IBM's artificial intelligence program Watson has been training to be a doctor over the last few years, applying its machine learning skills to genetics and cancer. But apparently the AI likes to cook in its spare time. In a just-announced collaboration with Bon Appetit, Watson is using the 9000 or so recipes in the magazine's database to generate new recipes based on available ingredients and a suggested cuisine style. The AI uses both the magazine's archive and its own database of flavor compounds to determine what ingredients will go well together, and comes up with surprising new combinations. For more on how this works, check out the IEEE Spectrum article about IBM's cooking initiative for Watson from last year's special issue on food and technology.

The typical ways in which patients get matched up with clinical trials aren't exactly state of the art. At hospitals, clinical coordinators painstakingly sort through patient records, looking for people that fit the requirements of a given experimental treatment; meanwhile, patients bring their own Internet research to their doctors, asking if some new drug might help them. The Mayo Clinic is now seeking to improve this process by putting IBM Watson on the job. The artificial intelligence known as IBM Watson can scan enormous troves of written information thanks to its natural language processing skills, and its machine learning programming means it quickly gets better at using that information to complete a given task. Most famously, it quickly got better at answering Jeopardy questions, and tromped the human competition in a 2011 exhibition match.

Twenty patients with an aggressive form of brain cancer will have a new doctor on their medical team: the learned geneticist known as IBM Watson. In a collaboration announced today between IBM and the New York Genome Center, IBM's Jeopardy-beating AI will analyze the genomes of those 20 patients in hopes of providing insights for their oncologists. IBM has been promoting its AI as a killer app for health care, thanks to Watson's natural language processing skills and machine learning abilities. Over the past two years Watson has been engaged in a separate project at New York's Memorial Sloan-Kettering Cancer Center, in which doctors are training the AI to understand the language of medicine. In that project, Watson is being taught to read patients' records and search the medical literature for relevant suggestions on treatment.

Over the last eight years, we have been working on the problem of case-based reasoning (CBR) for medical diagnosis. Through a succession of research projects, we developed a system that used physiologic causes to match findings in cases, evaluated the system on 240 cases, and developed a system that divides cases and memory based on the diagnostic units in the case. Each of these steps has been a significant advance toward diagnostic systems that can effectively learn from experience. Still, it is clear that CBR has not reached its potential to effectively handle the case material and work in concert with a model-based program.

FindMe systems originated in work I did with Kristian Hammond when we were both at the Computer Science Department of the University of Chicago. These systems use case-based reasoning as a way of recommending products in e-commerce catalogs and provide critique-based navigation as a primary user interface. One interesting outcome of this work has been to emphasize the complexity of the common-sense notion of similarity demanded by a user of such catalogs as compared to the metrics used by many CBR systems.

For AI to become truly robust, we must further our understanding of similarity-driven reasoning, analogy, learning, and explanation. Here are some suggested research directions.This article is part of a special issue on the Future of AI.

Case-based reasoning is a problem solving paradigm that in many respects is fundamentally different from other major AI approaches. Instead of relying solely on general knowledge of a problem domain, or making associations along generalized relationships between problem descriptors and conclusions, CBR is able to utilize the specific knowledge of previously experienced, concrete problem situations (cases). A new problem is solved by finding a similar past case, and reusing it in the new problem situation. A second important difference is that CBR also is an approach to incremental, sustained learning, since a new experience is retained each time a problem has been solved, making it immediately available for future problems. The CBR field has grown rapidly over the last few years, as seen by its increased share of papers at major conferences, available commercial tools, and successful applications in daily use.

IBM Watson Health has announced a joint initiative with the US Food and Drug Administration to study the use of blockchain technology to share health data to ultimately improve public health. At first, the two-year collaboration will focus on oncology data, pulling together and exchanging data from a variety of sources including that from clinical trials, genomic data, EMRs, and from miscellaneous Internet of Things data from wearables, apps and connected devices. IBM and the FDA will look at how the technology can facilitate information exchange across a spectrum of data types, including clinical trials and real world data. For example, patient-generated data from connected devices could provide clinicians with more insights into population health, potentially offering up research opportunities and ways to leverage large quantities of data into biomedical and healthcare industries. At the core of the collaboration is blockchain technology, which allows secure data sharing between organizations more freely and has been increasingly favored among industry leaders.

Illumina and IBM announced that they would be bundling IBM's Watson Genomics product with Illumina's TruSight Tumor 170, a tool used to help match very sick cancer patients with drugs that might help them. The move is the latest effort by DNA sequencing companies to try to get doctors outside major cancer centers like Memorial Sloan-Kettering Cancer Center in New York or M.D. Anderson Cancer Center in Houston to try to scan patients' DNA. The idea is that the DNA test results can be used to help patients who don't have any options find medicines–either approved or experimental–that might help them. So far, this is considered standard practice for late-stage non-small cell lung cancer, but not for cancers in general. Illumina says the sale of DNA sequencing machines for use in cancer represents about 10% of its annual sales, or about $240 million.