September 18, 2025 • 48 mins
Andra Keay, Managing Director of Silicon Valley Robotics, has been at the forefront of robotics research, commercialization, and policy for decades. A self-described “techno-pragmatist,” Andra has her pulse on the robotics industry, in the San Francisco Bay Area and around the world. Importantly, she is conscious and intentional to try and get ahead of the big issues we face down the road with the societal shifts coming as millions of robots, humanoid and other form factors, populate more of our world and interactions.
Andra is pro RoboTopia, and against an AI apocalypse. And robotics offers an insight into the world of AI, as they are AI with arms and legs, removed from the black box behind the screen, but out among us, where the consequences of a hallucination or miscalculation can have physical implications. Industrial robots were behind barricades and safely screens. Next gen Humanoids will work in homes, senior care facilities, factories, and other places with direct interaction with people.
Andra is a frequent industry speaker, and in fact, we’ve shared a few panels together over the years, but this is the first time she’s visited Turn The Lens so we could really get into it without restrictions.
And the timing couldn’t be better. Humanoid robots are having their moment. Figure just raised a $1B Series C, with a $39B post money valuation. Customers are moving from pilots to commercial engagements. Capacities are compounding at an exponential rate.
Oh Yeah, and did I mention LLMs and their like have transformed the robot training paradigm.
The next great wave of technological change is upon us. Humanoid Robotics, AI with arms, legs, and an ability to navigate the world, and do things.
If you follow no one else, follow Andra to keep up on this part of our rapidly changing world. Subscribe to Robots & Startups on Substack.
Andra Keay: Robo-Pragmatist, Humanoids, Technological Shifts, Laws | Turn the Lens with Jeff Frick, Ep42
#AndraKeay #Humanoids #Robotics #AI #TechnologyShifts #Automation #Ethics #RoboticsLaw #TurnTheLens #JeffFrick #FutureOfWork #Innovation #TechPolicy #Simulation #GenerativeAI #GenAI #RoboticsIndustry #FiveLaws #Society #Interview #Podcast #TurnTheLens
YouTube -
https://www.youtube.com/watch?v=lp46AO7aC_4&list=PLZURvMqWbYjk4hbmcR46tNDdXQlrVZgEn
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
So I will count us down
and we will go
in three,two, one.
Hey welcome back everybody.
Jeff Frick here
coming to you foranother episode of
‘Turn the Lens’ and I’ve got
what’s turning out to be apretty frequent guest.
What's interesting is
I've got to interview her
a lot of timeson other shows
and this isthe first time
I've actually had her on my show.
I think it's likethe fourth time
I've interviewed her.
(00:20):
So I'm really excitedto welcome in
through the magicof the internet
She's Andra Keay
the Managing Directorof Silicon Valley Robotics.
Andra, great tosee you.
So good to be here Jeff.
And you’re right, too manytimes on other stages.
It's really good tobe here with you.
It took a while
but we finally got ittogether so
I have a longlist of topics
which I always haveevery time we talk.
(00:42):
And I don't know if it was you that said it
or somebody said it that really
‘Humanoids are having a Moment’ right now.
What is going onon the humanoids.
You know, we've
they've always kind ofbeen there.
They're kind ofspacey and
you know, they're kind ofthe cool things
that people dreamabout and draw cartoons.
But what's happeningin humanoids
that’s so specialright now that's
different than whatit's been in the past.
(01:02):
Well once again, it’s taken 20 odd years
for humanoids to become an overnight success
but we really needed a certaincritical tipping point to reach
for humanoids to have this moment.
And those of us that have been enthusiastically following
companies like Agility Robotics, Apptronik
(01:23):
and some of the
1X and some ofthe other companies
from Firia, UnitreeDeep Robotics.
We've watched them
get commercialtrials
and in the caseof a couple of them
go from commercial trials
to actual commercial contracts.
(01:45):
And now that isa significant step.
And now that's openedthe doors to say
there may be a businesscase for humanoids now
that wasn'tpreviously possible.
It's possible now.
And here's the thing.
If you remember what a mobile phone looked like in the 19
(02:08):
oh, let's see, I think it was around
I was working infilm and television
and I think I got my hands on
one of the first ones
that was a kindof a mobile car phone
in around 1987.
[Andra] And it was[Jeff] That’s about right
I had the big giantMitsubishi brick.
You could jam it in your car.
You could pop outthe brick
(02:28):
and bringit inside
and plug itinto the wall.
Now, they never got bigger, heavier or more expensive
And that's what we need to think about
with humanoids today.
They are never goingto be any worse
than the current stateof technology.
They are going tocontinue to improve
they're going toget more capable
(02:50):
and they're going toget more affordable.
And these are the thingsthat are going to mean
it's not just 1 or 2 things
that humanoidrobots can do.
They're goingthey're here to stay.
Because the range of use cases for them
is pretty broad.
I think I've come up withabout 11, maybe 12 different
(03:10):
business modelsbusiness use cases
for humanoid robots.
And part of it is what you might say
being labor.
But a robot is capable
of more than simply
you know, lookinglike a human
and being a human replacement.
You know, in some ways
robots are far lesscapable than humans
(03:33):
but in other ways they are far more capable.
They're capable of doingrepeatable precision tasks
and now they are capableof being flexible
[Jeff] Right[Andra] as well.
So there's a rangeof different activities
where we can start to augment.
It's about doing somethingperhaps better
doing somethingin a way
(03:54):
that we weren't able to do before
and about being ableto do it affordably.
And the big reasonthat humanoids
are very appealingto many people
one anthropomorphically where we're just predisposed to
find things that are like us more appealing
Right, right.
But we also don't need to
create infrastructureand change our world.
(04:16):
The primary reason for creating a humanoid robot
is that it can usehuman tools.
It can operatein human
spaces designed for humans
operate equipmentdesigned for humans
and it's also much easierfor humans to train them
whether it's throughremote teleoperation
having that direct kinematicmapping to the shape.
(04:39):
So this translation
if my handsdo this
then that robot's hands do this.
If I reach upit reaches up
and it's really a parallel mapping.
And that makes humanoidsbetter in many ways
than saying, no let's create a custom built purpose robot
(05:00):
that could potentially do thisjob much, much better.
And these are thethings that you know
I see as being moreof the future thing
when we're talkingmaybe about
creating completely lights out operations of something.
Right
Then we wouldn't need it to be
if we’re custom designing
the full automation
the full building,the full work flow
(05:23):
then no, we wouldn't need to have it
being a humanoid,
but there's clearlycases for both
and I think that there'ssignificant demand
for both sorts of robotic automation
Right
humanoid and non humanoid.
And with suchincredible demand
I predict that there will be1 million humanoid robots
(05:44):
in commercial operation by 2030
And that's simply based onprojecting the plans of the
top half dozen or so
humanoid companies,the ones that are
seriously gettingtraction and
some of the analystshave been looking
further down the trackto 2035, 2040 and beyond
(06:07):
and seeing an incredible step change
like it might take us
five plus years
if we count perhaps the lastfive years of development
10 years to have gotten to1 million humanoids
it will onlytake us
In 10 years, we'll probablybe at 100 million.
Right, right.
So wasit just a
(06:28):
was it just a combination of kind of
classic technology curve effects
on the price of hardware
the sophistication of software
the speed of processors
and the connectivity of cloud
just kind ofall those things
you knowkind of moving
that it's gotit to this.
Or was there
was there a singleevent or was there a
a Cambrian event, I think you like to use the phrase that
(06:50):
that happened to changekind of the trajectory.
Certainly for roboticstechnologies in general.
There's been a Cambrianexplosion event
that we startedto see in the world
from 2010 onwards.
It's been buildingfor quite some time
but it's what I callthe ‘Robotics 2.0’.
(07:10):
It's about theability of robots
to do real worldreal time navigation
and you can see that as being
navigation like an autonomous mobile robot.
Sidewalk delivery robot
mobile robot in a factoryor self-driving car.
But you canalso see it as
three dimensional navigation
like a drone, somethingin the air or underwater,
(07:33):
or a robot armmoving in all directions
but capableof responding
sensing what is actually in space around it
and respondingin real time
if there's somethingout of plan.
So, for example
if a pedestrian comesin front of the vehicle
if somebody reachesout their arm
(07:55):
to stop therobot arm
any of thesethings that are
interrupting theplanned action
now a robotcan sense it
think about whatneeds to happen
to avoid to prevent collision
or just to redirectaround something.
Could be as simple as
(08:15):
bags of things fallingfrom the shelf.
Right
But previously
any robot technology wasonly as good as
‘Stop! There's an obstacle.’And then you had to wait
for some kind of helpto reset or restart.
Right?
Now it's not just aboutstop and go,
it's about go around.
It's about make changes in plans
(08:38):
in a real world,real time fashion.
So that is a fundamental step change
in terms of competencies
that have opened the doorfor an incredibly wide range
of use cases forrobotics technologies
and that's robotics technologiesacross the board.
What you were askingbefore that, though, was,
is there, that same thingfor humanoids?
(09:01):
Certainly they needthe ability to navigate.
Arguably humanoids are doingmore complex navigation.
The bodies of humanoidshave far more actuators
far many pointsof movement
and there are robots
that are movinginto social situations
in some casesfor humanoids.
(09:21):
And we're still not really greatat working that out.
So, there’s a lot of complexitiesahead for humanoids.
