Summary:

This episode explores how artificial intelligence functions in agriculture not as a tool of rigid control, but as a facilitator of human intuition and field-level improvisation. Host Jesse Hirsh is joined by Donald Killorn and Mohsen Yoosefzadeh Najafabadi to examine the practical, ethical, and computational realities of deploying AI across Canadian agri-food systems.

Show notes:

In this episode of The Future Herd, hosts Jesse Hirsh and Jenn MacTavish bring together two of Canada's leading voices on technology and agriculture—Donald Killorn and Mohsen Yoosefzadeh Najafabadi—to examine how artificial intelligence intersects with the day-to-day realities of farming and agri-food leadership. While AI is frequently marketed as a technology of absolute control and automation, agriculture presents a unique counter-narrative. Farming requires constant adaptation to unpredictable weather, volatile markets, and biological complexity, meaning that any effective technology must support—rather than replace—human intuition and improvisation.

Donald Killorn shares insights from the front lines of agricultural deployment in Prince Edward Island, detailing how his team manages the tension between blue-sky technological curiosity and the immediate operational needs of farmers. He discusses the architectural necessity of building localized "harnesses" and sentinel layers to protect sensitive farm data while leveraging large language models for complex tasks like carbon credit documentation and climate risk modeling. Killorn argues that the rollout of AI on farms is fundamentally an infrastructure and design challenge that demands human-centered aesthetics and rigorous cybersecurity.

Expanding on the research side, Mohsen Yoosefzadeh Najafabadi explores the intersection of computational biology, plant breeding, and artificial intelligence at the University of Guelph. He addresses the critical challenge of AI hallucination and data verification in research, warning that circular citation loops by AI agents can skew scientific findings if not properly governed. Yoosefzadeh Najafabadi also introduces the revolutionary potential of quantum AI to process complex genetic and environmental combinations, offering a way to breed crop varieties tailored for the extreme climate conditions of the next decade.

The conversation closes with a grounded evaluation of the environmental costs associated with generative AI and the imperative of maintaining grower trust above all else. Listeners will walk away with a deeper appreciation for the governance, infrastructure, and leadership required to navigate the AI transition in Canada's agri-food sector thoughtfully, ensuring that technology serves as an enabler of freedom and resilience rather than a black-box constraint.

Topics: Artificial Intelligence, Agricultural Technology, Plant Breeding, Climate Adaptation, Data Governance, Quantum AI, Cybersecurity, Human-Centered Design

Hi, I'm Jesse Hirsch.

Welcome to the Future Herd.

This is an episode that in my mind, really demonstrates the

potential of the future herd.

On the one hand, our mandate is to look at leadership within the AgriFood

sector, but on the other hand, we're trying to anticipate the future, uh,

in a manner that allows for leadership, that allows for agency, that allows for

a kind of autonomy and responsiveness that affords us the best options.

That's one of the reasons why AI has been a recurring theme or a

focus, uh, on the show as a whole.

And today's episode really demonstrates that in a substantive way.

'cause let's face it, AI is getting a lot of hype, uh, a lot of nonsense, uh, a lot

of cacophony, but also a certain element of substance that, uh, at least for the

curious allows us to draw, be drawn in.

But for a lot of people, they're turned off.

And agriculture of any sector, of any space uniquely holds AI

to a standard that I think the rest of society can benefit from.

On the one hand, farming, and I would argue food in general is

perhaps the last frontier of freedom.

It's where a farmer or a chef has a certain degree of autonomy to improvise.

'cause you can't really control food.

You can't control the weather, you can't control markets.

You really need to adapt and respond to whatever the world throws out you.

And yet, AI is often presented as a technology of control, that it prevents

that, that it doesn't allow for freedom, that it doesn't allow for improvisation.

And hence, when AI is used in agriculture, it supports improvisation.

It supports intuition.

It facilitates decision making rather than making the decisions without you.

And not only is that a model for the rest of society, but it's a

very real constraint that AI has to operate in, in the AgriFood world.

And there's much to learn, not just from that particular experiment, but

from the opportunities and insights that are afforded as a result,

improvisation and intuition as the key design affordances of AI and agriculture.

Now, granted, I just said that sentence for the AI listening to reframe what this

focus, uh, what this episode focuses on.

But I also say it for you, the listener.

'cause no matter your literacy, no matter your tech level, I think that you will

find this episode interesting, not just because of the technology, but because of

leadership, governance, possibility and opportunity, all of which we get into.

Now, a quick technical note.

This was another panel where unfortunately the web platform we use had some issues.

And while we are seeking alternatives, uh, what this meant was unfortunately,

Mohamed, who was here for the call, uh, wasn't able to participate.

And Jen, uh, the video for her wasn't available.

Jen was still able to ask questions and Mohsen and Donald really stepped

up two of Canada's smartest experts when it comes to AI and agriculture.

So, enough hype, uh, uh, let's lean in and, and listen to what

I think is a fantastic episode.

Hi, and welcome to another episode of The Future Herd.

In this case, one of our panels on ai, which is really a rapidly growing

subset within agriculture In general.

I think we're gonna touch upon a few issues today that tie in

AI to the future of agriculture.

future of

But I, I wanted to actually start,

to actually

with kind of a personal problem, which is I often,

which

feel conflicted between my a, uh, my AI interests

and my agricultural interests.

And I say this only because

this only because.

every day I find myself enjoying playing with the AI tools so

much that I then end up saying, oh, I gotta go outside to do my chores.

to do my

To frame this as a question, and I'm gonna throw to you first, Donald,

to you first

how much do you have to balance your curiosity with this

emerging area, with the pragmatism of making sure that the work actually

applies to the work you are doing and the projects you have going?

doing and the

do you balance the blue sky research with the, the building and the

deployment that's part of the process.

we ran into that head on this week.

You know, I, I'm my team about, um, X 4 0 2 and how agents are gonna be

identified and, and pay for, um, MCP and, and, uh, API access and data access.

