kind of re prompt it.
And we do this with individuals as well, like in
a project with a vendor or with a consultant, right?
Like we deliver something, hey, check this out.
Is this what you wanted?
And no, actually, could you modify this a little bit?
We're going to do the same thing with these prompts.
Sure.
So it can write emails? Oh, yeah.
It can draft blog posts.
Whether or not you should use that.
Obviously, Google's downranking a bunch
of content right now.
I'm a big fan of the using your own transcripts.
Go interview your subject matter expert for 30 minutes
or less and then write a blog about it
right after you've done your keyword research.
I definitely think there's a use case there.
So it can do some of these things.
And then we've talked about let's talk a
little bit more about what it can't do.
So in our world, we're consulting, right?
We get on calls with clients every day
and try to understand what their business, how
they run their business, what's your process, how
do you bring a sale in the door?
How do you invoice a customer?
I don't see that going away.
Do you see these AI models being able
to ask the questions that we're asking?
Or let me go a step further, build the
relationship with the client that keeps them coming back?
We've got some clients that are on our
33rd, sow yeah, can AI do that? Yeah.
Well, maybe a better question to ask
Mel is what can't it do yet?
Because all of these models are built off
of data that humans have generated, right.
We're having conversations online and we're building
relationships, maybe not in the real world,
but in an online fashion. Right.
People are building relationships
on Twitter and Facebook.
I guess it kind of goes back to
like, can I build a relationship online with
you without any interaction, like human interaction?
And if you can say yes to that question, then I would
say that AI will eventually be able to do that as well.
Then you tack in that human interaction
and things start to get really blurry.
Or you start to say, as of
right now, it cannot do this.
And one of those things is building relationships.
We always say we win on
relationships because we're personable, right?
We interact.
We understand where you're coming from.
We've had similar experiences, right.
We talk about the weather.
I mean, I don't want to go get drinks with Chachi
PT when it comes down to it after a conference and
we want to go hang out with partners and clients.
I'm not going to lie though, I did ask it to
give me the recipe for a Whiskey Smash and it did.
I'm sure it crushed it get it
because it knows that data, right.
But if you ask it to make you a Whiskey Smash,
obviously it's not going to be able to do that yet.
But that's one of the interesting things about this is
as AI starts to take hold of a lot of
things that we're constantly doing, the more that you dig
in into these creative things like podcasts and webinars and
YouTubes where there are these human elements, those things are
naturally going to be pushed up and not know elevated
is kind of the right thing because the human can't
be replicated yet in a sense. Right?
So YouTube anywhere where there's video content, audio content,
to an extent, those things are going to still
be things that the AI cannot do.
Yeah, that's an interesting take for do
you remember our finance webinar that we
ran, how AI was impacting finance professionals?
And Scott, he put together this, he gave a prompt to some
AI video tool and I think he even could pick out the
hair color and kind of the voice and gave it a script
and it was decent, but definitely AI generated, right?
Like you knew that wasn't a person and it was
kind of tongue in cheek to do that, right?
Like we're going to promote a
webinar about AI using AI.
But to your point about that, with this video
content, podcast things where we are still throughout this,
we're a few episodes in, we've had a few
like, oh, cut that, do that.
We're flawed, we make some
mistakes, but that's also relatable.
And that's where I think naturally there's going to
be kind of like a distrust of absolutely.
The first AI YouTube star.
Well, that's already a thing. That's a thing.
It is already a thing.
I need to get out from under my rock.
In Twitch, there's a guy that is programming and maybe
part of his stitch, right, is that he's programming and
he's streaming the programming, but he has this AI chat
buddy that he's now turned into audio.
And he will just ask it a question
like, hey, what do you think about this?
Should we do this?
And he is enhancing his own chat
bot while he's doing this on Twitch.
And the chat bot is written in a
way where it's kind of super witty, super
sarcastic, like, no, that's a terrible idea.
You're not any good at programming.
Well, it's only one step further
to put video to that, right?
To then basically take the game.
It's a game right now, it's graphics and look
at some of the games that are coming out.
It's near picture perfect and I think
it's only a matter of time.
I said, you can't do that yet.
