Latest / a16z AI Policy Brief / Fixing the Front Door to Government

Transcript

Fixing the Front Door to Government · a16z AI Policy Brief

0:00
SPEAKER_00: Welcome back to the AI Policy Brief.
0:02
SPEAKER_00: I'm Matt Peralt.
0:03
SPEAKER_00: Today I'm joined by Ben Suppel,
0:05
SPEAKER_00: Head of Global Policy at Eleven Labs,
0:07
SPEAKER_00: a leader in voice AI and one of the fastest growing AI companies today.
0:11
SPEAKER_00: And yet, it's got a policy function small enough to count on one hand.
0:15
Ben shares what it's like to navigate AI policy as a small team
0:19
and makes the case
0:19
SPEAKER_00: for why clear,
0:20
consistent rules matter more for startups than for companies
0:23
with large legal and
0:24
SPEAKER_00: policy teams.
0:26
We also talked about Voice AI as a new front door to government
0:29
services and Eleven
0:30
SPEAKER_00: Labs worked with governments to improve outcomes for citizens.
0:33
SPEAKER_00: Ben is living the little tech experience,
0:36
building and shipping at startup speed while navigating the
0:38
laws that will shape
0:39
SPEAKER_00: what gets built.
0:40
SPEAKER_00: Here's my conversation with Ben Supple.
0:43
SPEAKER_01: Most businesses, most governments have automated systems today.
0:48
SPEAKER_01: It just so happens that that automation doesn't work very well.
0:52
Our tagline when we announced the program was fixing the front
0:55
door to government.
0:57
SPEAKER_01: The technology is there today to make a much better citizen experience.
1:03
SPEAKER_01: I certainly value the state and local perspective,
1:07
SPEAKER_01: but from a very pragmatic and practical point of view.
1:11
SPEAKER_01: You know,
1:12
SPEAKER_01: it's just impossible for a team of one,
1:14
SPEAKER_01: a small startup,
1:15
SPEAKER_01: to keep up with all the proposals.
1:17
What the EU does very well is that they have a very
1:21
comprehensive and consistent framework.
1:23
And maybe it's more burdensome than other frameworks and there
1:26
are challenges and
1:27
SPEAKER_01: costs to comply.
1:28
SPEAKER_01: But at least you have that goalpost and you know what you need to do.
1:39
SPEAKER_00: Ben Zappel, Head of Global Policy at Eleven Labs.
1:42
SPEAKER_00: Welcome to the AI Policy Brief.
1:44
SPEAKER_00: Awesome.
1:44
SPEAKER_00: Great to be here, Matt.
1:45
SPEAKER_00: Great to see you again.
1:47
SPEAKER_00: Look forward to diving in.
1:48
SPEAKER_00: So what is Eleven Labs?
1:50
Can you talk a little bit about the company and what you guys
1:52
are building?
1:52
SPEAKER_01: Yeah,
1:53
SPEAKER_01: so Eleven Labs is amazingly a four-year-old company,
1:56
SPEAKER_01: founded in 2022,
1:57
SPEAKER_01: launched its first products in 2023.
2:01
The origin story is that you had two friends from Poland who
2:04
grew up together,
2:06
SPEAKER_01: who studied math and sciences,
2:08
were educated in the UK and one went to Google to work on AI
2:12
research the other was
2:13
at Palantir and they would get together on the weekends and do
2:16
side projects and
2:18
the early motivation for Eleven Labs was that when you try to
2:21
watch a foreign film
2:23
in Poland they don't have sophisticated dubbing so typically
2:26
they have a single
2:27
voice actor reading every part so like a male voice reading
2:31
every part in the movie
2:32
SPEAKER_01: and you ask anyone in Poland it's an awful experience
2:36
SPEAKER_01: So yeah,
2:36
SPEAKER_01: they tried to create a few synthetic,
2:39
SPEAKER_01: natural,
2:39
SPEAKER_01: human-sounding voices,
2:41
SPEAKER_01: did an excellent job at that,
2:42
SPEAKER_01: and Eleven Labs was born.
2:44
And I think they've always been focused on how do you make
2:48
human technology
2:49
SPEAKER_01: interaction more natural,
2:51
SPEAKER_01: more human-centered,
2:52
SPEAKER_01: and voice and audio was kind of the first problem area to focus on.
2:57
SPEAKER_01: So the company has, I guess, four main buckets of work.
3:00
SPEAKER_01: First is the foundational research.
3:02
SPEAKER_01: We build our own STT and TTS models, I think, best in the world.
3:07
SPEAKER_01: And, you know, you don't have to trust me.
3:09
SPEAKER_01: You can look at the leaderboards.
3:14
SPEAKER_01: You can access our voice library with over 10,000 voices.
3:21
You have a voice cloning feature if you wanted to clone your
3:24
voice and then narrate
3:25
SPEAKER_01: an audio book or create and scale content.
3:28
a professional dubbing and production studio so if you're like
3:32
a YouTuber in Ohio
3:33
and you want to reach a global audience you can dub your
3:36
content very quickly
3:37
across you know 30 plus languages and then we've recently
3:40
launched like music
3:41
SPEAKER_01: creation sound effects we
3:43
SPEAKER_01: We have an image generation,
3:45
video generation platform where you can stitch together our
3:48
voices with videos.
