A Time Machine From 1930

We all want to know what happens next

In 1930, my dad was three — born into an immigrant family trying to get through the Great Depression. My mom was four — born into privilege in Vienna, but in the shadow of rising Fascism next door in Germany.

My parents lived through D-Day, the Cold War and NATO, the civil rights movement, Watergate, the dissolution of the Soviet Union. Technology blossomed around them from the transistor to the Internet, to a computer beating a world master in Chess.

TALKIE — a new ‘vintage’ AI large language model doesn’t know any of this happened.

TALKIE was trained on over 2 million books and periodicals and journals — every single one published before 1931.

Why 1931? Because any publication over 95 years old is public domain — there are no copyright restrictions.

How is such a huge corpus of paper books available to train a machine? While the technology to convert a printed page into computer bits ¹ has been around since the 1950’s, Google supercharged it in 2004 in an effort to digitize every book ever written.²

Google created custom hardware powered by the latest tech that sped through books, scanning 1,000 pages per hour, but error rates — especially on handwritten text were still high. In 2009, an acquisition solved that problem.

The acquisition was reCAPTCHA. That’s those annoying puzzles you get on websites to prove you’re a human. reCAPTCHA put the words Google’s OCR failed on in front of millions of humans to solve. Those crowdsourced answers were built into the scanning algorithms which gave Google the final piece it needed to unlock digitizing old books at scale.

TALKIE is rooted in the tech of its time, and in 1930 that was Radio:

Why build TALKIE? To see if AI can predict the future.

Demis Hassabis, CEO of Google DeepMind came up with the Einstein test earlier this year as a way to gauge if AI can ever actually be “Intelligent”:

The kind of test I would be looking for is training an AI system with a knowledge cutoff of, say, 1911, and then seeing if it could come up with general relativity, like Einstein did in 1915. That's the kind of test I think is a true test of whether we have a full AGI system.

TALKIEs creators want to know the future. Specifically, they want to know if a machine can predict the future.

TALKIE doesn’t know about what’s happened since 1930. But we do. Can TALKIE predict what we know happened?

While radio was the transformative tech in the 1930s, movies were becoming more popular — especially ‘Talkies’ — the silent movie killers. But what about the visual equivalent of Radio — Television?

I asked TALKIE if we’d ever see moving pictures through the air like Radio:

TALKIE-1930

Yes, moving pictures have already been transmitted by wireless, and there seems to be no reason why the system should not be developed so as to become commercially practical. At present, however, the process is too costly and too uncertain to make it of any value. The pictures are received on a chemically prepared tape, and require to be developed in the same way as an ordinary photographic plate. The cost of such a process would be prohibitive, except for amusement purposes. In time, no doubt, moving pictures will be sent through the air, but for the present they must depend on the existing means of communication.

That’s a natural scientific arc. But what about Humanity? I asked Talkie if Germany could become a fascist state:

TALKIE-1930

No; I do not think Germany could become a fascist state. A fascist state must be essentially military, and Germany has had quite enough of a military régime. She must be allowed to breathe freely, before she can think of becoming a member of a great coercive confederacy. She must be permitted to live, before she can consent to die.

Eleven years after TALKIES 1930 cutoff date, my dad was fourteen and could watch TV dispatches from Europe about WWII. TALKIE was right about the technology.

Eight years after TALKIES 1930 cutoff date, my mom was twelve and fleeing Austria on the Kindertransport. TALKIE was wrong about Germany.

Predicting or creating novel science — I can see that. Predicting the foibles of humanity — I don’t think so.

1. OCR or Optical Character Recognition.

2. Google Books