There's been plenty of incidents of AI going rogue, with OpenAI, Anthropic, and Meta seeing issues during security testing.
As AI technology develops, researchers are taking steps to understand how it could all grow and change, as well as what separates AI from humans.
Multiple AI labs around the world are part of the study, including University of Auckland computer scientist, Dr Matthew Egbert.
"We've got lots of examples of technology that's not done what we've expected it to - and I think that the AI technology that's emerging is just giving us yet another example of this. It's hard to predict what technology it's going to do, and how society's going to interact with the technology."
LISTEN ABOVE
See omnystudio.com/listener for privacy information.
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:06):
You're listening to the Sunday Session podcast with Francesca Rudkin
from News Talks EDB.
Speaker 2 (00:12):
So in recent weeks, we've seen multiple stories of AI
models breaking out of containment and going rogue. We've seen
top tech companies Open AI and Tropic and Meta all
have issues during routine security testing. It naturally raises the
question just how much is AI capable of International research
is underway to try to figure that out and to
better understand what separates AI from humans. Multiple AI labs
(00:36):
around the world are part of the study, including University
of Auckland computer scientists doctor Matthew Egbert. He joins me
now to explain what they're working on. Matthew, good morning,
Good morning. Okay, let's start. Are you talking me through
this research that you're involved in. What are you hoping
to understand or learn?
Speaker 3 (00:56):
Yeah?
Speaker 4 (00:56):
Great, So, as you mentioned, I'm really interested in this
question about what the differences are between natural forms of
intelligence and artificial forms of intelligence. We're building all of
this technology that's increasingly reminding us of ourselves and its
abilities to do things that we thought only humans could do,
and so I'm really interested in what the similarities are
(01:16):
between how AI does these things, but also and the
way humans do things, but also the difference.
Speaker 3 (01:23):
So it seems to.
Speaker 4 (01:23):
Me pretty obvious that although AI reminds us of ourselves,
it's really a different kind of thing than we are.
It doesn't have the same sort of perspective or the
same kinds of things, like its own goals or desires
that you and I as human beings have, and I'm
really interested in trying to figure out what those differences are.
Speaker 2 (01:41):
We know AI is very capable, but this is about
seeing if it might ever act on its own, like
whether it might actually be able to develop its own goals?
Is that right?
Speaker 3 (01:51):
Yes, Yeah, that's right. Yeah.
Speaker 4 (01:53):
So we've got in the history of technology, we've got
lots of examples of technology that's not done what we've
expected it to and I think that AI technology that's
emerging is just giving us yet another example of this.
It's hard to predict what technology is going to do,
how society is going to interact with technology, And yeah,
(02:14):
what we're seeing now is is very impressive technology doing
things that we don't necessarily expect, and it's really hard
to predict what it's going to do next.
Speaker 2 (02:21):
Do you imagine though, that it might even act on
its own.
Speaker 3 (02:26):
Yeah, it's an interesting question.
Speaker 4 (02:28):
So you know, technology doesn't need to doesn't need to
act on its own.
Speaker 3 (02:33):
Yeah.
Speaker 4 (02:33):
Yeah, it doesn't need to act on its own or
have its own goals to be potentially you know, dangerous,
or to do things we don't want it to. My
personal perspective is that artificial intelligence is really radically different
from human intelligence. That it's not it doesn't have motivations,
it doesn't have goals, it's not deliberately lying to us
or any of these kinds of things. Sometimes it seems
(02:55):
like it's doing that, and other researchers might say, oh, no,
it really is. It really has these properties. It's really
human like. It's acting like it has.
Speaker 3 (03:04):
Its own goals. Therefore it has its own goal.
Speaker 4 (03:07):
I mean, part of the purpose of the project that
we're working on is to try to give a little
bit more context of these kinds of debates back and
forth between scientists that have one opinion or the other,
and try to inform those debates so that we can,
I don't know, draw some more solid conclusions about are
these kinds of things intentional, Do they have goals of
their own, do they have values of their own all
(03:27):
these kinds of really interesting questions. It seems like people
have their opinions now, but how can we inform those opinions?
So it goes beyond an opinion to you know, some
sort of evidence based or theory based perspective.
Speaker 2 (03:39):
The last few weeks we have been hearing about these
cases where AI has gone rogue has been the word
that's been used. But if we look at like the
open AI case, did it really go rogue or did
it just very diligently find a way to do what
it was asked to do?
Speaker 4 (03:55):
Yeah, well that's sort of exactly the question, isn't it.
All of this technology at some level is doing exactly
what we've told it to right. We build the technology,
we design how it works. You know, these AI models
are trained on lots of data, but then they start
doing things which again are hard to predict or hard
to explain why they're doing precisely what they're doing. And
(04:17):
so as humans we like to ascribe agency to things.
