From sexy news this is connections.

I'm Evan Dawson.

Our connection this hour was made in 1985, when a student named Kevin Sirus graduated from RIT and promptly began a career that would put him in the center of the tech world.

So race has almost too many accolades to list.

Inc. magazine Entrepreneur of the year featured on CNBC, CNN, time magazine.

He's often described as one of the innovators who paved the way for Siri.

He's often called the father of the digital assistant, and now, decades after graduating from it, he's back in Rochester and still working tirelessly in tech.

He has 94 patents at my last count, and was working on a 95th in the artificial intelligence space legitimately.

Kevin is a titan in AI in the tech world, but this will probably be a different kind of conversation for Kevin.

I've watched him in dozens of interviews, mainly on Tech Podcast, where he is kind of a deity, and here on connections listeners probably know that I've become a bit of a skeptic of AI, and so maybe Kevin can offer a kind of intervention for me.

Or maybe I will convince Kevin and his colleagues to just stop.

Probably not.

No, that's not going to happen.

So how are we going to live with AI in the future?

A big question that Kevin Surace has asked all the time is, what's it going to do to the workforce?

What is it going to do to my job or my kids jobs?

He and I were talking about that before the program began.

We're going to talk about work productivity, what might change, and a lot more with Kevin Surace, who is a Silicon Valley pioneer.

It's nice to have you in the studio.

Thank you for being here.

Thank you for having me.

So you really are the father of the digital assistant.

How did you do that?

I assistant voice.

I all came out of a project in a public company called General Magic that spun out of Apple Computer.

Actually came out of Apple in the early 90s.

And one of the things we developed were the first AI assistants.

And you could talk to her.

Her name was Mary, and you had a on virtually full company Mary.

Not Siri.

Mary, not Siri.

Siri didn't happen yet.

Siri was too generous.

Kind of two generations later, but a complete copy of what we built.

So what we built ended up having 11 or 12 U.S. patents or so and, it was called portico.

And my talk and magic talk and GM had it for the GM OnStar virtual advisor.

And so we hosted all of those.

We built an operation center.

There was no cloud at the time.

You had to build your own.

And, she was amazing, actually.

I mean, if, by the way, she would answer your phone and if she knew who you were, she'd say, oh, Evan, I recognize you.

Would you like to, book a time on Kevin's calendar?

I still can't do that today.

And this was 1990, 1998, 1998, 1998.

Okay, so pretty sophisticated.

Oh, yes.

So why aren't you known as the father of Siri?

Did they steal it from you?

Are you like the Winklevoss twins with Mark Zuckerberg?

Well, it's complicated in Silicon Valley.

You publish these patents, right?

And then lots of startups read the patents and go, we should do that.

And actually, the people in Syria and the people before Siri, tell me in some other, companies, a lot of those people came out of Netscape and a couple of other companies that knew me.

Like, I know that these people.

Right.

And they did exactly what you do in Silicon Valley, copy and copy is the best form of flattery.

Right.

And they copy what you did.

You listed all, you teach them how to do it in the patents and they do it.

And so you go, well, why didn't the startup get sued or something?

Well, there's a story behind that.

So Nathan Myhrvold used to be the CTO of Microsoft, and he was a friend, and Nathan bought all of the voice AI patents, the voice, user interface patents that we did a general match.

He bought them and he sat on them in a company called Intellectual Ventures.

And, and so he didn't go after those startups.

They have no money.

He waited until Apple bought one of the startups.

Then he knocks on the door and says, excuse me.

By my addition, you owe me 187 million today and probably a few billion over the coming years, right?

So that's how that works.

And it always works that way.

And so it's this is standard fare.

Kevin, Siri's my guest and he has been so graceful.

And I just want to say what a what a good and graceful guest Kevin is, you know, coming back from a long paternity leave.

We had Kevin booked for today, but turns out we had also booked the Rochester superintendent, which listeners might have heard that we mentioned he was going to be on the program today.

The Rochester superintendent, the new superintendent, doctor Eric J. Rosser, will be on the program likely next week.

Now he's had to move that conversation.

And now Kevin is back here.

By the way, thank you for being so flexible.

My pleasure.

I know you're I know you're very busy.

So happy to be on and when when you had Mary in the late 1990s and you see where this is going.

I mean, I don't know if you remember exactly what you're thinking the late 90s, but you certainly had a better idea for where technology could take us.

We were not a very the lay people are not very sophisticated about where we think it's going.

We have a lot of sci fi movies.

But in the late 1990s, how much was I?

Did you think it was going to dominate our future?

Did you think it was going to be a race to the top with that?

And how did you view I.

Well, there's a lot of ways to answer that.

First of all, on the Mary product or portico, you know, the first time we demoed that on stage, people nearly fell out of their seat because I was talking to Mary and she was talking back to me, and I could interrupt her, literally could interrupt her and say, no, but read me my email right now, okay?

Kevin.

No problem.

And she'd do that and you'd go, what is this right now?

It turns out the number of things she could understand was in the thousands, right?

The number of things she could say was in the thousands or millions.

But it wasn't trillions like an Lem today.

But we imagined that there would be a day that the features we're seeing today would occur.

But a lot of things had to happen for that to happen.

Right?

And the biggest was that compute horsepower had a double every year and a half or so, which it has since then.

You needed immense amount of compute horsepower to do the kinds of computations that we're doing today and have her respond quickly.

We could have had her respond in 30 minutes and have a lot of the understanding we have today, but to respond in a second or two, that's tricky.

So I'll tell you, you know, our vision was to have something like we have today.

And when people first saw GPT 3.5, I think it was two and a half, three years ago.

Right.

And they were blown away.

In the AI world, we weren't as blown away because we had been working with earlier models of transformer since 2017.

