Monday May 20th, we hosted an event around the theme of Artificial Intelligence.

AtonRâ Partners had the great honor to welcome two exceptional speakers who shared with us their vision on AI. Luc Julia, head of Innovation at Samsung, co-creator of the Apple’s Siri system and author of the book “L’Intelligence artificielle n’existe pas”, and Jonas Richiardi, Clinical Research Leader at the Department of Radiology of the Centre Hospitalier Universitaire Vaudois (CHUV) in Lausanne. He is currently developing novel statistical learning and signal processing methods to enable useful prediction with biomedical data (in particular high-dimensional time series, imaging, liquid biomarkers, and genomic data), with a focus on the brain. Mr Richardi is also an independent board member of AtonRâ Partners. Mr. Julia gave us a clear view on what he thinks about Artificial Intelligence on a macro level while Mr. Richiardi showed us a concrete case of the use of Artificial Intelligence (deep learning) in his lab at the CHUV.

You can watch Luc Julia's interview in French: 

Both Mr. Julia and Mr. Richiardi share the same vision, that we should not be overly concerned about AI or more commonly called by Mr. Julia Augmented Intelligence rather than Artificial Intelligence, as we, humans, adapt, innovate and are able of transferring knowledge from one discipline to another while a machine is rather good (in many occasions much better than humans) at performing one and one only specific task. Human intelligence is “creativity” and “failure”. AI does not satisfy our definition of intelligence (assuming there is one) but does augment human for recognition tasks. Mr. Julia downplayed the hype around Artificial Intelligence by saying that when it comes to what is commonly called AGI (Artificial General Intelligence), we still don’t have the right technologies and algorithms in place. While Mr. Julia is a big backer of AI, he is also worried that, with too much excitement around this field, future R&D projects might see a sharp decline when real-world results will be far away from overpromises made by some of the world’s top scientists. In his book, Mr. Julia, calls this period “winter time” for AI, where R&D money will drain from such projects and future developments will be negatively impacted.

Mr. Julia made also a case on the high energy consumption of such systems by explaining the hardware and the power needed by Google’s Deep Mind to defeat the world’s professional player of the Go game back in 2017. In fact, 1920 CPU’s (central processing units or more commonly the microprocessors executing your PC functions and programs) and 280 GPU’s (graphics processing units or more commonly the graphics cards in your PC which are performing graphics functions and which are very good in performing parallel computing) were needed to beat the world’s number one player, representing 1MW of power consumption vs. a mere 20 watts (50’000 times less) for the human brain.  Mr. Julia further stated that fully autonomous driving in all conditions (level 5 automation) will never come true despite what some advocates might say. Autonomous driving, in fact, will do the job 98% of the time and will be tremendously better than humans at driving but will be hopeless for the remaining 2%.

While Mr. Julia spent some time in demystifying AI and its intelligent capability, Dr. Jonas Richiardi showed to the audience that AI can still tremendously help human without actually replacing it. By showing his research field in medical imagery assisted AI at the Lausanne (CHUV) hospital here in Switzerland, Mr. Richiardi showcased how pattern recognition can help during the three stages of brain imagery (Acquisition, Diagnostic, Prognostic) by providing greatly enhanced results such as speed or prediction quality.
 

• In Acquisition: Rapid prediction of myelin content in the brain >50% more exams per day.
• In Diagnostics: Diagnosis of multiple sclerosis < 10 seconds vs. > 10 minutes with 93% accuracy.
• In Prognostics: Prediction of the future evolution (normal or abnormal) of diseases such as Alzheimer with 85% accuracy.
   

There is a dramatic need for productivity enhancement in the radiology area as the amount of data radiologists has to study is compounded (on average +10% Y/Y) every day while the available time spent studying such images decreases. The problem is further intensified by reimbursements that are under huge pressure (-30% on average according to the 2018 US Medicare physician fee schedule).

Mr. Richiardi commented on the reclassification (lower risk classification) of the rules by the FDA for new medical devices and more specifically devices using AI, thus easing the barriers to entry for many of the upcoming devices as the regulatory process gets faster and smoother. Furthermore, modern medical devices learn continuously from the data they acquire and their algorithms are constantly retrained. Regulators (FDA, CE...) must provide a framework that keeps these changes into account. Big healthcare conglomerate, relying more and more on third party companies in order to be always innovative and up to date, are opening up huge market opportunities for startups and small companies. Like Apple, these companies give access to their algorithms to third-party applications to enlarge and innovate their platform. Without investing in marketing and sales, small companies get quick access to the market and when their products are disruptive they become M&A target as well.

Mr. Richiardi further commented that radiologists will survive because AI will not be able to do everything for them. On the other hand, what is sure is that radiologists who do not use AI will disappear! AI is about recognizing things. It will always need human intelligence (and stupidity) to work.

When asked about his thoughts on Siri, Mr. Julia stated the following: "we knew deliberately it will not work out, but Apple insisted to release it despite the 20% margin error". It is tricky to read a book when 20% of the words are missing. To make it nice, we decided to add the "stupid intelligence" app to Siri. Even if Siri is not able to give you a correct response it will still answer you something funny.

Mr. Julia is although very bullish on IoT (Internet of Things) overall and all the voice assistants like Amazon’s Alexa as in order to have your audience engaged in today’s age you need Alexa.

He further added that the future of AI is probably lying into "small data" where edge computing becomes increasingly important. The “edge” subject is something that we already covered in our past monthly investment reports and is the subject of a deeper study by our company and engineers’ colleagues. With edge AI, algorithms are processed locally (no need of any live connection to the internet) on a hardware device. The algorithms are using data (sensor data or signals) that are created on the device and are particularly important where low latency and security (self-driving cars notably) is required.