So in one of my classes, I build these really
simple robots. They're just a couple of wheels and motors
hooked up to light sensors, and there's like almost no
AI to the middle. They're extremely simple things, but as
soon as you start playing with them and experimenting with them,
you start ascribing agency to them. You think, oh, it's
angry or oh it's afraid, and these kinds of things.
(04:40):
And we really tend to think of other things, even
things we know to be inanimate, as being lifelike or
having intentions or emotional states. And we're doing that also
with AI now, the way we do it with so
many other things.
Speaker 3 (04:54):
And so this idea of going rogue.
Speaker 4 (04:56):
All the language we're using for describing AI is again
this very sort of human anthropomorphic language right where we're
describing it in terms of escaped.
Speaker 2 (05:04):
Yeah, yeah, it's escape. But actually I don't think the
AI never left open Aiyes, machine, it just access some
information it shouldn't have.
Speaker 3 (05:14):
Yes, that's right.
Speaker 4 (05:15):
So we're giving the AI different abilities to you know,
go out onto the internet retrieve information, and now not
just perhaps go out to retrieve information, but to be
able to take actions in a sense, right to do
things online, to cause things to happen. Yeah, And because
it's such new technology, it's not clear what will happen
when we give that kind of path to to these systems.
(05:36):
And it's it's important we you know, think long and
hard about how we give power to these systems and
try to do so in as careful way as we
possibly can.
Speaker 2 (05:45):
And that's a really interesting point. So and it brings
us back to this research. So how important and I guse,
how urgent is this research?
Speaker 3 (05:54):
Yeah, an another good question. In a sense, it does
seem quite urgent.
Speaker 4 (06:00):
And what seems to me particularly urgent is sort of
the idea around caution, around the use of the technology
and trying to be deliberate about what we're doing. But
as I mentioned before, I think all new technology is
hard to predict in terms of how it's going to
work and what it's going to do, and so it's
(06:20):
just important that we think about these things and try
to try to be intelligent about it, even though it's
going to you know, behave in ways that we have
hard time predicting.
Speaker 2 (06:28):
Matthew, I sort of I can't help but wonder, considering
that sort of humans have always been concerned our inventions
could get the better of us for the centuries, should
at least sort of started thinking about this a bit
more earlier.
Speaker 4 (06:43):
Yeah, I mean that would be great, but we have
to sort of you know, sure the cards that we're given.
Speaker 2 (06:49):
I'm sure there's people like you who have been.
Speaker 4 (06:53):
Yeah, I think there are people that have been interested
in the fundamentals of cognition, you know, for thousands of
years and trying to understand these things that we're trying
to study. So, just to be clear, our project is
not explicitly about figuring out how to wrangle AI or
to ensure that it does the right thing. It's more
about again, trying to figure out these fundamental differences between
(07:15):
human intelligence, human minds even and artificial intelligence.
Speaker 3 (07:21):
And yes, it would be great if we had a great.
Speaker 4 (07:23):
Theoretical understanding of these things before we built the technology.
But one of the things that happens is as you
build the technology, you learn about the subject matter.
Speaker 3 (07:31):
Right, There's this strong.
Speaker 4 (07:32):
Connection between building technology and doing research where each sort
of drives the other. So it's very difficult, even impossible
to think it all through and know exactly what you
want before you've done the research, before you've built the technology.
Speaker 2 (07:44):
Yeah, and it's moving so fast, isn't it.
Speaker 4 (07:47):
It's moving so incredibly fast, And I guess this is
maybe one of the things that I find the most
concerning because, as you rightly pointed out, we've got technology
that's given us cost to concern repeatedly in the past.
But what's new or what continue The trend that continues
is this speed of development. Right, We're getting this new
technology that's more and more powerful and it's coming faster
(08:08):
and faster.
Speaker 3 (08:10):
And how as a society do we deal with that?
Speaker 4 (08:12):
It seems like the things that we have set in
place are accelerating that advance, for better or for worse
the other in a society, I think we need to
think a little bit more about how can we slow
that down if we want to or do things in
a more considered way.
Speaker 3 (08:27):
But man, it's a challenging.
Speaker 2 (08:29):
Problem, it certainly is. And is that something that you
hope will come out of this research that maybe it
will help in terms of thinking about regulation and laws
and things for AI.
Speaker 4 (08:38):
Yeah, so we're not going to be directly contributing to
you know, regulation rules or anything like that, but we
do hope to provide some theoretical understanding of what we
mean by these ideas of agency and goals and purposefulness
that allow people to talk about these things in a
more informed way and by understanding the subject matter better
(08:59):
and then hopefully be able to produce better forms of
regulation and laws.