But a lot of these things we would talk to in Python, not in English.

And finally, everyone could talk in English and it changed the game like 6 billion people want.

I could speak to a computer finally, right.

I've been speaking to computers since the 90s.

Right.

So so we've watched the technology move forward.

It's the most exciting time, I think, to be alive, but it's the most exciting time in tech we have ever had.

It's our Apollo program.

I can certainly understand that.

And so the one thing that stands out most to me, perhaps most often when I read about what is happening now that was happening in the 90s, the difference that I'm hearing is this in the 90s, you had Mary, and people were blown away when they saw what Mary could do, but you could explain whatever Mary was doing.

You could kind of black box it and figure it out.

I'm sure you saw the stories everyone was talking about grok and anti-Semitic explosion this summer.

And grok is a, for listeners who who don't know grok, you know, grok has become this popular piece of technology that people often consult as a kind of adjudicator of fact.

I mean, even on social media, people say, hey, grok, is it true that this bill will do what Republicans or Democrats say?

It will, and grok will have an answer, and it's very sophisticated.

But then grok all of a sudden gets some questions, and spits out these really anti-Semitic answers and is praising Hitler.

And the creators are going, whoa, we don't really know why.

I have no idea where that came from.

You know, we're sorry about this.

We'll try to figure it out, but I don't know that you can figure it out.

They're admitting they can't.

Sam Altman is saying his newest models.

You got to be really careful in the sort of building personal relationships, because that's going to be dicey.

Kids are going to grow up with AI friends, people are going to AI romantic relationships, and we're not exactly sure how that's going to go.

Are you comfortable that we're going to move through a society where even the smartest people, literally like yourself, can't always tell us why your creations are doing what they're doing?

Well, I can tell you why it's doing it.

What?

We can't.

Everyone, actually, everyone in the know knows why.

So even the grok thing, even the grok.

Okay, yes, we can explain what we can't.

What we can't do is dig in and say at that moment of time, why did it respond that way?

But fundamentally, we can say this.

We have trained the largest foundational models, these these, lumps, right, including grok and GPT and Gemini and anthropic and a few a clod.

Right.

We've trained them on over a trillion phrases of the English language that we could find wherever we could find it in the net, which means it learned from fiction, it learned from fact, it learned from Hitler's writings.

It learned from everything, of course, learn from everything, everything out there.

And, all these models do is put a waiting on the next word that comes after the last word.

That's all they're really doing.

They're saying.

So, for example, if I said it was my birthday today, it's not.

And I said, Evan, it's my birthday today, what would you respond with?

And I knew it wasn't your birthday.

No.

You know, it's pretend it was my birthday.

Oh okay.

Happy birthday.

That's it.

There's only two words.

It could be happy birthday.

Happy birthday Kevin.

There are.

There's no other phrase.

And you've learned that because you've heard that phrase thousands of times.

Yes, this works the same way.

And all.

It's so.

So it's not lamb.

Jones.

It's happy birthday.

Right?

It's.

Those are the only two words that make sense.

And so these get probabilities on these words.

And so you're just putting the highest probability word after the last probability word.

At any time these models will grab something that it read somewhere.

Misinterpret what you meant for some reason.

Right.

And quote unquote hallucinate.

But no, it's actually putting one word after another.

You know, like someone who doesn't really know what they're doing and just read the entire English language.

That's what it does.

Right?

So so how have we stopped that from happening more in GPT five?

Well, we stopped it from happening by hiring more than 10,000 people around the world, mostly at a company called scale AI, to put rules or guardrails around these things.

So in GPT 3.5, you were able to say, I love you.

And it would come back and say, well, I love you too.

And you'd go down some whole road of how this thing loves you.

And they did make up a name for it.

So when they stopped that like in a week and said, this isn't going to work well, right.

So you just put rules around it.

So now they have literally millions of rules and and it would be hard to get an open AI model or hard to get GPT five to respond the way grok did, because they put so many rules to keep it from doing it.

You have to jailbreak the thing.

Now.

It grok the particular owner of grok.

We know who that is Elon Musk.

Elon, Elon has decided I'm not going to put many rules around this at all.

And whatever she says, she says or he says, he says, right.

Whatever it says, it says, I don't care.

That's a choice.

It's not a choice I would make.

It's not a choice.

Any other company is made, but it's a choice.

He's made.

He doesn't care if it swears at you.

He doesn't care if it goes, you know, and it's anti-Semitic or not.

My problem.

It's just a model that built itself on the English language and whatever spews out, spews out.

And, you know, that's a choice.

Not illegal.

May not be moral or ethical, but it's not illegal.

Do we have to be better as consumers about what products we're using then based on the guardrails, don't we?

Of course.

And there may be times you want to use something without any guardrails, because you literally want the thing to dive into some love scenario.

Let's let's say, right.

So there may be a time that you wanted to do that.

There may be a time you want to dig into things that GPT five would never talk about.

For example, how to build a nuclear bomb.

GPT five will not tell you.

I haven't tried it on grok, but I suspect it'll tell you.

Oh, great.

Well, I could have also googled it, right?

It's not the biggest secret I understand, but but, but those are the kinds of choices that were made by the model makers.

And Elon, you know, lives in a whole different space that apparently doesn't even believe in PBS.

So.

Okay.

Well, just just to note that, right?

Yeah.

No, no, that that differentiation of guardrails is very interesting to me.

And I hope that we value guardrails as both in the tech world, the people creating these products and the consumers of them.

And I hope that we ask the right questions so we know what our kids are using, what we are using.

We don't want to get caught off guard with stuff that is either untrue or really dangerous.

So let me just say I appreciate that.

