Technology & Security Podcast with Dr Miah Hammond-Errey

Dr Miah Hammond-Errey (00:00)
My guest today is Rohan Samaraweera. Thanks for joining me on the Technology and Security podcast, Rohan.

Rohan (00:06)
thank you. Thanks for having me.

Dr Miah Hammond-Errey (00:07)
Rohan Samaraweera has just founded Emrest, an AI advisory firm.

He spent 25 years working in intelligence, national security and technology, and most recently he led enterprise AI and analytics capability for the Department of Home Affairs. Prior to that, he spent 16 years at the Australian Signals Directorate, including Time Overseas Embedded with Partners. He spent a year at the Harvard Kennedy School Belfast Centre on an Australian government fellowship researching AI and national security.

I'm coming to you today from the lands of the Gadigal people. I pay my respects to elders past, present, and emerging and acknowledge their continuing connection to land, sea, and community. Rohan, what is the tone of AI conversations inside government at the moment?

Rohan (00:46)
at the moment I there is a lot of conversation about the adoption of AI. there's been a big change since the emergence of ChatGPT or release of ChatGPT a couple of years ago. prior to that we were very much focused on building in-house capability in predictive analytics, natural language processing, computer vision.

natural language processing, sorry, I think I already said that. a network graph. but with the event of LMMs and Chat GPT Claude now our people have access to that technology and really want it. so the focus has been has has changed from building in-house capabilities to how do we get these

commercial capabilities into our agencies and how can we use them in supporting the national interest.

Dr Miah Hammond-Errey (01:37)
I read somewhere that you're a huge part of Delivering ERMIS, a national AI screening platform, which was recognized with the APS Data Science and Analytics Award in 2024. Can you tell us a bit more about that?

Rohan (01:50)
Yeah, so Hermes was a predictive model. so we applied experience and capability that we that was developed and developed prior to me actually joining the department as well in the visa domain in applying predictive analytics to operational border operations. my predecessors had built capabilities in the Visa risk domain system.

Hermes was an application of similar approaches to international mail, following on the government getting industry data in the international mail space for you know the use of governing the the border. so Hermes was applying fairly traditional predictive analytics, we were basically taking

all of the combined insights and detections of data of border officers and some of those and you know that have led to results and using that to train models to predict the potential risk of illicit drugs coming through or illicit items coming through the international mail space. so it's an application of of tradition traditional

predictive analytics. the yeah, it it is just one yeah, so

Dr Miah Hammond-Errey (03:03)
What what was the impact? 'Cause w like if anyone

I mean, for the listeners like who haven't been to an international mail sorting center, I have had the pleasure earlier in my career. They're chaotic, right? Like there's thousands of packages coming through and being sorted and organized. So yeah, like what was the impact coming out of this platform?

Rohan (03:23)
The case of Hermes was that it did result in bringing or or successfully identifying two or three hundred, sorry, three or four hundred million dollars of street value drugs that were entering or going to Australia. we undertook some more analysis supplying really, you know, a lot of law enforcement agencies you always say, hey, we we took out a billion dollars of of drugs or four three hundred billion dollars in this drug bus.

you know, what does that actually mean? and we applied or did some research in applying w what that social impact was. So we can actually get some some some metrics that mattered. and using some of the research from New University of New South Wales and Pennington Institute where they've actually calculated the social harm of specific illicit drugs down to the gram, we were able to to

provide a rough estimate of impact of just that one targeting method upwards of around two two billion dollars over you know a period of three or four years. so that certainly is quite a motivating achievement for the data science team. and you know it ha has had was a way that we could show

what that impact is for for the investment

Part of the challenge that we have, particularly with data science and AI, is that if we're successful, we do disrupt the status quo. And what that also means is that it's often hard to quantify what that return on investment is. it's often a case that the return on investment actually falls outside of

you know, your area. So if the ABF is really good at stopping drugs, you know, two billion dollars, they're not seeing seeing that saving on their balance sheet. In fact, they may have to, you know, increase investment to achieve that kind of impact in the community. and the actual cost saving of that yeah ach achievement may be maybe on on someone else's balance sheet.

in you know in this case state territory health and law enforcement areas. and that's a hard sell to take that to federal government and to the Department of Finance to to say, well here's our return on investment because a lot of these AI capabilities and a lot of the focus around AI capabilities is how do you gonna reduce costs within your organisation.

whereas I think a broader view is much more important for the national interest and we haven't found a good way to to account for the returns you know f for for the country.