3:50
SPEAKER_01: So that's one, the creative platform.
3:52
SPEAKER_01: Two is the API business.
3:53
So third party integrations with our API,
3:56
big companies who want to leverage our voices.
3:58
SPEAKER_01: And then third, which we're really excited about is 11 agents.
4:02
We found that a lot of peer companies who were building agents
4:06
were using our
4:07
SPEAKER_01: voices and trying to stitch them together with their orchestration layer.
4:11
SPEAKER_01: A16z
4:24
SPEAKER_01: So can you talk a little bit about
4:46
You gave some examples of why someone would come to you looking
4:50
to use your services.
4:51
SPEAKER_00: Can you give a few more?
4:52
Like what are the things that you're seeing in the ecosystem
4:55
that are driving
4:56
SPEAKER_00: people to use 11 Labs technology?
4:58
Yeah,
4:58
so I think,
4:59
you know,
4:59
voice is kind of the natural mode of communication.
5:02
SPEAKER_01: We've adapted to technology with keyboards and texting.
5:06
And now I think voice and voice AI has gotten to a point where
5:09
you can just have
5:09
SPEAKER_01: that natural conversation and unlock different experiences.
5:13
What are the kinds of businesses that are like looking to use
5:16
some sort of voice
5:17
integration and therefore coming to you and wanting to use your
5:21
technology for
5:21
SPEAKER_00: that?
5:22
runs the gamut some of the big tech players Meta you know we
5:26
both had our time
5:26
there for Instagram Reels for instance we have a partnership in
5:30
India where we are
5:31
dubbing all Instagram Reels in India across all Indic languages
5:35
or I think a dozen
5:36
SPEAKER_01: Indic languages
5:37
SPEAKER_01: You know,
5:38
SPEAKER_01: your traditional customer service use cases,
5:41
we're all familiar with calling our bank or calling our
5:43
insurance company,
5:45
SPEAKER_01: and you're met with this rote,
5:47
SPEAKER_01: very antiquated,
5:48
SPEAKER_01: automated experience,
5:50
SPEAKER_01: rigid,
5:51
SPEAKER_01: menu-based,
5:51
SPEAKER_01: you know,
5:52
SPEAKER_01: select one.
5:52
SPEAKER_01: A16z
6:17
SPEAKER_01: A16z
6:32
so we have a voice library of over 10,000 voices where you can
6:34
kind of narrate that
6:35
content so what does the competitive landscape look like like
6:37
the way that you
6:38
described value proposition sounds compelling and seems like a
6:41
potentially
6:41
SPEAKER_01: significant market which I assume is what's driving your
6:59
SPEAKER_00: revenue growth and your valuation.
7:01
SPEAKER_00: What does the competitive landscape look like?
7:03
Like it seems like something where other companies would be
7:05
trying to offer comparable services.
7:07
Yeah,
7:08
so super competitive,
7:09
super dynamic across every aspect of our business.
7:13
SPEAKER_01: The model layer,
7:14
a lot of the big frontier AI labs are also building their own
7:17
voice models,
7:18
SPEAKER_01: have those in production today.
7:20
SPEAKER_01: From an agent's perspective,
7:22
SPEAKER_01: obviously you have countless,
7:24
SPEAKER_01: you know,
7:24
SPEAKER_01: agentic products in the market.
7:27
SPEAKER_01: I think we both partner with a lot of the Frontier AI Labs,
7:30
SPEAKER_01: other tech companies who use our voices,
7:33
kind of understand our deep focus and kind of market lead and
7:37
don't feel the need
7:37
SPEAKER_01: to replicate that in-house in all instances.
7:41
So kind of a frenemy dynamic with some of the big tech
7:44
companies and labs.
7:46
but yeah we have to move fast we have to you know ship product
7:49
at a rapid clip
7:51
certainly can't rest on our laurels but I think where we've
7:54
found success is that
7:55
we're deeply focused on just a couple problem areas we're not
8:00
trying to achieve AGI
8:02
we're focused on AI transformation how do you turn content into
8:06
voice into audio
8:08
and how do you make that experience as good and rich as
8:10
possible what does that
8:11
competitive dynamic feel like on a daily basis like you're
8:14
you're talking about
8:15
SPEAKER_00: You know you're up against
8:36
SPEAKER_00: A16z
8:47
SPEAKER_01: Yeah,
8:47
SPEAKER_01: you know,
8:47
SPEAKER_01: from a policy perspective,
8:49
like there's a lot of common cause with the bigger labs and
8:52
what we're trying to
8:53
SPEAKER_01: achieve from a policy perspective.
8:55
I think watching what OpenAI and Google are doing from a policy
9:00
and kind of
9:00
SPEAKER_01: regulator engagement standpoint,
9:02
what they're doing from like a policy campaign or social impact
9:05
perspective is
9:07
SPEAKER_01: really instructive for us.
9:08
SPEAKER_01: And we can learn a lot from those counterparts.