Let me also ask you that as we go into more sophisticated models, are you comfortable with the word thinking when people say, well, what did what did GPT think about this?

And I know people we've talked to on this program, people who have named their GPT and and talk about GPT like it is anthropomorphized.

I mean, like, are you is that a why is it thinking?

Well, when we developed portico, we partnered with a team from Stanford called Reeves and Nass, and they wrote a book called The Media Equation in the early 90s in the Media Equation.

Basically, it was a summary of work they had done with Stanford students to have them interact with a variety of media, including computers.

And there were even experiments, for example, where they take a computer and wrap it in blue in another computer and wrap it in pink, and they would tell you that the pink computer was female and the other computer was male, and they'd have it interact with some English on the screen.

Right.

And and in fact, very quickly the students treated the male computer as if they were to bar, you know what I mean?

And they treated the female computers as they were.

If it's a guy, they were trying to pick her up, right?

I mean, that's so amazing.

So this was an experiment.

It was an important experiment because it said that you could take humans that grew up around other humans, and immediately they would treat virtually anything else like a human, because that's all they know how to do.

And so we naturally anthropomorphize these things.

And when we came out with Mary, immediately people said, Mary, will you marry me?

Mary, I'm in love with you.

Mary.

Now it's obvious this thing is a bunch of subroutines.

Ray, stop saying this.

But instead of putting rules that stopped her from responding, we put in rules that made her respond that said, oh, thank you, but I'm already married and stuff like that.

And she'd have all these randomized responses back just to play the game.

Like, obviously she wasn't available to marry you, but I'll play along.

Right?

Okay.

Most of the model makers have decided to not play along and just stop that.

They don't like it, and that's a choice.

So that's kind of where that all comes from.

Yes.

People anthropomorphize everything.

If you left this mike around the house long enough, someone's going to name the microphone.

Sure, Mike.

Probably.

Mike.

Yes.

But is it healthy to name your GPT, or do you think?

Look, we still need to understand the boundaries of what we're working with.

I'd say our our work, with Portico and Mary, said that it wasn't healthy or not healthy.

Some people really anthropomorphized, morph ized it and interacted with her as real human.

Others, you know, treated her like a computer.

But most people did treat her like a female, like a an assistant or a secretary, and actually say thank you and please.

And we had to deal with all that.

Oh, thank you so much for getting that.

You're welcome.

I mean, that has to be programed in.

Right.

So, I think people are going to do it.

I think we, you know, it's kind of one of these like, people ask and I know what's on your mind.

Gee, should we slow this down?

Should we stop these models?

Should we train left the station like the models are here?

They're getting better every week, and no one's going to stop them, because technology never stops and technology always wins.

And we have thousands of years since the invention of fire to prove that that's the case, whether we want it or not, whether you want it or not, it doesn't.

We may or may not want nuclear bombs, but once we figured out how to make one, however, many companies of countries have figured it out.

They figure it out.

Right?

And so and so part of your argument is we can prefer a world without AI.

We realistically are not going to have one.

That's right.

And we have to figure out how to live with it, how to be more productive with it, how to figure out how to have careers with it, how to be more productive workers, how to protect our careers, how to understand our relationships, all of those things.

Can I give you an example?

Yeah.

From the late 80s, in the late 80s, Microsoft comes out with something called Microsoft Excel.

Now, when I say Excel, we all know what Excel spreadsheet.

But there was no there was Lotus one, two, three slightly before that.

But basically the invention of the spreadsheet that plowed into every finance department in every radio station, every manufacturer, everywhere, right.

Every finance department now had that instead of pencils and ledger books.

And there were many people in finance that said, I am never going to use that thing.

It's blasphemy.

I've got pencils, pencils and ledger books and I'm going to keep adding my numbers.

And other people said, this is manna from heaven.

It's not going away.

I'm going to use Excel and I'm going to be the king of Excel.

I'm going to be the robot overlord of Excel.

Those who took on Excel excelled in their careers, and those who stuck with a pencil and ledger book left.

But because who would hire them?

And I can say that today because it's obvious no one would hire anyone that says, hi, I don't use Excel.

I'm sorry, this is a finance department.

I don't believe in it.

Get out.

I mean, there's no job, right?

And that's where we're headed with AI.

We are entering an era, even right now, where if two kids come out of RIT, let's say right, and you've got one that says, listen, my professors in my courses told me, don't use I don't use it for my homework, don't use it.

So I really don't know too much about it.

I played with a little, but you know, I really nose to the grindstone.

They go, okay, good interview.

And the next kid comes in and says, I'll be honest with you, I got straight A's, but let me tell you, I helped me all along the way.

I use that everywhere.

Not only that, I've got five tools you've never heard of.

They're going to increase the productivity in this company if you hire me.

Who gets hired?

Number two, door number two wins every time.

What the companies want today, they want more AI knowledge coming in the door, not people coming in and say, I don't use this stuff.

So the best thing you can do for your career is become the robot overlord, become the expert on every possible tool you can, and then people start relying on you.

Management starts to rely on you, everyone relies on you, and no matter what happens, you're likely to be very important to that company.

So I think everything you said is correct.

I also want to try to challenge as much of as I can and then get your take on something.

Okay, so let me start with this.

You said on a podcast in May the following I'll quote you from May.

You said every team had better be using AI, and if you know what you're doing, the top models today will just outperform any humans you have.

End quote.

Now, to be fair, you go on to say that the AI is only outperforming humans with the assistance of the humans.

The humans using the AI absolutely is outperforming the human, not using the human.

Using Excel is outperforming the accountant who uses the human who uses fire outperforms those who don't.

Okay, and that is the core of one of the arguments for that you make about why AI is what it's going to do to the workforce.