But the tipping pointscame on us really quickly.
There have been a whole sea of things
Moore's law in terms of computing technology.
The same thing having an impact on sensing technology,
the same thing happening in terms of
(09:42):
just us knowing howto build robots better
so that we can now builda smaller, far better robot.
And the role that AI andsimulation have played.
And so what I see now is it's been
exponential growth meetingexponential growth.
Right, right.
That was this incrediblespeed up in competencies.
(10:06):
So as roboticshas met
generative AI
or large scale learningand large scale
not just language modelsbut multi modal models.
Right.
And that’s very muchimportant for robotics.
And that hasmeant simulation.
So what we're seeing now
and this goes back to say Google arm farm
(10:27):
in the 2014 period where
a whole roomful of robot arms working
individually on the same tasks
would come up with
a whole lot more scenariosthan one robot arm.
But then those same strategies
(10:51):
would be tested in simulationand through AI
and that would multiplythe range of options
and the abilityto find the most
successful pathways out of them
which would then be fed back into the real
action of the robot armsthe next day
So you had this to and fro happening
between what the robots were exploring during the day
(11:13):
informing what theywere learning
and trainingduring the night.
Right.
And then going backinto the real world
during the day.
So you startedthis incredible
flywheel of feedback loops
and we've just multiplied
in a sense the number of feedback loops
and the transitions from simulation to real world
(11:33):
back to simulation to thereal robotics deployments
back to simulationand so on.
So it's compounding.
We're seeingincredible
exponentialgrowth compound.
Lots of greattopics here.
Lots of greatpaths of
of development.
So let's just jump ina little bit deeper
to the one you just touched on
which is the change in training.
And as you
(11:54):
you've kind of touched on briefly
we dig in,there's two things
that I saw at least at the show.
One is kind of LLMsand not necessarily
always largelanguage models,
but kind of the roboticsversion of those
for different typesof activities
changing the way thatthese things are trained.
And like you saidit's about
variability to change and being able to adapt
(12:15):
to not exactly follow the scriptand just say yes or no.
And then the other one is this
as you mentioned briefly
kind of this teleoperation piece
where you're integratinga human's behavior
into the training processto accelerate.
So very different way than
you know kind of a classicfactory implementation
(12:35):
where you're justmapping out,
you know, the pointat the end of the
laser welder to put a car together.
This is a really different way
to think about training
and opens up both speed and flexibility of the training.
It’s completely different.
Absolutely.
And there's several different waysthat training is happening
for roboticsand
(12:58):
I've been listeningwith interest
to differentdiscussions around
is this the right way?
Is this the way that
training is going to behappening in the future?
And at the momentI would say
the general agreement is that
we need all ofthese methods
and quite possiblyall of them together.
So there's real worldteleoperation
(13:20):
then there's having a person beside the robot
doing pose andmovement training.
There's programingtraining
there’s using Englishto instruct
there's trainingfrom videos
there's trainingthrough simulation
and there's training through
(13:40):
extracting ideas from
a range of other waysthat you’re mining data
around how these thingshave happened.
And it's opening
I think many people's eyes
to the fact that
we have very few.
We have limited insights
(14:01):
into how humansoperate in the world.
You know we've been able to
take manygreat strides
with artificialintelligence
through text based learning.
Right.
Even if we count videobased learning and
there's very much a limitto what's on the internet
about a whole range of things
like how a particular job
(14:22):
happens in a factory for example.
And we're onlyjust starting
to see some companies instrument
things like production lines
to learn exactly howdifferent tasks are done
and what'sthe range
between a task beingperformed successfully
or really, really well
and being performednot up to scratch?
(14:43):
And how we can learnacross those things
and improvethose outcomes.
It's a little bitlike when
robotics wentinto warehouse
into logisticsand factories.
There was a lot of feedback that
within theUnited States
up to 75%of factories
were not digitized
in terms of their inventories and operations.
(15:03):
It was kind of movingfrom an Excel spreadsheet
to a printout to somebody recording things
on clipboards to transferring them back in.
There wasn't a kind of
a digitizedreal time
base of knowledge
and a lot of these thingsyou needed to build
in some of thesefoundational steps
(15:24):
before you could go andadd more sophisticated
automation or autonomousmobile robots, for example.
Right.
There are someinstances where
the robots are able to step
past the lack ofprevious digitization.
But in terms of training
one of the big reasonsthat I think
humanoids are really having the moment
(15:46):
is because they'regoing to be ideal
platforms fordata collection
about doing all of these human type tasks in the world.
And they'll be developingthis digital library
of what is the rangeof how you do things.
And even simpletasks for example
the number of different door closures
(16:09):
that there are the worldis almost infinite.
They open outwards,they open inwards,
they open sideways,they open dual doors.
You push, you turn, you clip, you [psst]
There are so many differentways that we as people
[Andra] can navigate doors[Jeff] Right.
That's such abasic thing.
(16:29):
And yet collecting enough of that information
to have the sort of
materials that we need to do
successful training.
Well, maybe we do needrobots out there
jut exploring and discovering
a whole lot ofthose edge cases
that they're still having trouble with.
Right, right.
(16:50):
So, you've covered ethicsa lot in your time.
You've spoken on ethicsand robotics a lot
and some peoplemight say
robots are kind of theembodiment of software
or some people might say these days
it's AI with arms and legs.
As the robots move from the factory floor
out into this world
what are some of theethical considerations
(17:11):
both historicallythat
that have notreally changed,
but really noware new things
now that they've left the confines
behind the yellow tapeand the glass doors.
That’s a reallygreat question.
And sometimesI like to say
I'm on the side of robo-topia
because I don't wantto be on the side of
AI Apocalypse.
(17:31):
I can see a lot of thingspotentially going wrong
in the use of AI
and one of the reasonsthat it is perhaps
what I consider to be more dangerous
is because at some levels
it's cheap and easy and invisible,
and so wedon't see
(17:52):
what might be happeningunder the hood
and therefore we are notpaying the attention
that we need topay to it.
Whereas robots are inherentlymore visible being physical
and they cost moreto deploy in the world
we're not going to be
spamming the worldwith robots
and that makes ustake more care.
(18:14):
So I think thatthere are inherently
some guardrailsbuilt into robotics
that AI might not have.
And yet there is an incredible overlap
between the issues.
And oftentimes peoplethink of robotics only
in terms of physical issues.
Where as
we’re going to see the same
financial and emotional issues as well.
(18:37):
One of theareas that
really got mestarted in robotics
and that I am mostpassionate about
is the impact of thetechnology on society
and on all partsand levels of society.
There are ways that we can address
these thingswith, say
ethical or user centered design.
There areother ways
(18:57):
we're talkingabout
democratizationof technology
getting it out intomore people's hands.
But fundamentallywe also need to
broadly understand whatare the potential issues
of this technologymoving into society.
And we don't needto be too scared
because we've dealtwith new technologies
(19:18):
multiple timesas societies.
And there are
guidebooks forthe sorts of ways
that we as society deal
with majortechnological shifts.
I talk about it veryoften in terms of
here are my ‘5 Laws of Robotics.’
and they'rethe antithesis
(19:38):
of Isaac Asimov's‘3 Laws of Robotics’
which many peoplestill look to.
They express our needsand desires from robots.
We don't want robotsthat can harm us
or harm anybody else.
Those particular rules thatAsimov came up with
I always thought that they were examples of rules
that you could neverdeploy successfully
(20:00):
because all of his stories
were about how they went wrong.
But a lot of peoplestill look to them as
these are thesorts of rules we need.
It reflects on thethings that we want.
The things that are our priorities
but it doesn't reflect on
Pragmatically, how wouldwe actually go about doing this.
(20:20):
And how is this possible
with a roboticstechnology?
So I like to consider myself
a technopragmatist.
Not a techno optimistor a techno pessimist
but I believethat
we're going to be moving forward
technologicallyspeaking.
it's almostan imperative
but that doesn't mean
(20:41):
that we should do it thoughtlessly.
I want to beone of the people
that's movingforward
helping to put a brake onwhen it's needed
in the right spots.
And I want to sharethat information
and that abilityto see
where the potentialproblems are.
So my ‘5 Laws for Robotics’
start by soundinglike Asimov
because I sayrobots should not kill.
(21:04):
The thing is
this is one of our mostfundamental laws, rules
moral principles across society.
So we actually haveframeworks to help us
monitor that or control that.
We have laws.
If a robot is going to be acting lawfully
(21:24):
then it would notkill people.
Physical safety, though isjust the starting point.
See, the next thingfrom that is
robots should be designedto be law abiding.
Now that starts to raise the question of
but whoseresponsibility is this?
Is thisthe responsibility
of the peoplebuilding the robot?
The peopleoperating the robot
(21:45):
or the peoplearound the robots?
You know, where is theresponsibility sitting?
So we need robotsto be abiding by laws
and we have a lot of past record of this.
For example, Who
where is the liability foran automobile accident?
And we've come upwith different ways
of assigningresponsibility
(22:07):
based on thecircumstances.
Have we modified thatfor Waymos though?
I was gonna
it's funny youjust said that
because I was going to say,
you know, if a Waymogets in an accident
as a proxy robot
now you've got who wrote the software
who builtthe thing.
I mean, you got a wholenother layer of complexity on
whose responsibility.
Be curious. I don't know if that's
(22:28):
gone to court to kind ofhave some, precedent set
or maybe it has.
