Andrew McAfee, Author, MIT, @amcafee
State of Understanding a Decade Ago
- Book: The New Division of Labor.
- Dealt with the question: “what are humans good at, and what are computers good at?”
- Give all the rote work to computers, and leave the pattern-matching and complex communication to humans. Example: driving a car in traffic.
- Andrew then related his experience of riding in a Google self-driving car through 3 phases of personal experience: raw abject terror (first 10 minutes); passionate interest in what was going on (next 20 minutes); mild boredom (rest of the ride). My own thought at this point: “I’ve seen the future, and it’s really boring”
- Andrew then went through the example of IBM’s Watson computer participating in the game show Jeopardy! Watson versus people in 2006 was terrible. Watson today is now as good as – or better than – the best human champions. Andrew included a photo of Ken Jenning’s funny parenthetical comment on his last question against Watson: “I for one welcome our computer overlords.” Indeed!
Minds and Machines
- We need to rethink this combination…machine abilities are growing to match those of humans.
- How did this happen? Andrew alluded to Hemingway’s quote (regarding going broke): first it happened gradually, then it happened suddenly.
- A rough calculation of the “tipping point,” using Kurzweil’s first half/second half of the chessboard square-doubling analogy; 1958 was the first year the BEA measured computing power doubling every 1.5 years. Thus, 1958 + 32*1.5 = 2006
A Change in Approach
- Rules-based approach is inferior and doesn’t work very well (i.e. learning a language as an adult using verb conjugation books). There are too many rules to learn!
- Kids learn language through listening and absorbing inductively what’s going on. Humans are pre-wired for language.
- “We know more than we can tell” – Michael Polanyi
- The game of Go is way more complex than chess, and to date computers have not been able to beat the best human players. However, this will likely change before the end of this year. How? Because we’re going to give computers a goal of maximizing the score in a game, via trial and error. An example of this was shown with the game Breakout
- Let’s do a self assessment. Compared to the people around me, I’m “” (score yourself on a scale of 1-100)
- I have good intuition; I make good predictions; I’m a good judge of character
- Now average the 3 values
- We’re bad at self-judgement and are predictably irrational
Geeks Versus HIPPOS
- Geeks (who are evidence and data-driven) versus HIPPO (HIghest Paid Person’s Opinion)
- Robert Parker is the HIPPO of the wine world
- Orley Ashenfelter (a wine geek) came up with a remarkably accurate algorithm that made wine HIPPOS largely irrelevant
What Do Humans Still Bring to the Table?
- We have advanced social skills
- We have good intuition
- We have creativity