And of course, some jobs get eliminated, but generally speaking, it's not going to wipe out all jobs.

It might, in your words, create more jobs.

It gives workers a chance to be more productive.

And they're going to have to learn how to be more productive.

When I hear that, I wonder if that is looking at 2025 reality and I'm worried about 2035 2027.

So let me tell you more.

Artificial intelligence is today, as we said, it's the worst it will ever be.

It everything it's doing right now and it's already really sophisticated at myriad tests.

It's only going to get better.

And to think that it will just remain a tool that humans will use instead of replacing humans.

That seems naive to me.

I understand that there will be jobs where you use the AI to do better, but I got to think that companies are going to do whatever they can to actually eliminate the need to pay humans and use AI entirely if they can.

And it sounds like every year they're going to be able to do that more and more.

No.

Yeah.

That that is that is correct.

But, but but let's let's get a baseline here.

What did let's talk about American companies first.

Right.

Because this is the what we know the best.

Let's what did American companies do over the last 20 years?

They shed every job they could from America overseas that they could afford to do for variety reasons.

So let's look at customer support, all customer support before 20 years ago was done in the United States.

You'd call and you get someone in Georgia, you get someone in Texas, you get someone somewhere in Rochester.

Perhaps when Kodak was here, they probably had customer support.

You'd get a human in the U.S they cost, who knows, ten, 15, $20 an hour at the time.

Okay.

Over the last 20 years, virtually 100% of call center jobs got outsourced to India, Vietnam, Philippines, you name it.

Okay, that's because we went from maybe $10 an hour to maybe a dollar an hour over there.

Well, so what did American corporations do?

They said, I can save 90% of the labor costs by going offshore.

Right.

What are the first jobs that are going to be automated by AI?

It's the things that we already know are highly repetitive that we already sent offshore.

And now we're going to take a dollar an hour and make it $0.10 an hour.

And of course, everyone's going to do that.

So the first jobs that are being impacted over the next five years, it's call centers, it's software, QA, it's all kinds of stuff we sent offshore by the millions.

And this is creating a whole backlash offshore, by the way, and we see it in the industry.

Great survey recently that 31% of employees anonymously admit to sabotaging the AI, initiatives of their company, and that's impacting 80 to 90% of AI initiatives, telling the boss that it doesn't work when it does because they're trying to save their job.

And this is much more prevalent offshore, because you've got people in these countries who've worked very hard and they're feeding their family, doing this one repetitive task, it's all they know.

Probably didn't even go to college, but they do this one repetitive task, and AI comes in and just demonstrates doing a thousand times faster than what they do.

And so they're going to sabotage it.

And they put stuff in there and they tell their bosses it doesn't work.

And the word sabotage comes from France when workers, when they were automating French factories in the first industrial revolution, threw their shoes.

They're supposed into the equipment to sabotage the equipment and stop the automation.

So here we are again.

We've got automation coming in.

It's taking certain people's jobs, mostly not in the US, mostly offshore, and you've got people absolutely sabotaging it.

And we have data to show that now.

But you don't think I mean, it's not just offshore, right?

I mean, don't you think in the next five years, two years, we're going to see a lot of jobs that are at risk?

Yeah.

Only because, again, U.S. corporations already decided to shed those jobs.

So shedding them a second time is no big deal, right?

That's going to happen.

And mass and we're talking about maybe 5 to 10 million workers in highly repetitive jobs offshore.

That's going to happen.

And mass before we see something and mass here, except we are seeing here entry level positions in certain areas like coding, be at the lowest, opportunity that we've seen in, you know, because AI is being taught to, to write its own code.

Yeah.

Exactly.

Right.

Code quite well.

Okay.

Under the advice of a senior coder.

Okay.

So basically what's happening is I thought I needed to hire ten junior coders that I be training and they can't do much, but they'd code up some simple things.

And now I just have I do that work so I don't need to hire the junior coder.

I'm not getting rid of them, but I don't need to hire them.

I'm not going to fill those positions.

Now that creates a problem 510 years down the road, who's going to be the senior architect if nobody came in as the junior person to work their way towards senior architect?

Right.

So we do have those fundamental issues we have to deal with.

And the Excel example, Adam in Canada wants to jump in.

Hello, Adam.

Go ahead.

Hello Great Dane.

Great to hear you guys today.

A great discussion.

Quick comment about, the concept of jumping on board with something.

Your analogy, with Excel is fantastic because I love, recovering it by myself.

I was in the field for 20 years when the internet was just becoming the internet and adopting Excel was an absolute glorious thing.

Saved me tons of time.

I use it to this day, I'm an energy user, etc.

but it was well tested and came out with pretty much exact mathematical connotations, which was what it.

Oh, we lost you for application.

Oh, am I still there?

Yeah, you're kind of in and out.

Go ahead and make it tape.

Okay.

My apologies.

Anyway, Excel is a great thing.

It was well tested.

I think that I not adopting something that isn't completely well tested yet.

It doesn't necessarily have accuracy that can be relied upon 100% like Excel can.

So those of us who say, well, let's wait a little while or let's tweak it some more before we jump on board.

I don't believe that we're going to get lost in the, you know, in the world with the way the example was set with Excel, that if you didn't adopt Excel, you were definitely, you know, far behind, which made perfect sense.

But I don't think in this case it is I daresay that, you know, faster is it always better, to get things done accurately, I think is most important, even with your, call center analogy.

I know for a fact for for my own experience having called, automated systems, they tend to, to not understand the nuances, perhaps that a human might be able to interpret and understand.

So just being careful, I think it's wonderful stuff.

But I think it's just we're we get very excited about using things and getting it on there without thoroughly being sure we understand it completely, which obviously is what this discussions about.

