It's not the silicon computer, it's the meatware computer
Yet another missed point about AI and Big Data.

Via MindMatters, Gary Smith discusses how Big Data helped and failed in the 2008 and 2016 campaigns.

In 2008:

The Obama campaign put every potential voter into its database, along with hundreds of tidbits of personal information: age, gender, marital status, race, religion, address, occupation, income, car registrations, home value, donation history, magazine subscriptions, leisure activities, Facebook friends, and anything else they could find that seemed relevant…

After Obama secured the nomination, the fund-raising continued. For the full 2008 election campaign, Obama raised $780 million, more than twice the amount raised by his Republican opponent, John McCain. McCain didn’t have a realistic chance of winning, and he didn’t—with only 173 electoral votes to Obama’s 365.

Obama didn't need the data. He had intrinsic advantages. He was attractive and confident, and above all he KNEW what the people wanted to hear. Obama beat Hillary in the primaries because Hillary was SCREEEEECHING the standard D talking points over and over. Obama beat McCain in the general because McCain was speaking the standard Repooflican talking points over and over.

= = = = =

In 2016:

That should have worked for Clinton too, surely. Her team of sixty mathematicians and statisticians created a program called Ada to replicate the success, as widely expected. But, as Smith explains to Marks:

Smith: Some of the stuff that made a difference, they couldn’t put in a computer. Like the enthusiasm factor. When Bernie Sanders and Donald Trump gave speeches, tens of thousands of people showed up and yelled and screamed and were excited. And when Hillary Clinton gave a speech, a couple of hundred people would show up and sit politely. And you couldn’t put that in a computer.

Wrong. Audience size is a simple number. You CAN put it in a computer, and the computer CAN calculate the national average audience size. Hillary's staff DIDN'T put it in the computer, didn't write the code to evaluate it. Here again the computer wasn't needed anyway. The difference was so fucking obvious that you didn't need refined statistical evaluation to see it.

[This is broadly and importantly true. In science and in life, statistics are totally unnecessary. A distinction or measurement that only appears after statistical evaluation is NOT a meaningful difference. If you have to use stats to get the answer, you don't have an answer.]

The real difference was the same as 2008. Bernie and Trump were speaking clearly and confidently, saying things that agreed with human experience. Hillary was SCREEEEEEECHING the standard old Dem talking points. Bernie would have won if Hillary's team hadn't MEDDLED AND HACKED the primary results.

In both cases the data was totally irrelevant. But on another level all recent campaigns were in fact human vs machine. Hillary and McCain and Romney were repeating the same meaningless noises over and over and over like a busted washing machine. Obama and Trump and Bernie were acting like humans, appearing to observe a real problem that real humans had experienced, and appearing to offer rational solutions to the real problem.



In the end both Obama and Trump turned out to be fake humans. Both reverted to machine behavior after the election. But we're only talking here about campaigns.

= = = = =

Larger sidenote: I've been noting lots of AI pointmissers lately. All of them are intentionally giving the computer too much credit. In some cases the pointmisser wants us to be SCARED so the pointmisser can intimidate us into submission. In other cases the pointmisser wants us to be IMPRESSED so the pointmisser can sell us a fraudulent product.

Elon wants to IMPRESS us with autonomy so we'll buy his shares. Hillary wants to SCARE us with the unstoppable force of the computer so we won't notice that she's a dumb low-IQ drunken thug.

There's nothing new about Big Data, and nothing autonomous about computers. Computers handled millions of verbal and numerical data points in the 1890 Census, and Hollerith machines have been figuring and printing mass quantities of checks and other documents ever since then.

This year's model can handle more data and more sophisticated algorithms than the 1890 Hollerith, but it's still just sorting and counting. It's still mechanically performing a task assigned by the HUMAN who hires the programmers. We should always give full credit or blame to the HUMAN who sets the goal.

Labels: AI point-missing, Blinded by Stats

  ¶ 11:26 PM