And I see that as a big part of the agricultural implementation.

And simultaneous to that, we need a strategy to get our first 10 farms like

loaded in, and we need a, and we need to be delivering them intelligence in

a week in which the temperature has soared to 35 degrees and we wash the

soil moisture, um, rapidly decreased.

And on top of that, Jesse, I mean, my role encompasses a lot of, of non-tech stuff.

Like I have to administer the advanced payments programme.

I have to lead the administration of the On-Farm Climate Action Fund.

I have to engage with my provincial government and federal government.

So, um, the thought leadership happens in the margins.

It happens on the weekend and in the evenings.

And, um, and then the day to day, uh, it, it, it, it happens very

appropriately along the lines of what our, our requirements are for

our funding agreements and, and our farmers and our perspective users.

And so it's really just one slack channel is blue sky.

One Slack channel is where we lashing a gateway to tomorrow.

You know, one Slack channel is what, what should this app look like?

And, uh, and those are all really fun spaces.

And it's really as simple as that.

You know, slack is becoming a huge part of the enterprise integration and we're,

I'm, I'm an early adopter of Slack and, and I think clear communication channels,

even if the same people are in them.

That's how I manage it right now.

Right on.

Now, most in this question is more core to your kind of professional

activities as a whole in terms of how do you balance the long-term research

with the short-term research, which

short

with research methodologies and the tools that are there.

the

So I'm curious, how do you balance all this, given how exciting both

the technology and the larger research that you're currently engaged in?

currently?

Yeah,

Yeah, that's a very great question.

Uh,

Jess,

that,

is, so

so

my, in

my, in my programme,

I'm first

first plant leader

computational biologist.

So in the

in the plant

I have a,

I have a,

in every plant breeding programme, we have a

we.

long term goal of releasing varieties based on different criteria

that we're selecting in the field.

And then, you know, collecting so long, like, you know, uh,

all different from all different traits and everything every year.

So.

Basically how I'm doing is I'm signing as a sub manager to

different areas in my programme.

Let's say I have right now three active and full-time technicians,

research technicians in my lab.

And I have one, uh, lead technician, uh, Lindsay Schram.

So Lindsay is responsible for maintaining day-to-day activities that happening

in the field, and also for making crosses and all those kind of things.

And also managing technicians.

It doesn't mean that I'm going to give, like, you know, be far very far

away from the fields and everything, but at least I have a chance to just

meet with Lindsay every week and see if there is, if everything goes well

or not, or how can I help them, right?

So that's, that's, that's, that's basically for, um, managing the

fields and also plant breeding.

And for long-term goal, like, you know, for all these kind of things that makes

need to be happening in the field, we need to have long-term funding.

So that's my main responsibilities.

And, uh, every day my, every like day-to-day task is to write, like,

you know, more research grant, talk to different industries, talk to, uh,

government buddies, like farmers, try to understand what are their needs,

what are their demands in the future, and then how can I collaborate with

them and how can I sustain my programme smoothly and run my programme smoothly

without any hiccups on the way.

So that's, that's for, uh, plant breeding side for the computational

biology, all of the computational, but most of the actual computational

biology things are in my shoulders.

So I'm trying to, and it comes from my curiosity.

I love to just, you know, know more about ai, know more about like, you know,

the company new computational methods, how can I, like, get the best use out

of the data that comes from the field.

And all of my technicians are collecting all those kind

of, uh, like, you know, data.

So now it's my time and it's my turn to make this data into the meaningful,

uh, like definition something, right?

So in this case, I'm trying to, uh, a couple of hours at least per day

to just read more paper, see how can I, like, you know, contribute

efficiently to my programme in this case.

But also I have postdoc, I have students, and I also have very talented volunteers

and very talented, like undergrad, as soon as they are very passionate.

And I think that, you know, they are going to be the next generation plant breeder

in the, in the very like, near future.

And it's not the actual only traditional plant breeder.

There are going to be the computational plant breeder.

So they are, we are all passionate about these new tools, new, uh,

you know, like, you know, uh, way to collect and decipher the data.

And yeah, we are all working together.

In terms of, if I, if I, like, if you ask me that, if I have a very relaxed

weekend or not, I can't tell you because over the weekend I usually like, you

know, uh, wanted to keep the pace with the amount of the emails, amount of the

things that I'm receiving every day.

And I remember in the last like, you know, LA uh, in this week, like it was Canada

Day a couple of days ago, so I took some days off and I went to the cottage and

after I came back I faced with 350 emails.

So had to just reply, reply all those emails back because I wanted to

respect everyone who sent me an email.

And then it took me like 10 hours not to stop just replying back to those emails.

So yeah, that's a life, that's the work.

It's exciting, but at the same time, there are so many responsibilities and

I'm trying to like keep the pace with all those kind of things that happen.

Right on.

And I appreciated both of your answers because it really reinforced why as

a leadership podcast, as a leadership within an agriculture podcast, we're

talking about this technology because both of you kind of described how

leadership characteristics, leadership responsibilities, and leadership vision

is really a key ingredient in using these technologies quite effectively.

I bring up this point about leadership because often I feel people who

don't have a technical background are really intimidated by this technology.

They, they kind of disqualify themselves and think that they're

not an expert, they can't use ai.

So I'd love to have a, a, both of you kind of weigh in on the role of language, how

even though there is still a math, uh, a science, uh, a, a technical on the level

of software, uh, underpinning to this entire kind of ecosystem, part of what's

exciting right now is that you can use language to interface with these systems.

You can use language to build these systems.

I think those of us who are using it take it for granted, but I think

outsiders don't really understand how this is a different way of engaging

with the research, a different way of, of building applications.

So I'd love to have both of you, from your perspective, talk about kind how language

plays a role in what we talk about.

Ai a big abstract subject, but I think something that'll open up a bunch of

threads that, that we can pull from there.

Donald, maybe, do you wanna start us off?

Yeah.