So I just want to say these example was a great one, except that it only applies if the product is thoroughly tested and performs 100%.

So thank you so much, Adam.

Thank you for the phone call.

What do you think, Kevin?

Sure.

A lot to digest, but real quickly.

Look, we solved math, with Excel, and people never go back to that and realize after Excel showed up, did any of us really have to do long division again in our head or with a pencil?

Except to teach a student?

The answer is no.

It would solve that right?

And we accepted the fact that math was solved, done, solved, done.

But it took another 40 years to solve language because language is much more complicated.

And so lems solve large language models.

Solve language, and language is a perfect as it is.

But what I look at these models and say I can ask lots of questions, I can poke it and I'll give you an example and do all kinds of things.

Is it more accurate than if I had a human sitting here?

Faster and more accurate.

And the answer is yes.

Not perfect, but more accurate than any human I could hire for probably any amount of money.

It's, you know, it's like 100,000 professors wrapped up in one, not dissimilar to the way that self-driving cars right now actually would have a better safety record than human beings.

They do.

The data are very clear.

Yeah.

I've always said, and I think we had a conversation ten years ago on this program where I said, you could have what do we have?

30,000 people die in car crashes a year in this country, 40, 30 to 40,000.

It's a lot.

It's tens of thousands of people.

And then there's hundreds of thousands, if not millions more injured and maimed in car crashes.

Because we're not paying attention, we're falling asleep.

We're not good at it.

We're looking at our phone.

Whatever.

Yeah, exactly.

Yeah.

Numbers have gone up because of the problem with the the self-driving car is there are still a human lawyers.

And you can tell a society we had five people die compared to 40,000 because self-driving cars were better.

But if you're one of those five families who lost somebody, you're going to take them to the cleaners and sue them for everything they're worth.

And the litigious nature of this makes it very, very difficult for society to want to number one, give up control driving, which people like to do.

There's a lot of identity with that.

But also just I mean, it may be better, but people want to do it.

Yeah.

You know, this is this is one of the banes of technology.

Technology comes along that's far better than humans.

But we expect it to be absolutely perfect because Excel was perfect and we've had technologies that are perfect.

Therefore, the self-driving car can never make a mistake.

But of course, that's impossible because there are situations where no matter what you do, you're either going to kill the driver or run over the little old lady, right?

So you have to make a choice by the way, the the trolley problem.

Yeah.

Yeah.

Yeah.

And the self-driving car, manufacturers have already made the determination that if they had to, they would save the driver at all cost.

Yeah, that you had to make it, or else who would get it?

The thing about.

Right.

And that's another important point.

I think Adam's phone call is a good one, because he's not wrong to say in mathematical terms, you can either be precise, incorrect, or you're wrong in language terms in these culture.

And there's all these other changes coming.

It's so much less precise.

It is.

But, but, but but look, I'll give you an example of how I use it every day.

So I get, you know, call it five requests from the press a day to comment on a variety of things going on, I, etc..

Right.

I can't answer them all, but it used to be I'd answer only a few a week because they come with 5 or 6 questions and they they post something, they come with a handful of questions and I would struggle not with my opinion, but I would struggle with getting each reply, each answer correct in a way that's quotable by the press.

Right?

Because that just takes time and I might spend an hour on it.

What?

Well, today what I do is I write gibberish of two paragraphs, basically of here's my opinion on this topic.

This is what I want to say about it.

Now take my opinion and write the five responses in my voice boom, boom, boom, boom boom.

And it's like I may change a few words here and there, but it is my opinion.

Notice that I installed the AI to write those sentences, but its prose is better.

I'm not an English major, right?

I'm an engineer.

It's prose is better than mine will ever be.

Oh, but this is where I.

This.

I admit I struggled with this because, one of your presentations you talked about, writing really good prompts, the importance of understanding how to write good prompts to get a better return from AI.

So if you're writing a presentation, as you said, you said, look, I'm if I'm not an English major and I know that that's not my top skill set, I is very capable of producing better quote, quote unquote, better work or a tighter, more cogent presentation.

You, based on my opinion.

Right.

Okay, okay.

I, I to be clear, this is why I want you to convince me.

Because if I'm in the audience and and I know that you're giving a presentation and it's all your own warts and all.

Okay?

And if I know that the presentation you're giving me is based on a really detailed, not just a three word prompt, 500 words really deep, and mine are hundreds of words.

Okay.

Yeah.

More.

Right.

So but still it was created.

The final product was created by I might be sitting there going, I don't want the AI presentation.

I want Kevin to race his presentation.

Yeah, but it is Kevin's racist presentation if I use the AI is a total cheat and said, I don't know anything about this topic, right.

A presentation that's a different story.

But I have done my own research, by the way.

I have my own experiences, right?

I bring my own experiences to it.

And I'm saying, here's how I feel about this.

Now.

Create, for example, a slide that has five bullet points that summarize these 500 words down to the five bullet points that would be most interested to an audience of these engineers, let's say, or finance people or whatever.

Boom boom boom boom boom.

Now I can take those or leave those, but frankly, they're really good.

And that's saying but you're still telling me that's still your work?

it is.

It's not the AI opinion.

It had no opinion.

I said, summarize what I gave you and put it into five bullet points.

Now, I could spend two hours and put that into the bullet points as well, and I may eventually get to something pretty similar, but I'm not actually adding value, taking my set of opinions and getting it to five bullet points.

That's not my value add, right?

My value add was I took 40 years of technical experience.

Right.

And and and and opined on that.

I hope that makes sense.

Yeah, it does make sense.

But tell me if I'm doing my job right.

I've never used AI to write a single word that I've said on this program.

I write the only scripts.

The only real scripts I write are the first two minutes of every show.

