One point O’Neil made in the conclusion that really hit home was that the negative impacts of many of the algorithms she described tend to be worse for groups that already experience a great deal of inequity, and thus it reinforces their situations and makes it even harder for them to escape. While not quite the same, this reminded me of a recent article I read on data cascades that focused on how data quality issues can grow as the usage of the data cascades through downstream use cases. A simple analogy is the old telephone game in which a relatively short message is relayed one by one via a series of people. The errors introduced at each step may be small but they compound as the number of people grows, often resulting in a very different story after a relatively short number of retellings.
One aspect of the relevance here is that data provenance is often not well understood, leading to bad assumptions about the appropriateness of the data used to train a model. Data sets are sometimes used without an understanding of the noisiness of the data or the conditions under which it was collected. Inappropriate usage invariably leads to incorrect conclusions. But at a more conceptual level, it made me realize that as bad as the direct inequity of some of these algorithms may be, the indirect inequity that is compounded by certain groups’ past experiences and the combined impact on their likely future experiences is even worse.
]]>I was fortunate enough to attend a book reading by Cyrus at my neighborhood library in Oakland that was attended by none other than Lou Katz, who makes an appearance in the next to last chapter. I met Lou at an OpenOakland hack night many years before but had not known that he had founded USENIX.
]]>Although it was published several years after Bostrom’s Superintelligence, I recommend reading Human Compatible first. Russell covers similar ground with respect to the problem of control over a superintelligence but in a style that I think most interested readers will find easier to follow and more insightful. If you then want a more expansive coverage of the risks and challenges posed by a superintelligence, as well as the likely ways one might emerge, then Superintelligence is your book.
One of Russell’s key points is that we shouldn’t think of the control or value problem in the same way we might think of solutions for a more traditional machine learning algorithm or even a narrow AI. It’s simply not possible for us to specify a priori a loss or cost function to optimize. None of us would qualify as a superintelligence, and I don’t know about anyone else, but I don’t think I could come close to specifying the loss function I might be trying to optimize. Russell’s proposal is that rather than trying to directly seed a superintelligence with the optimal values and objectives, we should instead ensure that it strives to achieve our objectives after it first learns them. History is rich with tyrants and terrorists whose objectives we obviously wouldn’t want to be learned by a superintelligence but a more representative sample might just work to ensure that we don’t get broken down into piece parts as part of Bostrom’s paperclip factory nightmare.
]]>Your score from taking the quiz at the beginning of Factfulness is both all-important and of little importance, depending on the perspective. One of the core premises of the book is that we have a much more negative view of the state of the world, especially of countries that are typically referred to as “developing”, than is the reality. In that sense, the results of the quiz are all important. But in the many areas in which most of us will have little prior, detailed knowledge, it is easy to be off in your guesses and to do poorly. I did very well on the quiz but it wasn’t because of any special knowledge I had other than knowing that core premise in advance.
Regardless, it’s a delightful, well-researched book that I found to be highly educational.
]]>Polson and Scott have done a fantastic job of this at an obviously much greater scale. I love the historical stories they have woven into the explanations. Storytelling can be a great way to communicate information but sometimes all you remember is the basic plot line. The authors deftly used historical precedents to build a foundation for the technologies they were explaining, giving the reader an easier context in which to understand newer and more complex concepts. And for that reason, I would also strongly recommend this book to engineers and even to experienced practitioners, because the stories are also pretty damn interesting.
]]>If you’re comfortable with command line tools, exiftool provides a handy way to clear or set metadata for PDF files and other file types.
$ brew install exiftool
$ exiftool yourfile.pdf
$ exiftool -Creator="" yourfile.pdf
Not all behaviors are so revealing. I have to admit I’ve added many books I aspire to read to my Goodreads “Want To Read” bookshelf, or perhaps even want others to think I’ll read, but am actually quite unlikely to ever read. Case in point, Deep Learning by Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Netflix queues/lists are notorious for being filled with documentaries that go forever unwatched. The behaviors to pay attention to are books read and movies watched. The types of books that are more likely to be started but never finished is another interesting area for research.