Many court cases.
One was in an Uber vehicle driving autonomously in Phoenix
and a pedestrian was killed
Now, in this instance,the safety driver
who was there as the responsible person
(22:48):
was the one that washeld accountable.
But I think that was largely a victor
victory for the
companies with the best paid lawyers
versus what would be repeatable case law.
Right.
And there's no
and there’s no safetydrivers anymore.
We've moved pastsafety drivers.
And this does mean that
people in the legal professionand legal scholars
(23:10):
have not been thinkinglong and hard about this
[Jeff] Right, right[Andra] in the last 20 years,
And really diving into it
Where we'reat is that
these things haven'thappened frequently enough
for their to besome more
established, precedentsand understandings.
Go to numberthree.
The third law.
Robots need to begood products
because the samething happened
(23:30):
in a non-fatal accidentin San Francisco
and it caused the completeshutdown of Cruise,
which was GM's self-driving vehicle play
It killed the company.
Yeah. Well.
Arguably in that case
they weren't ready for prime time
because when the data came out
on the numberof interventions
(23:52):
they were so farfrom autonomous
their intervention ratewas like once
every two miles or something.[actually 4 - 5 miles]
It was like
you guys aren't readyfor prime time.
Put those things away.
Without taking
without kind of litigating that one further
becausemaybe they shouldn't
have been ableto be testing.
Right, right.
Maybe we should be talking towhoever was certifying that
or allowing that.
But these thingscan become kind of
(24:14):
Jenga towers and
All I'm saying really is that
there is this kind of hierarchy.
If you startby saying
robots should not harmshould not kill
they shouldobey the law
then that's talking about where
law is one of ourmajor frameworks
for dealing with this.
And then if we talk about
they need to be good products
(24:34):
then we recognize that tobe commercially viable
there is a really strong incentivefor robotics companies
to be incorporatingthis early.
And certainlyI see this
with the roll outof humanoids.
The CEOs of the humanoid robot companies
want to see standards.
They want to see benchmarks for what is safe
(24:55):
They want to be able to assuredeployments and customers
that they are as good as
is reasonable to expect
Right
from this technology
that they're
they're doing everything right.
So they want tocreate these benchmarks
so that the industrycan move forward
and so that theyas companies
(25:15):
can be successful.
Certainly a dependency right
for masscommercialization.
Absolutely.
And this is where I say my laws.
It's about technopragmatism,
because each ofthese laws
speaks to how and where
we should look tobe implementing this.
But the next tworeally deal with
financial andemotional harm,
which we rarely talk about when we talk about robots.
(25:37):
And yet to methey are potentially
deeper, darker problems waiting for us
if we don't factor that in.
And yet it really plays intobeing good commercial products
Robots should be identifiable.
And to me that means that any,
any mobile robotis like a vehicle
(26:00):
and it needs to havean identification number
it needs to havea registration.
And then you can look at
what the process isthat go behind that.
And we accept thatin other vehicles
if it's out there
then it has to go throughcertain licensing
Right
and registrationprocesses.
Now then you goone beyond that
(26:20):
and it speaks to theyshould be transparent
and they should be truthful
and truthful
not just speakingthe truth
but transparentas in
is this robot speaking to meas an autonomous entity?
Is this a remote operatorspeaking through?
Is this a commercial script?
(26:41):
And it's designed forwhat particular outcome?
If a robot is being nice to me
we are pretty goodat starting to assess.
Well, we're okaywith assessing
if someone's being nice to me.
What doesthat mean?
What are theygaining from that?
And potentially,how are they
trying toexploit me?
(27:02):
Right.
Is this is harmful for me?
Now with robotics
just like with AI only that's more invisible.
There are many levelsat which that interaction
can be being steered orguided for other purposes.
Right.
And that can be assimple as saying
robots and AI are going to be really good at
being salespeople.
(27:23):
They might knowour particular
interests in history
or they mightbe able to
sense and detectour eye gaze
and our interest levels
things like pupil dilation
all of thosesignals of interest
that as people we oftenunconsciously pick up
these things can be picked up by
(27:44):
by camera technology today.
They might alsoknow exactly
what to offeras the next step.
The advertisingindustry is really good
at coming up with sort of psychological profiles
of who arethe people
that they want to have buying things
and how toappeal to them.
I think we need to applyyour rule number five to
(28:06):
and number four,actually to social media.
Maybe we wouldn't have
some of the problems that we have in social media
But it’s funny, I can't help but think, right.
One of the bigpotential markets for
the humanoidsis senior care
and taking careof people
in assisted livingsituations or whatever.
And, you know, I can seepeople with dementia
(28:27):
asks, you know, where Susie,
and Susie might be dead.
Susie, you know, maybewas an old friend,
you know you start thinking about who's
what are the differentpriorities for the robot
and the caregiversin that situation
in defining the answerto that question
both to be transparentbut also to be
(28:47):
careful, safety
appropriate for whateverthat medical situation,
may require, which might not be just
bald faced, you know,black and white facts.
And I think that there are a lot of parallels in health care
and the way that we'vedeveloped guardrails
(29:08):
in that so things likeinformed consent
and developing consent plans
ahead of timefor example
in some situationsyou could see that
a medical providermight say
it is the betteroutcome for this
particular personsay with dementia if
there is a robot caregiver
(29:29):
that can take on whatevercharacter they think it is
and that will keep themfar more comfortable.
Or there could be a casewhere there is a strong
desire from the personas an individual to say
I never want to be lied to regardless
and that that mightthen take precedence.
(29:50):
And we need to work these sorts of things out.
But this is reallygreat because
I did the five laws
and this is based on some of the best think tanks around
is some of the best work.
And I’ve taken part in a lotof these different activities.
Most of themI kept thinking,
how on earth are youever going to apply
these ideasor ideals?
(30:11):
And I very muchlike the direction
that this Five Laws are in terms of
pragmatic abilityto be deployed.
But I went beyond that and I went
How would I thinkabout that?
And I've got five things thatI think could be answers
One is we talked about already
the robot registrylicensing.
(30:34):
A second one
we've talked aboutin AI ethics cases
which is algorithmictransparency
having things like model cards
having things where the hidden workings
have to be
detailed and accessible.
If people need tolook up these things.
A third one
taking fromhealth
(30:55):
is to have independentethical review boards
and maybe it meansthat you're not looking
at something on acase by case basis,
but you're saying
in these circumstances
the best outcomefor people
is if a robot is allowed to behave like this.
Whereas in all of theseother circumstances
it's far betterif a robot
does the opposite.
(31:16):
Right
So we'll build upthese things if we have
I think, independentethical review boards.
Beyond that,I think there could be
a very interestingrole for the idea of
a ‘Robot Ombudsman’
to represent concerns that
individuals and or groups of society have
that they don't have
(31:38):
necessarily theability to voice.
They don’t necessarily know who to take these concerns to
or may not want toor be scared to.
But if you can resortto an ombudsman
who can thensay aggregate
these concerns and say, okay
maybe asa state
(31:59):
or as a companyor as a country,
this needs to becomesomething
that we develop policyor legislation around
because it is having an impact
on parts ofour society.
And we we're lettingtheir voices be heard.
Finally, I thinkrewarding examples
(32:21):
of what good robots are
making it something that's desirable
for companiesto create.
If we have ways of
let's not just punishpoor behavior
or poor corporatebehavior
but let's reward good robots
good robot design
and good applicationsof this technology.
(32:45):
Those arelofty goals.
I hope we do a better job
than we've done againon social media.
And the other one is privacy
where, you knowwe've just not been
not been diligentin keeping up with
you know, somethingas simple as
the cookiepolicies.
And, you know, you getGDPR in the EU
where they get a whole bunch of countries
that can cometo an agreement
but we still fall backon our fundamental
(33:06):
states versus nationalhead banging
and can't evenget a national
‘Disclosure When Breached’ regulation done
So we’llfingers crossed.
At least you gota framework which I like.
Getting ready for thisI came across
one of your older interviews
and you talked about robots
being the AI canaryin the coal mine.
(33:26):
So what is specialabout robots
and their connectionwith AI that gives them
the potential togive visibility
into positivesand negatives
that maybe, as you said
were kind of hiddenbehind a screen
inside of a computer before.
Exactly.
Robots are the physicalembodiment of AI
(33:47):
and becausethey're so visible
and because theyare capable
of physical damage
we have much greater.
We take a lot more care
with what's happeningwith robots and
so I think that will allow people to understand
(34:10):
what some of the problemswith AI might be,
where AI is that invisiblepotentially toxic gas
that might beall around us
and we might notbe aware of it.
And because robots are the
physical embodiment of AI
when we see things happening
that we don't like
(34:31):
it can inform us as to where these things
might also be happening invisibly.
It's really interestingthat, you know
we reactmore viscerally
and maybemore aggressively
to physical harmthan emotional harm
because you can see it, right?
They can put a picture of it
on the frontof the newspaper.
If there's a crash
if there's a crash inan autonomous vehicle.
(34:52):
Where the emotional harm
and some of the other problems
that can come again
just picking onsocial media
because I like topick on social media.
Aren't necessarilyas visible
kind of in your face.
So maybe they don't get thepriority that they should have
in terms of addressing them.
So a very different situation.
You're all over the place
in Bay areaSilicon Valley Robotics.
(35:13):
You got a greatnewsletter.