And sometimes, I mean, I do a lot of research and background work to make sure I've got different ways to pivot information that will be helpful.

I mean, I've got stuff I'll never use, and so I've got that for myself.

And sometimes I shape it in my own words, but I've never put it into AI to help me with that.

Should are you telling me I shouldn't?

So I think the audience doesn't want me.

Oh well, I how many programs I've been on.

That the host, they've got their notes and stuff, but they will take like 20 pages on me that or somewhere and say summarize and give me the top five questions I should ask Kevin today.

You know, boom, boom, boom, boom.

You can ask them or not ask them.

Doesn't matter.

Right.

But you also could read all of that and try to get to those five.

But you're going to take several hours to do that.

And you may not have those several hours.

And you want the show to be as good as it can.

So if you don't think this is being used, I do think it's being used.

Oh, I definitely, of course, of course is being used.

I agree with you.

But but you're the one who kicks that off.

You're the one who uses or doesn't use those.

You know those questions, right?

I mean, in the end, you control it.

And that was one of your early points, is that we control these things right now.

They they sit there and go, I want to service you.

What do you want?

What do you want, Kevin?

What do you want?

It's just waiting, right.

Just listening.

And then again, I will put in.

We could talk about music, all this stuff.

I'm doing music and stuff like I will put in, maybe, you know, an entire chapter of a book that I've written, an entire chapter and say, what's wrong with this?

What didn't I do right, poke holes in it?

What am I missing?

What do I need to add?

How could I rewrite some sections of this so it would be more poignant for this particular audience?

This is amazing.

It's like hiring, more than a ghostwriter, you know, an incredible person.

And and the I didn't write anything, right.

I wrote everything, but I realized my prose cannot possibly.

Because now your prose might be excellent because of what you went to school for.

I didn't go to school for that, but it took me three years.

I mean, I was published, I have one book.

It's nothing in your field, Kevin.

It's, it's on the Finger Lakes wine industry.

It was published in 2011.

It took me three years.

You could argue I could have done it faster, but I wouldn't have wanted to do it another way.

And.

And you know what's great about II?

Anyone, including a music or a publisher, can use it or not use it.

It's your choice.

But take it or use it.

So you've pointed to Suno in the past?

Yeah.

So who knows, maybe the most sophisticated, one of the more sophisticated music program it is.

And so I think about creation of art.

So take writer's block.

Anybody who's had to write a presentation, you're writing a book, you're writing poetry.

Everybody's had writer's block.

Musicians, songwriters have writer's block.

I know a ton of people.

My twin brother quit college to be an architect, to go write music and play music and did that for years.

He has a ton of half written songs.

He's never finished, and conceivably he could put those into.

You can put the half written songs now, just recently into, you know, and it'll it'll finish it for you and it'll do an arrangement and a baseline.

And I'm saying I don't think that would be his song.

It is.

I don't think so.

I think I came up with the melody, I know, but he didn't write the last verse.

He didn't write the bridge.

But but let me but let me, let me give you some, some other examples.

Right.

In the music world.

And I'm a musician also, I have a concert coming up at, October 26th.

What do you not do?

While I'm conducting that concert.

So I'm putting it together.

I don't play football.

Okay, so, so here's so, so so, look, I am using, pseudo and other AI stuff, in music for a lot of things.

So, for example, if I had a melody and I loved the melody, I would often, as a musician, might have to then go to an arranger, because maybe I'm not an arranger or an orchestrator, and I'd have to have them put in.

Strings are put in this, or put in that.

Right.

Maybe that would have to happen.

Maybe I would collaborate with someone and we'd say, I just can't get this bridge right.

Help me with that.

Give me some ideas.

Right.

That's why it would take months to get even.

A demo of a song took months and sometimes weeks.

Well, now I can have a demo of my song, whatever that means.

I'm the creator.

I'm not exactly the writer of every note, but I'm the creator.

And I can have sono do the orchestration for me, place a singer in there and the whole thing sounds so good.

It's the best demo of A.

It's better than going to the recording studio and hiring a bunch of people and doing it right, and then I can have 20 demos of that song in the next hour and decide which direction I want to go.

Then I can rip it apart and change it and move the chorus and move the bridge.

Now, I presented this recently to a group of musicians, and I showed how, for a Broadway musical potential brand new musical, I could go from I don't have an I want song, which a very big, important song in the musical.

It's how we get to to care about the main character, right?

I want, I want to be married, I want to live, I want to get cured of cancer, whatever it is.

Right?

And it brings the audience up.

And I said, I don't have an I want song, so take this script that is written, tell me where the I want song goes.

Who should sing it based on the last 30 years of Broadway hits, what I want songs and then here are the lyrics.

I had a different model write the lyrics for me.

I had GPT five write the lyrics and then I fed the lyrics in to Suno and said, this is what this song represents, and this is the the type of music I want.

This is what I want it to sound like, and it's got to be a Broadway soaring ballad and blah blah blah.

A minute later I had something that was so good in front of people, right?

I did it in front of a bunch of musicians, and the room was still and cut in half, half, half.

The people said, I'm a musician and I have to start using this.

This is a game changer for what I do, right?

And the other half had some people screaming at me, rightly so, saying, I've spent 45 years learning my craft and this thing just did it in, you know, in a minute.

We're losing our humanity.

They might say, oh, I hey, here's what I would say.

Do we use autotune in every single recording today?

Of course we do.

Why?

It cuts down the time in the recording studio by 80%.

We've been using tools.

We've been using EQ, we use all kinds of things like band, drum track, GarageBand, drum track.

We've been using every TV show you watch has digital instrumentation.

There are no musicians playing that.

There's one person that played it all out on a keyboard.

They had literally a day to do a 30 minute TV show.