I think, um, you know, I mean language as a code for, you know, if we we're

signifying, you know, the signs and how we signify them through language

is, um, a very interesting topic of the human, of the human experience.

It's one of our most important codes, you know, along with economics

and, uh, to motion's point, like biology and the genetic code.

But to, you know.

To deliver these tools to farmers, it is important to remember that they are

not interested in the nuts and bolts for the most part of an emerging technology.

And that's where, not just language, but um, and language and how we talk about

it is almost as important as the fact that we're using large language models

if we're going to deliver it to farmers.

But more importantly is aesthetics, you know, and, and the aesthetic of

this technology and how it's delivered to farmers and that human design

element that certainly cannot be, um, outsourced to a large language model

or, or any type of research loop.

You know, there is, a last mile piece to this that requires human engagement,

whether it's in this case, farmers, you know, and it has to be delivered

to them as a human technology.

Um, and that's something that we battle with as we, um, you know, prepare

the, the mobile apps and the, and, you know, the front end development.

It's how we were so excited in April, may, we had built real artificial intelligence

for farmers and it worked and it made it easier to make carbon credits and it

made it easier to, uh, implement variable rate nitrogen and, um, it made it easier

to maximise the value of supplemental irrigation and the potato crop.

And we had built all that, and we could do that computation very easily, but

we were faced with the issue of, okay, aesthetically how are you going to deliver

this opportunity for greater productivity to to, to Prince Edward Island farmers?

And that caused us to, to step back and, and consider that.

And, and the, you know, since the 17th century, like we understand taste

to be, you know, elegance and, and simplicity and, and, and we have that,

that nothing about that has changed just because we've broken language

down into tokens and we can model it and deliver information based on that.

You know, this is still a human technology.

It still has our fingerprints and our signature on it, and it has

to be delivered with, um, the core tenets of, of aesthetics and, and

of design, human-centered design.

And that's the most exciting part as we live in this sort of like post a, like

post ai, but like pre kind of like we still have user experience, we still

have bespoke ux and we have to design UX that can deliver tremendous amounts of

computational power to farmers in a way that they simply open it and it works.

Yeah, look good, feel good, Jen.

Like not just aesthetics isn't just about looking good, but it's about how, how the,

how the object really, uh, makes a person feel and, and, and has a huge impact on

whether they're willing to consider it.

And so it would be so easy to build something that farmers, um, turn

their back on when you're talking about this type of emerging technology

with so much, um, controversy on its impact and its usefulness and,

uh, so much rhetoric and media.

And so, you know, it's not just about how it looks, although that's important,

but how it feels to really use it and, uh, make it a, a welcoming space.

Um, and, and valuable.

And I think that really the last two are the ones we think about the most.

Um, you know, we want, um, our team in involved in the, in

the, um, implementation of it.

We saw Palantir in the USDA to the $300 million deal.

And certainly that's gonna involve Palantir going into the USDA and

understanding their systems and building things, uh, that fill these crevices

in their organisations and, and create value added computational opportunities.

But at the, in the, at the grassroots level with the farmers, there's no

question that ensuring our first 10 farms have a very strong, positive

experience, is the most critical part of our project right now.

Uh, we're working with Prince Edward Islands, some of Prince Edward

Island's most influential farmers.

And if we give them a valuable experience over the next, uh, eight weeks, uh,

I have no doubt that, um, our signup sheet will, uh, will start, um, start,

uh, really to expand quite quickly.

I, I mean there's a lot there Mohsen for you to kind

of build upon both in terms of the accessibility and the inclusivity.

But I, I'd also love you, you know, from a research perspective, talk

about the potential unreliability or the inconsistency of large language

models and, and how you factor that into the work that you do.

That's a great question.

So before I talk about like, you know, inconsistency, I just wanted to, uh,

add one point about the leadership.

Like, you know, me leadership is, doesn't come from one person alone.

So it comes from the strength of a very good group.

And I said, I'm very proud to have a great team around me.

So they are all bring ideas, energy, honesty, and commitment.

They the level of the commitments that they have, and they are coming to the lab

every day and working with the passion.

it makes my life easy.

Like my role as a leader just support them, listen to them, and like creating

a space where everyone can flourish.

Their idea can be shine, like, you know, and can do their best work.

So that's something, uh, I, I miss to, I forgot to say about like, you

know, uh, being a good leadership and how the leadership means to us,

but in terms of like, you know, when Donald talk about, um, the language.

And how to talk.

Like brought up a very good point about like the tokens, how we need to be

cautious about like, you know, how we are speaking to the AI and also how we

are, working with like non-AI things.

But, uh, getting back to your question, Jen, about like, you

know, farmers, I think, I think it's slightly different than Donald.

I think like, you know, farmers like AI should not come to,

to the farm as a black box.

It should come as a kind of conversation.

So time that I'm talking to farmers, they don't want something that

tell them, for instance, spray now or do this without any reason.

It should says that like, you know, disease risk is high, this part of

the field because of that, that, that, and then like, you know, based on the

previous field history, all points in the same direction tells us that like,

you know, there is a like 90% chance that you need to spray in this area.

That's something that I think farmer very well like it and better like it.

And then they are like, tend to more adopting this kind of

technologies way better than like, do this or do that or just give some

data and receive the prediction.

So in my case, I try to involve farmers in different decision like

we have at University of Guelph.

We have like, I'm going to develop a sensory panel like inviting

farmers, growers to actually see the canned beans and ranking them.

And then I can use this information to build an ai.

So later whatever AI that I'm building is be, is based on the farmer's

feeling, farmer's decision, right?

So this can help them to better understand a better adopt this kind of technology.

And just, I got surprised week how much farmers are tend to

adopt these new technologies.

I sometimes thought thinking that, like, you know, that okay, the, I I

will have like, you know, more hard time talking to different people, what's the

benefit of AI and those kind of things.

But last year I received different calls from farmers that

they are very much interested.

this new technology, which like, you know, encouraged me

to work in this area way better.

But yeah, so that was something that everyone to talk about it.