A few sections of the book are not only NSFW, but were a little uncomfortable for me to read on a plane with strangers at my shoulders. Internet searches tell us a lot about the true nature of people’s interest in racism and pornography. One of his more notable discoveries was the correlation between areas with high numbers of racist searches and areas in which Obama performed significantly worse than John Kerry. Having grown up in the South, I know how prevalent virulent racism remains there. So, it’s unsurprising to me that Internet searches tell us how deeply racist the South remains. But, Internet searches with geo-located IP addresses tell us that racists also span the country in large numbers, even in very liberal areas.
]]>The second part of the book explains the ACA in rigorous detail. While increasing access to healthcare for more individuals was a very important part of the ACA, less than 25% of the ACA is actually about access. While more of the ACA should have been about controlling costs, the reality is that many compromises were made in order to pass any bill and the Republican-controlled government has actively worked to weaken most of the parts of the ACA that do control costs. The rollout of healthcare.gov was initially a disaster, though the site now works extremely well, due in large part to a lack of oversight of the private firms hired to build the federal exchange.
Emanuel closes the book with a look forward (as of 2013-14) on the future of the ACA and other potential healthcare reform. His optimism is so far not being delivered upon, as the opportunity for improving the ACA has been shut down by desperate attempts to destroy it, no matter how destructive the impact will be on the most vulnerable Americans. But with the growing popularity of the ACA now that more people know what it actually contains, I share his hope that it will be constructively improved.
]]>My employer Castlight Health is mentioned in a brief discussion of companies hired by employers to help their employees understand in advance the likely costs for specific procedures performed by specific providers. In my 6+ years at Castlight and in my current role as CTO, I’ve had a front seat view of the crazy, complicated mess that is healthcare pricing in the US. I can confirm Elisabeth’s claims concerning the lack of correlation of cost to quality, unjustifiably wide variance in amounts charged, and unpredictability of prescription drug pricing. In our experience, the strongest correlation to cost is often the market power of the negotiating medical provider.
As much as I enjoyed the book, I found a few parts frustrating. She writes critically of hospital’s efforts to learn from the travel and hospitality industry to create a better patient experience, while at the same time emphasizing that hospitals are often criticized for providing poor patient experiences. That criticism could still make sense if they’re spending most of the money on the wrong things, but I haven’t seen conclusive evidence of this. While it’s easy to criticize the fancy lobbies, appearances do matter. Otherwise, the better hotels wouldn’t bother, either. And she sometimes writes critically of companies trying to mitigate the screwed up aspects of our healthcare system because they have their own financial motives. But that is how these firms manage to stay in business and pay their employees. We can’t all be non-profits, though in the healthcare business, being a non-profit doesn’t necessarily mean you don’t rake in revenues far exceeding the required expenses.
]]>After changing my password on a work laptop running Sierra, I finally rebooted after about 10 days. After a long series of prompts indicating different processes that couldn’t access the keychain, I finally gave up on fixing it and created a new login keychain. After another reboot, the prompts stopped appearing, though I had to re-enter a lot of credentials. All seemed good until I launched Evernote. The main window would not appear and all of the menu items were grayed out except for a sub-menu, but nothing was selectable from it.
After looking in the Evernote log file I found a message like the following appearing repeatedly.
"No auth in keychain. We can not login to” …
My subsequent Internet search returned people posting on Evernote forums and other places about this problem, along with the above error message, but no solution. When I looked in Keychain Access, I found that there was no longer an entry for Evernote. Thinking more about it gave me the idea that Evernote might have a bug that put it in a state in which it couldn’t find the keychain entry, but wouldn’t prompt me to provide credentials to add a new entry.
While trying to fix the problem I repeatedly killed Evernote and EvernoteHelper. That led me to poke around with EvernoteHelper in the OS X menu bar. I clicked on the display monitor icon to save a screenshot to Evernote. This brought up an Evernote login dialog. After I logged in, the main window in Evernote suddenly appeared and a new entry for Evernote appeared in Keychain Access. Problem solved.
]]>