Dr Miah Hammond-Errey (05:50)
Yeah, I mean that's a such a important segue because obviously recently the Prime Minister has announced that they'll have a new office of AI And so that that's a that's a new function that one hopes might be able to look across the whole of government and see some of those benefits and public value across

you know, multiple competing areas and domains. Because as you say, to achieve genuine public good, we can't, you know, we can't balance the sheets in one agency. They're gonna be across state

territory, you know, a range of different public services and and to to make choices about the investment in specific platforms or applications, we actually really need to see what the whole downstream impact is, not just in the specific location the investment might be being spent in. Do you think it's a good outcome? in terms of driving driving a better understanding of that kind of broader picture.

Rohan (06:48)
I hope so. but it's it's gonna be a challenge because

Dr Miah Hammond-Errey (06:51)
Yeah.

Rohan (06:51)
it does it's going to take a bit of creativity and a bit of a broader look at or asking a much more you know applying curiosity towards these capabilities. and often you know one of the challenges

I've faced is you often get asked, you know, what is the return on that the work, you know, on the resources that you're you're putting in. and often the answer to that is actually outside of my remit or outside of you know the the team that's developed it or even the agencies that's developed its control. so there is a need for greater coordination in terms of

broadening some of those relationships and partnerships, So that you can get those required metrics, you can close that feedback loop. One of the biggest challenges and I think frustrations for those that have been building data science capabilities and AI capabilities, both in in industry and the public sector,

is that the investment to close that feedback loop. in some cases is not there. a lot of the systems that we rely on as a nation or as an economy are ancient systems that that you you know you might have heard the term or or phrase system archaeology. and you know that takes you know quite a bit of creativity

some McGivid solutions to really be able to you know solve that problem in terms of getting getting that feedback loop closed, which is critical for the governance of AI models going forward.

Dr Miah Hammond-Errey (08:17)
Robo Debt is a stark caution retail of failed automated government decision making. did that failure change about how you think about automation and AI?

Rohan (08:26)
so Robo debt for me there's a lot of argument about whether it was AI or not, but it doesn't really matter because what it essentially gets down to is it's the decisions and the infrastructure around, you know, a a s technical process that was put in place. the structure and governance wasn't

you know, to to the standard that we need for AI was certainly not applied for Robo debt. and for me, as someone trying to uplift government standards within my remit, Robo Debt was not only just a cautionary tale, but it was also a great support in terms of being able to have those conversations with, you know

or not just stakeholders but also with my own team about why they needed to do more in terms of applying a process, a framework and a and and working through steps and not just going out and building like a cool capability that's you know supporting a particular national security challenge or or administrative challenge, but actually thinking about, you know, what is the purpose of and why are we doing this.

And is it legal? does it meet and does it comply with current policy and legislation? Robo debt absolutely did raise a lot more awareness, a much you know, pen you know, across a broader range of of people and actors involved. so while the

instance of Robo debt has caused a lot of damage. it has helped that conversation move in the right direction in terms of putting governance and standards on the agenda and establishing much more a much more robust conversation

and a a lot more willingness to really you know improve governance and manage risks in a more accountable manner.

Dr Miah Hammond-Errey (10:24)
I'm gonna zoom right out. to what's called the contest spectrum. What's a cooperation, competition or conflict that you see coming in the Australian government use of AI or AI policy in the next twelve months?

Rohan (10:36)
So in Australian policy

In terms of cooperation, we're going to see a lot more cooperation between I guess non-AI superpowers, so between ourselves and other states.

Dr Miah Hammond-Errey (10:50)
Everyone that's not the US and China.

Rohan (10:52)
Yes. Exactly. so and and and that's going to be at at a at a working level.

