The Rise of the Deep: Eric Topols Deep Medicine To Stand The Test Of Time – Forbes

Posted: October 4, 2019 at 10:41 am

This post was added by Alex Diaz-Granados

Deep Medicine to stand the test of time

We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten. Don't let yourself be lulled into inaction, Bill Gates

Since the massive public shows of artificial intelligence capabilities pioneered by IBMs DeepBlue outperforming a human chess player and Watson beating human players in Jeopardy, the word artificial intelligence started making headlines. Many of these capabilities failed to produce tangible practical results, giving the AI skeptics the platform to resist the adoption of AI in their organizations. But this situation changed dramatically in 2014 when deep neural networks started outperforming other algorithms and even humans in image recognition and then in other tasks.

While many concepts in deep learning date back to the mid-20th century, it was the 2014 ImageNet competition that demonstrated that the error rates of the deep learning systems fell below human error rates. This was the real revolution and renaissance in AI. Suddenly, deep learning systems started solving very practical problems in many areas including image, voice, and text recognition.

The Atari games show by DeepMind and its astonishing acquisition by Google made the word Deep a popular prefix for corporate, service, and product branding. Deep Genomics, Deep Chemistry, Deep Pharma and this brings us to Eric Topols book - Deep Medicine.

Eric Topol chose the best possible name for the book on artificial intelligence in healthcare. It contains the key word from the most disruptive technology and a hint of depth.

But writing a book like Deep Medicine is a gargantuan task. It is important to understand the healthcare industrys main problems and how messy this industry is to appreciate the importance and gravity of the book.

Healthcare is a very broad field with many segments and AI is a very broad term with many underlying technologies and data types. Another problem is that most of the AI scientists nowadays publish their findings on ArXiv, a manuscript repository that is not peer-reviewed. Many of these AI scientists have no experience in drug discovery, or even basic biology or chemistry. This naturally generates many poor-quality papers on toy data sets by the AI scientists. There are also many papers where data sets with good-quality well-annotated data are used by the bioinformaticians venturing into AI that do not demonstrate the true power of deep learning. Most of these papers will never be translated into real-world applications but are often presented as a finished product. Finally, most startups do not publish at all and it is only possible to judge them by examining their press releases, websites, people, deals, and investors. The industry looks like martial arts before MMA. There are boxers, kickers, and grapplers but only very few mixed martial artists who can understand both the AI components and the practical biology, chemistry, or digital medicine components.

Another problem that I see is the clear split in attitude among the pharmaceutical industry experts. Since the 2014 revolution in deep learning, the pharmaceutical industry split into the three camps:

AI skeptics, who doubt everything about AI and present every bad paper as an industry fiasco and point to IBM Watsons early challenges in the healthcare domain as evidence that AI will never work in the medical field. They look for every opportunity to downplay AIs value and potential. These skeptics are typically biostatisticians, bioinformaticians, and computational chemists, who changed their company titles to be more in-line with the trends. Also, this group includes third-tier AI scientists who do not have any other way to make a name for themselves but to appear as, and appeal to the AI skeptics by dissecting and downplaying the potential of the AI-parts of the research papers.

AI optimists, who take an optimistic yet cautious view on the state and the potential of AI. They typically have a balanced view.

AI hypers, who present every advance as a breakthrough and are very vocal about that. Hypers are primarily people associated with technology companies, including technology giants that invest billions of dollars into the underlying technologies and platforms. These hypers demand more funding and more resources for the grandiose projects intended to solve intelligence.

And finally, in this industry, nobody reads the research papers in depth. For example, very often the heads of AI companies and business development executives ask some of the original inventors of critical technologies about what technology they are using, what problems can they solve and how different they are from the others without taking the time to read the published literature. They do not have the time.

You need to be a real pro in many areas of science and technology to be able to structure a book like that. Deep Medicine solves most of the above problems with a very deep dive into both technology and medicine. Dr. Eric Topol is a medical doctor and a scientist with the h-index of 223 with over 250,000 journal citations. He is in the top 10 most-cited scientists in medicine and is in top 100 top-cited scientists of all time.