I think it's only a matter of time before
you've got YouTube series that are not animated.
They look real life.
And it looks like this individual is actually doing
these things, but it is all like, programmed.
Okay, take a walk with me.
I'm going way out into the future.
I think it might only be one or two.
No, you've got me thinking about this, like,
personifying or putting video to something like that.
Could I report?
Is there a world in which
people report to report to what?
An AI manager.
Like a manager where I'm on a zoom call
jumping in on my one on one with this?
Now you need to get into the
fast moving pace of the news.
There's already a guy that is running a company and
the CEO is, I think, OpenAI's large language model.
I think he's really just typing questions
like, what decision should we make?
CEO chatbot, right?
And it's popping it out and he's literally
taking the answer and doing the thing.
He probably has to do some
modification, right, to make it work. Right.
But I think there will be companies that are
run by AI at some point in the future.
Not yet.
Speaking of the news, let's kind
of look at some headlines.
So pull this one up here.
AI's dirty secret.
Meet the hidden human workforce behind
the Boom in artificial intelligence.
This is a interesting take, right?
Because, Mel, you and I were talking about this before
we started recording this idea that it's kind of taboo
to have something else do the work for you, right?
Because I'm paying you to do the work, you're
having something else do the work for you, right?
And you're delivering it as your own work.
But what's really interesting there is
we do this in businesses, right?
Like, we hire people to do work, we sell
it as our own and then we deliver it.
I mean, it's really no different.
I think the real tabooness is that the amount of time
that it took for the chat bot to do it was
like a minute and it would have taken you an hour.
But you sold that for the same amount, right?
Like, there's this extreme difference in time
or cost to actually generate that deliverable.
And that's where a lot of these questions pop up.
It's like, well, now, are
you price gouging your customer? Right? Right.
Or do you have so much margin that it's now unethical?
But one of the things that I saw here, this
was actually on Hard Fork, one of their recent episodes.
They were talking about Amazon mechanical Turk.
It's a mini job site where you can ask people to
classify an image or write me a headline for this topic.
It's basically like you can now ask a
lot of these questions of the chat bot.
Most of it is geared towards researchers that are
saying, well, how do you feel about this?
Or give me your hot take on this one blurb
and they pay you like five cents, ten cents.
But now a lot of those answers are being generated
by Chat GBT and these people were just copying and
pasting the answers in and they're making five cents.
And it goes back to that question of like well, I
really wanted the human element to answer, not the chat bot.
And that's where I think there's this preconceived
or pre communicated notion that I want Mel
to write the content, not the chat bot.
And if you do enough upfront, like setting
expectations and there's mutual trust, that'll be okay.
But where you don't have that context, like hey, please
don't use GBT or let me know if you do,
let's just be open and clear and honest.
Humans are naturally inclined to find the
most efficient way to do things.
It's just natural for us to do that.
This will continue to be a good problem or a bad
problem, depending on which side of the aisle that you're on.
I believe that talking about we
keep going back to generating content.
And this actually came up, I was on a webinar last
week, someone asked how long should a blog post be?
Right? How many words?
I'm thinking, well, it varies based on the content type
or the subject and what you're trying to do.
But we kind of know some
things about how content is ranked.
And someone, this post I saw said it's not
2000 words, it's not what SEMrush says, it's how
long it takes to tell the story.
And if the chat bot doesn't have the context
or the right, like if that's not something that's
readily available or out there, I mean, think about
the companies that let's say they're going through kind
of like a rebrand or they're trying to kind
of position our startups, right?
Like how do you tell that story?
Yeah, unless I guess you are started as an AI.
You are run by a CEO that is an AI bot.
That's something that happens
in collaborative workshopping, right.
And figuring out, like know, spending time
with our founder of I Can.
I am kind of like Mel GPT in the sense, like
if we're going to sit down and I'm going to write
something that Scott would write, whether it's an email communication or
a post on LinkedIn, he has high confidence in me that
for the most part, I'm probably use the words that he
likes to use or would use.
And it's conversational natural, and it would
seem as though he wrote it.
But again, if we're not feeding that into some
other tool, I don't know, job security for me.