9:11
I would say though that when we're small and scrappy you have
9:14
the advantage like
9:14
you don't have hundreds if not thousands of policy people and
9:19
20 layers of approval
9:22
before you can do something like when I have an idea and one of
9:26
my early passion
9:27
projects was launching 11 labs for government you get like five
9:30
people in a room
9:31
you're like I think we have killer products to solve these
9:33
challenges that
9:34
government face like let's get together and put something out
9:37
in the market
9:38
and you know I had obviously worked at WhatsApp and advised
9:41
OpenAI for a while and
9:42
like launching a program of that scale and complexity in a big
9:46
company that can
9:47
take six months if not a year right with working through the
9:51
bureaucracy and all
9:52
the approvals and so I think in some ways it allows us to move
9:55
faster the other
9:57
advantage is that like you know the metas of the world they
10:01
have their reputation
10:02
SPEAKER_01: those views are pretty entrenched among policy makers
10:06
I go to DC and it's like 50-50 in terms of the people who've
10:09
ever heard of 11 Labs.
10:11
And so that first meeting can kind of set the tone of a future
10:14
relationship and you
10:15
SPEAKER_01: can set the policy narrative,
10:17
SPEAKER_01: set the relationship on your terms.
10:21
I think like I haven't seen intense competition from a policy
10:23
perspective but
10:23
certainly as you know like over time you know especially if
10:26
you're looking at a
10:27
SPEAKER_01: global remit
10:37
you can only go so deep and do so much in India and Europe and
10:41
Brazil parachuting
10:43
in from California and I think the governments over time want
10:47
to see that local
10:48
commitment and like have that local person on speed dial so
10:51
hopefully we'll get to
10:52
SPEAKER_01: that point in the future
10:53
you lead the team and you were the first hire of the team you
10:56
are the team I know
10:58
obviously these discussions were before your time at the
11:00
company but do you have a
11:01
SPEAKER_00: sense of why 11 Labs decided to start a policy function now?
11:04
SPEAKER_01: So yeah I report up to the legal team our general counsel was kind of
11:09
Dual-hatted doing policy and legal work,
11:13
doing a bit of comms on the side.
11:15
SPEAKER_01: There's a ton of excitement among governments to talk to us,
11:18
SPEAKER_01: to work with us on government adoption,
11:21
SPEAKER_01: on AI transformation,
11:23
SPEAKER_01: socially beneficial use.
11:25
So I think he got to a point where he's like,
11:27
I can't be both a GC and a head of policy.
11:31
And so I need someone who can come in and kind of,
11:34
you know,
11:35
handle that scope.
11:38
and I think they also have grown from a small scrappy startup
11:41
to now an 11 billion
11:43
dollar company like we're bigger than Snapchat we're bigger
11:46
than American Airlines
11:47
SPEAKER_01: right so you need to build functions and build that muscle while kind of
11:53
maintaining that startup culture or flat structure or limited
11:56
bureaucracy so I
11:57
think that was the kind of early vision for policy and also to
12:00
bring in people
12:02
who've done this for bigger companies and can kind of bring
12:05
that big tech skillset
12:06
SPEAKER_00: I think it can be a really hard stage at a company's growth from a policy
12:09
perspective when you have this outsized perception or outsized
12:14
sense of like the
12:15
potential for the company to grow but you're still a small
12:19
legal team you're still
12:20
a policy team of one and that gap can be really hard like
12:23
policymakers are
12:24
expecting you to be available to talk to them and they're
12:27
expecting you to have
12:29
Lots of programmatic options available to them and points of
12:31
contact from a
12:32
customer service perspective and that you've thought deeply
12:34
about every policy
12:35
SPEAKER_00: position that might implicate your company.
12:37
SPEAKER_00: Obviously,
12:38
it's really hard for you to do that given the resources that
12:42
you have available to
12:43
SPEAKER_00: you.
12:44
SPEAKER_00: So how do you prioritize?
12:45
SPEAKER_00: Like how do you decide on a Monday morning?
12:47
SPEAKER_00: Here are the set of things that I'm going to focus on,
12:48
but also the public policy function that 11 apps is going to
12:51
get done this week.
12:52
SPEAKER_01: So certainly there are days that are daunting.
12:55
SPEAKER_01: It's a big scope.
12:56
SPEAKER_01: There's a lot of demand.
12:57
We both had the experience at Meta when you can kind of call
13:00
the privacy policy
13:02
expert or call the India policy team and you had all these
13:05
regional and subject
13:06
SPEAKER_01: matter experts at your disposal.
13:09
so I think for me it's you know how do you ruthlessly
13:11
prioritize I have kind of
13:12
four focus areas policy development through the lens of like
13:17
what does go-to-market
13:19
want to do and where does the business want to grow and how is
13:23
policy an impediment
13:24
to that so if we want to do more in India how do we build the
13:26
right ecosystem we
13:28
want to do more in ed tech how do we make sure we comply with
13:31
you know child safety
13:32
SPEAKER_00: rules and so forth
13:34
so that's one kind of just showing up in our key markets like I
13:37
can't be in every
13:39
country where we do business but what I said in my initial plan
13:42
was let's go to the
13:43
top five let's spend time in Washington and London and Delhi
13:46
and Brazil and you
13:48
know kind of build those foundational relationships third I
13:52
think is you know on
13:53
the government adoption side and working hand in glove with
13:55
go-to-market and then
13:56
SPEAKER_01: fourth is social impact and socially beneficial use
14:00
of those four buckets I think I've been more focused on
14:03
proactive partnerships
14:05
working with government to help them achieve their goals the
14:09
social impact side of
14:10
SPEAKER_01: the house and less on policy shaping or
14:15
SPEAKER_01: AI Policy Brief
14:30
SPEAKER_01: A16z
14:47
AI Policy Proposals But I think we have the right extended team
14:51
in place that if
14:52
SPEAKER_01: something is breaking and impacts us negatively or positively,
14:56
SPEAKER_01: we'll get a phone call and you'll let us know where we need to engage.
