Bits of Books - Books by Title


How People and Machines Are Smarter Together

Nick Polson and James Scott

More books on Inventions

AI based on decades old ideas, but needed three enabling tech advances. Fast computers, huge troves of data as knowledge got digitised, and cloud computing (which provided the infrastructure).

We panic about evil robots, mainly bc we've seen so much film/TV featuring them. But we are a long way from creating a robot with general intelligence. For the foreseeable future, oru 'smart' robots are smart only in their specific domains.

The production team behind House of Cards shopped their concept to all the major studios, each of who wanted to see a pilot episode first. "How can we sure there is an uadience for something as sinister?" Well, Netflix was sure. Their data told them they din't need a pilot to demo that there was a market. So Netlix took the next step from just being a channel to deliver movies, to rebuilding its business around what they could predict from user data.

This also marked a step for all internet users. Until then the most impt algorthm was the one behind Google Search. But the key algorthms of the future would be suggestions, rather than search.

Urban legend around statitician Abraham Wald. The US Navy needed to know where best to armour protect their planes in WW@. Basically it resolved into whether should armour fuselage, engines or cockpits.The story you'll commonly read about on internet says that planes returning to base had lot more bullet holes in fuselage than anywhere else, so the miltary blockheads were proposing to armour there. But it took Abraham Wald to point out the flaw in their reasoning - that they were just looking at the survivors, and not seeing the ones hit in engine or cockpit that never made it back.

But the miltary weren't dumb - like anyone, they could see that cockpit or engine were more vulnerable points. What they needed from Abraham Wald was a way to calculate without having the data from the planes that never returned. Wald came up with concept of conditional probability - the probability of an event (A), given that another (B) has already occurred.

And this is the (simplified) basis of Nteflix suggestion algorthm. Uses its data that shows viewers who saw the movie you just watched, moved to such-and-such a film (and styaed to watch that one right through, and gave it a postive vote).

Past data not very helpful in predicting US presidential elections, bc there's only been 56 of them. But the billions of photos stored online produces many useful data points.

56% of American teens talk on phone while driving. In 2015, 2715 American teenagers died, and 221,000 went to hospital. Half of all crashes with teenage drivers were single car accidents. Our grandkids will react to the knowledge that we permitted these drivers in the same way as we react to the way our grandparents used to drive drunk.

The problem with interpreting medical data. Take for example a mammogram for breats cancer. 1% of 40 yo females have breats cancer. The test has 80% success rate, so if test 100 40 yo women, 10 of them will have BC, and the test will find 8 of them. But the test has a 10 false-positive rate. So out of that 1000 woman sample, 100 will be wrongly flagged as having BC. This means that if the test says a woman might have BC, the probability is only 7.6$ that she actually does. But doctors, presented with those stats, estimated that a flagged woman had a 70% chance of having BC. Conceptually biased by the "80%" success rate of the test, instead of anchoring idea that 99% of the sample are actually healthy women, and in fact the odds are overwhelming that the test has failed.

The other area of conspicuous misunderstanding of statistics is with investing. There are a tiny number of people like Warren Buffett and Peter Lynch who have successfully picked stocks that delivered results better than stock market index.

The problem comes in identifying these very few winners among the huge number of losers. The odds are terrible - you have virtually no chance of picking a winning investment manager, bc you have no reliable data on which to base your choice.

IBM's Watson competing in Jeopardy! was asked for a rhyming phrase for "a boxing hit below the belt" The correct answer being 'low blow', but the computer came up with 'wang bang', a phrase which was not in its library, and which it must have come up with on its own.

Books by Title

Books by Author

Books by Topic

Bits of Books To Impress