You keep up on alot of good things.
What's happeningin the funding world?
Has the fundingfigured out that
this is not yesterday's robots
and the opportunityand the technology
has progressed to a point where
they shouldn't be making a comp to
you know, when they were looking
at the little Sony dog
or Pepper or some of those early
cute iterationsof things.
(35:35):
A lot of the investmentcommunity has
worked out that
not only robotics isready for investment
but also
a lot of otherdeep tech.
And what I thinkis most exciting is
is in the lastten years.
And this is in spiteof there being a
whole lot of obstacles in
(35:56):
funding in venture at the moment.
But we're seeing acompletely new class
of investor, and that's the engineer,
that's the scientist,that's the investor
that can have
better appreciation
of a complex systemthat they're investing in.
(36:17):
And if youlook at
the background
experience and educational background
of investors that were ableto do really well during
I suppose, the riseof social media
and the internetand smartphones
many of them were
(36:38):
coming from financial orliberal arts backgrounds
as opposed tobeing engineers.
You didn't necessarily need to understand the technology
to understand what might bemaking good business cases.
Whereas now I think
there's so much of this
(36:59):
deep tech and roboticsis a great example of it
because roboticsis a complex system
and it really is something that is new
is new as this really complex robotics 2.0.
So we need to haveinvestors that are capable of
(37:19):
understanding thecomplex systems.
Now, I don't mean that everysuccessful investor in robotics
necessarily has aphysics degree
or has had a backgroundas an engineer,
but there's certainly a lot more investment firms
that are focusing on these technologies
that do have a strongbackground in those areas.
(37:41):
And this is actuallymore similar
to the first waveof venture capital
in the semiconductorand early computing days.
The deeper techkind of
kind of deeper techfoundational investment.
Another thing I wantto get your take is
I know you have an opinion on it is autonomy
And, you know,you said earlier on that
(38:01):
one of the bigbreakthroughs was when,
you know, a robot couldmake a course correction
when there was somethingthat was interrupt
interrupting its standard path.
And for mejust in terms of fun
when Skydio introducedthe autonomous drone
and we've talked about this before
where I no longerhad to actually
pilot the drone,
but now I'm reallyinstructing the drone
as to whatI want it to do,
(38:21):
whether that'sin a program
or coordinatesor mapping it out
whether it's, you know, inspect a
a transmissionwire
or inspectan ugly factory
that I don't want toclimb all over the silos
but it changedthe relationship
with the operationof the thing
where now it’s you're
you're giving it instructions to fulfill a task
as opposed toactually operating.
(38:42):
Where do you see autonomy
and how is that goingto change things?
And then I just
just to throw it inwith Waymo
was that a long timeor a short time
they’ve been working onautonomous vehicles.
You knowit's like
they've been talking about it
well guess what.
It's here.
At least30 years.
Is that short or long?
I think it's appropriatefor the difficulty
(39:02):
that we're facing.
And I think it'sit's not a mature tech
by any means, yet.
But it's starting to be
let's see, it wasvery, very much
kept away from the rest of the world
and limiteduse case scenarios
or research only for the last 30 years.
(39:22):
Whereas what weare now seeing is
we have the
early adopters, the early use cases and
I was a little hesitant
to put myself intothe autonomous taxis
around the Bay area.
Even knowing
that some companies havedefinitely got, you know
(39:43):
much better track recordsthan other companies.
I was still not wantingto be the first person
to make that leap
but I've startedusing them recently.
Yeah.
And I'm.
I thinkthough
that it's an interesting area
(40:04):
because whenever there is going to be a problem
and arguably there will be fewer problems
than with human drivers,
it's just there goingto be different problems.
Oh, they're completelydifferent problems.
The problemswith human drivers
have nothing to dowith driving at all.
It's distraction.
It's I had a fight withmy spouse in the morning.
(40:25):
My boss pissed me off.
You know,
I got a bunch of billsin the mail this morning.
It’s I got threetexts coming in.
My kid’s havinga bad day at school.
Those are the things that
make human driverssuch bad drivers.
But we think that we can understand those
and, or maybe predict if somebody’s
acting under theinfluence of something
(40:46):
that's going to make them a bad driver.
Now, we don'thave that insight level
into that occurring withan autonomous vehicle.
And so I think we've still gotto go through that
shaking outperiod
where we start to be.
We might find that there is a
couple of things that as
(41:06):
experienced usersof autonomous technology
if we see this happen
then we go, ‘Oh,I'm putting pause.’
I don't likethat happening.
We know that that mightbe an indication
that things aren'tworking perfectly, seamlessly.
So we haven't yet
worked out thesekind of codes.
(41:30):
Codes of conductcodes of behavior
and codes of how of reaction to the technologies
I thought you were goingto go somewhere else
when you said you knowlet's talk about autonomy
because one of the things is
we're seeing increasing autonomy
and yetthis is just
shifting the levelas you talked about
(41:51):
we're no longergiving step by step.
We're now expectingwe can say,
Okay, take me from this point to that point.
Right.
We're reaching a point which is
democratizing the technology
in the sensethat it's no longer
something that you needto program or code.
It's something where youcan make a request
Right
(42:12):
Do this, do that.
And that's much more like
when, for example, someonethrows a ball to us,
we don't say, okay,
now I want to extend my elbow
I want to move myshoulder up 90 degrees.
I want to open my fingers,rotate my wrist, etc.
We automatically do that.
We just needed to say ‘catch that ball’ and
(42:34):
I think it's really going to democratize the technologies
in the ability for themto be out in the world
and being used by a wide,wide range of people.
If we can reacha point where
English becomes the common operating system
What a concept.
We're getting a lotcloser to that.
Right?
That's one of the things that'sbeen happening pretty rapidly.
(42:54):
Is really exciting.
Yeah.
So we're gettingto the end of our time.
I know one of the topicsyou are very passionate about
and speak often aboutis the demographic trend.
And where robots play
in the realityof kind of
some of the demographic trends
that we're seeing interms of
just having enough peopleto do the jobs, period.
especially in more developed countries, right?
(43:15):
The birth rate is not sustainingor it's heading in the
in the negativedirection.
So the demand for these
both in terms ofgreat opportunities
and all thethe crappy jobs
that traditionally robotshave been targeted for
but it's going to goway beyond there
in terms of the opportunityfor them to fulfill
really a real lack of just flat out, people and labor.
(43:36):
I'm going toparaphrase a
tweet that became viral from an author
and I've forgottenher name
but she said
I don't want AI to do my art and write my stories.
I want to dothat while
AI does the dishesand takes out the trash.
(43:56):
We need robotsfor all of that.
Do the dishes,take out the trash.
All of those jobsthat particularly
factoring in unpaid labor
that so many of usdo as parents
or with aging parents
all of the caretakingfor society
(44:17):
as our societyis aging
on so many ways.
We have a loneliness epidemic.
We have agingpopulations.
We have a lack of labor
in most all of the dirty, dangerous and dull jobs.
People are saying
I would rather do anything else
than dothat.
So acrossthe board
(44:38):
there is no way we can
we can maintain the level of society
without developing robotsthat can take some of the
some of theload off us.
Yeah.
All right, Andra.
Well, I'll give youthe last plug.
Give a quicklittle plug for
for your newsletter.
And what's going on with
with Circuit Launch in the Bay areafor our Bay area folks.
(45:01):
Thanks, Jeff.
Well, once a weekI put out a newsletter
sometimes evenmore often
‘Robots and Startups’ on Substack.[https://robotsandstartups.substack.com/]
And those are mytwo favorite things
especially whenit comes together.
And that can include things like
what are upcoming conferences or events?
Not always in the Bay area, mind.
I attend a lot of roboticsevents around the world
(45:22):
and I like to just bring all the news
that interests meabout robotics
into the weeklynewsletter
and I'm very pleased thata lot of my audience are
people in roboticsand they say
It’s just a one
one thing that theyread each week
that makes themfeel like they’re
up to speed withwhat's happening
and Circuit Launch,my favorite place.
(45:43):
I'm hoping that we seemore Circuit Launches
not just inthe Bay area,
but in Australiaaround the world.
It's the hardwareacceleration
particularlyfor robotics
but for biotech, hardware,electronics, and IoT
that is an open ecosystem of acceleration.
So it's not about giving you money and taking equity
(46:05):
it's about making it really easy and affordable
for you to have a fullcommunity around you
With accessto the prototyping
and small batch manufacturingequipment that you need
and being in the company of other people
building similartechnologies.
So I am now at CircuitLaunch Mountain View
most ofthe time
(46:25):
and Circuit LaunchOakland is of course
the firstCircuit Launch.
And as I said
in 2026 I'm hoping thatI get to spend time
In some other greatCircuit Launch locations
and I'd love to show you around.
And we have lots of eventslike we have the
Robotics Network event often
often on the firstWednesday of each month.
(46:46):
But we havethings like
Robotics and GenAI Hackathons.
We have discussion meeting groups
we have workshops
and we just loveto be a place
where the communitythat is into deep tech
comes, hangs outand helps
(47:06):
get their deep tech built.
It's a great community.That's how we met.
And Andra’snewsletter is fantastic.
I'll have links inthe show notes
and if you're in the Bay area check it out.
Like she said,there's a lot of events.
There's a beerand robots thing
I think once a month
so it's definitely worthchecking out.
Well Andra I still have pages of notes
we could go for three hours
but I think we’re going to have to save it for our next
(47:27):
get together whichhopefully won't be
in the not too distant future.
Yes, let's make surewe have more discussions.
Absolutely. All right.
Well thanks again, Andra.
Thanks somuch Jeff.
She’s AndraI’m Jeff.
You’re watching Turn the Lenswith Jeff Frick.
Thanks for watching.
Thanks for listeningon the podcast.
Catch you next time. Take care.
Okay. Super.
Thank you.
(47:48):
Hey, Jeff Frick herewith a special shout out
to the podcast audience.
Thanks forlistening in.
You can find transcriptsand show notes
at www.TurnTheLensPodcast.com.
That's one word.
www.TurnTheLensPodcast.com.
Thanks for listening in.
Share, subscribe
and smash thatnotification bell.
Have a great day.
Take care. Bye bye.
So I will count us down
and we will go
in three,two, one.
Hey welcome back everybody.
Jeff Frick here
coming to you foranother episode of
‘Turn the Lens’ and I’ve got
what’s turning out to be apretty frequent guest.
What's interesting is
I've got to interview her
a lot of timeson other shows
and this isthe first time
I've actually had her on my show.
I think it's likethe fourth time
I've interviewed her.
(00:20):
So I'm really excitedto welcome in
through the magicof the internet
She's Andra Keay
the Managing Directorof Silicon Valley Robotics.
Andra, great tosee you.
So good to be here Jeff.
And you’re right, too manytimes on other stages.
It's really good tobe here with you.
It took a while
but we finally got ittogether so
I have a longlist of topics
which I always haveevery time we talk.
(00:42):
And I don't know if it was you that said it
or somebody said it that really
‘Humanoids are having a Moment’ right now.
What is going onon the humanoids.
You know, we've
they've always kind ofbeen there.
They're kind ofspacey and
you know, they're kind ofthe cool things
that people dreamabout and draw cartoons.
But what's happeningin humanoids
that’s so specialright now that's
different than whatit's been in the past.
(01:02):
Well once again, it’s taken 20 odd years
for humanoids to become an overnight success
but we really needed a certaincritical tipping point to reach
for humanoids to have this moment.
And those of us that have been enthusiastically following
companies like Agility Robotics, Apptronik
(01:23):
and some of the
1X and some ofthe other companies
from Firia, UnitreeDeep Robotics.
We've watched them
get commercialtrials
and in the caseof a couple of them
go from commercial trials
to actual commercial contracts.
(01:45):
And now that isa significant step.
And now that's openedthe doors to say
there may be a businesscase for humanoids now
that wasn'tpreviously possible.
It's possible now.
And here's the thing.
If you remember what a mobile phone looked like in the 19
(02:08):
oh, let's see, I think it was around
I was working infilm and television
and I think I got my hands on
one of the first ones
that was a kindof a mobile car phone
in around 1987.
[Andra] And it was[Jeff] That’s about right
I had the big giantMitsubishi brick.
You could jam it in your car.
You could pop outthe brick
(02:28):
and bringit inside
and plug itinto the wall.
Now, they never got bigger, heavier or more expensive
And that's what we need to think about
with humanoids today.
They are never goingto be any worse
than the current stateof technology.
They are going tocontinue to improve
they're going toget more capable
(02:50):
and they're going toget more affordable.
And these are the thingsthat are going to mean
it's not just 1 or 2 things
that humanoidrobots can do.
They're goingthey're here to stay.
Because the range of use cases for them
is pretty broad.
I think I've come up withabout 11, maybe 12 different
(03:10):
business modelsbusiness use cases
for humanoid robots.
And part of it is what you might say
being labor.
But a robot is capable
of more than simply
you know, lookinglike a human
and being a human replacement.
You know, in some ways
robots are far lesscapable than humans
(03:33):
but in other ways they are far more capable.
They're capable of doingrepeatable precision tasks
and now they are capableof being flexible
[Jeff] Right[Andra] as well.
So there's a rangeof different activities
where we can start to augment.
It's about doing somethingperhaps better
doing somethingin a way
(03:54):
that we weren't able to do before
and about being ableto do it affordably.
And the big reasonthat humanoids
are very appealingto many people
one anthropomorphically where we're just predisposed to
find things that are like us more appealing
Right, right.
But we also don't need to
create infrastructureand change our world.
(04:16):
The primary reason for creating a humanoid robot
is that it can usehuman tools.
It can operatein human
spaces designed for humans
operate equipmentdesigned for humans
and it's also much easierfor humans to train them
whether it's throughremote teleoperation
having that direct kinematicmapping to the shape.
(04:39):
So this translation
if my handsdo this
then that robot's hands do this.
If I reach upit reaches up
and it's really a parallel mapping.
And that makes humanoidsbetter in many ways
than saying, no let's create a custom built purpose robot
(05:00):
that could potentially do thisjob much, much better.
And these are thethings that you know
I see as being moreof the future thing
when we're talkingmaybe about
creating completely lights out operations of something.
Right
Then we wouldn't need it to be
if we’re custom designing
the full automation
the full building,the full work flow
(05:23):
then no, we wouldn't need to have it
being a humanoid,
but there's clearlycases for both
and I think that there'ssignificant demand
for both sorts of robotic automation
Right
humanoid and non humanoid.
And with suchincredible demand
I predict that there will be1 million humanoid robots
(05:44):
in commercial operation by 2030
And that's simply based onprojecting the plans of the
top half dozen or so
humanoid companies,the ones that are
seriously gettingtraction and
some of the analystshave been looking
further down the trackto 2035, 2040 and beyond
(06:07):
and seeing an incredible step change
like it might take us
five plus years
if we count perhaps the lastfive years of development
10 years to have gotten to1 million humanoids
it will onlytake us
In 10 years, we'll probablybe at 100 million.
Right, right.
So wasit just a
(06:28):
was it just a combination of kind of
classic technology curve effects
on the price of hardware
the sophistication of software
the speed of processors
and the connectivity of cloud
just kind ofall those things
you knowkind of moving
that it's gotit to this.
Or was there
was there a singleevent or was there a
a Cambrian event, I think you like to use the phrase that
(06:50):
that happened to changekind of the trajectory.
Certainly for roboticstechnologies in general.
There's been a Cambrianexplosion event
that we startedto see in the world
from 2010 onwards.
It's been buildingfor quite some time
but it's what I callthe ‘Robotics 2.0’.
(07:10):
It's about theability of robots
to do real worldreal time navigation
and you can see that as being
navigation like an autonomous mobile robot.
Sidewalk delivery robot
mobile robot in a factoryor self-driving car.
But you canalso see it as
three dimensional navigation
like a drone, somethingin the air or underwater,
(07:33):
or a robot armmoving in all directions
but capableof responding
sensing what is actually in space around it
and respondingin real time
if there's somethingout of plan.
So, for example
if a pedestrian comesin front of the vehicle
if somebody reachesout their arm
(07:55):
to stop therobot arm
any of thesethings that are
interrupting theplanned action
now a robotcan sense it
think about whatneeds to happen
to avoid to prevent collision
or just to redirectaround something.
Could be as simple as
(08:15):
bags of things fallingfrom the shelf.
Right
But previously
any robot technology wasonly as good as
‘Stop! There's an obstacle.’And then you had to wait
for some kind of helpto reset or restart.
Right?
Now it's not just aboutstop and go,
it's about go around.
It's about make changes in plans
(08:38):
in a real world,real time fashion.
So that is a fundamental step change
in terms of competencies
that have opened the doorfor an incredibly wide range
of use cases forrobotics technologies
and that's robotics technologiesacross the board.
What you were askingbefore that, though, was,
is there, that same thingfor humanoids?
(09:01):
Certainly they needthe ability to navigate.
Arguably humanoids are doingmore complex navigation.
The bodies of humanoidshave far more actuators
far many pointsof movement
and there are robots
that are movinginto social situations
in some casesfor humanoids.
(09:21):
And we're still not really greatat working that out.
So, there’s a lot of complexitiesahead for humanoids.
But the tipping pointscame on us really quickly.
There have been a whole sea of things
Moore's law in terms of computing technology.
The same thing having an impact on sensing technology,
the same thing happening in terms of
(09:42):
just us knowing howto build robots better
so that we can now builda smaller, far better robot.
And the role that AI andsimulation have played.
And so what I see now is it's been
exponential growth meetingexponential growth.
Right, right.
That was this incrediblespeed up in competencies.
(10:06):
So as roboticshas met
generative AI
or large scale learningand large scale
not just language modelsbut multi modal models.
Right.
And that’s very muchimportant for robotics.
And that hasmeant simulation.
So what we're seeing now
and this goes back to say Google arm farm
(10:27):
in the 2014 period where
a whole roomful of robot arms working
individually on the same tasks
would come up with
a whole lot more scenariosthan one robot arm.
But then those same strategies
(10:51):
would be tested in simulationand through AI
and that would multiplythe range of options
and the abilityto find the most
successful pathways out of them
which would then be fed back into the real
action of the robot armsthe next day
So you had this to and fro happening
between what the robots were exploring during the day
(11:13):
informing what theywere learning
and trainingduring the night.
Right.
And then going backinto the real world
during the day.
So you startedthis incredible
flywheel of feedback loops
and we've just multiplied
in a sense the number of feedback loops
and the transitions from simulation to real world
(11:33):
back to simulation to thereal robotics deployments
back to simulationand so on.
