Oct 7, 2022
This podcast is about big ideas on how technology is
making life better for people with vision loss.
In 1997, Gary Kasparov lost an epic chess rematch to
IBM’s supercomputer Deep Blue, but since then, artificial
intelligence has become humanity’s life-saving collaborator. This
episode explores how AI will revolutionize vision technology and,
beyond that, all of medicine.
Karthik Kannan, co-founder of AI vision-tech company
Envision, explains the difference between natural intelligence and
artificial intelligence by imagining a restaurant recognizer. He
describes how he would design the model and train it with positive
or negative feedback through multiple “epochs” — the same process
he used to build Envision. Envision uses AI to identify the world
for a blind or visually-impaired user using only smartphones and
smart glasses.
Beyond vision tech, AI enables faster and more
effective ophthalmic diagnosis and treatment. Dr. Ranya Habash,
CEO of Lifelong Vision and a world-renowned eye
surgeon, and her former colleagues at Bascom Palmer,
together with Microsoft, built the Multi-Disease Retinal Algorithm,
which uses AI to diagnose glaucoma and diabetic retinopathy from
just a photograph. She acquired for Bascom Palmer a prototype of
the new Kernal device, a wearable headset that records brain wave
activity. Doctors use the device to apply algorithms to brainwave
activity, in order to stage glaucoma, for example, or identify the
most effective treatments for pain.
Finally, AI revolutionizes drug discovery. Christina
Cheddar Berk of CNBC reports that thanks to AI, Pfizer developed
its COVID-19 treatment, Paxlovid, in just four months. Precision
medicine, targeted to a patient’s genetic information, is one more
way AI will make drugs more effective. These AI-reliant innovations
will certainly lower drug costs, but the value to patients of
having additional, targeted, and effective therapies will be
priceless.
The Big Takeaways:
- Natural vs. artificial intelligence, and the
“restaurant recognizer.” Karthik Kannan, CEO and co-founder of
Envision explains the difference between natural and artificial
intelligence by describing how humans recognize restaurants in a
foreign city and comparing that to how he’d train a “restaurant
recognizing algorithm.” Here’s a hint: the algorithm needs a lot
more data.
- Sensor fusion AI. AI developers are
interested in using different types of sensors together to give the
algorithms a sense of the world closer to human intelligence. One
example is the use of LiDAR in the Envision app, in addition to the
phone camera.
- Transhumanism. Humans don’t have LiDAR.
Does that mean AI will surpass human capability? Karthik offers
that some radiology AI have higher accuracy than human
radiologists, but he thinks it will be much more of a partnership
between the human and the machine.
- Multi-Disease Retinal Algorithm. Dr. Ranya
Habash and her colleagues at Bascom Palmer Eye Institute worked
with Microsoft on an AI diagnostic tool. They fed the algorithm
86,000 images of eyes, labeled with relevant diseases, and taught
the machine to diagnose eye disease with just a photograph, making
remote diagnosis not just possible but inexpensive.
- The Brain-Machine Interface. Dr. Habash
wrote a grant that earned Bascom Palmer a prototype of the Kernal
device, a helmet-like device that measures brainwave activity.
Doctors used this device to create a “brain-machine interface”
which advances brain research on glaucoma, diabetic retinopathy,
Alzheimer’s, and pain management.
- Bias in AI. Karthik Kannan reminds us that
the biggest threat that humanity faces from AI is bias encoded in
the algorithms. This is a real harm that humans have already
experienced, and AI engineers need to be extremely sensitive to
ensure they are not encoding their own biases.
- AI for Drug Discovery. Christina Cheddar
Berk, a reporter for CNBC, shares how the pace of drug discovery is
set to speed up, thanks to AI algorithms and supercomputing power
that can cycle through millions of possible chemical compounds per
second to I.D. effective options. Pfizer used a similar process to
develop Paxlovid, in a process that took only four months.
Tweetables:
- “The secret sauce is always in the data.” — Karthik
Kannan, CEO and Co-Founder of Envision
- “Human intelligence is so holistic. We have so many
sensors on our bodies. […] Whereas an AI is taught only images.” —
Karthik Kannan, CEO and Co-Founder of Envision
- “I know what’s going to work and what’s not going
to work within thirty seconds of seeing it. […] They need to show
up with a smartphone. Then I’ll take them seriously.” — Dr. Ranya
Habash, CEO, Lifelong Vision
- “I don’t think there’s anything more powerful in
medicine than to be able to treat a patient and get rid of a
problem that is plaguing them so much.” — Dr. Ranya Habash, CEO,
Lifelong Vision
- “If you can measure it you can control it.” — Dr.
Ranya Habash, CEO, Lifelong Vision
- “It strongly takes over the bias of whoever is
actually feeding the data […] and I think that has much, much more
potential for harm than an AI taking over humanity.” — Karthik Kannan, CEO and
Co-Founder of Envision
- “The value for patients of having those additional
therapies available; it’s hard to put a price on.” — Christina
Cheddar Berk, reporter, CNBC
Contact Us:
Pertinent Links: