Eyeworld

MAY 2018

EyeWorld is the official news magazine of the American Society of Cataract & Refractive Surgery.

Issue link: https://digital.eyeworld.org/i/978371

Contents of this Issue

Navigation

Page 18 of 86

16 May 2018 EW NEWS & OPINION Insights by J.C. Noreika, MD, MBA to monitor political activity and suppress dissent. Searchable Log of All Conversation and Knowledge (SLACK) scans an employee's profes- sional communications and social media profiles. Computer vision can analyze processes in the clinic and operating room such as compliance with hand washing requirements. Ignaz Semmelweis can finally rest. Karp identifies a fundamental question regarding AI. She asks, "Where does it start and where must it end." The higher the stakes in forecasting future outcomes, the greater the urgency for technology not answering but asking appropri- ate questions. Rest easy. AI will not replace ophthalmologists. (Accenture's Job Buggy agrees.) It may make us more efficient, productive, and better diagnosticians by augmenting skills learned through study, observation, and experience. Google's Gulshan understands: "Hence, this algorithm is not a replacement for the compre- hensive eye examination, which has many components, such as visual acuity, refraction, slit lamp exam- ination and eye pressure measure- ments." And no binary code will ever replicate the support, compas- sion, and connection of an empathic physician. That Latin exam? My clandes- tine effort at primitive AI failed miserably as the runes carefully inked into my left palm smeared indecipherably. Caesar's Gallic Wars had claimed another victim. EW References 1. Chen JH, et al. Machine learning and prediction in medicine – Beyond the peak of inflated expectations. N Engl J Med. 2017;376:2507–2509. 2. Suich Bass A. Non-tech businesses are beginning to use artificial intelligence at scale. The Economist. March 31, 2018. 3. Beam AL, et al. Big data and machine learn- ing in health care. JAMA. 2018;319:1317– 1318. Editors' note: Dr. Noreika has practiced ophthalmology since 1981. He has been a member of ASCRS for more than 35 years. Join the discussion on this article and others on the EyeWorld blog at blog.eyeworld.org. Contact information Noreika: jcnmd@aol.com interpret retinal images of diabetic patients. A team headed by researchers at Google found that their deep learn- ing algorithm could detect patients with referable diabetic retinopathy, diabetic macular edema, or both with approximately 90% sensitivity and 98% specificity. Fifty-four U.S. eye specialists defined the standard referable diabetic pathology as "moderate or worse" retinopathy and/or maculopathy. Another study reported that computer vision tech- nology performed equally well as 21 board-certified dermatologists assess- ing digital images to identify benign and malignant skin lesions. Machine learning may soon support clinicians by providing reliable and reproducible differential diagnoses based on components of the electronic health record such as patient history, lab work, imag- ing studies, and genetic analysis. It might uncover pathologic "zebras" lurking in the high grass of medical acumen. Long eyelashes and blue sclera? Kabuki syndrome makes the list, its probability ranked. As algo- rithms become more sophisticated, they may "listen" Alexa-like to a physician's conversation and alert her to omissions when, for exam- ple, discussing therapeutic options. Computer vision can exploit facial recognition tools to discern a pa- tient's level of interest, attention, or confusion. Algorithms can analyze voice patterns to detect one's level of empathy; Metlife and Humana em- ploy an app called Cogito to identify and address an employee's "compas- sion fatigue." Sounds Orwellian. Conse- quences will be both calculated and unintended. 3 China applies AI customer satisfaction while reducing cost. Sound familiar? Like most things technologic, the concept isn't new. Academics first convened to discuss artificial intelligence in 1954. ENIAC, the computer's primogenitor, wouldn't be powered down for another year. AI is now possible because of the incalculable amount of data avail- able, the unimaginable power of today's hardware and the evolv- ing sophistication of algorithmic architecture. Sundar Pichai, CEO of Google, enthused, "AI will have a more profound impact than elec- tricity or fire." But AI is a lot more abstruse than electricity or fire; it lags far behind the sophistication of human thinking. Gurdeep Singh, Microsoft Corporate vice president, called AI platforms "idiot savants." They can detect the smallest defects in manufactured goods "but have trouble with things that people find easy, such as basic reasoning." 2 AI and machine learning are used interchangeably. These binary creations help define a matrix of al- gorithm-based programming. An al- gorithm that exhibits the capacity to "learn" with less human instruction or specification advances its claim further along the AI continuum. Where the demarcation point lies is not defined. However, the far right side of the sequence includes appli- cations that embrace deep learning. These ultra-complex, multilayered networks are modeled on our brain's neural system. They require hun- dreds of thousands of examples—big data—to uncover intrinsic patterns correlating with the object of inter- est. Computer vision that controls self-driving automobiles is an exam- ple. Computer vision is also used to Is artificial intelligence (AI) the key to greater medical productivity, efficiency, and value or an over-hyped dissembler introducing more cost and complexity than benefit? T here was a day when "artifi- cial intelligence" was a crib sheet carefully concealed to abet one's performance on a Latin mid-term. "Mundus vult decipi, ergo decipiatur" or for the great unwashed among us, "The world wants to be deceived, so let it be deceived." The riddle of artificial intelligence's value hinges on its capabilities and limitations. Will we be hoodwinked by the hype or, to paraphrase Gertrude Stein, is there a there there? 1 I recently heard Athena Karp, founder and CEO of HiredScore, a startup that produces software employing AI algorithms to scan job applications and vet candidates whose skills and temperament may fit an organization's needs. This is no small task; Johnson & Johnson receives 1.2 million applications per year for 25,000 job openings. Guess wrong and it costs as much as 20% of the position's salary to recruit a replacement. HiredScore's innova- tive contribution encompasses not just selection but espouses compas- sionate rejection. Its clients convey decisions to applicants within days. The software suggests improvements to or better matches for an individ- ual's skill set. Its database retains candidates' information to be mined if future opportunities present. AI is poised to transform human resource management, supply chain logistics, and customer service. The McKinsey Global Institute predicts that $2.7 trillion of economic value will be created by AI's effect on mar- keting, sales, and supply chain man- agement alone. The finance industry is betting heavily on its promise to increase efficiency, productivity, and Smartest bot in the room J.C. Noreika, MD, MBA

Articles in this issue

Links on this page

Archives of this issue

view archives of Eyeworld - MAY 2018