Artificial intelligence is redesigning healthcare. How do CIOs get a grip on what’s practical versus theoretical in the dynamic space of machine intelligence? It starts by framing the A.I. landscape.
Pattern-recognition algorithms can transform horses into zebras; winter scenes can become summer; artificial intelligence algorithms can generate art; robot radiologists can analyze your X-rays with remarkable precision.
We have reached the point where pattern-recognition algorithms and artificial intelligence (A.I.) are more accurate than humans at the visual diagnosis and observation of X-rays, stained breast cancer slides and other medical signs involving general correlations between normal and abnormal health patterns.
Before we run off and fire all the doctors, let’s better understand the A.I. landscape and the technology’s broad capabilities. A.I. won’t replace doctors — it will help to empower them and extend their reach, improving patient outcomes.
An evolution of machine learning
The challenge with artificial intelligence is that no single and agreed-upon definition exists. Nils Nilsson defined A.I. as “activity devoted to making machines intelligent, and intelligence is that quality that enables an entity to function appropriately and with foresight in its environment.” But that definition isn’t close to describing how A.I. evolved.
Artificial intelligence began with the Turing Test, proposed in 1950 by Alan Turing, the scientist, cryptanalyst and theoretical biologist. Since then, rapid progress has been made over the last 75 years, advancing A.I. capabilities.
Isaac Asimov proposed the Three Laws of Robotics in 1950. The first A.I. program was coded in 1951. In 1959, MIT began research in the field of artificial intelligence. GM introduced the first robot into its production assembly line in 1961. The 1960s were transformative, with the first machine learning program written and the first demonstration of an A.I. program which understood natural language, and the first chatbot emerged. In the 1970s, the first autonomous vehicle was designed at the Stanford A.I. lab. Healthcare applications for A.I. were first introduced in 1974, along with an expert system for medical diagnostics. The LISP language emerged out of the 1980s, with natural networks integrating with autonomous vehicles. IBM’s famous Deep Blue beat Gary Kasparov at chess in 1997. And by 1999, the world was experimenting with A.I.-based “domesticated” robots.
Innovation was further inspired in 2004 when DARPA hosted the first design competition for autonomous vehicles in the commercial sector. By 2005, big tech companies, including IBM, Microsoft, Google and Facebook, were actively investing in commercial applications, and the first recommendation engines surfaced. The highlight of 2009 was Google’s first self-driving car, some three decades after the first autonomous vehicle was tested at Stanford.
The fascination of narrative science, for A.I. to write reports, was demonstrated in 2010, and IBM Watson was crowned a Jeopardy champion in 2011. Narrative science quickly evolved into personal assistants with the likes of Siri, Google, Now and Cortana. Elon Musk and others launched OpenAI, to discover and enact the path to safe artificial general intelligence in 2015 — to find a friendly A.I. In early 2016, Google’s DeepMind defeated legendary Go player Lee Se-dol in a historic victory.
The business of artificial intelligence
What will 2017 have in store for artificial intelligence? With the history of A.I. behind us, we can now determine how A.I. could potentially help advance our healthcare capabilities through four actions:
- Identify which foundational A.I. component we’re interested in exploring.
- Select the framework for artificial intelligence that best tackles our business problem.
- Evaluate the technology stack for the best problem solution.
- Engage a digital healthcare company to provide the capability (grouping component capabilities).
The foundation of A.I.
The foundation of A.I. has been defined into four essential classifications. Next, identify the classification that has the greatest ability to advance your current business model.
- Ones that think like humans.
- Ones that think rationally.
- Ones that act like humans.
- Ones that act rationally.
Normally A.I. is considered an alternative or replacement for replicating intelligent behavior. This replication could potentially surpass human abilities. However, to date, high-performance A.I. has only performed in narrow fields, such as gaming, facial recognition and driving cars.
The full spectrum of A.I. is much broader than the narrow fields we read about in the daily headlines.
The framework of A.I.
Prior to outlining the technology stack and the domains where artificial intelligence offers value, we need to review the framework of artificial intelligence. The section of the broad categories of A.I. will accelerate where A.I. adds value and, more specifically, how we as CIOs can tap directly into that value for our organizations.
Artificial intelligence is broken out into 10 functional areas:
- Deduction, reasoning and problem-solving: performing sophisticated mental tasks.
- Knowledge representation: usable information about real-world objects in a domain that A.I. is used to solve various problems.
- Machine learning: a rational agent that perceives environmental factors and takes actions to maximize its chance of success toward a specific goal; computer algorithms that improve automatically through experience.
- Robotics, motion and manipulation: robotic movement from one place to another and the identification of efficient paths, manipulating physical objects (e.g., doors, stairs, spaces).
- Planning: the ability for agents to set goals and achieve them.
- Natural language process: interpreting written and spoken human communication.
- Perception and computer vision: improving the capability of computer systems to use sensors to detect and perceive data.
- Social intelligence: agents that can recognize, interpret, process and simulate human affects.
- Creativity: the combination of the theoretical (philosophical and psychological) and practical (generate outputs) to formulate a solution thought to be creative (creating something new).
- General intelligence: autonomous thinking, emotional intelligence, learning intelligence; machines that can perform any intellectual human task.
