- [Pascal] And we're back here at
Singularity University Hub.
On the show floor at the summit in San Francisco where,
as people tell me because I haven't been out of here,
it's actually pretty foggy.
We're here today with Abi Ramanan, who was part of our
global solutions program two years ago,
if I'm not mistaken right, 2015.
- Co-founded a company called ImpactVision about
which we will talk quite a bit.
I'm super stoked to have you here.
- Thank you so much for having me. I'm thrilled to be here.
- You also co-founded two other companies-
- Yes I did. Smaller, not really technology oriented,
social enterprises
- And we'll get there in a minute.
- Yep.
- ImpactVision uses hypo spectral imaging, to tackle the
$1 trillion dollar problem of food waste correct?
- Yep, exactly.
- So talk a little bit more about what is it, how did you
come up with the idea-
- Yep absolutely. So I've been working in or around food
for awhile. I started a food business focusing specifically
on supply chain waste, but looking more at the consumer end
of things and creating a secondary marketplace for surplus
products and trying to attribute a commercial value to that
to start to challenge consumer perceptions around surplus.
And the reason why I was really excited to go to SU
is because I've always been more community focused,
kind of grass roots, and the opportunity to address some of
the challenges in the food system in a more systemic way,
looking at closer to harvest and using technology was
really appealing. And so we actually ... it just all started
we had a latch off by someone who was making satellites with
hyper spectral capabilities, and he actually encouraged
the class, look for on earth applications, the sensors
are becoming smaller and cheaper and we're going to start
to see, every industry wants non evasive information.
It's all kind of great value,
particularly for products which
are perishable, have a short shelf life, being able to do
that type of analysis has a lot of value.
And we've changed the core concept of the business, has
changed very little since that summer two years ago which
is to provide this analytics layer to interpret this world
of new information from these sensors.
- So let me, I know that the audience of course surely
knows what hyper spectral imaging is.
I have to admit when we first met, I was in that lecture,
I was like oh that's interesting. I kind of knew somewhat
what it is, but I didn't really fully rock it.
Would you mind explaining to the audience a little bit,
like what is it, what is exciting about hyper-spectral,
how does it work, and how do you use it
in your particular case.
- Yeah absolutely. So hyper-spectral imaging combines
two different technologies: spectroscopy, which is a really
well established technique, technology that's existed for
about 60 years. And that's the process by which you acquire
chemical information from a single pixel.
And hyper spectral imaging combines that technique
with computer vision. So you're basically acquiring
chemical information across hundreds or thousands
of pixels. So why is this-
- By looking at something right? By looking-
- From an image, by measuring reflectants.
You're looking at reflectants across hundreds of
continuous wavelengths, as opposed to just the three
channels which you and I, or a traditional camera
processes light. And what this allows you to do is access
information in different parts of the visible spectrum,
near infra red and other parts of the electro magnetic
spectrum, where information exists in the world, it's just
we're not able to see it.
And so then we make software that
gives information about the tenderness of meat,
the ripeness of fruits, the freshness of fish.
Which either today are measured by destructive tests,
visual inspections, or aren't measured at all.
And this is one of the reasons why there's a lot of waste
and fraud in the food system, in the supply chain,
because it hasn't yet had to benefit from these types
of digital technologies other industries
have benefited from. And I mean, food is central to
everything from water to energy, it's also very emotive,
it crosses cultures, and I think it's a tragedy that
is has not had the benefit of technology and that's why
we're trying to apply these tools to address some
of those problems.
- So to explain this to my grandma, so what you're saying
effectively if I get this right, please correct me,
it's like I can point a camera let's say,
at an apple and can determine
how fresh that apple is. Or if it has gone bad.
- Exactly. So part of it is around being really specific
but basically that's the premise. You take an image
and you're able to understand information about certain
quality parameters.
We try not to talk too much about things like texture,
taste, freshness, because those are more subjective
kind of human concepts, but we look more at like
shelf life in the context of pH as a measurement.
Or tenderness in the context of pressure that's applied
in newtons to cut meat.
So how it actually works, let's take tenderness of meat.
We take a hyper-spectra image of a steak, we then carry out
the destructive measurement which is you've measured the
pressure applies in Newtons to a piece of meat.
You repeat this process a few hundred times,
you build a model that correlates information from the
image, references it against the ground truth measurement.
After you've done that certain amount of times, the system
has learned to make that correlation by itself and you
don't need to use that destructive measurement again.
And at that point, that can be integrated in a distribution
center, within a company's supply chain processes, and they
don't need to rely on those destructive tests anymore.