I think there's been a lot of cooperation certainly in the AI governance space. But where cooperation can provide a lot more value is in the deployment space. So that is in the realm of data collaborative model developments on the and in you know outside of the frontier models. Let's just look at the other other side of the coin in terms of that building

you know more specific sovereign capabilities. Certainly saw that from my time in government where

AI and data science provides real opportunities for partnerships where you know each of our each of our governments, each of our our you know, the systems and and functions that are managed by government, a lot of the data collected is is you know it's like having a sensor. so there is opportunity for partners to benefit from threats that we are much more

aware and developed to protect against and for a vice versa exchange of of that protection. so I I do see particularly as a bit more concern over the concentration of power in the frontier model space. I I do see that that concern will drive

more conversations for a broader base cooperation, really to diversify risk and support sovereignty and agency.

Dr Miah Hammond-Errey (12:27)
you're leading right to my next question. you've written that access to Frontier AI is a dependency, not an asset. I agree, but talk us through your thinking here and why it matters.

Rohan (12:36)
So there's two ways of looking at that problem. One is obviously at the national state level in terms of having the ability and the agency to have not just have access to the technology but also be able to use it. The same argument is true when it comes to outsourcing, you know, even at a domestic level.

when we're in government the question that that we all have to ask is if I lost my key vendor or if my contract workforce you know left what would my capability look like? so in terms of th how I see sovereign capability

a licence is not sovereign sovereign capability. It's having that deployment muscle, that deployment infrastructure to be able to perform your function is the sovereign capability. and we need to we need to to s to safeguard that agency and be able to to have the ability to to not

Be, I guess, you probably heard the term vendor lock-in. you know, it's the same at the state level in terms of having access to the sovereign models. you know we need to understand what those risks are, and we need to know how we could operate if we didn't have them.

and at the same time we need to have a plan B for for those models as well. Well not for for other models, sorry, for for those capabilities as well. One thing I would try and do would be just on a

I guess an operational level is I did reduce our reliance on contract the contract workforce. And where I did have a key contractor, having them shadowed by a junior APS person to to manage and to develop their capabilities and to learn how to run those systems was really important. so we're going to need to build that kind of thinking in you know, across all of our engagements. and it possi and not just to

with government as well. I think if you're a company you you don't want to have your you don't want to have your competitive advantage completely dependent on access to a model or a product or a system that you have no control over but and that could be taken away from you you know without any any warning.

Dr Miah Hammond-Errey (14:59)
Yeah, absolutely. coming from a national security perspective, it obviously comes fairly easily to us to understand that the notion of critical dependency. I think it's taken industry quite a while to sort of re-shift its focus to see to see how vulnerable it really is if it's reliant and Mythos and Fable being cut off among many other examples. Even, you know, the crowd strike outage, for example, provided

tangible proof of what happens when you just don't have access. obviously there's a huge discussion in Australia at the moment about how we can increase, you know, some people like the term agency. I don't because I think it implies that it implies a simple trade-off, which

it allows you to ignore many of the fundamental inputs to that capability or agency. So, you know, when we talk about something like data centers, we often see people say, well, if we have data centers here, then we will have more leverage. Now I think data centers are a good thing and I think we need data centers to be a part of, you know, an evolving AI economy, but

The way those data centres are structured and the access that Australians and Australian capabilities have to those is actually more important than the data center itself, if that makes sense. I've written about it in terms of the tech stack, but how do we think about securing aspects of AI or technology stacks for Australian sovereignty?

Rohan (16:19)
So that's a really good point. and particularly on the data centers. you know, having them on on on our sovereign territory certainly is going to enable the traditional view of data sovereignty, which is you you want to have your data stored on on your onshore. You don't want to be giving sensitive data and and hosting that offshore on a foreign owned

platform in a f in a foreign location, but that only supports, I guess, part of the picture. in this space it's what kind of stakes do we have in terms of using those capabilities for our own purposes. what kind of agreements, you know, can we put in place and can we negotiate?

with the foreign investment that comes in to make sure that there is a portion that's being provided to support Australia's national security, to support our innovation ecosystem, to support our in institutions, our academic institutions and their research programs as well. you know, there is a

a massive competition for for getting, you know, GPUs and compute. and you know the

The the hyperscalers aren't building data centres in Australia to support our sovereign capability. You know, they're building them for their own purposes, and they have their own interests. And yeah,

Dr Miah Hammond-Errey (17:34)
Yep. They're commercially viable.