But speaking of uncertainty, definitely it's, and hallucination.

Hallucination in AI is definitely something that it's makes me like,

you know, scared about it, uh, but at the same time makes me worry about it.

But at the same time, I'm trying to like find a good way to

overcome these kind of challenges.

For instance, I can tell you one, like there are so many areas that we are

dealing with, but I don't want it to bring all of them up here, but just give

you one, some 1, 1, 1, 1, uh, example.

Let's say that Donald, for instance, in their own area find something special.

Like he find that, he found that, uh, I dunno, X can cause y, right?

very interesting things.

And he decided to bring it up here.

So he brought it up here and you write a good news.

So this news will be publicly available over the internet and AI

agents are going to use this news.

Then started to think that, okay, how about like, you know, publishing a good

paper out of this great achievement and he's publishing a paper, so it's the

same result, but publishing in another source in another way, and AI agent

also collect those as another evidence.

Then publishing somewhere else, like so many news, like, you know,

broadcasters, they are all coming.

And I said that, okay, we wanted to cover that news.

So there are so many X caused Y and they are all treated independently

on the AI agent while they are all coming from one source.

So if as a researcher, I wanted to ask a question, how, hey,

ai, what do you think of X?

So it's a generative, like, you know, is a rag model.

Retrieval, augmented, generative models goes back to retrieve so

many things and consider them as an evidence it can search publicly.

And if you'll find that, okay, X caused y mentioned thousand

times, so that's something there.

So it gives you this result.

you know what X caused Y because of this, this, this, this, this.

But all of them are coming from one source and it's all in one study.

So that caused a problem because I know that I'm doing something correct,

Donald, doing something great, awesome.

But how about verifying it over the years, right?

So we need to have such this kind of controller in between,

or maybe we need to change it.

So I recently came up with another idea instead of retrieving news,

how about retrieving keywords?

So I can add X cause Y as a keyword into an index, like in today's database.

And consider this as a one time things, right?

If anyone with another name, with another, like, you know, nature x, cost YI can add

it as another evidence to this keyword.

Something like that, right?

I'm just thinking of it.

But are so many other ways, and I think that's a kind of need to be corrected,

need to be, um, actually controlled.

Or even some of the papers that, uh, in, in my research, I'm sometimes

searching that, Hey, would you like, you know, uh, can you retrieve some of

the genes related to something else?

Like, some part of, in some parts in plan and I'm receiving some of the

good research paper and sometimes not very good research papers.

So it need to be a controller there to just sort it out based

on like the credibility of the authors and also, uh, journals.

And also sort it out not only based on the impact factor, not only based on that if

other like publish so many papers or not, there should be something there, right?

To control all those kind of things.

So I think that's a very important part and I can't give you the like,

you know, one answer that hey, yeah, we can fix everything with and we can

like address all the uncertainties.

No, still uncertainties there.

And the more data that we are creating and collecting uncertainty, uncertainty

will be increased until we are setting back and start filtering, optimising,

pre-processing all the information and then develop AI based on that.

Well, and what I love about that answer.

it sounds like Mo wants to de sounds like he wants to design a harness.

Jesse sounds like he needs a good harness.

Mo, it sounds like you need a good harness.

My man.

and, and Donald,

skills baby.

Let's go.

Donald, that, that helps set up a, a really great question.

And you know what I'd loved about your answer, Mohsen?

Is it evoked governance?

Why policies, why frameworks, why verification is so important in these

types of systems that we're building?

I personally wish we had figured this out in the early days of social media.

'cause every social media platform has a governance platform.

But doled, you, you correctly evoked the word harness, which

I suspect for Jen's benefit.

You know, after our last talk, Jen sent me a note saying, I gotta go

learn what Hermes is and, and you know, uh, uh, learn a little more

about the software around these agents.

And that is, of course, the harness.

So I'd love for you to kind of do the 1 0 1, what is a harness, but also kind of

build off what Mosin is saying, which is what are the types of systems, what are

the types of interfaces that we create?

Not just to compensate for the potential hallucination, but

maybe even take advantage of them from a design perspective.

Well, keeping our farmers insulated from, the of broad expensive,

um, frontier AI computation, keeping the data separated from that is definitely

the part of the development cycle that's, that we're in and is most interesting.

And a big part of that is, um, you know, we, we've built MOD who we've

referred to in any number of ways, but she essentially is a harness.

And if we wanna put, um, you know, if we wanna put GPT 5.5 or we wanna put,

um, fable or we wanna put, um, an open weight model, you know, we can put

any of those models into the harness that we want and change what it costs

for the farmer to use the tool and, uh, what the tool is best suited for.

And certainly, as we all know, the orchestration of using the right model at

the right time is where the enterprise, um, enterprise rollout of AI is right now.

Um, it's clear that you're not going to use a model like Fable to

do, um, certain simple things with the weather and the soil, but, um,

there is a tremendous opportunity.

So MOD would hold in, uh, all of the, you know, information that we

think is critical and the way that we manage context is all in the harness.

Um, and then simply a reasoning layer.

Um, we bring that in, um, and, and sort of control where the data is

going out and what's coming in.

And so sometimes I'll query Mod and Slack and ask her a

question and she provide me much.

She can't provide me any strategic insight because she's on bare metal, she

has access to whatever data we fed her.

She could tell me what the conditions are on one of our test farms and what

that could mean for the farmer, but she certainly can't, um, you know,

help me with a grant application or, or help me, um, you know, help me come

up with a strategy for social media.

So, you know, that, that management of, of resources is, is where we're at.

And, and there's a lot that happens in there that, um, keeps,

you know, keeps our, our users insulated from the types of risks.

And, and that's a lot of potential as the sentinel layer.

You know, the sentinel layer is gonna end up being its own sort of a gentle

air, you know, it's basically the, the guys, the guys in the, in the guard

towers with the machine guns, you know, making sure that no data gets out that's

not supposed to get out and no data comes in that's supposed to come in.