15:00
SPEAKER_01: So yeah, it's been fun.
15:01
SPEAKER_01: It's been dynamic.
15:02
SPEAKER_01: You know,
15:02
hopefully we can kind of prove the concept of what a policy
15:05
team does and get some
15:06
SPEAKER_01: reinforcements over time.
15:07
SPEAKER_01: But as a team of one,
15:08
SPEAKER_01: I think we've done some good things and I've learned a ton kind of,
15:12
SPEAKER_01: you know,
15:13
SPEAKER_00: embedding in more of a startup culture.
15:15
SPEAKER_00: You're based in California.
15:17
and that means there are lots in your team of one that means
15:19
there are lots of
15:20
policy jurisdictions that you're not present in you're not
15:23
physically present in DC
15:25
full-time you were mentioning Delhi we saw each other at the AI
15:28
Policy Summit in
15:29
Delhi and you were there but you obviously are not there
15:31
full-time same goes for
15:33
Brussels I think that can be really hard from a policy
15:36
perspective because you know
15:37
all politics is local policymakers really like that there's
15:40
someone local in their
15:41
SPEAKER_00: jurisdiction who they can call when they've got an issue
15:44
SPEAKER_00: A16z
16:05
SPEAKER_00: DELTA could be really frustrating to policymakers.
16:07
SPEAKER_00: It seems like it probably is really exacerbated for you.
16:10
SPEAKER_00: Again,
16:10
SPEAKER_00: you're trying to have a sizable presence as you possibly can,
16:13
but you're not going to have the kind of presence that larger
16:16
companies are able to
16:16
SPEAKER_00: have.
16:18
Does that affect the nature of the back and forth that you have
16:21
with policymakers?
16:21
SPEAKER_00: Like when you show up in Delhi, you're talking to them.
16:24
How are they thinking about your company and how do you deal
16:27
with the local presence issues?
16:29
SPEAKER_01: Certainly working across time zones,
16:31
I think it's a 13 and a half hour time zone difference between
16:34
California and
16:35
SPEAKER_01: India.
16:36
And,
16:36
you know,
16:37
asking policymakers to talk to you at 7pm,
16:41
8pm at night is not ideal.
16:43
At the same time,
16:44
they see that we're a three,
16:46
four year old company and growing fast.
16:48
SPEAKER_01: And,
16:48
SPEAKER_01: you know,
16:49
SPEAKER_01: there are a lot of companies,
16:50
SPEAKER_01: I'm sure,
16:50
SPEAKER_01: you know,
16:51
SPEAKER_01: in your network that are much more
16:54
SPEAKER_01: A16z
17:02
SPEAKER_01: The other thing I would say is that, you know, we're still a small team.
17:06
SPEAKER_01: We're not super stovepiped yet.
17:08
SPEAKER_01: So like if there's a person in Poland who's interested in policy,
17:13
SPEAKER_01: but maybe he works in finance or works in marketing,
17:16
in some cases we'll just tap someone else in who's not a policy
17:20
person per se,
17:21
SPEAKER_01: but interested and can engage on the company's behalf.
17:24
and then as I mentioned you know how do we get the right
17:27
consultants work with the
17:28
right trade associations how do we make sure that those trade
17:32
positions are
17:33
representative of our view and not just the big tech view but
17:37
yes definitely
17:38
definitely a challenge to work across all those jurisdictions
17:42
especially in this
17:43
SPEAKER_01: current kind of geopolitical climate where there's such an obsession with
17:48
Sovereign AI and you have to kind of claim your allegiance to
17:52
one country or region.
17:54
SPEAKER_01: So yeah,
17:54
showing up and showing commitment and that each country and
17:57
region is special while
17:59
SPEAKER_01: not actually having a physical presence is certainly a delicate balance.
18:03
SPEAKER_00: Then you've talked a bunch about growth.
18:05
SPEAKER_00: You're at $330 million in reported revenue in 2025.
18:09
SPEAKER_00: You got there after five months.
18:10
SPEAKER_00: Five months before that, you were at $200 million.
18:12
SPEAKER_00: This is an incredibly rapid pace of growth.
18:15
SPEAKER_00: What's it like to be at a company that's growing that fast?