Rohan (17:38)
exactly. Exactly. So, you know, we how do we have stake in that regard? and Australia's had some good historic experience in negotiating treaties with

with our allies in terms of making sure that we we have states, you know, where we we do in some cases have Australians you know working embedded in highly sensitive important programs for our allies where we are a true partner. the question is how do you do that with a foreign commercial entity? because commercial entities can pull out of Australia for commercial reasons as well.

And the other question is who does that negotiation? Is that something that government does?

it's quite a a challenging proposition, where you do see that diffusion of power. where, you know, twenty years ago it would be between state actors and that's a lot more predictable. But now that we've got a lot more

Dr Miah Hammond-Errey (18:28)
Arguably.

Rohan (18:30)
Yeah. Arguably. and we we're whereas now we've got companies, multinationals, international companies that have a lot more power than they did, you know, twenty or thirty years ago.

Dr Miah Hammond-Errey (18:43)
So how do you see that countries like Australia can engage either in the current, you know, AI or even broader technology economies and future ones without simply replicating those dependencies on foreign platforms and companies?

Rohan (18:56)
Australia has to build its own assets and build and protect its own assets. So we're talking data, the you know human expertise and that deployment infrastructure to be able to apply AI for our own commercial security and yeah service purposes.

in terms

Dr Miah Hammond-Errey (19:22)
Everyone's focusing so much on

LLMs and frontier models. I'm not discounting that, but rather than focus

Rohan (19:28)
Yeah.

Dr Miah Hammond-Errey (19:28)
There there is a really big opportunity here for us to focus on building small, reliable AI systems that actually work for the purpose that we want them for. you know, offering,

small companies and research institutions the compute capacity to innovate new and exciting things. We don't have to compete at the Frontier LLM space. Yet there's just so much focus about that rather than where can we start investment.

Rohan (19:57)
Well, you've basically just identified the main frustration for a lot of the data science, artificial intelligence community. I I do often have people reaching out from small, you know s small startups and and companies that are doing you know, trying to build service industries using AI, using small language models, building their own in house capability.

and they do feel like the the focus has has has moved away and it is a bit of a distraction I think from the main game. and I I would agree with you in t in terms of focusing in on you know what data do we have that's unique and Australia has a lot of very good

data in a whole range of domains and sectors where we are world leaders. We have great talent in the country. we do have a lot of really good talent in the AI AI domain. and when it comes to infrastructure.

you know, we we have the deployment infrastructure in in Australia to solve Australian problems. the LMMs have have take have have stolen a lot of the limelight and a lot of d the discussion and a lot of the attention and it's taking it away from

areas of focus where we were doing some hard work. Well the the country, you know, people working on AI and data in in Australia have been working on maturing the the capability and building the deployment I guess muscle here. and the conversation ha

has sort of shifted to, you know, w a lot of the debate in the media has is about, you know, how do we get access to these these big, big systems. But at the end of the day, LMMs aren't going to grant visas. they're not going to identify drugs in mail, and they're not going to improve the precision of a of a guidance weapon in the defence space. work has to be done in in other areas.

for high trust sovereign systems that solve specific problems that we need solved.

Dr Miah Hammond-Errey (22:04)
Yeah, absolutely. And it it's such an important point to bring it back to the purpose rather than trying to retrofit an existing capability onto something to make it AI scalable. The idea is actually to solve the problem by using AI that works for the problem that you have. And it it's obvious it sounds so simple, but so often in the technology space, we get caught up in what's shiny and new rather than

focusing back on the essentials of what what is the purpose? What are we actually trying to solve here? And that is such a critical point because so many of these, so many of the difficult problems that we need to solve and the public good outcomes from government are not about things that you can retrofit an LLM or some other, you know, model. They're something that you need to purpose build something for. I run a startup and we have a scalable AI

component to our platform, if that component isn't available, we can use other other capabilities internally. if you rely on that too much, you're then in a position where your scalable platform is now critically dependent. that doesn't bode well.

If we want to shape Australia's AI future, what do we need to think about and put in place now?

