Hacker News Reader

Comment by mountain_peak | Hacker News Reader

For the past month, I've been buried in Archive.org to collect data for an alumni article I'm writing up. The historical figures I'm researching seem to have 'come to life' for me based on their own writing and how others described them. I'm pretty much convinced that with a large enough corpus, we'll be able to have a decent conversation with historical figures, ask questions, etc. For anyone participating in social media since 2010, we'll likely be able to interview their digital "ghost" in the not-to-far future for all sorts of insights - maybe even for clues in solving crimes. If they have a large enough audio and/or video presence, we'll be able to see them and hear their own voice as well.

View on HN

Yes. Though, of course, you'll only be able to have conversations with the aspects of their personality that show in the surviving documents.

As a trivial example, Angela Merkel is/was well-known for making pretty good jokes with journalists when off-the-record. Any journalist who let that part of her personality shine through, wasn't invited again.

Thus the media portrayal, and most interviews with and speeches by her are much drier.

(For a personal example, if you train on my work emails, you get a different personality than if you train on my text message, and that's still different from training on how I talk with my mouth all day.)

Similarly, we'll be tempted to ask virtual famous people from the past about events that happened after they died, but their responses will be extremely speculative.

I've seen studies where Google broadly 'categorized' personality types from certain time periods (e.g., a decade) and from a similar area (e.g., the US) as having very similar experiences. I know HN is a microcosm of the general public, but even here, I find dozens (maybe hundreds) of people who were exposed to the same 'input' as me tend to have the same long-term views and interests.

Certainly that doesn't help 're-imagine' someone as unique as Angela Merkel, but if her Google searches were known, maybe her humourous personality would shine through if they were combined with traits common to Germans born in the 50s? Maybe someone asking questions to a digital ghost would also have to specify whether they want to speak with Angela the person (limited information) vs. Angela the politician (lots of information)?

Much of what I've discovered and correlated on the people I'm researching would have remained undiscovered unless someone had the time to sift through literally thousands of physical journals as opposed to a month of Archive.org queries on my part. The digitization and open access to old books uncovered new connections and information. It's fascinating!

Of course, part of the dream (or nightmare) of machine learning is that with enough training data we can build models that can recover hidden aspect of your private personality from eg video footage of your public personality.

Really silly example, perhaps your humor and wit (or lack thereof) actually show through in the way you smile? Of course, the correlations the machines will actually uncover won't be as transparent nor legible to humans as this.

That Black Mirror episode was disturbing.

First time I've heard of this series, but it sounds like something I'd really enjoy - thanks for the tip! I guess I'll know the episode you're referring to when I come across it.

Reply on news.ycombinator.com