They send the music off, they got their X $100 for that.

That's just how it's done.

It's been done that way for a decade, right, since we had digital audio workstations.

So we've been using these tools.

They're just getting better, and now we've got a tool that will arrange and create a demo song for us from from our inspiration.

And, it doesn't create things on its own.

You have to inspire it.

That is amazing.

And some people are going to love it.

Some people are going to hate it.

My my word to them was it's here anyway.

Like it's not going away.

Yeah that I get and so I'm so late for our only break.

I wish we had both hours with Kevin to race this remarkable guest who is a true Silicon Valley pioneer and someone who has been really at the forefront of tech innovation AI for decades now.

When we come back from our very brief only break, I will just one more question and sort of the way people use it.

And then I want to get to your advice for the audience when it comes to thinking about their worries about the future of jobs, how you would encourage people to use AI, and why you would encourage people not to think that it's going to wipe out all the jobs of the future.

So let's talk about that on the other side of this only break.

Coming up in our second hour, six months ago, Senator Chuck Schumer convinced his party to help support keeping the government open.

Well, six months of funding is about to run out at the end of this month, and they've got the same decision to make.

Now, the Democrats have seen how the second Trump administration is governing.

What should they do?

That same debate is happening, shut down the government, block what they're doing, or make sure government continues to run.

We'll talk about it next hour.

Support for your public radio station comes from our members and from Mary Carolla Center, supporting residents to become active members of the community, from developing life skills to gaining independence.

Mary Carey, Yola center Transforming Lives of People with disabilities More online at.

Mary Carrie, ola.org and jazz cast pros and easy hosting the Aligned Entrepreneur Pot Athon, a live podcasting event for business owners with a focus on ways to succeed without self-sacrifice.

Saturday, September 27th from 9 to 4 in studio A at Z.

Tickets at Jazz Cast pros.com.

This is connections.

I'm Evan Dawson.

Let me read a couple of emails from Sheila and John.

Sheila says Evan, I'm an I am an English teacher.

Your guest is right.

Language is complicated.

Anyone who's ever tried to use speech to text or answer a prompt, who has any kind, of a speech impediment knows that these little A's are not perfect and sometimes frustrating.

If you substitute the words secretary for AI and taking your draft notes, typing them up and making them make sense, you are having someone else do your work.

It might be aligned with your opinion, but it is not your words.

It's like hiring a ghostwriter.

The problem is, as more people rely on AI for writing, they become dependent upon it and disconnected from the ability to write clearly and concisely.

In engineering, the ability to communicate ideas and plans is just as important as in journalism, and we should not relinquish our own control of our words to AI.

That's from Sheila.

And here's what John writes.

John writes for a moment, I want to grant the idea that I won't replace humans, just offer humans the opportunity to do things better, more efficiently, more productively.

This is where I think people in many fields are too often self-selecting in efforts to understand the impact of AI.

I don't have any doubt that I can make a person smarter, better educated.

You can take what you know in any subject and level up with AI, but that is just not how most people are using AI.

They're using AI to think for them.

They're using AI so they don't have to learn the thing themselves.

Does your guest think that's incorrect?

Yeah.

So, there was a great study out of MIT just on this, that came out just a couple months ago, and it showed there are two kinds of people.

There are people who are using AI as a total cheat, and they're not writing their opinion.

They're not doing anything.

You're saying, give me the following and they're turning it in.

They're not even reading it.

They're not editing it.

They're not doing anything that isn't a good use of the tool.

That will not make you smarter, right?

That will make you and the and the data shows actually they lost their ability to critically think in a matter of months.

It's very fast.

Then there's another set of people that use it like I do.

So AI with AI can output ten times what I used to output in music, in text, in video, in illustrations.

Right.

It's nothing.

I didn't know how to do those by hand, but they took a long time.

And now the amount of output that I've got is incredible.

Now, I am really thoughtful about what I'm putting in, which is hundreds of words of prompts.

Right?

And what comes out.

I read every word and edit it, and what I find is I'm learning from that cycle.

So there's this whole set of people that are getting smarter with AI.

The brain is clicking better, and I think, I think mine is clicking as best as it's ever clicked.

Right.

Mostly because I am producing much more content, reading much more content and, and executing much more.

John says definitely, but that is not how most people in society are going to use AI, and it's going to dumb society down.

Well, I don't know about most because I don't think we have those stats.

It is true that some people are going to use AI as a cheat, just like some people cannot do math anymore at all because they've relied so heavily on Excel and the calculator.

They can't even add, you know, two and five anymore.

Well, and that that piece in The Atlantic that last week, written by a high school student named Ashanti Rosario, she writes about her debate club, which she says has been infiltrated by teams using AI to hone their arguments instead of learning to think quickly and pivot for themselves and more.

Damningly.

She writes that students now work hardest to figure out how to get AI past the sensors, to beat the system and cheat and she concludes with this quote the trouble with chat bots is not just that they allow students to get away with cheating, or that they remove a sense of urgency from academics.

The technology has also led students to focus on external results at the expense of internal growth.

The dominant worldview seems to be why worry about actually learning anything when you can get an A for outsourcing your thinking to a machine?

End quote.

Great question.

That sounds like a five alarm crisis.

Look, I'm on the board of our I.T.

So we talk about this all the time, right?

We see this.

There is no question that some students, by the way, I won't tell everyone how to do it.

It is extremely easy to get the model to ultimately give you something that cannot be detected.

Absolutely.

Just just get past that.

Right.

But but the second thing is, is that the smart professors are saying use AI all you want.

Please use it, because you've got to be ready to be employed and employers are going to want that skill.

However, when you come into the classroom, I'm going to ask you to stand up and defend that in person, that work in person.