Rohan (23:15)
we we need to think about our how we how we invest in AI. Australia's been a

a bit of has been a laggard when it comes to investing in in AI compared to a lot of our friends and allies, not just in the US and China, you know, not just the US and China, the big superpowers, but other countries are investing billions of dollars in deploying and building AI capability and putting billions of dollars towards AI. we have not invested

we need to think about

investing in building in in matching the good work we've done in terms of governance with investment in being able to make it enable it. you can't just create a governance regime and say, right, everyone comply to this. Sometimes that compliance is going to take investment in terms of training people. Sometimes it's gonna take investment in terms of potentially replacing an old obsolete system that has really bad data.

it's going to require investment to close that feedback loop and do some re you know change the architecture in which that that operates as well as invest in new capabilities to measure what those those impacts and results are

we certainly need to look at support how we support our educational institutions, how we bed down governance infrastructure and standards, and how we measure results. the other point I would make would be

how we invest in developing an ecosystem or an innovation ecosystem that is able to

get success in deploying ai and creating real impacts in our community, our society and and our economy as well.

Dr Miah Hammond-Errey (24:56)
What are some of the interdependencies and vulnerabilities between AI and security that you wish were better understood?

Rohan (25:01)
I wish that it was better understood what the impact of our human workforce is in in AI. There's a lot of talk about automating whole jobs and and reducing the numbers of our of our people, not just in government but in in companies as well. And we've all seen the stories of the startup or the

technology company that sacked a whole bunch of people then asked them to come back. I wish people understood that they're that the I wish it was better understood that

people are a critical part of operating and running AI to to achieve outcomes. and it's only you know if you deploy AI against a particular national security problem, what happens, we all know. you know n your adversary will change tactics and there'll be a new pattern and that pattern may not be something that

It may be something that AI sees, but it may not be something that AI necessarily, can i identify. I'll I'll give you an example. so hanging out at the International Mail Centers with with ABF staff, one of my visits,

The ABF officer picked up a package off the conveyor belt and just stuck it straight onto the relevant pile for a partner to come and take away. and without even opening it. And I asked him why he did that, and he said, you know, that that is this substance, it comes from this person. I see them come through all the time, I just pick them up and stick them straight onto the pile.

And sure enough he was right, it was the exact substance that that he said was gonna be in it. and you know that was yeah, you know, his insight in terms of of picking them up.

Now I had to ask a question, like, if you're always picking up those packages, why are they still being sent? And the answer to that question is probably that there's a whole lot more packages like that that are going getting through the system when that particular ABF officer isn't on, or if it goes through another mail center where that particular I guess, signature or that particular t style of package is not not not necessarily understood.

So, we built a system in terms of being able to identify or at least provide a you know, explore the the idea that you could actually identify packages by sight So imagine that officer picks up that package, logs it, and then we have a system set up that that computer vision system that goes and identifies all those different packages. The adversary is going to change.

But it's probably gonna be the border officer or someone else on that border officer's team that's going to either see what that change is, or alternatively, you might see a reflection through a different channel. those are out of the box events or impacts where human problem solving and thinking is going to be key.

so we need to value our our human expertise. and I guess my frustration with the the current debate and the current focus on automation is that it it is missing that critical part that I as someone who's been part of a team that's been building and deploying these capabilities.

you know sees that criticality of of having humans in that as part as a as a core part of that of that that the AI you know AI infrastructure.

Dr Miah Hammond-Errey (28:22)
which brings you me perfectly to another segment, which is emerging tech for emerging leaders. What then are the limitations of AI you wish leaders currently understood to help bring in this human perspective?

Rohan (28:33)
you ask very good questions, Miah Yeah, so so the the limitations of of of of AI really get down to that they're only as good as the data that they're trained on.

as well as how they're tuned and how they they are constructed. and that really does get down to the skill of the the people building those capabilities, as well as the data sets that we are able to have access to and the quality of those of that data. and the provenance and you know of that data as well.

The other limitation is that perhaps I'm speaking from from my experience within you know very specific channels and very specific capabilities,

and they duck can't necessarily adapt to a you know black swan event or to external, you know, sharp changes that that might come through. they might pick up, you know, patterns over time but

those big shocks are are you know where human expertise comes to a fore.

Dr Miah Hammond-Errey (29:36)
What does meaningful AI avis oversight actually require from government and how is it different to business?

