Princeton Election Consortium

A laboratory for innovation in democracy. Since 2004

Aug 13: Biden 357 EV (D+6.1% from toss-up), Senate 52 D, 48 R (D+4.0%), House control D+3.5%
Moneyball states: President AZ NV AR, Senate MT KS AK, Legislatures KS TX NC

Auto-Crat (TM) 1.0

February 5th, 2017, 11:48am by Sam Wang

Here is an idea that, if done well, subverts the use of Twitter as a political tool by making the content into a joke.

We hear about technology as a destroyer of jobs: automated factories, and maybe someday self-driving cars. These trends hit a whole sector of Trump’s supporters. But could the need for Trump himself be eliminated?

Above is an example of a Trump tweet simulator. The idea is fairly simple. Trump’s tweets have many distinctive characteristics: “Sad!” “Overrated!” and the like. It does not seem like such a stretch to take all of Donald Trump’s tweets (and maybe utterances too) and train an algorithm to spew out text that has the same statistical properties. This is exactly what Justin Stanley has taken a Trump lexicon and used it to train a neural network. His network generates a random sequence of characters like the one above.

It is clearly a work in progress. But it could be great as a real tool. It’s almost as coherent as Trump; add a few rules, and it could be a competent emulator. Plus, it is less likely to cause international incidents.

If Trump’s 140-character pearls can be emulated by a computer algorithm, that could be a convenience for Steve Bannon. After all, puppet Trump is fairly expensive to operate, even with Russian subsidies. With the help of Auto-Crat, Bannon could reduce Trump’s hours to executive-order signing ceremonies.

Tags: President

14 Comments so far ↓

  • Dana

    You can also draft a text message by mashing the predictive text and it will sound just like Donald.

  • LondonYoung

    We need more from you on a topic like this Sam. You may be more qualified than any other human being to tell us: Can silicon chip compute power now compete with the human brain in this case – this is, to pass the Turing/Tweet test?

    Being serious here …

    Next step – can we train a neural net to win an election. I believe Stern in his recent interview where he said that Trump stumbled onto this …

  • Amitabh Lath

    The anxiety generated by the automation and internetization of the economy is a real fear. It does not feel like the usual economic cycle bust and boom. Maybe at a state U I am more exposed to it than Sam is. A lot more students are opting to go into K-12 teaching; lower reward but safe and secure. I do not know if this leads to more Trump support but it does not lie along the usual liberal/conservative fault lines.

    PS: I’m sorry I have not commented recently. Shock, not just at political situation, but at the failure of instincts for data analysis. As a student I chuckled at scientists who bought into cold fusion. Now I am them.

    • Erik Johnson

      Nothing especially new about technology increasing productivity. From fire onwards, a significant percentage of advances in technological application have favored greater economic productivity. It is a simple cost-benefit analysis, less calories expended to sustain human life, more availabe for the pursuit of happiness.

      The problem ultimately comes down to one question. Who is the beneficiary of this increase in productivity?

      Keynes famously predicted that the 20th century would see a massive boost in per capita productivity. He also predicted the increase would mean a work week as short as 25 hours. He was wrong about how that second part would turn out, but he didn’t have to be. Had that productivity increase benefited everybody roughly equally we would be in a very good situation today, but it didn’t. Consequently, economic recovery is anemic, nobody can afford the former levels of consumption.

      Instead, the gains were absorbed almost entirely by the very richest people in this country, creating a situation of unprecedented inequality. This in a society where money buys influence has seen a system rigged for the rich become even more sharply tilted in favor of one social stratum over any other.

      Instead of putting people out of work, technological advance should be increasing workers’ pay at the same rate it increases their productivity. Everybody’s cut grows, everybody wins. Henry Ford understood this, Donald Trump and his ilk do not.

    • Bulgakovs Cat

      Eric, 1.5 million long haul truckers will lose their jobs over the next 4 years– displaced by smart trucks.
      How is tech advances gunna help them?
      Amitabh, good to see you– I was worried.
      It wasn’t your failure– everyone from Gelman to Wang told us it wasn’t Brexit– but really it was.
      Even down to the shy-trumpsters– white suburban women leery of tellin a pollster they were voting for president pussygrabber.
      If the gods of the abaci were wrong what hope for us mere mortals?
      We were laid low by liberal confirmation bias and a crap candidate.
      Hilary is just awful– her arrogance, hubris, and sense of entitlement and destiny — but worst of all her choice of VP — someone guaranteed to never overshadow her.
      I voted for her because I feared Trump but I guess we didn’t fear Trump enough as a tribe.

  • Lorem

    “add a few rules, and it could be a competent emulator”

    I’m not sure to what extent I should just take the whole post as a joke, but not really. The imitation of text with a character level rnn is pretty well known and known to work well. Getting things to make semantic sense (or understanding what makes sense in general) is not currently working at all to my knowledge (as it requires overall world understanding). “A few rules” would be very little help, unless the plan is to limit vocabulary to only a few combinations.

  • pechmerle

    Here is a similar idea: create your own executive order, complete with photo of DJT holding it up.

  • Paul could Trumpify your own tweets by hand

  • pechmerle

    For those who might be interested, oral argument on the stay (TRO) of the immigr. exec. order out of federal court in Washington State before the Ninth Circuit will be live-streamed at 3:00 PST on Tues. Feb. 7. The live stream should be linked here:

    Naturally, this will get technical with a lot of case precedents tossed around, but the questions that the panel puts to the attorneys can be interesting. Note that appellate judges often ask probing questions to see if the attorneys have answers, without the questions actually indicating the questioner’s own view.

    If you don’t catch it live, the audio will be posted shortly after conclusion of the hearing, i.e. sometime after 4:00 p.m.

  • ArcticStones

    Maybe more useful would be: an organized, searchable database of all Tweets from @realDonaldTrump. Perhaps augmented by quotes with substance (when it’s linguistically possible to parse them, mind you).

    Let me give a modest example:

    “I don’t know Putin, have no deals in Russia, and the haters are going crazy – yet Obama can make a deal with Iran, #1 in terror, no problem!” (Twitter, 02.07.17)

    Journalists might quickly discover these quotes in such a database:

    “I do have a relationship, and I can tell you that he’s very interested in what we’re doing here today.” (Interview, MSNBC, Nov. 2013)

    “I was in Russia, I was in Moscow recently and I spoke, indirectly and directly, with President Putin, who could not have been nicer, and we had a tremendous success.” (National Press Club, May 2014)

    And of course there is this from his son:

    “Russians make up a pretty disproportionate cross-section of a lot of our assets. We see a lot of money pouring in from Russia.” (Donald Trump Jr., real estate conference, 2008)

    I suspect that such a database could be a treasure trove of unsanctioned facts.

Leave a Comment