Absolutely.
Maybe until he starts to we start to
put his podcast out, it goes back to
this idea of original content, right.
A lot of what Chat GPT and these
large language models are doing is they're regurgitating
things that have already been written.
Well, if you're going to use
that to post content, you're already
regurgitating something that's already out there. Right.
Because of what we do.
We do a lot of things that have never been done before.
Well, Chat GBT might have an idea on how
to do it, but we are now the thought
leader and the experts in that space because we
were the, quote, unquote, first to do it.
There's no internet documentation of
how this should work.
It's the individual that now knows how to do
it that we can derive content from and be
the first to market, if you will. Right, right.
So anywhere you have this original content,
original ideas going back to being elevated,
those things are going to be elevated.
Things that are regurgitated over and over
again are naturally going to move down. Desuppressed.
Yeah.
Just a couple of days ago, there was
a headline, AI is already linked to layoffs
in the industry that created it.
I think this particular headline, as I was reading
some of it, this entails the individuals that were
gathering the data to train the model.
Once the model has been trained, those
folks, they've effectively completed their job. Right.
And they're moving on to something else.
The people that are not being laid off are the ones
that built the model or they're building the next model.
So this is kind of like one of
those catchy headlines that it's like, oh, it's
laying off everybody, start worrying about your gig.
But I do envision and we have seen this, like,
this idea of AGI where it can self improve, right.
It can go into its own code base,
problems or opportunities to improve itself, and now
write that code and enhance itself.
That's where you start to get maybe a little concerned.
And again, you have to check the work. Right.
It's not going to be perfect.
So there's always going to be some level of review
where you really need to get worried, and this is
easily a decade away, where it's going to start writing
its own code over and over and it never sleeps
and it does it 24 hours a day.
It's going to be that superhuman, all
knowledgeable, all powerful, blah, blah, blah that
everybody's worried about, but it's always going
to be contained within a box.
You've got me thinking about all the
things that, again, it can't do.
So, like, do I need to go pick
up fly fishing and go find it? Oh, absolutely.
Some outdoorsy hobby.
Go grab an axe, start chopping wood.
Okay, there you go.
I think we could come up with a pretty
hefty list of things that still need to be
done to make the world go round.
You take any trade and if you don't have
any trade skills, maybe you should start picking those
up just so you can have a backup plan.
I wish I remembered who the
company was that advertised this?
I think I told you about it.
One of the best ads I'd seen.
A building was going up, right?
And on the side of it, where they
had already had finished out the concrete, it
said, Chat GPT can't finish this building.
We're still hiring.
And it's out there hiring
laborers and construction workers.
I thought that that was a
very clever but interesting play.
There are obviously machines that can build.
Yeah, well, this is where you get into that.
AI is directing. Right.
It's the project manager. It's the CEO. Right.
The AI was the architect on the building. Exactly.
That's where there's a delineation.
And not to get too far into the future, I
mean, it then is like, all knowledge workers are, AI
right, and I need some hands to do some work
are humans, and now the humans are, like, pushed down
the Dodom bowl a little bit.
I don't think that will ever be a massive
issue for us in our particular generation or maybe
our kids, but I think way in the future. Right.
What did we learn through COVID?
The human connection that was, like, what came out of it was
like what we all lost a few years, it felt like.
And then everybody kind of realized,
man, I missed being around people.
I missed having even just we are an in office culture.
We work in office.
I love being able to jump into
a conference room and whiteboard stuff out.
But even just in general, like, going to
the grocery store, not having my groceries delivered.
Right.
To somebody who you still do that?
No, I've only ever done it twice. Really?
To the grocery store?
No, had them delivered.
I actually love going to the grocery store.
I never wish that a bot will replace that for me.
And I make my own list.
Although somebody in our team actually said one of her
use cases when we kind of opened it up for
lunch and learn, she said, I told Chachi PT that
I wanted to eat let's insert keto.
And I asked it to make me a grocery list, and it
came back, and I said, I don't like those two things.
Refactored it.
People are having it, right?
Recipes, workout plans, things like that. Sure.