18:17
It's a lot of fun and certainly meta like we grew fast we
18:21
shipped products fast I
18:22
had you know some exposure to to open AI and and I'd say this
18:27
is like a step change
18:30
more intense and faster and some of that is just because we're
18:35
small we're a
18:36
SPEAKER_01: startup we don't have layers of bureaucracy as I've mentioned
18:41
and I think the competitive environment right it's like you
18:44
know every week people
18:45
are shipping better models I think people inherently understand
18:49
that to win an AI
18:51
you need to be multimodal you need to have a solid voice layer
18:56
so yeah I think it's
18:56
intense in terms of the competitive dynamics in terms of the
19:00
team size but we also
19:02
have a killer product and I think the product kind of sells
19:05
itself you know I had
19:06
SPEAKER_01: this experience at WhatsApp where I think WhatsApp
19:09
achieved like a billion or a billion and a half users before
19:13
they hired their first
19:13
marketing person and I'm like that's kind of cool like it sold
19:16
itself to over a
19:18
billion users before you needed a marketing team and I feel
19:21
like you know Eleven
19:23
Labs is in that same vein where it's like once you've
19:25
experienced it once you've
19:27
had a conversation with a voice agent who understands
19:31
everything you're saying when
19:32
SPEAKER_01: you're driving to work
19:33
with a lot of background noise and they not only understand you
19:37
they're like
19:38
emotionally aware and they understand when you get frustrated
19:40
and they change the
19:41
cadence at which they speak and they can actually resolve the
19:44
issue for you without
19:46
having to redirect to someone else and wait for 30 minutes it's
19:50
game changing you
19:51
know it's kind of like the chat GPT moments like I didn't think
19:54
I needed this but
19:55
SPEAKER_01: now that I know it exists like I can't work without it
19:57
and I feel like our technology is in that category where like
20:01
once you've had some
20:02
of these conversations with a lab's voice agent it's game
20:05
changing you know like
20:06
you know what I don't know if I want to be redirected to a
20:09
human operator like this
20:10
was this was really cool and it totally understood what I
20:12
needed it is so funny
20:13
what you say about redirected from an AI tool to a human
20:19
because there's so much
20:20
SPEAKER_00: conversation now in policy debates about
20:22
SPEAKER_00: AI Policy Brief a16z
20:45
Part of the idea of the product is to take some number of
20:47
conversations that would
20:48
be between humans and move at least some percentage of them
20:51
some of the time to
20:53
SPEAKER_00: between a human and an AI agent.
20:55
SPEAKER_01: I think most businesses, most governments have automated systems today.
21:00
So like the front door to government or the front door to
21:02
customer service at a
21:04
SPEAKER_01: business is already automated.
21:06
It just so happens that that automation doesn't work very well
21:10
and it only deflects
21:12
SPEAKER_01: and redirects and it puts a big strain on human operators.
21:16
I think there's always last mile support where you will need a
21:20
human operator to
21:21
SPEAKER_01: jump in and to work with you hand in glove.
21:24
but what we found and we've all experienced this is in peak
21:27
times when you're
21:28
calling your airline you're calling your tax authority there
21:32
just aren't enough
21:34
butts in seats so to speak enough human operators to address
21:38
the demand and so
21:40
imagine if you had an AI assistant that could do 50% resolve
21:43
like the basic issues
21:44
SPEAKER_01: that 50% of people are calling about
21:47
But yes,
21:48
I definitely think you want that human oversight in the system
21:52
design.
21:53
You want that,
21:53
you know,
21:54
kind of phone a friend option to a human operator.
21:57
SPEAKER_01: I think that's required both from a policy perspective,
22:00
SPEAKER_01: but also a lot of companies like when I was at WhatsApp,
22:02
SPEAKER_01: you had to have a human escalation path to the product flow.
22:07
SPEAKER_01: Otherwise, you couldn't launch it.
22:09
SPEAKER_01: So we certainly value that.
22:11
SPEAKER_01: But I think what we've seen is that
22:14
You think of government use cases for instance and how many
22:16
people don't apply for
22:18
SPEAKER_01: benefits who are eligible because they get used,
22:22
SPEAKER_01: they get frustrated,
22:23
SPEAKER_01: they don't want to wait for two hours.
22:25
SPEAKER_01: And so imagine if you had an AI
22:29
SPEAKER_01: who could work with you and resolve,
22:32
SPEAKER_01: you know,
22:33
50% of those issues and get people the kind of outcomes and the
22:36
services they need.
22:37
SPEAKER_00: So you lead the policy apparatus.
22:39
SPEAKER_00: So let's dive in a little bit deeper on policy issues that you're facing.
22:43
What are the policy topics and dynamics that are keeping you up
22:48
at night right now?
22:50
SPEAKER_01: So the first thing I would say is I think when people hear AI startup,
22:54
their mental model goes to OpenAI and Google and Meta and the
22:59
litany of policy
23:00
SPEAKER_01: challenges we've all read about in the headlines.
23:03
SPEAKER_01: So first, I think it's important to kind of situate us in the AI stack.
23:08
SPEAKER_01: Like we're not, as I mentioned, racing toward AGI, right?
23:13
We're focused on transformation scenarios,
23:15
transforming speech to text and text to speech.
23:18
So think of us less as like an open AI where you're worried
23:21
about catastrophic risks.
23:23
You know,
23:23
can you use this tool to create a bioweapon or change national
23:27
security outcomes?
23:29
SPEAKER_01: And more like an Adobe for audio, right?
23:32
SPEAKER_01: Where we're providing that creative suite.
23:35
SPEAKER_01: But the spectrum of policy risks is very unique and very different.
23:39
SPEAKER_00: And I think in some ways more limited than some of the big frontier labs.
23:44
SPEAKER_01: So what are we focused on?
23:45
SPEAKER_01: I think issues on,
23:46
SPEAKER_01: you know,
23:47
SPEAKER_01: digital replicas,
23:49
SPEAKER_01: voice cloning,
23:50
SPEAKER_01: things like the No Fakes Act at the national level,
23:53
SPEAKER_01: Elvis Act at the state level.