AI before and after deep learning

Before the revolution in deep learning IBM invested substantial resources in promotion of AI in healthcare and generated enormous publicity striking big deals with major hospitals and pharmaceutical companies. In the book Topol dissects or maybe vivisects is a more apt term the IBM Watson controversy. He writes that, while IBM initially promoted the AI and super computer technology that powered Watson as capable of better-than-human ability to make diagnoses, further studies showed the technology was deeply flawed.

IBM Watsons experience with MD Anderson, one of the countrys leading cancer centers, was a debacle noteworthy of many missteps, Topol writes. One of the most fundamental was the claim that ingesting millions of pages of medical information was the same as being able to make sense, or use, of the information.

Watsons poor introduction caused many people, including technology experts, to doubt the promise of AI to improve not just the medical field, but also other areas.

Provides a very balanced view

The ability to point out the flaws of past AI implementations without necessarily following those criticisms into cynicism and Luddite-ism is the real strength of Topol and his book, specifically. Perched wisely between utopian optimism and dystopian pessimism, Topol takes a middle position that benefits readers by providing them with a thorough, accurate and well-balanced view of the state of AIs role in next generation medical technology.

Deep Medicine will stand the test of time

Nowadays I truly enjoy reading books from the dotcom era dating to 1999 and 2000 when experts tried to provide an accurate account of the status of the new industry which just emerged and try to forecast into the future. The technology was changing so quickly back then that it was difficult to own a laptop for over a year. The Internet revolution happened and things got exponential. I was studying computer science at Queens University and remember that feeling of missing out on the revolution. Starting or joining a dotcom company was so tempting. The books published back then did not stand the test of time. To quote Bill Gates again: Most people overestimate what they can do in one year and underestimate what they can do in ten years.

But it is not the case with Deep Medicine. It provides a very balanced view. Industry insiders will immediately notice that the author started writing the book a few years and some of the companies mentioned in the book made substantial progress from where they started or changed their business models even before the final publication. This shows how quickly the industry is evolving. And that is one of the reasons the book is valuable. Since it was published exactly five years after the deep learning systems first outperformed humans in image recognition and the field exploded, it provides a rare record of the birth of the industry. Most of the startups leading the field today already made their first steps and demonstrated some capabilities. So for everyone planning to start a company in AI for healthcare, the book will provide valuable background on how it all started. One interesting fact about the industry is that over the past two years the number of AI startups entering the industry declined.

One story I am intimately familiar with was Insilicos Cornucopia of Meaningful Leads study, which was submitted in June and accepted in November 2016. To my knowledge, it is the first peer-reviewed paper on the application of Generative Adversarial Networks (GANs) in drug discovery. And this story is covered in Deep Medicine.

Back in 2016 no one in the pharmaceutical industry believed that this technology could be used in drug design and conference presentations, while entertaining, were met with heavy skepticism. But in 2018 the wind started changing, more scientists joined the effort and the validation experiments are slowly converting the skeptics who are willing to take the time to bridge the gap between the novelty of the AI methods and very basic validation into believers.

Another startup by a deep learning pioneer Brendan Frey called Deep Genomics.The name appeared in Deep Medicine more than once. There are many companies claiming to do AI, Deep Genomics does exactly what their name states and more - credible application of deep learning to a variety of biological problems. In September 2019 Deep Genomics announced the nomination of their first therapeutic candidate. This is no simple feat as they also claim to have discovered a rare disease target and went end-to-end in less than two years.

The main problem with deep learning in healthcare is that while the advances in high-performance computing and data management made it possible to train the systems very quickly, the time it takes to test the output is very long any orders of magnitude longer than in other industries dealing with pictures, videos, and text or even in robotics. It takes a very long time, effort and money to validate molecules in cells and animals and requires close collaboration between the domain experts in both AI and biological, chemical, and medical sciences. Failures are very common and painful.

Deep Medicine not only provides a helicopter view of AI for drug discovery and development but goes further into the limitations, regulatory and ethical considerations. It talks about many aspects that go beyond the capabilities of state-of-art AI, such as compassion and empathy. Like Kai-Fu Lees AI Superpowers, Deep Medicine is a must for every bookshelf. It is also available in audio format for busy professionals who like to absorb content while commuting to work.

Recommended as:

Read more from the original source:

The Rise of the Deep: Eric Topols Deep Medicine To Stand The Test Of Time - Forbes

Related Post
This entry was posted in Medical Technology. Bookmark the permalink.

Comments are closed.