But no, I still want to take my list in the grocery
store, and I want to go up and down the aisles, and
I want to spend time you know what I'm waiting for, Mel?
Something that it can't do yet, hopefully, is work out for,
like, go work out for me so I don't have to.
But I'm already pretty good at not working out.
Maybe one day maybe one day it'll do the work for me.
That'll be a later episode. Yeah.
Up next up next all right, so speaking of
up next, we're going to move on to our
I'm calling this our ask Me anything. All right?
But I'm just asking you. Thanks.
Mel, should we trust AI to predict natural disasters?
Let me ask you first, are you a weather guy?
Like, do you pull out your phone, like,
look at that, like the night before?
Do you plan outfits around weather?
Do you plan really?
I tend to wear the same clothes, t shirt and shorts.
And is it is a summer in Texas right now?
That's probably a fair play, but what I will say so
you said you need to get your head and kind of
get up to speed on some of the news. Right.
Weather is one of those dark spots for me in
my life that I just kind of roll the dice.
Now, depending on if I'm traveling somewhere, I
do want to know, am I expecting snow?
Yeah, I want to be prepared.
I'm also a type A planner, so I check it when
there is a event or something that I really feel otherwise.
The other day, drove straight into a hailstorm on my
way to work and the radio warned me and everything.
It was like, here's Mel driving into the know.
It's going off on the radio and it says every
county that the golf ball size hail was going through.
And I was like, I'm just going
to take my chance, roll the dice.
And I was warned.
So I just kind of wanted to predicate that on.
Do you particularly watch the news
now we're talking about natural disasters.
I think that's something that hopefully
everybody knows when it's going on.
Although I have not been privy to tornadoes in
my immediate vicinity and that's just, again, my own
kind of getting stuck into what I'm doing.
But do you think that AI is going
to start predicting those types of disasters?
What's really interesting and the weather people
are going to kill me out there.
But what's really interesting is that a lot of
the predictions are based on data that already exists.
Right.
We don't necessarily know what the water
temperature is going to be tomorrow.
We can make a prediction. Right.
And what ends up happening is you have all of this
data that is now instantly old every second the weather is
changing, but you have a ton of data to look at.
The problem with predicting in a true
artificial intelligence manner is that there's so
many variables that go into that, right.
Like all across the world, you've got water
temperatures, you've got heat indexes, and I'm going
to melt my own area of knowledge here.
But I do know that because there is so
much data and there's so many interactions in the
client that it's going to naturally be difficult for
the AI to take ingest all that data and
instantly tell you, Mel, you're on the road.
And if you keep going at 70 miles an
hour, you're going to hit golf size ball. Hail.
Those things are easily decades away.
Just because there's just so much data to consume. Yeah.
Now, can it potentially predict stuff?
We're already doing this, right?
We're already predicting the path of the hurricane.
Right.
We can already tell that these counties are
at risk for hurricanes or for hail.
We're already kind of predicting that.
When I think about AI, I'm thinking more
about like, okay, I'm on my way home.
Do I need to be worried about hail?
It will hail hit my car if I continue?
Like, that's the kind of prediction that I'm
thinking about, because then I want to make
sure that I don't do that. Sure.
That kind of prediction is way off, way down the road.
Yeah, there's just been some recent headlines
about that, so I was just curious.
I do have to give a nod to someone
is quoted in this article that I'm reading.
It's from the Washington Post.
And she uses, in the last twelve months,
we have had a tsunami of demonstrations of
different AI methods being used for forecasting.
First I thought, are we talking about tsunamis?
No, she was just using that
as like a tsunami of information.
She's so punny. So punny. Love good pun.
Bad joke.
All right, well, I think that brings
us to the end of today's episode.
So, again, I think we have more we can probably
revisit this one as tools continue to come out easily.
Yeah.
What it can and can't do, let's revisit.
That should be a section of our content. What can it do?
It's like can it blend well? Blend tech. Yes.
The goat of content marketing.
All right, well, we want to know what your
questions or thoughts are out there to our listeners,
so please email us your take at thejunction@bendtechnology.com thanks
for tuning in, and until next time, keep exploring,
stay curious, and embrace the power of AI because
it is here today. Keep it automated.