- So ultimately as a very simplified headline I guess,
you're getting rid of the Best Buy Date right?
Because you're giving me a real Best Buy Date.
- Right or we're giving it a much more accurate use.
And today those dates are set in a really regressive way,
and so you're actually losing a lot of value of the product
and then you have issues with markdowns in store shrink,
and in store waste and all these kinds of things which
are partly due to a lack of information or poor quality
information and information that's only
obtained on a sample basis.
- Got you. Fascinating.
So where are you roughly today? And I'm curious like what
do you see as like the longer term,
not just for your particular company,
but in hyper spectral imaging generally,
what is the longer term trajectory you see in the next
5 or 10 years? Like what do we need to look out for,
what should we get excited about?
- Yeah, so a lot of people within the community, which is
still fairly small, it's still a fairly emerging technology
particularly for industrial applications.
Well established in space.
They talk about it being similar to the next GPS, so I think
within the next couple of years we can, max five years
within 10 years I think we probably will have completely
new ways of processing information, smartphones probably
won't exist, but on a shorter timeframe, we have a
partnership with a company for example that's developing
consumer-grade prototypes to start mass production in a year
and a half to two years.
They will cost around $200 dollars when produced, so within
a couple of years, sensors integrated into smartphones
consumers will have access to this technology.
B to C spectroscopy devices are already
available on the market.
The reason why this is kind of valuable, is like an
evolution of that technology, is because you can only tell
so much from a single pixel and it's great for some
applications when you're measuring homogenous products,
but you want to look at complex matter like meat, if you do
one pixel, it could be a lean pixel it could be a fat pixel.
You're not able to say anything about intra muscular fat
for example. For that you need to measure the distribution
of parameters and that's what the computer vision element
allows you.
So the hardware is becoming a commodity, three or four
companies in the last year are doing innovations within
that space, and so increasingly the complexity is in the
analytics where before you had servers that could do
processing and normalization of images and all that stuff.
All that's going to have to happen in the cloud and
partly in the device, and that's I think what a lot of the
development over the next couple of years needs to happen.
But yeah, within two years, with not every application in
the world but for some applications you'll be able to go
to the supermarket, take a photo of fish and find out
the species. That's feasible.
- Wow.
- Yep.
- That's amazing. That's awesome.
- Yeah, we're very excited.
- I can tell. We first met at the global solutions program,
2015. We just wrapped up 2017 on Friday.
Tell us a little bit of your journey.
For you coming out of the program
to the point where you're now the CEO of a company in
a super hot emerging, small community field.
- Yeah absolutely, so I didn't go to Singularity with a
background in technology, and I think it is of great
testament to SU that has enabled someone with my background
who had domain knowledge, had started businesses before,
but to be able to go on to start a company in quite a
technical field, I think SU's quite unique in the world
for being able to give people that opportunity so I think
that's been really transformative.
I do love working in technology. There are challenges
of course, I'd always been interested in technology
but I studied social science and so it's allowed me to
go on to start a technology company and travel all around
the world, get a really deep understand of certain trends.
I think a really cool thing about SU is that you are
essentially building a business for some point in the future
and finding a way to sustain yourself during that initial
couple of years can be challenging but you rarely see many
other places that are kind of teaching you to build for
a certain inflection point or look at certain trends,
and that's probably the biggest thing I took from SU,
project into the future and create something for a point
at which everyone has 5G and 1 million people are going
to be coming online and computation and image processing
and all these things are becoming more and more widespread
so look at how you can utilize all of that
and that's the really valuable.
I think, without sounding like I've been too indoctrinated
because I do think there are absolutely limits to technology
but I think that's a little bit around that mindset shift.
- What was most surprising for you in this journey
for you personally?
- I started this thinking I will get questioned a lot about
being a non-technical founder, and I thought the technology
would be the majority of it.
In fact, it is very much that 80/20 rule.
I very rarely get questioned about the technology in
a capacity that I can't answer it.
In fact I overcompensated, and now people think that
I have a PhD in imaging.
- I could clearly tell. Like our opening clearly was
a PhD in imaging.
- Whereas the business component, the value proposition,
what is the return on investment going to be for companies
that have single figure profit margins,
haven't updated their software since the 1970s,
and would typically rather make
an investment in a better pH meter
than a sensoring software.
Food industry doesn't do software as a service.
So that by far has been the biggest surprise.
I thought I knew food and I absolutely didn't.
- Wow. Talking about new food, you had two companies.
Finally enough through the two years we know each other,
I just now learn their names.
One is called Papi's Pickles and Day Old.
- Day Old, yeah.