Speaker 2 (09:04):
Doctor Matthew Egbert, thanks so much for talking us through
all that. Really appreciate it. I'm sure that we will
speak to you again soon. It is twenty three past nine.
Speaker 1 (09:13):
For more from the Sunday Session with Francesca Rudkin, listen
live to News Talks it Be from nine am Sunday,
or follow the podcast on iHeartRadio
You're listening to the Sunday Session podcast with Francesca Rudkin
from News Talks EDB.
Speaker 2 (00:12):
So in recent weeks, we've seen multiple stories of AI
models breaking out of containment and going rogue. We've seen
top tech companies Open AI and Tropic and Meta all
have issues during routine security testing. It naturally raises the
question just how much is AI capable of International research
is underway to try to figure that out and to
better understand what separates AI from humans. Multiple AI labs
(00:36):
around the world are part of the study, including University
of Auckland computer scientists doctor Matthew Egbert. He joins me
now to explain what they're working on. Matthew, good morning,
Good morning. Okay, let's start. Are you talking me through
this research that you're involved in. What are you hoping
to understand or learn?
Speaker 3 (00:56):
Yeah?
Speaker 4 (00:56):
Great, So, as you mentioned, I'm really interested in this
question about what the differences are between natural forms of
intelligence and artificial forms of intelligence. We're building all of
this technology that's increasingly reminding us of ourselves and its
abilities to do things that we thought only humans could do,
and so I'm really interested in what the similarities are
(01:16):
between how AI does these things, but also and the
way humans do things, but also the difference.
Speaker 3 (01:23):
So it seems to.
Speaker 4 (01:23):
Me pretty obvious that although AI reminds us of ourselves,
it's really a different kind of thing than we are.
It doesn't have the same sort of perspective or the
same kinds of things, like its own goals or desires
that you and I as human beings have, and I'm
really interested in trying to figure out what those differences are.
Speaker 2 (01:41):
We know AI is very capable, but this is about
seeing if it might ever act on its own, like
whether it might actually be able to develop its own goals?
Is that right?
Speaker 3 (01:51):
Yes, Yeah, that's right. Yeah.
Speaker 4 (01:53):
So we've got in the history of technology, we've got
lots of examples of technology that's not done what we've
expected it to and I think that AI technology that's
emerging is just giving us yet another example of this.
It's hard to predict what technology is going to do,
how society is going to interact with technology, And yeah,
(02:14):
what we're seeing now is is very impressive technology doing
things that we don't necessarily expect, and it's really hard
to predict what it's going to do next.
Speaker 2 (02:21):
Do you imagine though, that it might even act on
its own.
Speaker 3 (02:26):
Yeah, it's an interesting question.
Speaker 4 (02:28):
So you know, technology doesn't need to doesn't need to
act on its own.
Speaker 3 (02:33):
Yeah.
Speaker 4 (02:33):
Yeah, it doesn't need to act on its own or
have its own goals to be potentially you know, dangerous,
or to do things we don't want it to. My
personal perspective is that artificial intelligence is really radically different
from human intelligence. That it's not it doesn't have motivations,
it doesn't have goals, it's not deliberately lying to us
or any of these kinds of things. Sometimes it seems
(02:55):
like it's doing that, and other researchers might say, oh, no,
it really is. It really has these properties. It's really
human like. It's acting like it has.
Speaker 3 (03:04):
Its own goals. Therefore it has its own goal.
Speaker 4 (03:07):
I mean, part of the purpose of the project that
we're working on is to try to give a little
bit more context of these kinds of debates back and
forth between scientists that have one opinion or the other,
and try to inform those debates so that we can,
I don't know, draw some more solid conclusions about are
these kinds of things intentional, Do they have goals of
their own, do they have values of their own all
(03:27):
these kinds of really interesting questions. It seems like people
have their opinions now, but how can we inform those opinions?
So it goes beyond an opinion to you know, some
sort of evidence based or theory based perspective.
Speaker 2 (03:39):
The last few weeks we have been hearing about these
cases where AI has gone rogue has been the word
that's been used. But if we look at like the
open AI case, did it really go rogue or did
it just very diligently find a way to do what
it was asked to do?
Speaker 4 (03:55):
Yeah, well that's sort of exactly the question, isn't it.
All of this technology at some level is doing exactly
what we've told it to right. We build the technology,
we design how it works. You know, these AI models
are trained on lots of data, but then they start
doing things which again are hard to predict or hard
to explain why they're doing precisely what they're doing. And
(04:17):
so as humans we like to ascribe agency to things.
So in one of my classes, I build these really
simple robots. They're just a couple of wheels and motors
hooked up to light sensors, and there's like almost no
AI to the middle. They're extremely simple things, but as
soon as you start playing with them and experimenting with them,
you start ascribing agency to them. You think, oh, it's
angry or oh it's afraid, and these kinds of things.
(04:40):
And we really tend to think of other things, even
things we know to be inanimate, as being lifelike or
having intentions or emotional states. And we're doing that also
with AI now, the way we do it with so
many other things.
Speaker 3 (04:54):
And so this idea of going rogue.
Speaker 4 (04:56):
All the language we're using for describing AI is again
this very sort of human anthropomorphic language right where we're
describing it in terms of escaped.
Speaker 2 (05:04):
Yeah, yeah, it's escape. But actually I don't think the
AI never left open Aiyes, machine, it just access some
information it shouldn't have.
Speaker 3 (05:14):
Yes, that's right.
Speaker 4 (05:15):
So we're giving the AI different abilities to you know,
go out onto the internet retrieve information, and now not
just perhaps go out to retrieve information, but to be
able to take actions in a sense, right to do
things online, to cause things to happen. Yeah, And because
it's such new technology, it's not clear what will happen
when we give that kind of path to to these systems.
(05:36):
And it's it's important we you know, think long and
hard about how we give power to these systems and
try to do so in as careful way as we
possibly can.
Speaker 2 (05:45):
And that's a really interesting point. So and it brings
us back to this research. So how important and I guse,
how urgent is this research?
Speaker 3 (05:54):
Yeah, an another good question. In a sense, it does
seem quite urgent.
Speaker 4 (06:00):
And what seems to me particularly urgent is sort of
the idea around caution, around the use of the technology
and trying to be deliberate about what we're doing. But
as I mentioned before, I think all new technology is
hard to predict in terms of how it's going to
work and what it's going to do, and so it's
(06:20):
just important that we think about these things and try
to try to be intelligent about it, even though it's
going to you know, behave in ways that we have
hard time predicting.
Speaker 2 (06:28):
Matthew, I sort of I can't help but wonder, considering
that sort of humans have always been concerned our inventions
could get the better of us for the centuries, should
at least sort of started thinking about this a bit
more earlier.
Speaker 4 (06:43):
Yeah, I mean that would be great, but we have
to sort of you know, sure the cards that we're given.
Speaker 2 (06:49):
I'm sure there's people like you who have been.
Speaker 4 (06:53):
Yeah, I think there are people that have been interested
in the fundamentals of cognition, you know, for thousands of
years and trying to understand these things that we're trying
to study. So, just to be clear, our project is
not explicitly about figuring out how to wrangle AI or
to ensure that it does the right thing. It's more
about again, trying to figure out these fundamental differences between
(07:15):
human intelligence, human minds even and artificial intelligence.
Speaker 3 (07:21):
And yes, it would be great if we had a great.
Speaker 4 (07:23):
Theoretical understanding of these things before we built the technology.
But one of the things that happens is as you
build the technology, you learn about the subject matter.
Speaker 3 (07:31):
Right, There's this strong.
Speaker 4 (07:32):
Connection between building technology and doing research where each sort
of drives the other. So it's very difficult, even impossible
to think it all through and know exactly what you
want before you've done the research, before you've built the technology.
Speaker 2 (07:44):
Yeah, and it's moving so fast, isn't it.
Speaker 4 (07:47):
It's moving so incredibly fast, And I guess this is
maybe one of the things that I find the most
concerning because, as you rightly pointed out, we've got technology
that's given us cost to concern repeatedly in the past.
But what's new or what continue The trend that continues
is this speed of development. Right, We're getting this new
technology that's more and more powerful and it's coming faster
(08:08):
and faster.
Speaker 3 (08:10):
And how as a society do we deal with that?
Speaker 4 (08:12):
It seems like the things that we have set in
place are accelerating that advance, for better or for worse
the other in a society, I think we need to
think a little bit more about how can we slow
that down if we want to or do things in
a more considered way.
Speaker 3 (08:27):
But man, it's a challenging.
Speaker 2 (08:29):
Problem, it certainly is. And is that something that you
hope will come out of this research that maybe it
will help in terms of thinking about regulation and laws
and things for AI.
Speaker 4 (08:38):
Yeah, so we're not going to be directly contributing to
you know, regulation rules or anything like that, but we
do hope to provide some theoretical understanding of what we
mean by these ideas of agency and goals and purposefulness
that allow people to talk about these things in a
more informed way and by understanding the subject matter better
(08:59):
and then hopefully be able to produce better forms of
regulation and laws.
Speaker 2 (09:04):
Doctor Matthew Egbert, thanks so much for talking us through
all that. Really appreciate it. I'm sure that we will
speak to you again soon. It is twenty three past nine.
Speaker 1 (09:13):
For more from the Sunday Session with Francesca Rudkin, listen
live to News Talks it Be from nine am Sunday,
or follow the podcast on iHeartRadio
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by Audiochuck Media Company.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com