If you stand there and defend what you wrote and that's what you find out did, were they really engaged in that output or were they not engaged?

And and here's what I can say.

No one at work, you know, we you all used to write blog posts, let's say.

Right.

Right.

Okay.

No one is going to be expected to hand write a blog post 2 or 3 years from now.

In fact, it might be scorned upon.

It's like.

I don't think so.

I'm not paying you for that.

You have five minutes to write this blog post, not 50 minutes.

Chop, chop.

So?

So some skills are going to be rarely used.

There's there's no question.

So.

But we still want thinking to happen in the blog.

We have to have thinking.

You've got to have critical thinking.

So I think we can get our students to critically think, even if they're using all that to do their homework.

Remember when we used to tell students, don't use a calculator?

Of course they use the calculator, but eventually they would be in class and have to take a test and they couldn't.

And then you found out who really knew something.

There was a time.

Look, when I went to write, a lot of our homework wasn't even graded.

You did it.

However you did it.

You did it with people.

We did it in study halls.

We did it with us.

We did it with calculators, whatever.

And then we had to take a test.

And you had nothing around you but your brain, right.

And the test?

We had five tests in a in a session.

And that's it.

You were graded on the five tests.

And so we're going to go back to that.

So let's let's conclude with this I think the concerns about education you're saying they're valid.

But we're going to see a change in an evolution in how students are tested if we're testing.

Well, yes.

To show that we are not just allowing the outsourcing of our thinking entirely to machines, you can outsource your thinking, but then you better read every word, learn every word, go research.

If it's right, can you defend it?

So there's it.

Just change the work that you're doing.

You might not have written every word, but you're going to have to know every word.

So in our last three minutes, give people a prescription for dealing with the future that they're worried about for their jobs.

They want a job proof their careers or AI proof their careers.

They're concerned that age is coming for their jobs.

Why do you want to leave with people?

Well, you know, job one is become the robot overlord because they'll always have jobs.

Okay.

And this is true when, when we invented the wheel and when we invented Excel, and, you know, again, the people who learned Excel stayed with it, right?

Do you remember when we got PCs on our desktop?

You're maybe a little young, but.

But that didn't happen until the, you know, really the late 80s and into the 90s that all of a sudden everybody had a PC on the desktop.

And then the secretarial pool was gone.

Right?

We actually lost a really good resource there.

Right?

It was gone because we now had to all type our own stuff, and most of us don't even type that well.

But now we had to write.

And so that's what you've got to look at this.

There are some jobs that are going to go away.

There's no question they're really tasks.

But the people who stay become the experts in using the new tools, a variety of tools, right from GPD five to Gemini to things that aren't LMS, but there are 6000 funded companies out there that make vertical oriented tools to solve supply chain and other things.

Look, if you're in the supply chain world and you're not leveraging some supply chain AI, your competitor is right there.

So it's a given that you're going to have to.

And that may mean that you don't need as many analysts in your supply chain team.

But the few analysts that go, I'm going to own this tool.

I'm going to be the expert on it.

They get to stay now.

Again, the outcome of technology has always been like a GM.

When they put in the first robot for the first robot, 1961.

Today they have millions of robots.

They employ more people than they've ever employed.

Why?

Because that by robots sizing their factories, they drove down the cost of a car to, you know, 20 something thousand instead of $200,000.

That's a huge difference.

If you made it all by hand today, no one would afford it.

But at 20,000, almost everyone can find a way to buy or lease a car or get a used one or whatever the case is.

So what happens is technology drives down the cost of products, goods and services.

And as it does, it drives up the demand.

And when it drives up the demand, you need more people, but they're in much different roles than they were in before.

So on a car line, I may not have someone turning that screw or welding that that seam, but I do have literally thousands and a plant of thousands of people in a plant that manage the robots and repair the robots.

Why?

We'll be good.

I don't need you to.

Well, but I need you to know how this darn robot works.

Right?

So the jobs shifted, and that's what we're going to see here to become the robot overlords.

You.

When you are, you're going to be working past your retirement.

Is there a common question?

We got this out.

Is there a safest field to go into?

Yes.

Plumbing, Hvac, repair, electrician these are very important.

And robots can't do them and probably won't do those for a few decades.

Okay.

And you, Kevin told me before the program for kids like my son who want to get into the law.

You said we're not.

We're probably not in our lifetimes going to have only machine lawyers, although you think we actually could.

We could we probably could today and we could have AI judges.

But that's just not going to happen because humans won't accept it.

They want other humans, you know, sitting in that box and helping them.

So, no, you're going to have lawyers for a long time, but once out.

Sad to say, one thing I one thing I did not hear you say endorsed, though, is the idea of giving up on the importance of critical thinking, of leveraging knowledge, just doing it your way, thinking.

Use your critical thinking more, create more, and critically think about what you created and if you can create ten times more and you're critically thinking about everything you are creating in partnership, that's amazing.

I get a bunch of half written songs myself.

You tell me I can use soon to finish them.

You can.

I don't know about that, man.

Send them to me.

I'll use them.

Thank you for being here.

I really appreciate that.

Kevin Torres really is a legend in the field.

I'm not just saying that.

I mean, if you don't know his name and you're not in tech, boy, his resume is remarkable.

It's a great resource.

I really appreciate that conversation.

We've got more coming up in just a moment.

This program is a production of Sky Public Radio.

The views expressed do not necessarily represent those of this station, its staff, management or underwriters.

The broadcast is meant for the private use of our audience, any rebroadcast or use in another medium, without express written consent of Sky is strictly prohibited.

Connections with Evan Dawson is available as a podcast.

Just click on the connections link at WXXI news.org.

The working life of Thurgood Marshall literally changed this country.

He understood how.