Rohan (29:42)
I

think both both government and business need to embed AI governance into the their capabilities as they're being built.

it's very difficult to to apply a standard and then retrospectively fix fix all of the infrastructure you put in place that might not be up to speed or might not be up to standard. So it's really important to embed AI assurance at that ground level and to do it from when you are you know starting to you know right to the point of of identifying the purpose that you will want to apply AI AI against.

Dr Miah Hammond-Errey (30:16)
2024 Independent Intelligence Review called for an AI leadership focus and an uplift in awareness and education across the community. What's happening in this space and what do Australian and Five Eye AI governance forums and frameworks look like?

Rohan (30:31)
When it comes to executive AI literacy, there's been a lot of development in terms of creating a training for a senior executives to learn about AI.

term in terms of frameworks, there's been a lot of work in developing standards, a lot of engagement across the APS in building AI frameworks.

Dr Miah Hammond-Errey (30:55)
I'm gonna go

to a segment called grounded. What should we stay focused on and connected to this year given the pace of change?

Rohan (31:01)
I think we need to not get too head up in trying to keep track of you know all of the AI models that are that are going out and all of the announcements. and I I think we need to really focus in on ourselves in terms of our own our own capability.

For you know leaders, one of the biggest problems that we or biggest challenges that we we face is we we aren't terribly good at identifying or or articulating what our strategic challenges are in a way that can then start that discussion about AI. There's a lot of discussion about how do we apply AI.

and there is a lot of a lot of work going into identifying use cases. a lot of the time those use cases that that come out are quite peripheral to our actual core strategic problems and we are can be we we are timid when it comes to applying

AI against some of those key strategic problems that we we do need solved.

Dr Miah Hammond-Errey (32:05)
I want to go

to another segment on alliances. What relationships should Australia and Australian companies focus on?

Rohan (32:11)
we should be focused on building an innovation ecosystem that provides that infrastructure for advancing AI across the various different domains. So rather than focusing in on, you know, can we get access to Chat GPT or Fable or Mythos, we we need to be building these broad based communities where we have government universities,

corporates, we have an entrepreneurial community, and we have finance as well, all operating and we have you know to build that infrastructure so that we have the mechanisms in place to to advance AI, to develop and build new capabilities, to commercialize capabilities here in Australia.

We we need to have a market here to to build capability. and there's there's definitely a lot that large institutions, whether it's governments or potentially our financial sector, you know, can invest in in companies that are promoting and developing innovation.

Dr Miah Hammond-Errey (33:10)
Let's go to another segment. It's called disconnect. How do you wind down an unplug?

Rohan (33:14)
Finally, an easy question.

I love cooking and I also like a lot of your former guests very much enjoy getting out into the outdoors. I like bikepacking and hiking.

and I'm a bit of a gear nerd when it comes to that in terms of I have way too many bicycles and I have way too many tents, but they're all designed to do very, very specific things and go to very very specific places, having the right tool for the right job. the other thing I'm doing a bit more is I'm actually I wouldn't say torturing, but I I am trying to teach my kids using the Socratic method.

So that they are yes,

Dr Miah Hammond-Errey (33:47)
They would absolutely say you're torturing them, just by the way.

Rohan (33:51)
they would, they would. So where I ask my sons lots of questions, to help them guide, you know, guide them through getting to the right answer. I think that's really important, particularly where increasingly we don't have to dig too hard to get

answers and information because technology's so good. so building that that muscle so that that people can do it. You know, my my sons can do it on

Dr Miah Hammond-Errey (34:15)
I run a cognitive readiness platform helping leaders to think you know, improve their cognitive performance in under pressure. And absolutely that is something that I advocate for.

I wanna go to one of my last two segments. It's called Eyes and Ears. What have you been reading, listening to or watching lately that might be of interest to my audience?

Rohan (34:33)
I do listen to a lot of podcasts and occasionally those podcasts lead me to authors. so the newest author that's come popped up on my list has been a a man called David Epstein. He's written a book called Inside the Box, which really talks about how it constraints can enhance creativity. really, really interesting.

premise and something that resonates with with me in terms of some of the capability some of the experience I've had professionally, where some of our best results have been in very, very resource constrained environments.

and the the really good case study that he uses is actually Doctor Seuss. I did not realise that the Cat in the Hat was written on a bet to where Doctor Seuss was told you're gonna write a book and it has to be fifty words or less. he said, Right, I'll take that and he wrote The Cat in the Hat and we have, you know, one of the best selling books in, you know, human history.