So it's compounding.
We're seeingincredible
exponentialgrowth compound.
Lots of greattopics here.
Lots of greatpaths of
of development.
So let's just jump ina little bit deeper
to the one you just touched on
which is the change in training.
And as you
(11:54):
you've kind of touched on briefly
we dig in,there's two things
that I saw at least at the show.
One is kind of LLMsand not necessarily
always largelanguage models,
but kind of the roboticsversion of those
for different typesof activities
changing the way thatthese things are trained.
And like you saidit's about
variability to change and being able to adapt
(12:15):
to not exactly follow the scriptand just say yes or no.
And then the other one is this
as you mentioned briefly
kind of this teleoperation piece
where you're integratinga human's behavior
into the training processto accelerate.
So very different way than
you know kind of a classicfactory implementation
(12:35):
where you're justmapping out,
you know, the pointat the end of the
laser welder to put a car together.
This is a really different way
to think about training
and opens up both speed and flexibility of the training.
It’s completely different.
Absolutely.
And there's several different waysthat training is happening
for roboticsand
(12:58):
I've been listeningwith interest
to differentdiscussions around
is this the right way?
Is this the way that
training is going to behappening in the future?
And at the momentI would say
the general agreement is that
we need all ofthese methods
and quite possiblyall of them together.
So there's real worldteleoperation
(13:20):
then there's having a person beside the robot
doing pose andmovement training.
There's programingtraining
there’s using Englishto instruct
there's trainingfrom videos
there's trainingthrough simulation
and there's training through
(13:40):
extracting ideas from
a range of other waysthat you’re mining data
around how these thingshave happened.
And it's opening
I think many people's eyes
to the fact that
we have very few.
We have limited insights
(14:01):
into how humansoperate in the world.
You know we've been able to
take manygreat strides
with artificialintelligence
through text based learning.
Right.
Even if we count videobased learning and
there's very much a limitto what's on the internet
about a whole range of things
like how a particular job
(14:22):
happens in a factory for example.
And we're onlyjust starting
to see some companies instrument
things like production lines
to learn exactly howdifferent tasks are done
and what'sthe range
between a task beingperformed successfully
or really, really well
and being performednot up to scratch?
(14:43):
And how we can learnacross those things
and improvethose outcomes.
It's a little bitlike when
robotics wentinto warehouse
into logisticsand factories.
There was a lot of feedback that
within theUnited States
up to 75%of factories
were not digitized
in terms of their inventories and operations.
(15:03):
It was kind of movingfrom an Excel spreadsheet
to a printout to somebody recording things
on clipboards to transferring them back in.
There wasn't a kind of
a digitizedreal time
base of knowledge
and a lot of these thingsyou needed to build
in some of thesefoundational steps
(15:24):
before you could go andadd more sophisticated
automation or autonomousmobile robots, for example.
Right.
There are someinstances where
the robots are able to step
past the lack ofprevious digitization.
But in terms of training
one of the big reasonsthat I think
humanoids are really having the moment
(15:46):
is because they'regoing to be ideal
platforms fordata collection
about doing all of these human type tasks in the world.
And they'll be developingthis digital library
of what is the rangeof how you do things.
And even simpletasks for example
the number of different door closures
(16:09):
that there are the worldis almost infinite.
They open outwards,they open inwards,
they open sideways,they open dual doors.
You push, you turn, you clip, you [psst]
There are so many differentways that we as people
[Andra] can navigate doors[Jeff] Right.
That's such abasic thing.
(16:29):
And yet collecting enough of that information
to have the sort of
materials that we need to do
successful training.
Well, maybe we do needrobots out there
jut exploring and discovering
a whole lot ofthose edge cases
that they're still having trouble with.
Right, right.
(16:50):
So, you've covered ethicsa lot in your time.
You've spoken on ethicsand robotics a lot
and some peoplemight say
robots are kind of theembodiment of software
or some people might say these days
it's AI with arms and legs.
As the robots move from the factory floor
out into this world
what are some of theethical considerations
(17:11):
both historicallythat
that have notreally changed,
but really noware new things
now that they've left the confines
behind the yellow tapeand the glass doors.
That’s a reallygreat question.
And sometimesI like to say
I'm on the side of robo-topia
because I don't wantto be on the side of
AI Apocalypse.
(17:31):
I can see a lot of thingspotentially going wrong
in the use of AI
and one of the reasonsthat it is perhaps
what I consider to be more dangerous
is because at some levels
it's cheap and easy and invisible,
and so wedon't see
(17:52):
what might be happeningunder the hood
and therefore we are notpaying the attention
that we need topay to it.
Whereas robots are inherentlymore visible being physical
and they cost moreto deploy in the world
we're not going to be
spamming the worldwith robots
and that makes ustake more care.
(18:14):
So I think thatthere are inherently
some guardrailsbuilt into robotics
that AI might not have.
And yet there is an incredible overlap
between the issues.
And oftentimes peoplethink of robotics only
in terms of physical issues.
Where as
we’re going to see the same
financial and emotional issues as well.
(18:37):
One of theareas that
really got mestarted in robotics
and that I am mostpassionate about
is the impact of thetechnology on society
and on all partsand levels of society.
There are ways that we can address
these thingswith, say
ethical or user centered design.
There areother ways
(18:57):
we're talkingabout
democratizationof technology
getting it out intomore people's hands.
But fundamentallywe also need to
broadly understand whatare the potential issues
of this technologymoving into society.
And we don't needto be too scared
because we've dealtwith new technologies
(19:18):
multiple timesas societies.
And there are
guidebooks forthe sorts of ways
that we as society deal
with majortechnological shifts.
I talk about it veryoften in terms of
here are my ‘5 Laws of Robotics.’
and they'rethe antithesis
(19:38):
of Isaac Asimov's‘3 Laws of Robotics’
which many peoplestill look to.
They express our needsand desires from robots.
We don't want robotsthat can harm us
or harm anybody else.
Those particular rules thatAsimov came up with
I always thought that they were examples of rules
that you could neverdeploy successfully
(20:00):
because all of his stories
were about how they went wrong.
But a lot of peoplestill look to them as
these are thesorts of rules we need.
It reflects on thethings that we want.
The things that are our priorities
but it doesn't reflect on
Pragmatically, how wouldwe actually go about doing this.
(20:20):
And how is this possible
with a roboticstechnology?
So I like to consider myself
a technopragmatist.
Not a techno optimistor a techno pessimist
but I believethat
we're going to be moving forward
technologicallyspeaking.
it's almostan imperative
but that doesn't mean
(20:41):
that we should do it thoughtlessly.
I want to beone of the people
that's movingforward
helping to put a brake onwhen it's needed
in the right spots.
And I want to sharethat information
and that abilityto see
where the potentialproblems are.
So my ‘5 Laws for Robotics’
start by soundinglike Asimov
because I sayrobots should not kill.
(21:04):
The thing is
this is one of our mostfundamental laws, rules
moral principles across society.
So we actually haveframeworks to help us
monitor that or control that.
We have laws.
If a robot is going to be acting lawfully
(21:24):
then it would notkill people.
Physical safety, though isjust the starting point.
See, the next thingfrom that is
robots should be designedto be law abiding.
Now that starts to raise the question of
but whoseresponsibility is this?
Is thisthe responsibility
of the peoplebuilding the robot?
The peopleoperating the robot
(21:45):
or the peoplearound the robots?
You know, where is theresponsibility sitting?
So we need robotsto be abiding by laws
and we have a lot of past record of this.
For example, Who
where is the liability foran automobile accident?
And we've come upwith different ways
of assigningresponsibility
(22:07):
based on thecircumstances.
Have we modified thatfor Waymos though?
I was gonna
it's funny youjust said that
because I was going to say,
you know, if a Waymogets in an accident
as a proxy robot
now you've got who wrote the software
who builtthe thing.
I mean, you got a wholenother layer of complexity on
whose responsibility.
Be curious. I don't know if that's
(22:28):
gone to court to kind ofhave some, precedent set
or maybe it has.
Many court cases.
One was in an Uber vehicle driving autonomously in Phoenix
and a pedestrian was killed
Now, in this instance,the safety driver
who was there as the responsible person
(22:48):
was the one that washeld accountable.
But I think that was largely a victor
victory for the
companies with the best paid lawyers
versus what would be repeatable case law.
Right.
And there's no
and there’s no safetydrivers anymore.
We've moved pastsafety drivers.
And this does mean that
people in the legal professionand legal scholars
(23:10):
have not been thinkinglong and hard about this
[Jeff] Right, right[Andra] in the last 20 years,
And really diving into it
Where we'reat is that
these things haven'thappened frequently enough
for their to besome more
established, precedentsand understandings.
Go to numberthree.
The third law.
Robots need to begood products
because the samething happened
(23:30):
in a non-fatal accidentin San Francisco
and it caused the completeshutdown of Cruise,
which was GM's self-driving vehicle play
It killed the company.
Yeah. Well.
Arguably in that case
they weren't ready for prime time
because when the data came out
on the numberof interventions
(23:52):
they were so farfrom autonomous
their intervention ratewas like once
every two miles or something.[actually 4 - 5 miles]
It was like
you guys aren't readyfor prime time.
Put those things away.
Without taking
without kind of litigating that one further
becausemaybe they shouldn't
have been ableto be testing.
Right, right.