If we drill into machine learning, a subtype within artificial intelligence, we have three primary sub-classifications:
- Supervised learning: decision tree learning, inductive logic programming, association rule learning, and support vector machines.
- Unsupervised learning: clustering, sparse dictionary learning, similarity and metric learning and genetic algorithms.
- Reinforcement learning: Bayesian networks, deep learning, neural networks and manifold learning.
We live in a business climate where the norm is a continuous pressure to perform, deliver and innovate. CIOs are forever in search of the tool for competitive advantage. Attaining knowledge of the A.I. landscape and A.I. capabilities will drive more informed decisions, resulting in offering better services to consumers.
The technology stack of A.I.
The simple version of this stack is:
- Data collection: data sets and prep.
- Data infrastructure: technology and math.
- Deep learning: math, business technology, and design.
The evaluation of the technology stack is challenging and can result in the incorrect identification of capabilities that are too immature to fully be integrated into conventional business functions, processes and workflows. Spend the required time with your teams to properly evaluate where business and technical capabilities should be extended.
- Agents and conversational interfaces: Automat, Facebook CommAI, Howdy, Kasisto, KITT.ai, Flamingo, Maluuba, Octane AI, OpenAI Gym, Semantic Machines.
- Data science: Ayasdi, BigML, Dataiku, DataRobot, Domino Data Labs, Kaggle, Rapidminer, Seldon, SparkBeyond, Yhat, Yseop.
- Machine learning: Bons.ai, CognitiveScale, Context Relevant, Cycorp, Datacratic, Deepsense.io, Geometric intelligence, H2o.ai, Hyperscience, Loop AI Labs, minds.ai, Nara Logics, Reactive, Scaled Inference, Skymind, SparkCognition.
- Natural language: Agolo, Cortical.io, Lexalytics, Loop AI Labs, Luminoso, MonkeyLearn, Narrative Science, spaCy.
- Development: AnOdot, Bons.ai, Fuzzy.ai, Hyperopt, Kite, Layer 6 AI, Lobe.ai, Rainforest, SigOpt, SignifAI.
- Data capture and enrichment: Amazon Mechanical Turk, CrowdAI, CrowdFlower, Datalogue, Datasift, Diffbot, Enigma, Import.io, Paxata, Trifacta, Workfusion.
- Hardware: 1020 Labs, Cadence Tensilca, Cirrascale, Google’s Tensor Processing Unit, KNUPATH Intel (Nervana), Isocline, NVIDA DGX-1/Titan X, Tenstorrent.
- Research: Cogitai, Kimera, Nnaisense, Numenta, OpenAI, Vicarious.
Digital healthcare capabilities of AI
The A.I. foundation has been determined. The framework was selected. The A.I. technology stack was evaluated. We’re now ready to engage an emerging organization to help us achieve the expanded capabilities we have defined we require.
The following hot A.I. companies will help open the possibilities of how A.I. can generate new business models, fueling new organization growth.
- Imagia: artificial clinical intelligence to detect and predict cancer changes early.
- Butterfly: building a device that will make medical imaging accessible to everyone in the world.
- Deep genomics: predict the molecular effects of genetic variation.
- Mindshare: precision medicine through image-driven intelligence.
- Bay labs: bringing deep learning advances to critical unsolved problems in healthcare.
- Zebra: algorithms assist radiologists in detecting often overlooked indications.
- Behold.ai: an artificially intelligent medical imaging record platform that helps healthcare providers parse and process billings, claims, and medical images.
- Atomwise: the creator of AtomNet, the first Deep Learning technology for novel small molecule discovery, characterized by its unprecedented speed, accuracy, and diversity.
- OmeCare: merges artificial intelligence and deep learning with precision medicine.
- Advenio: provide artificial intelligence, deep learning, and machine learning-based computer-assisted detection (CADx) for diagnostic clinical imaging.
- Enlitic: uses deep learning to distill actionable insights from billions of clinical cases.
- Lunit: developing advanced software for medical data analysis and interpretation via cutting-edge deep learning technology.
- Sig tuple: builds intelligent screening solutions to aid diagnosis through AI-powered analysis of visual medical data.
- Insilico Medicine: artificial intelligence for drug discovery, biomarker development, and aging research.
- MedyMatch: bringing accuracy to physicians and capacity to healthcare to prevent chronic conditions and improve patient outcomes with the right treatment at the right time.
Assessing the potential of artificial intelligence to differentiate your organizational capabilities starts with an understanding of the A.I. foundation, the A.I. framework, the A.I. technology stack and the A.I. companies offering dynamic and useful interactions. This strong background will serve you and your team well, as you venture into the uncharted world of artificial intelligence.
Peter B. Nichol, empowers organizations to think different for different results. You can follow Peter on Twitter or his personal blog Leaders Need Pancakes or CIO.com. Peter can be reached at pnichol [dot] spamarrest.com.
Peter is the author of Learning Intelligence: Expand Thinking. Absorb Alternative. Unlock Possibilities (2017), which Marshall Goldsmith, author of the New York Times No. 1 bestseller Triggers, calls “a must-read for any leader wanting to compete in the innovation-powered landscape of today.”
Peter also authored The Power of Blockchain for Healthcare: How Blockchain Will Ignite The Future of Healthcare (2017), the first book to explore the vast opportunities for blockchain to transform the patient experience.