AiThority Interview Series With Christopher McCann, CEO and Co-Founder, snap40

At snap40, success rests with building a large, labeled and generalized data set, defining a particular problem and then extensive validation.

Know My Company

Tell us about your journey into the intelligent tech industry. What galvanized you to start Snap40?

I was a medical student. In-hospital, we still monitor around 90 percent of patients by manually going around to each patient every few hours and collecting vital signs. Every year, it takes years of nursing time in the average hospital. It means we detect deterioration a lot later than is ideal. If we could automate the capture of vital signs, if we could use technology to identify the deteriorating patient for the doctor, then that could save lives and it could reduce pressure on healthcare staff.

At the same time, my grandmother passed away after recurrent illness and chronic decline. The problem above is exponentially greater at home – we have no way to identify deteriorating health, no way to intervene preventively and thus we wait until the patient comes to healthcare e.g. visits their primary care physician or the ER.

Hence, snap40 — we want to bring healthcare to the patient before they know they need it.

What is snap40 and how does it fit into the modern healthcare industry?

Healthcare providers globally are facing the same challenges — an aging population that’s experiencing more chronic disease is pushing more healthcare into the community and outpatient environment. The US spends more on healthcare than any other nation, and yet that investment returns some of the poorest health outcomes. There is a focus by payers and providers on reducing cost, on treating patients preventively at the earliest, and lowest cost opportunity.

That’s where snap40 comes in. We’ve built the most accurate, most sophisticated wireless wearable available in the UK and the US. Through this, from the patients’ upper arm, we continuously and passively monitor the human body. We combine this with patient symptoms and other sources of health data, to derive an actionable insight that can be used by a physician to intervene earlier. We’re focused on how we bring the doctor to the patient before the patient knows they need help.

What does it take to start and succeed in a Deep Learning Tech startup ecosystem?

For us, success rests with building a large, labeled and generalized data set, defining a particular problem and then extensive validation.

In healthcare, the need for generalization is arguably the most acute. A simple example — the pulse oximeter, which is used to measure oxygen saturation. The output of a pulse oximeter is based on a calibration curve — a simple regression model based on training data captured from healthy volunteers with induced hypoxia. Historically, those healthy volunteers were white, young and healthy. Unsurprisingly, they didn’t work as well on patients who weren’t white, weren’t young and weren’t healthy.

Making something works on a healthy population is easy. Making it work on a sick, elderly, frail population is something else entirely. So success rests with building a data set that’s not biased and then defining robust, extensive validation.

How do you prepare for an AI-driven world as a business leader?

I meet health system executives near daily. Every one of them will say that data is one of their most important priorities. I think they are both right and wrong at the same time. Data is a priority, but data in itself is not useful. Knowledge is useful. That knowledge can be learned or inferred from data. AI allows us to process the huge amount of data being generated by companies and organizations and turn it into knowledge and signal.

How is AI/ML unlocking the capabilities in human intelligence?

The human brain is one of nature’s most remarkable inventions. To this day, we don’t know what caused — in a remarkably short space of time — Homo sapiens to grow such remarkable cognitive abilities. But the human brain is limited within a solid structure — the cranium. We can’t network. Our brains can’t expand. Computers can. Networks yield infinite computational and storage capacity. While computers excel in “rational” choice and outperform humans in structured, specific problems, the human brain still leads in terms of executive function, planning, empathy, and emotion. The human brain still excels in reasoning in loosely defined problem spaces. But AI can work hand in hand with the human mind.

Read Also: JumpStart 2019: 10 Ways Robotic Process Automation Will Mature In 2019

If we consider the impact of Deep Learning on radiology. The radiologist won’t be replaced by AI in the near future. But AI can help manage throughput for that radiologist. It can flag the 2 percent of images that require human input from the radiologist. We can focus down the attention of the human mind on the most important situations.

Would AI-based patient monitoring systems become the norm of daily life?

I think if we think forward 15 years, everyone will have their health monitored 24/7. Healthcare will come to us before we know we need it. Monitoring the entire human population is an unfathomably large amount of data. This will only be possible by using AI to identify the signal in the noise — the one patient who requires healthcare and the attention of the doctor.

How could they be interlinked with other technologies such as IoT, connected devices, and self-driving cars?

In the same way that the brain needs the human senses (input) in order to produce an action (output), then AI only functions with input data. In healthcare, real-time health data is essential to predicting deterioration. AI is just one part of solving any particular problem.

What are the foundational tenets of your AI/ML research for health? How could businesses and society benefit from your initiatives?

We focus on building a large, generalizable labeled training data set. We think a lot about how we can build that data set in populations that would typically be underrepresented in AI research but that are the most important in healthcare delivery.

Recommended: Context And Data Is King

We think a lot about how we best validate and test our models and how we do that in a way that’s robust and justifiable to regulators. There is minimal best practice in the healthcare space on model validation. We’ve had to define that best practice ourselves.

Our focus as a company is on how we turn all of the health data we are collecting into a meaningful action that can be taken by doctors and nurses. We also think a lot about how we can use those meaningful actions to empower patients — how we can we have them take the first action before we have to bring in the doctor? Making sense of large volumes of health data in order to deliver healthcare preventively is our mission.