Maybe we should be talking towhoever was certifying that
or allowing that.
But these thingscan become kind of
(24:14):
Jenga towers and
All I'm saying really is that
there is this kind of hierarchy.
If you startby saying
robots should not harmshould not kill
they shouldobey the law
then that's talking about where
law is one of ourmajor frameworks
for dealing with this.
And then if we talk about
they need to be good products
(24:34):
then we recognize that tobe commercially viable
there is a really strong incentivefor robotics companies
to be incorporatingthis early.
And certainlyI see this
with the roll outof humanoids.
The CEOs of the humanoid robot companies
want to see standards.
They want to see benchmarks for what is safe
(24:55):
They want to be able to assuredeployments and customers
that they are as good as
is reasonable to expect
Right
from this technology
that they're
they're doing everything right.
So they want tocreate these benchmarks
so that the industrycan move forward
and so that theyas companies
(25:15):
can be successful.
Certainly a dependency right
for masscommercialization.
Absolutely.
And this is where I say my laws.
It's about technopragmatism,
because each ofthese laws
speaks to how and where
we should look tobe implementing this.
But the next tworeally deal with
financial andemotional harm,
which we rarely talk about when we talk about robots.
(25:37):
And yet to methey are potentially
deeper, darker problems waiting for us
if we don't factor that in.
And yet it really plays intobeing good commercial products
Robots should be identifiable.
And to me that means that any,
any mobile robotis like a vehicle
(26:00):
and it needs to havean identification number
it needs to havea registration.
And then you can look at
what the process isthat go behind that.
And we accept thatin other vehicles
if it's out there
then it has to go throughcertain licensing
Right
and registrationprocesses.
Now then you goone beyond that
(26:20):
and it speaks to theyshould be transparent
and they should be truthful
and truthful
not just speakingthe truth
but transparentas in
is this robot speaking to meas an autonomous entity?
Is this a remote operatorspeaking through?
Is this a commercial script?
(26:41):
And it's designed forwhat particular outcome?
If a robot is being nice to me
we are pretty goodat starting to assess.
Well, we're okaywith assessing
if someone's being nice to me.
What doesthat mean?
What are theygaining from that?
And potentially,how are they
trying toexploit me?
(27:02):
Right.
Is this is harmful for me?
Now with robotics
just like with AI only that's more invisible.
There are many levelsat which that interaction
can be being steered orguided for other purposes.
Right.
And that can be assimple as saying
robots and AI are going to be really good at
being salespeople.
(27:23):
They might knowour particular
interests in history
or they mightbe able to
sense and detectour eye gaze
and our interest levels
things like pupil dilation
all of thosesignals of interest
that as people we oftenunconsciously pick up
these things can be picked up by
(27:44):
by camera technology today.
They might alsoknow exactly
what to offeras the next step.
The advertisingindustry is really good
at coming up with sort of psychological profiles
of who arethe people
that they want to have buying things
and how toappeal to them.
I think we need to applyyour rule number five to
(28:06):
and number four,actually to social media.
Maybe we wouldn't have
some of the problems that we have in social media
But it’s funny, I can't help but think, right.
One of the bigpotential markets for
the humanoidsis senior care
and taking careof people
in assisted livingsituations or whatever.
And, you know, I can seepeople with dementia
(28:27):
asks, you know, where Susie,
and Susie might be dead.
Susie, you know, maybewas an old friend,
you know you start thinking about who's
what are the differentpriorities for the robot
and the caregiversin that situation
in defining the answerto that question
both to be transparentbut also to be
(28:47):
careful, safety
appropriate for whateverthat medical situation,
may require, which might not be just
bald faced, you know,black and white facts.
And I think that there are a lot of parallels in health care
and the way that we'vedeveloped guardrails
(29:08):
in that so things likeinformed consent
and developing consent plans
ahead of timefor example
in some situationsyou could see that
a medical providermight say
it is the betteroutcome for this
particular personsay with dementia if
there is a robot caregiver
(29:29):
that can take on whatevercharacter they think it is
and that will keep themfar more comfortable.
Or there could be a casewhere there is a strong
desire from the personas an individual to say
I never want to be lied to regardless
and that that mightthen take precedence.
(29:50):
And we need to work these sorts of things out.
But this is reallygreat because
I did the five laws
and this is based on some of the best think tanks around
is some of the best work.
And I’ve taken part in a lotof these different activities.
Most of themI kept thinking,
how on earth are youever going to apply
these ideasor ideals?
(30:11):
And I very muchlike the direction
that this Five Laws are in terms of
pragmatic abilityto be deployed.
But I went beyond that and I went
How would I thinkabout that?
And I've got five things thatI think could be answers
One is we talked about already
the robot registrylicensing.
(30:34):
A second one
we've talked aboutin AI ethics cases
which is algorithmictransparency
having things like model cards
having things where the hidden workings
have to be
detailed and accessible.
If people need tolook up these things.
A third one
taking fromhealth
(30:55):
is to have independentethical review boards
and maybe it meansthat you're not looking
at something on acase by case basis,
but you're saying
in these circumstances
the best outcomefor people
is if a robot is allowed to behave like this.
Whereas in all of theseother circumstances
it's far betterif a robot
does the opposite.
(31:16):
Right
So we'll build upthese things if we have
I think, independentethical review boards.
Beyond that,I think there could be
a very interestingrole for the idea of
a ‘Robot Ombudsman’
to represent concerns that
individuals and or groups of society have
that they don't have
(31:38):
necessarily theability to voice.
They don’t necessarily know who to take these concerns to
or may not want toor be scared to.
But if you can resortto an ombudsman
who can thensay aggregate
these concerns and say, okay
maybe asa state
(31:59):
or as a companyor as a country,
this needs to becomesomething
that we develop policyor legislation around
because it is having an impact
on parts ofour society.
And we we're lettingtheir voices be heard.
Finally, I thinkrewarding examples
(32:21):
of what good robots are
making it something that's desirable
for companiesto create.
If we have ways of
let's not just punishpoor behavior
or poor corporatebehavior
but let's reward good robots
good robot design
and good applicationsof this technology.
(32:45):
Those arelofty goals.
I hope we do a better job
than we've done againon social media.
And the other one is privacy
where, you knowwe've just not been
not been diligentin keeping up with
you know, somethingas simple as
the cookiepolicies.
And, you know, you getGDPR in the EU
where they get a whole bunch of countries
that can cometo an agreement
but we still fall backon our fundamental
(33:06):
states versus nationalhead banging
and can't evenget a national
‘Disclosure When Breached’ regulation done
So we’llfingers crossed.
At least you gota framework which I like.
Getting ready for thisI came across
one of your older interviews
and you talked about robots
being the AI canaryin the coal mine.
(33:26):
So what is specialabout robots
and their connectionwith AI that gives them
the potential togive visibility
into positivesand negatives
that maybe, as you said
were kind of hiddenbehind a screen
inside of a computer before.
Exactly.
Robots are the physicalembodiment of AI
(33:47):
and becausethey're so visible
and because theyare capable
of physical damage
we have much greater.
We take a lot more care
with what's happeningwith robots and
so I think that will allow people to understand
(34:10):
what some of the problemswith AI might be,
where AI is that invisiblepotentially toxic gas
that might beall around us
and we might notbe aware of it.
And because robots are the
physical embodiment of AI
when we see things happening
that we don't like
(34:31):
it can inform us as to where these things
might also be happening invisibly.
It's really interestingthat, you know
we reactmore viscerally
and maybemore aggressively
to physical harmthan emotional harm
because you can see it, right?
They can put a picture of it
on the frontof the newspaper.
If there's a crash
if there's a crash inan autonomous vehicle.
(34:52):
Where the emotional harm
and some of the other problems
that can come again
just picking onsocial media
because I like topick on social media.
Aren't necessarilyas visible
kind of in your face.
So maybe they don't get thepriority that they should have
in terms of addressing them.
So a very different situation.
You're all over the place
in Bay areaSilicon Valley Robotics.
(35:13):
You got a greatnewsletter.
You keep up on alot of good things.
What's happeningin the funding world?
Has the fundingfigured out that
this is not yesterday's robots
and the opportunityand the technology
has progressed to a point where
they shouldn't be making a comp to
you know, when they were looking
at the little Sony dog
or Pepper or some of those early
cute iterationsof things.
(35:35):
A lot of the investmentcommunity has
worked out that
not only robotics isready for investment
but also
a lot of otherdeep tech.
And what I thinkis most exciting is
is in the lastten years.
And this is in spiteof there being a
whole lot of obstacles in
(35:56):
funding in venture at the moment.
But we're seeing acompletely new class
of investor, and that's the engineer,
that's the scientist,that's the investor
that can have
better appreciation
of a complex systemthat they're investing in.
(36:17):
And if youlook at
the background
experience and educational background
of investors that were ableto do really well during
I suppose, the riseof social media
and the internetand smartphones
many of them were
(36:38):
coming from financial orliberal arts backgrounds
as opposed tobeing engineers.
You didn't necessarily need to understand the technology
to understand what might bemaking good business cases.
Whereas now I think
there's so much of this
(36:59):
deep tech and roboticsis a great example of it
because roboticsis a complex system
and it really is something that is new
is new as this really complex robotics 2.0.
So we need to haveinvestors that are capable of
(37:19):
understanding thecomplex systems.