23:55
SPEAKER_00: And when you talk about those things,
23:56
are those things that you're generally comfortable with those
23:59
current frameworks?
24:00
Or do you think that those frameworks underappreciate either
24:03
elements of a company
24:04
SPEAKER_00: like yours or the kind of technology that you're building?
24:06
SPEAKER_00: Like, are they appropriate for voice AI generally?
24:10
SPEAKER_01: You know,
24:10
I think we need to study each example more and those I know are
24:14
still working
24:15
SPEAKER_01: through the process.
24:16
SPEAKER_01: So I think at the right time, we'll do the deep dive.
24:19
SPEAKER_01: Generally, I think we're pretty confident in our safety stack.
24:23
SPEAKER_01: And like at the end of the day,
24:25
SPEAKER_01: irrespective of the policy environment,
24:27
we want to be an enterprise grade company that can sell to the
24:30
biggest companies in
24:31
SPEAKER_01: the world,
24:31
SPEAKER_01: including in regulated industries and government.
24:34
And to do that,
24:35
you have to be high trust and above board on all the compliance
24:39
frameworks.
24:40
So generally I think we're in a pretty good position with the
24:43
safety approach and
24:44
SPEAKER_01: the compliance approach we've taken.
24:46
But yeah,
24:47
in terms of other issues,
24:49
you know,
24:49
think like robocalls,
24:50
AI disclosure and robocalls.
24:52
We have agent products that can actually proactively reach out
24:55
to people and
24:56
SPEAKER_01: execute those calls.
24:58
there are privacy dynamics where voice prints are considered
25:02
biometric information
25:04
that we pay attention to and then of course just like keeping
25:08
up with the broader
25:09
debates on you know a federal standard versus a patchwork of
25:12
state and local laws
25:14
SPEAKER_01: and I think for us that's you know
25:17
SPEAKER_01: AI Policy Brief a16z
25:41
SPEAKER_00: A16z
26:01
SPEAKER_00: A16z
26:27
SPEAKER_00: Obviously,
26:28
that list that you just went through is an incredibly broad
26:31
list of various
26:32
SPEAKER_00: different priorities,
26:33
SPEAKER_00: and it would take you a long time,
26:35
SPEAKER_00: a lot of effort to make sure that you're touching all of them.
26:38
SPEAKER_00: It sounds like at some point in the future,
26:40
SPEAKER_00: you're likely to hit this heightened disclosure threshold,
26:44
and that will put you in the same category as the companies
26:46
that you described as
26:46
SPEAKER_00: potential competitors,
26:47
SPEAKER_00: like the Open AIs and the Anthropics,
26:50
SPEAKER_00: but also
26:50
the Googles and the Meadows so can you say a little bit about
26:53
how that sits with
26:55
you like how do you think about this as you kind of go from
27:00
small footprint to
27:02
larger footprint that will come with heightened regulatory
27:04
obligations I think
27:06
we're you know fast approaching that threshold and that's a
27:09
good thing and the
27:11
I guess challenge with growing so fast is that you become
27:14
subject to different
27:16
SPEAKER_01: regulatory regimes.
27:18
So yeah,
27:18
I think as we achieve those thresholds,
27:21
we'll have to expand the team appropriately.
27:24
SPEAKER_01: I should say that I'm a team of one for public policy,
27:27
SPEAKER_01: but we also have a pretty big safety team.
27:29
SPEAKER_01: We have some safety policy people.
27:31
SPEAKER_01: We have a compliance team.
27:34
SPEAKER_01: AI Policy Brief a16z
27:52
SPEAKER_01: A16z
28:22
SPEAKER_00: A16z
28:35
SPEAKER_00: It's really interesting.
28:43
SPEAKER_01: I mean,
28:44
we're very much in some ways a European company with Polish
28:47
founders who care
28:49
SPEAKER_01: deeply about Poland and EU issues.
28:52
They were educated in the UK and kind of conceived as a company
28:55
in the UK.
28:56
SPEAKER_01: That's where we have one of our largest hubs.
28:58
SPEAKER_01: We incorporated in the US, a huge team in the US.
29:01
So I think like very authentically we are more global than a
29:06
lot of our competitors
29:07
SPEAKER_01: and that we have those roots in Europe and roots in the UK and the US and
29:12
SPEAKER_01: Yeah,
29:13
SPEAKER_01: I think for us,
29:13
SPEAKER_01: you know,
29:14
SPEAKER_01: often the growth path is like companies will start in California,
29:17
SPEAKER_01: they'll build for the US,
29:19
SPEAKER_01: and then over time they will localize for international markets.
29:24
SPEAKER_01: For us, I think Europe has always been a business priority.
29:28
SPEAKER_01: The UK has always been a business priority.
29:30
SPEAKER_01: So I think like we've had to
29:32
SPEAKER_01: So I think in many ways it works to our advantage.
29:35
SPEAKER_01: You know,
29:36
SPEAKER_01: I saw this at Meta and other companies where they felt very U.S.