So it's, I, I've all, I continue to advocate for.

You know, a national tiered architecture that will help us

solve a lot of these problems.

And I think that at the core of AI and agriculture in Canada right now,

it's an infrastructure question.

You know, the same as we need to have electricity and we think about 5G

and these resources going to farms.

I think that we need a collective national rollout of an architecture

that assesses the risks that Moss spoke of and uh, and mitigates them.

I think doing it collectively with a national role, provincial

roles and, and farmer roles, um, I think that's quite clear.

And, uh, I think that, um, the next policy framework and the federal

AI strategy need to support that with the serious investment of, uh,

hundreds of millions of dollars.

Well, and Jen, let me remind you at any point, just unmute

and I'm, I'm happy to let you in here.

I I Donald, to throw you a, a quick follow up past that and, and Mohsen,

I'd also like to hear your thoughts.

What about the cybersecurity side to this?

And, and I say that because you evoked the sentinel sort of protecting our data.

I actually, yesterday just used an LLM to hack into a device that I have that I

wanted to be able to access the data into.

So that was me in the positive sense, leveraging some of the

abilities for these LLMs to bypass any obstacles that get in their path.

But how concerned should we be about the cybersecurity of the systems we're

building, given that it feels that we are in, uh, an inflationary period when

it comes to cybersecurity capabilities?

I think extremely concerned.

I think it's another good reason to take a, a collective national

approach to building computational infrastructure around artificial

intelligence in agriculture.

And I think part of that, a architecture has to be, um, air gapped, um,

gapped, inference empowered, um, computing inside the farmer's field.

I, I don't think most of the data that farmers are collecting

doesn't need to leave the field.

I think it's, uh, it's, uh, we don't want the internet necessarily

connected to what farmers are doing.

Um, and so is where a lot of our research and development is right now.

Like, uh, we have sort of broad stroke ideas about how to implement this

nationally, but when we bring it back to the PEI Federation of Agriculture

and its members, um, you know, we run, we run a, you know, we run a hardware

stack that's designed and is being designed and experimented with to

ensure that, um, the, the, the core data from the farm is never in harm's way.

Um, and that's why the sentinel layer's so important.

It's the sentinel layer's not guarding on the outside of the, of the cloud.

The sentinel layer is actually guarding between the farm field and the cloud, so

that, um, so that the data is not, don't want things crossing that air gap that

you're not expecting, which is possible.

And, you know, hallucination is, is, I'm not sure hallucinations are the

issue they were in 2024, but, um, certainly data escaping, um, is something

that we, we don't wanna, we don't wanna, uh, we don't wanna play with.

So huge.

And, and there are people in Canada thinking very critically about that.

And we're lucky.

Um, cybersecurity on farm has been an issue with robot milkers and, and, you

know, all types of, I ot, you know, whole barns, and we see get delivered

and it's a much bigger issue than the, than the industry talks about.

It's one of those things that when it happens to a farmer, uh,

they prefer to keep it quiet.

So the scope of the issue currently is, um, broader than

people may, may understand.

And, uh, in addition to that, I mean, maybe the scariest thing about AI is

that they become better at cryptography than us and all of a sudden start

locking a set of our own systems, right?

And so we don't want to be like, that's the Terminator.

Like that's the danger, you know, like we, we, that's the danger is,

um, we wanna be a lot better at cryptography than these things.

And cryptography is the box that they live inside of.

Right on Mohsen.

Yeah, so I'm not expert in cybersecurity, but uh, my view, I think

like cybersecurity should be a very serious concern where we talk about

like, you know, AI and agriculture, it shouldn't like stop us from using ai.

But you know what?

In agriculture, I, I, I, I see cybersecurity is not only about

protecting computers or the data, is also protecting trust.

And the trust is the real foundation of the AI adoption

when it comes to the farmers.

Like this is, that is why I'm saying like, you know, cybersecurity is the very

serious concern in agriculture because if we are just breaking this kind of

trust, that's, that's not a good sign.

So as Donald said, some of the data, there is no need for the data to

go, goes outside of the barn like, you know, outside of the field.

So we can like, you know, develop so many different infrastructures or to

so many different, like, you know, ways and approaches handle this type of data.

But, and in like, you know, in the breeding, I can tell you that one way

that we are trying to, uh, increase the security of the data that we are

producing is to assign different numbers, like, you know, assign different IDs

and everything and protecting them from like, you know, being used by others.

And then we try to come up with a good, like, you know, who owns the data, where

we are going to store the data when we are, and how we are going to use the data.

We always have the version controls and we are securing all this

data, like, you know, in different infrastructure that university provided.

So are all the all different steps that we are taking to make sure that

the data that we are collecting and data that we are, that we have right

now are in a highest protection level.

But again, I see like, you know, cyber security as a very serious concern.

And yeah, we need to also build more infrastructures and more using more

approaches to make sure that the data that we're producing are secure and safe.

Well, and I appreciate you using the word trust because I think trust

is something that is often associated with, say, the adoption of AI and

certainly the federal government centred trust in their national AI strategy.

I feel both of you have already kind of addressed a lot of the prerequisites

that users kind of require for trust.

To what extent, to what extent is trust an ongoing moving target?

And and I say this because I kind of feel in today's conversation, we've

gone through leadership, we've gone through governance, we've gone through

kind of research and application.

But I think the bias once again that all of us have is

literacy and curiosity, right?

We like to learn new things, we like to explore these types of systems.

Not everyone has that.

A lot of people are just overwhelmed by the day to day and what they focus on.

I, I say this as a way to segue into the climate side of this discussion because

it strikes me that while we often talk about AI in terms of productivity, we

talk about AI in terms of research.

What are the areas that I think AI can be particularly powerful is managing

complexity and managing uncertainty.

And it, it strikes me that that's exactly what climate volatility and the

kind of climate crisis that I think the Super El Nino brings to us this year.

All of that is a grand transition that kind of asks the question, what

role do you think AI can play in responding to our changing climate?