I keep track of an economist called Eric Brinjolfsen and I I know that a lot of people in the AI space refer to to to him. He wrote a book called The Second Machine Age a few years ago, which really talked about and really explained how electrification

changed our society and and really improved living standards across the the the board. But it also talked about how it took 30 years to move from the previous industrial age to electrification, getting to the point where we have, you know, consumer el electric goods actually took 30 years because I did try to retrofit

electrification into their existing infrastructure and mills and whatnot. Which has a lot of resonance for me in terms of trying to put AI into these existing systems in using old old methods. so that that resonates with me.

Dr Miah Hammond-Errey (36:21)
My final segment is need to know. Is there anything I didn't ask you that would have been great to cover?

Rohan (36:26)
It it'd be it would have been I think it would have been great to cut go into a bit more depth into how how organizations innovate. Like the difference like between trying to innovate from the inside to to being you know, between entrepreneurship and entrepreneurship. I think there's some real real challenging insights in that space that

Dr Miah Hammond-Errey (36:44)
So take me there, what what what is the what is the difference between

Rohan (36:45)
would have been good good good to cover.

Dr Miah Hammond-Errey (36:48)
entrepreneurship and entrepreneurship and what can we take away from it?

Rohan (36:52)
there's a concept called the one no versus one yes problem or paradox. and that is that if if I was an entrepreneur, if I was, you know, Steve Jobs, all I need to get started is one yes.

And I can go out into the big white world and I can find that yes anywhere. So I can drop round to my mother's and say, hey mum, me and my other mate Steve, we want to rock up and can we have your garage? We want to set up our own computer company. And she says yes, then you know I that's a step forward. if I need funding, I can go to a VC. If they say no, I can go to another VC, I can find an angel investor. all I need to keep going is

Find yes, and that and the challenge is in this big wide world where you find, you know, where do you find that that yes? It's out there, you know it's out there, you just need to go and find it. When you're trying to innovate on the inside, you could be all it takes to stop you in your tracks is one no, and that one no can come from anyone, it can come from the secretary of your agency, it could come from your CEO.

at the high level saying, look, priorities have changed, we can't do that, or that it's no longer a strategic priority. Or it could come from like someone very, very junior, someone who doesn't give you access to a data set or doesn't give you access or system administrator doesn't give you access to a system. And so the challenge there for entrepreneurs is that we have to try and reduce the probability of those no's.

we have to be constantly, constantly problem solving and working through all of those all of those challenges in terms of of avoiding those no's as well as building a coalition of of partners who are willing to to help advance that capability. so there is a real

challenge. in you know, you often hear, you know, businesses saying, well well, you know, it must be easy in government. It it's actually quite quite a difficult proposition to to innovate in in an organization.

there is one more point though, Miah, actually, and that that's probably a really, really important one, which is we have a phenomenal problem and maybe this is the answer to the A the question about what I wish people knew about AI, and that is that I wish that

institutions understood that responsibility for AI is actually shared across a whole range of different areas. Who is responsible for AI is probably one of the hardest questions to answer in any organization. And the truth of the of AI adoption is that it it actually belongs to multiple people.

we have to get better at allocating the accountabilities across the spectrum when it comes to an AI capability and understanding what is this capability for? Who is the the business owner and user of that capability? How's it going to change our workforce? What's its impact going to be from a resource perspective?

cybersecurity, you know, all of those other impacts as well. and that's probably the biggest challenge when it comes to AI governance as it stands.

Dr Miah Hammond-Errey (39:57)
Rohan, thank you so much for joining me today. It's been such a pleasure to have you on the Technology and Security podcast.

Rohan Samaraweera (40:02)
Thank you very much, Miah really appreciated the opportunity.

Dr Miah Hammond-Errey (40:05)
TS listeners, I will be in Rome at the NATO Defense College doing some research until 2027, but I'll keep hosting technology and security, so listen out for my upcoming European guests.