- Tell us a tiny little bit about what were you doing
with those, what were they attacking, what was a problem
you were attacking.
- So Papi's Pickles, I'm Tamil so I grew up my whole life
learning about the conflict in Sri Lanka, I never felt
like there was much that I could personally do around it,
and then I actually was inspired by a company based in
San Francisco called La Cocina, and it was one of the
first incubators for female Mexican entrepreneurs to
start food businesses or work in food, which again is a
kind of very male dominated industry.
And that kind of spawned a lot of social enterprise,
food social enterprises around the world and I was really
interested in working specifically with Sri Lankan
migrants and refugee women who came to the UK during
the conflict.
Unemployment is really high in those communities and
migrant women and refugee women are some of the most
marginalized groups in society,
employment is really core to the process of integration.
It's a catering company and we cater weddings, pop-ups,
we do street food and do mini restaurants and basically
train women from those communities to become chefs
and earn an independent income and all the kind of
additional benefits that come from
participating in meaningful employment.
And the second one, Day Old is also to do with food waste.
We work with organic bakeries and collect the products that
they don't sell at the end of the day, package them
in beautiful Day Old branding, deliver it to offices
places like PWC and do the more softer approach to
awareness raising, and then donate the profits to
child hunger charities in London.
To highlight those twin issues, beyond just the model of
donating surplus food to people in poverty which doesn't
actually address either one of those underlying issues,
it's about creating value, creating a commercial product,
and reaching more mainstream audiences as well.
- That's awesome. There's a restaurant here in the Bay Area,
it's called The Mayfield-
- Yes I've heard of it.
- There's a bakery attached to it, and what I love about
The Mayfield is that you go there, kind of like their
second serving of the evening before they close shop,
they regularly bring you the remains of the day
from the bakery and just give them out their patrons
and say like, hey do you want to take a loaf of bread
because we were to throw it away.
- Yep. I've heard of the restaurant.
I didn't know they did that but that's great.
- Super cool. I'm curious kind of in the wrap up, so you've
been through GSP, outside of the two technologies
you now have a PhD in, which is hyper spectral,
and image as well as AI, what are you most interested in,
what excited you in terms of technology?
What do you think has the biggest potential for change?
- It's quite a big question.
- Of course.
- Within food, I think the alternative protein movement
is monumental. I think in 20 or 30 years we'll look back
and think it was absolutely abhorrent that we farmed animals
in the way that we do today. So before I spoke earlier today
Memphis Meats I think are absolutely incredible.
So there's kinds of various strands to this.
There's the kind of pea protein side of things, but they're
actually engineering tissue cultures in labs.
It's not synthetic meat, it's actual meat produced in I
think it's something like today, 23 calories in terms
of grain is required to produce one calorie of beef,
and they've got it down 3 to 1.
So in terms of energy efficiency, it's amazing.
So yeah, I'm really excited about that less the B to C
chocolate covered crickets thing, more looking at like,
there's a more interesting company called Gel-Tor,
they've created a synthetic or a biologic replacement
for gelatin. So I think what's really exciting is only
2% of potential plant based plants have been researched
that have properties to replace animal protein so
it's only kind of opportunity ahead.
So I think that's really exciting.
On a completely unrelated note, obviously a lot of the stuff
around neural ink and human brain interface and how to
increase the capacity of your brain.
You know, I don't think of the brain as a computer, but
I do think a lot of the work that going on around increasing
capacity of your brain, a lot of research into memory,
of course AI is super hyped, but I think neural networks are
really interesting. I was reading recently about a technique
called hierarchical temporal memory, which more mimics
the neocortex and the way that it's structured in terms,
I think anything that's looking in terms of memory and how
to create that and how to store information and process it
more effectively, I think is really interesting and more
opportunities in developing countries. I still think that
crowd sourcing or utilizing the power of people, particular
for small holder farmers and companies that aggregate simple
still SMS based systems, but have managed to meet their
tipping point where they're aggregating a lot of information
in terms of those kind of business models
I think is really interesting.
A lot of great stuff, like interesting stuff in energy.
Block chain as well, I still sometimes struggle to see
the absolute use case for information to be distributed.
I do get it, but I think it has a lot of value
in very specific- I think probably I need to understand
it a little bit better.
- So clearly you did your PhD in a whole bunch of areas.
That's amazing.
- I wouldn't say that.
- So with that I wand to wrap it.
This was Abi from ImpactVision. Check it out.
It's impact- Your URL is
- Impactvi.com
- Impactvi.com. So check that out.
Abi thank you so much for being here.
- Thank you so much for having me. This was great. Thank you
Không có nhận xét nào:
Đăng nhận xét