29:39
first from a policy perspective and certainly there's now I
29:41
think a lot of pressure
29:42
SPEAKER_01: to kind of
29:57
choose sides and where do you want to build but I think we've
30:00
been able to thread
30:02
that needle very effectively and have been able to move fast
30:06
internationally
30:07
without having to you know kind of go deep in one country
30:11
before we're allowed to
30:13
go elsewhere so navigating jurisdictions in a complicated way
30:18
is part of US and
30:20
Europe policy work right now but it's also part of federal
30:23
policy work vis-a-vis
30:25
SPEAKER_00: state policy work in the United States there's
30:28
Heard different numbers but 1400 1800 bills related to AI
30:32
introduced at the state
30:33
level again you're a small team a mighty team but a small team
30:37
how do you guys go
30:38
about trying to make sure that you are not disadvantaged by
30:43
policy developments at
30:43
SPEAKER_00: the state level
30:44
SPEAKER_01: Great question.
30:45
SPEAKER_01: And it is a tall task, no doubt.
30:47
SPEAKER_01: I'd love to say that,
30:48
SPEAKER_01: yeah,
30:48
I'm across all the proposals and I've gone deep and everything
30:52
in California and
30:53
SPEAKER_01: Colorado and elsewhere.
30:54
SPEAKER_01: But sadly, there's just not enough time in the day to do so.
30:58
SPEAKER_01: I think our approach is really partnering,
31:01
SPEAKER_01: as I've mentioned,
31:02
SPEAKER_01: with the tech nets of the world,
31:05
working closely with you all and other partners who have those
31:09
kind of state level
31:10
SPEAKER_01: trackers and can give us
31:12
SPEAKER_01: A heads up when something is percolating that might impact us.
31:17
As I said before,
31:18
it's like,
31:18
you know,
31:18
I've engaged a lot with the California government.
31:20
SPEAKER_01: I think they've taken a very thoughtful approach.
31:24
Like there's certainly room for local nuance and how do these
31:28
tools impact
31:30
SPEAKER_01: Californians in a way that may be different than other states.
31:33
SPEAKER_01: So I certainly value the state and local perspective,
31:37
SPEAKER_01: but from a very pragmatic and practical point of view,
31:42
It's just impossible for a team of one,
31:44
a small startup,
31:45
to keep up with all the proposals.
31:47
And what the EU does very well is that they have a very
31:51
comprehensive and consistent framework.
31:53
And maybe it's more burdensome than other frameworks and there
31:56
are challenges and
31:57
SPEAKER_01: costs to comply.
31:59
SPEAKER_01: But at least you have that goalpost and you know what you need to do.
32:03
SPEAKER_01: You guys recently announced an expansion in New York.
32:07
SPEAKER_01: What was the rationale behind that?
32:08
SPEAKER_01: I'm curious if that means you'll be
32:27
doing more policy work in New York specifically it's a state
32:31
that's important from
32:31
a policy perspective they moved a transparency bill that ended
32:35
up looking a lot
32:36
like California's actually they moved that last session there's
32:39
a lot of stuff on
32:40
the table now so I'm curious about the expansion and then how
32:43
that's going to
32:44
SPEAKER_00: affect your policy engagement as well
32:46
SPEAKER_01: Yeah, so New York has always been our primary U.S. hub.
32:50
I mean,
32:50
obviously a great talent pool in New York,
32:53
an incredible city,
32:54
thriving,
32:55
tech scene.
32:56
SPEAKER_01: Some of it is practical in that,
32:57
SPEAKER_01: you know,
32:58
SPEAKER_01: with a lot of folks in Europe and the U.K.
33:00
SPEAKER_01: managing across time zones pulls a lot of people eastward.
33:05
SPEAKER_01: I think,
33:05
SPEAKER_01: like,
33:06
fundamentally we outgrew our office in New York and had more
33:10
demand and needs than
33:11
we could satisfy and were able to work with the government to
33:15
kind of secure...
33:17
a bigger office space and you know worked with them to
33:20
articulate our growth plans
33:22
which includes hiring over a couple hundred folks investing in
33:26
more kind of tech
33:27
SPEAKER_01: engineering talent I think we committed to a sizable R&D investment
33:33
SPEAKER_01: You also released 11 Labs for Government recently.
33:36
SPEAKER_01: What's the idea of that program?
34:02
SPEAKER_01: So as I mentioned, this was something near and dear to my heart.
34:05
SPEAKER_01: When I was at Meta,
34:06
I kind of split time between public policy and then government
34:09
partnerships and
34:10
SPEAKER_01: government sales.
34:12
You know,
34:12
our tagline when we announced the program was fixing the front
34:16
door to government.
34:18
SPEAKER_01: And I think in most instances,
34:21
SPEAKER_01: when you have an important need of your,
34:24
SPEAKER_01: you know,
34:25
SPEAKER_01: government at the local level,
34:27
SPEAKER_01: you're picking up the phone and you're calling,
34:29
SPEAKER_01: right?
34:29
SPEAKER_01: You're calling the school district to see if your kid's enrolled.