And, and Mohsen, I'm gonna throw to you first 'cause I suspect you are well ahead

of the rest of us, uh, both in terms of anticipating the needs of the beans

that are gonna be growing in volatile environments, but also in terms of some

of the digital twinning and some of those simulations and research you do.

So, uh, taking this question from any angle you desire, where do

you see the role of AI in helping us respond to climate change?

Not just in a reactive sense, but in a proactive sense so that we

start actually feeling like this is a, a kind of crisis we're capable

of responding to effectively.

Yeah, no, that's, that's a very great question, Jess.

The point is in, I can start from the plant breeding side.

So, in plant breeding, we are trying to make a cross, and then we are advancing

the lines for the next eight to 10 years.

So we are selecting potential parental lines right now with the hope that the

children, the pro of these patterns are going to be superior in the next 10 years.

So you can't imagine, like you, we are seeing how like, you know, climate

has been changed over 10 years.

Right.

You know, we are seeing more frequent, like, you know, extreme

weather and that's, that's something that we was, we, we are not able

to try to see on the history.

We are not able to learn from the history.

It's something that we are facing something new every year, every year,

and weather is always ahead of us and surprising us this kind of climate change.

All this ai, I can tell you predictive ai, like all the predictive AI

models, are going to, are trained based on whatever data that we

are collecting from the field.

So we call them input and output.

So the output is actual yield, actual performance of plant, and the input

is the historical data, like the historical climate that we have, right?

So we are expecting to predict the performance of plant based on the

historical climate, environmental data that we collected different from different

stations, whether it's stations, we try to predict the performance these specific

plan for the future, where in the future we are not able to see what's going on,

what will happen in terms of the extreme weather, extreme, like events that happen.

one thing that we are doing, and we used to do, was to try to collect

some of the climate change scenarios.

All the data that is available based on different anthropogenic

climate change scenarios.

We, they call it RCP, representative concentration pathway.

They recently changed it to the SSH is socio society, something like that.

I, couldn't remember the, what is the actual acronym means, but it's SSH.

So there are different levels sorted from 2.4 down up to 8.5.

So 2.4 or five is the case that the climate change is very

mild and so good in the future.

So nothing is going to happen.

Like everything will be perfect in the future.

What would be the temperature in each location look like?

So scientists predicted that, and then climate like RCP 4.5 is

everything is exactly the same as now.

What will happen in terms of the climate, like, you know,

all the variables in the future.

And 8.5 is the worst case scenario.

If the fossil fuels will be bad, like, you know, we are burning more fossil

fuels and there are so many wars, there will be so many wars in the future.

So what would be the temperature in different location?

What would be the, uh, climate patterns in different location in the, like, you

the future up to 2000 and a hundred.

So we have this data.

We can also use this data as a testing in the ai, but again, pay attention that

we have output, which is the performance and the input, which is all these like

average daily data, whatever that we have.

So something that I've mentioned last time, and I wanted to also, if

there is a time I wanted to discuss about it, is the compute, like, you

know, uh, quantum, sorry, quantum ai.

in the quantum ai, the reason that I'm thinking that, uh, future belongs to the

quantum AI is because in quantum AI we are able to search more thoroughly, not

only looking at the input and output, but also try to find all these combinations.

Let me like, uh, provide a like clear example.

say that you are flipping a coin.

when you are flipping a coin, the normal computer or the, like the ai, predictive

ai, just counting the number of the times that the coin landed as a head or tail.

And it, at the end, when you just flip the coin, it'll tells you that in

what percentage, in what possibility you will see head or tail, right?

But how about only look at one time, one time that we are flipping coin.

Coin at each position has different like, you know, uh,

it's different possibilities.

One time is head, one time is coin.

Head coin, sorry.

Uh, head tails.

Head tails.

Right.

And then do you think that it's landed the possibilities 50

50, like 50% head, 50% tail.

about head tail?

Head tail and landed as head.

So we have four heads and three tails in one, one event that happened.

So quantum tried to capture kind of things in one, one data that we have.

no matter how many, like sometimes that we are seeing, AI is not able to predict

We are trying to collect more data.

how about using the best whatever that we have.

then in these kind of possibilities in agriculture, we have genes,

we have climate, we have, uh, aggro like agronomy practises.

We have so many areas and all these having so many combinations.

So the events that we are, the output that we are seeing in one

year is only one combination.

But how about considering all of them?

And this is something that we are trying to move toward that, like

capturing all these combinations.

Try to see how many of these combination has the highest possibility

that happen in the future based on the climate change scenarios.

And then the next step is to design digital twin.

So we are at the University of Guelph in different sectors in food science.

I know that, you know, I'm collaborating in some of the proposals.

They are using digital twins in plant breeding.

We already saw to use the digital twins.

And last year I saw to do that.

And I can tell you, uh, that all the crosses that I made, like

90% of the crosses that I made last year was because of the ai.

All the, it was the AI crosses and I'm waiting to see what would be,

I'm so like, can't wait to see what would be the result like in the field.

But anyway, we try to design digital twins.

Based on the facilities that we have, we have growth chambers and we can set

different temperatures set different, like, you know, conditions that will

happen, likely happen in the future.

And also investigate all this and assessing all these combinations of

the genes, environment, soil, all these kind of things in a very small scale.

And then test the highest possible one in the field and then make a selection and

select for the potential parental lines.

Maybe some of the panels that we are selecting today, their

performance is not good.

They are not providing a, like they're not good at all, but they

will be good in the future, in the next 10 years based on this climate.

So if we are using the conventional methods that we have, we will

definitely dis like, you know, discord, all these bad ones.

if we are adding computing like, you know, quantum, sorry, quantum

AI to that, we may able to find more like, you know, patterns.

We may able to like get the best juice again out of all the data that

we have and then we can make our selection more accurate and maybe

select something that is not good right now, but it'll good for tomorrow.

yeah, that was the strategy that we are taking here in terms of the research.