Now, I don't mean that everysuccessful investor in robotics
necessarily has aphysics degree
or has had a backgroundas an engineer,
but there's certainly a lot more investment firms
that are focusing on these technologies
that do have a strongbackground in those areas.
(37:41):
And this is actuallymore similar
to the first waveof venture capital
in the semiconductorand early computing days.
The deeper techkind of
kind of deeper techfoundational investment.
Another thing I wantto get your take is
I know you have an opinion on it is autonomy
And, you know,you said earlier on that
(38:01):
one of the bigbreakthroughs was when,
you know, a robot couldmake a course correction
when there was somethingthat was interrupt
interrupting its standard path.
And for mejust in terms of fun
when Skydio introducedthe autonomous drone
and we've talked about this before
where I no longerhad to actually
pilot the drone,
but now I'm reallyinstructing the drone
as to whatI want it to do,
(38:21):
whether that'sin a program
or coordinatesor mapping it out
whether it's, you know, inspect a
a transmissionwire
or inspectan ugly factory
that I don't want toclimb all over the silos
but it changedthe relationship
with the operationof the thing
where now it’s you're
you're giving it instructions to fulfill a task
as opposed toactually operating.
(38:42):
Where do you see autonomy
and how is that goingto change things?
And then I just
just to throw it inwith Waymo
was that a long timeor a short time
they’ve been working onautonomous vehicles.
You knowit's like
they've been talking about it
well guess what.
It's here.
At least30 years.
Is that short or long?
I think it's appropriatefor the difficulty
(39:02):
that we're facing.
And I think it'sit's not a mature tech
by any means, yet.
But it's starting to be
let's see, it wasvery, very much
kept away from the rest of the world
and limiteduse case scenarios
or research only for the last 30 years.
(39:22):
Whereas what weare now seeing is
we have the
early adopters, the early use cases and
I was a little hesitant
to put myself intothe autonomous taxis
around the Bay area.
Even knowing
that some companies havedefinitely got, you know
(39:43):
much better track recordsthan other companies.
I was still not wantingto be the first person
to make that leap
but I've startedusing them recently.
Yeah.
And I'm.
I thinkthough
that it's an interesting area
(40:04):
because whenever there is going to be a problem
and arguably there will be fewer problems
than with human drivers,
it's just there goingto be different problems.
Oh, they're completelydifferent problems.
The problemswith human drivers
have nothing to dowith driving at all.
It's distraction.
It's I had a fight withmy spouse in the morning.
(40:25):
My boss pissed me off.
You know,
I got a bunch of billsin the mail this morning.
It’s I got threetexts coming in.
My kid’s havinga bad day at school.
Those are the things that
make human driverssuch bad drivers.
But we think that we can understand those
and, or maybe predict if somebody’s
acting under theinfluence of something
(40:46):
that's going to make them a bad driver.
Now, we don'thave that insight level
into that occurring withan autonomous vehicle.
And so I think we've still gotto go through that
shaking outperiod
where we start to be.
We might find that there is a
couple of things that as
(41:06):
experienced usersof autonomous technology
if we see this happen
then we go, ‘Oh,I'm putting pause.’
I don't likethat happening.
We know that that mightbe an indication
that things aren'tworking perfectly, seamlessly.
So we haven't yet
worked out thesekind of codes.
(41:30):
Codes of conductcodes of behavior
and codes of how of reaction to the technologies
I thought you were goingto go somewhere else
when you said you knowlet's talk about autonomy
because one of the things is
we're seeing increasing autonomy
and yetthis is just
shifting the levelas you talked about
(41:51):
we're no longergiving step by step.
We're now expectingwe can say,
Okay, take me from this point to that point.
Right.
We're reaching a point which is
democratizing the technology
in the sensethat it's no longer
something that you needto program or code.
It's something where youcan make a request
Right
(42:12):
Do this, do that.
And that's much more like
when, for example, someonethrows a ball to us,
we don't say, okay,
now I want to extend my elbow
I want to move myshoulder up 90 degrees.
I want to open my fingers,rotate my wrist, etc.
We automatically do that.
We just needed to say ‘catch that ball’ and
(42:34):
I think it's really going to democratize the technologies
in the ability for themto be out in the world
and being used by a wide,wide range of people.
If we can reacha point where
English becomes the common operating system
What a concept.
We're getting a lotcloser to that.
Right?
That's one of the things that'sbeen happening pretty rapidly.
(42:54):
Is really exciting.
Yeah.
So we're gettingto the end of our time.
I know one of the topicsyou are very passionate about
and speak often aboutis the demographic trend.
And where robots play
in the realityof kind of
some of the demographic trends
that we're seeing interms of
just having enough peopleto do the jobs, period.
especially in more developed countries, right?
(43:15):
The birth rate is not sustainingor it's heading in the
in the negativedirection.
So the demand for these
both in terms ofgreat opportunities
and all thethe crappy jobs
that traditionally robotshave been targeted for
but it's going to goway beyond there
in terms of the opportunityfor them to fulfill
really a real lack of just flat out, people and labor.
(43:36):
I'm going toparaphrase a
tweet that became viral from an author
and I've forgottenher name
but she said
I don't want AI to do my art and write my stories.
I want to dothat while
AI does the dishesand takes out the trash.
(43:56):
We need robotsfor all of that.
Do the dishes,take out the trash.
All of those jobsthat particularly
factoring in unpaid labor
that so many of usdo as parents
or with aging parents
all of the caretakingfor society
(44:17):
as our societyis aging
on so many ways.
We have a loneliness epidemic.
We have agingpopulations.
We have a lack of labor
in most all of the dirty, dangerous and dull jobs.
People are saying
I would rather do anything else
than dothat.
So acrossthe board
(44:38):
there is no way we can
we can maintain the level of society
without developing robotsthat can take some of the
some of theload off us.
Yeah.
All right, Andra.
Well, I'll give youthe last plug.
Give a quicklittle plug for
for your newsletter.
And what's going on with
with Circuit Launch in the Bay areafor our Bay area folks.
(45:01):
Thanks, Jeff.
Well, once a weekI put out a newsletter
sometimes evenmore often
‘Robots and Startups’ on Substack.[https://robotsandstartups.substack.com/]
And those are mytwo favorite things
especially whenit comes together.
And that can include things like
what are upcoming conferences or events?
Not always in the Bay area, mind.
I attend a lot of roboticsevents around the world
(45:22):
and I like to just bring all the news
that interests meabout robotics
into the weeklynewsletter
and I'm very pleased thata lot of my audience are
people in roboticsand they say
It’s just a one
one thing that theyread each week
that makes themfeel like they’re
up to speed withwhat's happening
and Circuit Launch,my favorite place.
(45:43):
I'm hoping that we seemore Circuit Launches
not just inthe Bay area,
but in Australiaaround the world.
It's the hardwareacceleration
particularlyfor robotics
but for biotech, hardware,electronics, and IoT
that is an open ecosystem of acceleration.
So it's not about giving you money and taking equity
(46:05):
it's about making it really easy and affordable
for you to have a fullcommunity around you
With accessto the prototyping
and small batch manufacturingequipment that you need
and being in the company of other people
building similartechnologies.
So I am now at CircuitLaunch Mountain View
most ofthe time
(46:25):
and Circuit LaunchOakland is of course
the firstCircuit Launch.
And as I said
in 2026 I'm hoping thatI get to spend time
In some other greatCircuit Launch locations
and I'd love to show you around.
And we have lots of eventslike we have the
Robotics Network event often
often on the firstWednesday of each month.
(46:46):
But we havethings like
Robotics and GenAI Hackathons.
We have discussion meeting groups
we have workshops
and we just loveto be a place
where the communitythat is into deep tech
comes, hangs outand helps
(47:06):
get their deep tech built.
It's a great community.That's how we met.
And Andra’snewsletter is fantastic.
I'll have links inthe show notes
and if you're in the Bay area check it out.
Like she said,there's a lot of events.
There's a beerand robots thing
I think once a month
so it's definitely worthchecking out.
Well Andra I still have pages of notes
we could go for three hours
but I think we’re going to have to save it for our next
(47:27):
get together whichhopefully won't be
in the not too distant future.
Yes, let's make surewe have more discussions.
Absolutely. All right.
Well thanks again, Andra.
Thanks somuch Jeff.
She’s AndraI’m Jeff.
You’re watching Turn the Lenswith Jeff Frick.
Thanks for watching.
Thanks for listeningon the podcast.
Catch you next time. Take care.
Okay. Super.
Thank you.
(47:48):
Hey, Jeff Frick herewith a special shout out
to the podcast audience.
Thanks forlistening in.
You can find transcriptsand show notes
at www.TurnTheLensPodcast.com.
That's one word.
www.TurnTheLensPodcast.com.
Thanks for listening in.
Share, subscribe
and smash thatnotification bell.
Have a great day.
Take care. Bye bye.
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.
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
CrimeLess: Hillbilly Heist
It’s 1996 in rural North Carolina, and an oddball crew makes history when they pull off America’s third largest cash heist. But it’s all downhill from there. Join host Johnny Knoxville as he unspools a wild and woolly tale about a group of regular ‘ol folks who risked it all for a chance at a better life. CrimeLess: Hillbilly Heist answers the question: what would you do with 17.3 million dollars? The answer includes diamond rings, mansions, velvet Elvis paintings, plus a run for the border, murder-for-hire-plots, and FBI busts.