34:32
if you're struggling to figure out how to get a business
34:34
license you know you're
34:35
SPEAKER_01: calling the business department the tax authority
34:39
SPEAKER_01: AI Policy Brief a16z
35:03
it's a horrible antiquated in most cases automated system menu
35:08
based rigid
35:09
redirecting deflecting not actually taking action and resolving
35:13
any of your issues
35:15
SPEAKER_01: and I think the technology is
35:18
SPEAKER_01: A16z
35:38
that's a benefit from a company perspective but from you know a
35:41
citizen experience
35:43
and citizen outcome perspective like how many people give up
35:47
and don't get the
35:49
benefits and the support they need from governments because
35:51
they can't navigate the
35:52
process you know in California for instance I think we have
35:56
like 200 spoken
35:57
languages in the state huge Spanish speaking population with
36:02
our agents you can
36:03
SPEAKER_01: have a conversation live language switch
36:06
you know the agent can even pick up a local accent and speak to
36:10
you in Mexican
36:11
Spanish or Colombian Spanish so there are just like incredible
36:15
capabilities in
36:17
terms of personalization to help improve those citizen outcomes
36:21
and avoid the high
36:23
abandonment rates that we see too often I know the program
36:26
launched pretty recently
36:27
SPEAKER_00: are you seeing interesting use cases so far?
36:30
SPEAKER_01: Yeah, so we have a few existing government partners in Ukraine.
36:34
We've worked with the Ministry of Digital Transformation,
36:37
twofold project.
36:38
You had the minister who created the digital twin so he could
36:41
actually deliver
36:42
SPEAKER_01: updates in his authentic voice.
36:45
And then they have a citizen engagement platform called the DIA
36:49
app through which
36:49
SPEAKER_01: you can access dozens of services that was limited to text.
36:54
SPEAKER_01: And so we're powering the voice layer.
36:56
SPEAKER_01: AI Policy Brief a16z
37:11
SPEAKER_01: Amazingly,
37:12
I think 85% of calls are resolved by the voice agent Used to be
37:17
that 100% had to be
37:18
SPEAKER_01: redirected to a human Now only 15% So super powerful As you mentioned,
37:24
we just launched it So we're in a bunch of early conversations
37:27
I think there's a
37:28
SPEAKER_01: lot of excitement But in my mind,
37:30
I saw this evolution when I was at WhatsApp Where you had
37:34
almost no governments in,
37:35
SPEAKER_01: say,
37:35
SPEAKER_01: 2019 Delivering services through chatbots
37:40
by the end of 2020 and a lot of this was propelled by the
37:44
pandemic and like call
37:45
centers were overwhelmed and they couldn't keep up and you
37:48
needed to reach people
37:49
more proactively you know we had half of product volume coming
37:55
from governments by
37:56
like May 2020 in a five month period zero to 50% of product
38:00
volume in a five month
38:01
period and I think thousands of governments are now delivering
38:06
services through
38:08
SPEAKER_01: text, you know, AI powered text chatbots.
38:11
SPEAKER_01: Every single one of those I think can be improved.
38:15
SPEAKER_01: by building a voice layer.
38:18
SPEAKER_01: Think of the options for personalization, as I mentioned.
38:20
SPEAKER_01: You can only personalize so much over text.
38:23
SPEAKER_01: With voice,
38:24
SPEAKER_01: having that emotional intelligence,
38:27
SPEAKER_01: having local accents,
38:29
SPEAKER_01: switching across languages.
38:31
So I'm biased,
38:32
of course,
38:33
but I kind of see voice agents as that next big thing for
38:37
governments.
38:38
SPEAKER_01: So super excited to dig in.
38:40
SPEAKER_01: We have some early proof points, but a lot of work to do.
38:44
SPEAKER_00: Seems like an awesome project.
38:45
SPEAKER_00: We've talked a little bit about the growth of business.
38:48
SPEAKER_00: We've talked about the use of the product for government.
38:50
SPEAKER_00: Can you talk about just the general social value of voice as well?
38:54
Like what is the potential of voice for improving our economy
38:58
and our social life?
38:59
Yeah,
39:00
so super excited about the social value and exploring socially
39:04
beneficial use.
39:06
SPEAKER_01: Obviously,
39:07
SPEAKER_01: from an accessibility standpoint,
39:09
SPEAKER_01: there are a lot of people who prefer to engage with voice,
39:13
SPEAKER_01: and maybe they can't navigate complex web forms and so forth.
39:17
And so it gives them an extra tool in the arsenal to engage
39:20
with that content.
39:21
SPEAKER_01: We have some existing programs in place.
39:24
SPEAKER_01: So we launched a 1 million voices program, which is incredibly inspiring.
39:29
The concept there is that if someone has suffered permanent
39:32
voice loss due to an
39:34
SPEAKER_01: accident,
39:34
SPEAKER_01: due to a medical condition like ALS,
39:38
they can use our tools to create a synthetic version of their
39:43
voice and we give
39:44
them a free license in perpetuity to do that so they can
39:48
communicate naturally and
39:50
SPEAKER_01: authentically.
39:51
SPEAKER_01: AI Policy Brief a16z
40:05
and I think for the first time we worked with an individual who
40:09
was a singer in a
40:10
folk band and worked with him over several weeks maybe months
40:15
to recreate his
40:17
singing voice and we brought him on stage with his old band and
40:22
powered an AI
40:23
singing voice of his which then got picked up by national news
40:27
in the UK and
40:28
SPEAKER_01: created quite a splash so great foundation I think on you know the social
40:33
beneficial use side of the house and I think there's so much
40:36
more we can do and how
40:38
do we partner with governments right so there are AI for
40:41
science moonshots and AI
40:43
SPEAKER_01: for education and health care
40:45
SPEAKER_01: Ben, thanks so much for joining the AI Policy Brief.
40:49
SPEAKER_01: Thank you.
40:49
SPEAKER_01: Great to be here.