Which AI technologies are most likely to impact the Marketing and Sales businesses?

It’s already clear this is happening — look how much Salesforce is investing in AI. The purpose here is just the same as in healthcare. How can businesses use all of the data they are collecting on customers and potential customers to spot patterns, better identify ideal customer profiles, better identify ideal prospects, determine the interests of those prospects and the messaging required to increase conversion rates?

These act to reduce the acquisition cost of new customers.

What are your top predictions and must-watch AI/ML-related technologies for 2018-2022? How much of these technologies would be influenced by socioeconomic trends?

I think we’ll continue to see China dominate. Their investment in AI is vastly outstripping every other nation. I think we’ll continue to see lots of work on computer vision coming out of these companies but also a continued focus on building better hardware — I think we’ll see chips that are far more powerful than today’s GPUs (just look at Nvidia’s share price as evidence).

We’ll also see large adoption into healthcare — particularly in detection and diagnostics. We already see healthcare moving to new models of care, where there is earlier, preventive care. AI will support these new models.

Tell us about your AI and Deep Learning research programs.

We’re focused on how we make meaningful sense of the large amounts of real-time health data we’re collecting from patients, turning it into an action that can be taken by a patient, a doctor, in order to preventively manage a deterioration — before it becomes a deterioration.

We’re focused on building large, generalized labeled datasets that span both the objective and the subjective.

What’s the “Good, the Bad, and the Ugly’ about AI? How do you prepare for these situations at snap40?

Define a problem properly, extensive generalized data collection and extensive validation. And continue that cycle.

Do you think “Weaponization of AI” is the biggest threat to mankind now?

I think this ignores far greater threats. Climate change. Antibiotic resistance. Poverty. These are far greater and more immediate threats to mankind. AI has huge potential, but it’s so so far away from a Terminator-style scenario. That being said, I think the nature of our short-term political cycles, the brevity and polarization encouraged by today’s discourse, it means we’re ignoring these big challenges.

The Crystal Gaze

What AI and Machine Learning start-ups and labs are you keenly following?

For me, some of the most interesting work is coming out of China. SenseTime, who most people won’t have heard of, is doing incredibly advanced work on facial recognition and computer vision. They are just one of a number of major, very well-funded startups in China looking at Computer Vision. Partly, that’s because of the applications in security, particularly given the scale of domestic Chinese surveillance infrastructure. However, the applications beyond defense and security are just as large. Some of the most interesting techs we use in our daily lives has come from work in defense and security. SenseTime does a lot of research globally. For a company founded in late 2014, their scale of work is immense.

Closer to home, Darktrace in Cambridge, U.K., is doing some fascinating work using Deep Learning for cybersecurity.

What technologies within AI and computing are you interested in?

Clearly, I’m most interested in the applications to healthcare – but that’s not just because of what I do, but because I think healthcare is the industry that most needs positive technological change.

Doctors and nurses will tell you that when they see a patient, they intuitively know who is sick. They can’t tell you why — they just know. For me, Deep Learning offers huge potential to replicate that intuition — identifying relationships and connections within seemingly disparate data across all of the different sources it is now possible to collect from intonation of voice, movement of the human body, vital signs, medications and so much more.

Clearly, computer vision is already having a significant impact on radiology. What’s interesting to me, is how we can use this to empower patients. Can we give diabetics the ability to check for retinopathy, can we help them to identify those changes and thus incentivize them to better control their diabetes?

As an AI leader, what industries you think would be fastest in adopting AI/ML with smooth efficiency? What are the new emerging markets for AI technology markets?

AI is already widely deployed. I think there’s this perception that AI is new or that it means a sentient, self-aware computer system — but that’s not it. Ada Lovelace wrote about AI nearly a century before Alan Turing. Today, many of the products we use every single day are powered by some level of AI. Google, Netflix, Amazon, Siri, Alexa – all utilize AI.

I think we’ll continue to see adoption in industries where vast amounts of data are generated, where it’s beyond the human ability to interpret that level of data at speed or where its necessary to apply all pre-existing knowledge to that data and provide an output. For example, the management of cancer or a self-driving car.

I think we’ll continue to see healthcare rapidly deploy AI, but that will present regulatory and clinical challenges. AI is still largely maintained with decision support. The next step is for AI to start providing direct care e.g. changing drug doses, introducing new medications. A lot of validation and testing will be required to back that up.

What’s your smartest work related shortcut or productivity hack?

I think Paul Graham said it best in his piece on makers vs. managers. We live in a world now of constant noise — email, Slack, Facebook, Twitter, WhatsApp. It’s so many notifications. I’ve turned all this off and I like to try and build blocks of focused, uninterrupted time. For example, scheduling meetings at 11 a.m. and 2 p.m. – it just destroys the whole day.

Tag the one person in the industry whose answers to these questions you would love to read:

Eric Topol – he’s a physician, editor of Nature’s Digital Medicine and he’s spoken extensively about the challenges and potential of technology within healthcare.

Thank you, Christopher! That was fun and hope to see you back on AiThority soon.