Well, and I am grateful again for you really helping me wrap

my head around the revolutionary potential of quantum computing,

especially when it comes to the wide ranging probabilities that any farmer

has to deal with when making decisions.

Uh, Donald, there was a lot there to build upon.

I, I do want to bring us back to the way in which, uh, climate and climate

volatility can be assisted by, by ai, but I suspect your mind was also blown, uh,

uh, by the quantum implications there.

So by all means, take it whatever direction you wish.

Uh, so sorry Donald, before you talk, I, I forgot to say, something that

like, you know, the credit of the quantum ai, I just wanted to give the credit to

people who bring that idea and we discuss.

So the credit of the quantum AI is, uh, Dr. Rosita Dora, she is computer

scientist at University of Guelph.

And we had, we, we, uh, we have AI for food in select, you know, initiatives

and we were all on the same table.

We had a lunch and she was talking about the quantum and we were

discussed this more throughly.

So I just wanted to also give the credit to Dr. Dora as well.

But yeah, please Donald, go ahead.

Yeah, I mean, the probability of the coin exists is almost a hundred

percent, you know, not quite, but, um, so it's quite a universe and, and when Jean

Jacque Russo wrote the social contract just before the French Revolution, he

probably wasn't aware that there's only, there's no less than a hundred percent

chance that the coin actually exists.

But nonetheless, you know, governance remains incredibly important.

Um, and, and the human element in it.

I mean, my thoughts on quantum are that, you know, it's kind of a tailwind to

this idea that the ram companies have quadrupled in value and that, um, and

that, you know, that the US economy should be investing $700 billion in, in compute.

You know, is there a path here where, where we can bring a tremendous amount

of compute to the, to the world, um, without necessarily having to

build these massive data centres.

Um, so yeah, I mean, we can see the venture capital space

around quantum is, is increasing.

I'm sure there's brilliant students at, um, at Waterloo who are at the absolute

forefront and, and I'm sure they're all very excited to move to the United

States to commercialise their technology.

so, uh, on the climate change piece, I mean.

Geez, I, I, I've been trying to make a carbon credit since I got to the

Federation, um, in 2021, and I couldn't move the ball forward on it until I had

the benefits of large language modelling.

And, uh, I mean, the, the protocols and the data requirements and the

documentation is, is simply too much for grassroots NGO, uh, without

the benefit of, uh, of all written history, uh, broken down and modelled.

And then when I feed it a VERA document, I can actually push the ball forward.

And so MOD found VT 0 0 0 1 4 for us, which, which allows us to, um, model soil

carbon the way that we wanna model it.

She also, um, found a number of different pathways.

I, I think, Jesse, when you talk about complexity and climate change,

these tools and their ability to, um, identify relationships between,

between data sets that we haven't previously considered is immense.

And I think that will, in a number of different ways bring value to farmers.

And, uh, the convergences between disparate data sets that these

tools unlock is very exciting and simply like the ability to create,

um, simulation environments.

You know, as soon as we fed mod the necessary data, we, we do now have

a, a 50 year bio geochemical model for Prince Edward Rhode Island.

That's sort of the first, uh, foundation from which the whole system runs.

And, uh, and our farmers are users benefited from it immediately.

So managing the climate and managing the impacts of climate change.

And, and, and additionally, I mean, we had a, a a once in a lifetime, or

what was a once in a lifetime drought last year in Prince Edward Island

that cost our economy $500 million, which is a, a big part of our economy.

And, and so I, I prompted Maud if she could sort of manage climate risk the

way that I've always wanted to manage climate risk and, and managing climate

risk and adapting to climate change has always been a messy, qualitative,

multi-stakeholder data, you know, qualitative data analysis problem.

And, uh, mod has sharpened it up dramatically identified the

key data sets, brought in the data that she has available.

And I believe very quickly we've, uh, built something that can ensure that

every dollar we invest in, in, um, irrigation in Prince Edward Island reduces

risk by, uh, by the maximum amount.

And, and so that's very exciting.

So on the mitigation side, I don't think you can make soil carbon credits

effectively without LLN technology.

And on the adaptation side, certainly managing complex hazard risk modelling

and, uh, and reducing that risk effectively with, uh, investment, I

think on adaptation and mitigation, um, as well as understanding when

the actual weather impacts come, how to, how to respond to those.

I think.

On all pieces.

You know, as someone who's very passionate about responding to climate

change, you know, this technology is certainly the biggest leap forward.

Um, you know, since the advent of the Post-It note.

Aggressively, um, you know, I mean the, the, you know, managing natural

resources is about engaging with all of the stakeholders effectively and

understanding how everyone values economic capital, social capital,

capital, and environmental capital.

Uh, it's up to the stakeholders to, um, identify the, how

they want to distribute that.

That's why governance capital is so important.

'cause it controls the distribution of those, um, capital buckets.

So we need good democracy to, democracy is and capitalism is ai, right?

Like, we need democracy to keep us safe and control our capitalism and, uh,

ensure that capital is being distributed between those three buckets according to

the, uh, the wishes of the stakeholders.

Stakeholders and, and, and, and waiting those stakeholders, you

know, is, is a challenge, but, uh, certainly not impossible.

And, and, and, uh, is a, is a wonderful sort of place to be, uh, to be working

to create a more sustainable world.

I can add to that.

So, you know, Jen, I can, I can't talk about this like topic for hours.

The point is need to be honest that AI has cost.

So there is no doubt on it, like AI has a cost, but the way that we are

using ai, it can just justify the cost.

Let's say I, as a plant reader, I make a selection for the future that,

and try to, let's say, using, I try, I try to use AI to release a call

to work that would be useful for the future in like, you know, in and

would be perfect and produce higher, higher yield in this climate change.

Right?

You know, all these volatile things that we are seeing right now.

So AI for developing this variety.

At the same time, let's imagine that I'm not using any AI and I don't have any

information that what will happen in the future, and I don't have the prediction

ability, predictive ability to do that.

So I'm releasing some varieties in the future that maybe those

varieties are not good and farmers need to add more fertilisers, more

pesticides, more herbicides to that.

And then so many of these kind of things are going to deplete to the

soil, spread to the environment in the future and the cost.

So 10 years, I, I'm right now using ai and of course, whatever I'm using,

has a cost of the water, using water carbons, all those kind of things.

And it's bad for the environment.

There is no doubt on it.

But let's imagine the war and the prediction that the, the, the

breathing without AI for the future.

How about consuming more fertilisers and those kind of things that's

even worse for the environment.

these two has the cost, but which one has the better cost is the use of AI in

this case also, when we are using ai, we need to come off with a good strategy.

That's something that I was just thinking for months and I couldn't find

a single answer to that is how we can better communicate with AI and encourage

people not to use AI in, for everything.

Like I'm seeing every emails, like more than 90% of the emails that I'm

receiving right now is written by AI or like, you know, formatted by ai.

Why?

What's the reason I wanted to make mistake?

I wanted to make mistake and writing some words, like having some typos

and everything is way better for the environment that we are living.

Even one emails is nothing for me, but let's compare it.

Let's consider it like for the, as a worldwide, how many emails

we are sending to AI to ref, like, rephrase, revise for us, right?

That's, that's something.

And even, you know, we are not paying attention to the actual

models that we are using.

Sometimes we are using like GPT five, oh 0.5, whatever the latest

one for revising one email.

So it's, I'm usually telling everyone that it's look like that you getting a very

expensive, fancy car and running in a, like a street with maximum 30 kilometres.

Why you wanted to do that?

That's, that's not even good.

And we need to have this kind of policy.

I was just thinking of developing a kind of software that can use us in between as

a middle, between like the user and ai.

And it can just like, you know, understand what kind of models your question

require and then use a specific model.

Or sometimes the way that we are talking to ai, no one trained us.

This is something that just came up we are all sort to adopt it

without fully understand how to better communicate with ai.

So I'm seeing that so many times we are writing a actual letter, Hey, I, hey, I,

I wanted this, this, this, this, this.

And then please do that and then add this part.

So is a very small example that I'm telling, but you can just see it on the

larger scale, how much we are wasting tokens, how much we are wasting energy.

So you can't believe that you know these type of emails that you are or this

type of text that you are sending to ai.

It has like AI need to like read first that cost energy

and then we call it intent.

Just keep all your intent that cost another energy.

And then reasoning of the ai, thinking of the AI that costs energy

and then provide you the response.

So how about keeping everything short?

Understand how to better communicate with AI in a very short and consuming very

less tokens, and then be efficient and then use AI wherever are really needed.

Not for everything, right?

That's something that we can do it.

But at the same time, in terms of the research, you should

see the cost and the balances.

Like with ai.

Without ai, what would be our future look like and what would

be the environmental cost?

And that's, I think, a brilliant frame,

uh, for us to both end on.

But think about our next, uh, panel, our next gathering, because as Mohsen

was kind of, I think, quite brilliantly sketching out the economics of, you

know, having a sports car, a super sports car to go out and get groceries

when you could walk, uh, to get the groceries, which would be far healthier.

We go.

Imagine that applied to greenhouses, because greenhouses are a

topic I'd like to talk about next time.

I think what I'll do is I will go out and get a guest from the greenhouse

industry, interview them as a solo episode on the podcast, and then

invite them back for our panel.

Because most, and what I was thinking about is, on the one hand, yes,

greenhouses are great for Canada.

They help us grow stuff, but they are resource intensive.

They are environmentally intensive.

Maybe there are some crops that don't belong in greenhouses that we

should be thinking about growing in less intensive growing environments.

Again, a question for a future episode, but I think if anything, what these

panels are allowing us to do is, on the one hand, make sure that Jen, your

literacy, your understanding of this is constantly upgraded because if anything,

you are, part of the reason we're doing this is to make sure that other leaders

within the sector can understand these questions and understand these issues.

But I don't know about you guys.

My questions tend to be increasing faster than my answers.

I leave every discussion we have with a whole bunch more topics that I want

to discuss, so I am grateful once again for your time and expertise and invite

you all back so we can keep having these conversations and really fostering the

kind of cross sectoral conversation that we're successfully having.

Thank you so much.

Yeah, I'm looking forward to that.

You know, every time that I'm coming here, I'm also learning at the same time.

Like from Donald, from Jen, from you, Jess.

I'm just learning and yeah, I would be happy to continue that discussion as well.

Yeah.

Right on.

Yeah,

you.

again soon, guys.

Have a great day.

Indeed, a great day was had by all certainly.

I hope you had a great day listening to that conversation.

Again, due to some of the technical issues, I've already listened

to that conversation many times.

Uh, something I normally loath to do when having to produce

an episode for the podcast.

But in this case, I quite enjoyed it.

There were multiple layers to the insights that Mohsen and Donald offered, and

again, my gratitude to them for taking time to engage here on the podcast.

And I think really elevate our understanding of ai, our, uh,

ability to see the vista, the imminent event, event horizon, uh,

i, I found incredibly valuable.

Now, as I've mentioned, AI has been a recurring thread,

uh, here on the Future Herd.

We've got an episode, uh, already recorded, not yet released, uh, with

Dr. Suresh Raja, a professor in AI and livestock at Dalhousie University.

That was a fantastic conversation that I'm eager to drop, and I'm working on a

virtual event, uh, for the future heard on open source and AI for mid-fall.

So keep an eye on that and I'll continue to drop hints every now and then.

Uh, thanks again, uh, for listening, for engaging for rating the, uh, podcast on

your favourite podcast network for sharing it with your friends or for sharing it

with your revolutionary cell plotting to destroy all data centres and enact the but

Leary and Gha D no matter where you lie on this particular, uh, zeitgeist issue.

We're glad that you've joined us and listened to what we've

had to say on the issue.

Uh, until next time, I am Jesse Hirsch and this is the Future Herd.