I firmly believe in a ‘bite and chew’ approach. Watson is taking on too much all at once. IBM was definitely too ambitious on this one. And it seems like the algorithm’s structure is itself a major reason why it is so difficult to train (in addition to other aforementioned reasons).
To the question, “As a member of the executive team at a hospital, at what point do you decide that a collaboration has cost too much and yielded too little?” A possible answer is when you determine that a significant amount of additional cost needs to be invested to move the technology a step further in the face of a high-impact albeit smaller alternative that can begin to yield results today. (Think next best alternative)
Love this article! As I read it, I immediately thought of my dog, Nancy, her dog-husband and her two dog-children (yes I know it sounds funny but they literally live together like a happy family.) I’d totally 3-D print a new house for them, the next time they dig through the walls of their current concrete-based kernels. But why won’t I 3-D print one for myself too? Well, I don’t trust that I’d be safe in it in a certain country where crime rates tend to be high so that using concrete is the only safe way we know how to build. Irrational as it may seem, if Winsun ever seeks to expand to the developing world, which arguably needs it’s services the most, it will have to do a lot to convince the people that it’s houses are just as safe as those made from concrete.
Matt, this is really thought-provoking. Applying crowd-sourcing to the business of national security is almost like empowering everybody to become a vigilante of some sort. However, two generalizable observations greatly inhibit the usefulness of this approach:
1. People’s fears are easily driven or influenced by cultural/racial biases.
2. Lone wolves, like many regular people, are localized and thus interact with only a small portion of the population daily.
Thus, of the say, 30 people that a lone wolf interacts with in a week, 25 of them will probably be repeat interactions. Of these 25, only a few will ever become suspicious enough to raise an eyebrow at the lone wolf’s tell-tale behaviors. Of those who become suspicious, only one or two, will think it significant enough to warrant making a report. When the report hits the FBI analyst’s desk, the analyst, who is probably aware that the report may be motivated by biases, reviews it to try to assess if it’s a credible lead. Uh oh. Hold on. There are a million other lone wolf reports coming in from other localized situations, each of which is backed by only one or two “concerned citizen’s” suspicions, the analyst realizes. Not enough data points per report to even out the idiosyncratic noise associated with each report, the analyst concludes. Maybe machine learning algorithms can that can process humongous amounts of data can then be deployed to scour the online presences of the reported individuals for validating trends. But imagine the violation of privacy that will be associated with this, given that a huge portion of the reports are likely to be spurious. I don’t know man. Look’s super challenging and leaves me wondering if it’ll only lead to another social problem of privacy invasion and victimization.
This is an interesting piece. Ironically, it seems to me like the only way Bridg can truly maintain a competitive edge is to expand it’s human-based consulting services. It’s one thing for the algorithm to generate useful customer insights. Making decisions on those insights is another and this is where such consulting services may come in. As to the question of whether or not they are selling to a dying market, I don’t believe they are. Not all Brick and Mortar stores are bowing out to Amazon. Restaurants and fast food chains, for example, are certainly not. Seeing what they were able to achieve with Chipotle, I’m guessing there’s still more than enough opportunity for them to serve this current target market.
As I read this article, I began to see even how the application of this technology can be extended from maintenance into production optimization. Rather than rely on humans to observe well performance on a day-to-day basis, and decide on whether to make operational changes like beaning up/down wells or increasing gas-lift rates if applicable, the same mathematical models used by operators, production technologists and reservoir engineers can be coded into the machine learning algorithm such that, on a real-time basis, the algorithm can ‘decide’ on the right course of action and execute by sending appropriate control signals to the wells. This will free up time for humans to focus on more critical or strategic tasks.