While recording a podcast with the guys at DecipherSciFi about the twee(n) love story The Space Between Us, we spent some time kvetching about how silly it was that many of the scenes involved Gardner, on Mars, in a real-time text chat with a girl named Tulsa, on Earth. It’s partly bothersome because throughout the rest of the the movie, the story tries for a Mohs sci-fi hardness of, like, 1.5, somewhere between Real Life and Speculative Science, so it can’t really excuse itself through the Applied Phlebotinum that, say, Star Wars might use. The rest of the film feels like it’s trying to have believable science, but during these scenes it just whistles, looks the other way, and hopes you don’t notice that the two lovebirds are breaking the laws of physics as they swap flirt emoji.

Hopefully unnecessary science brief: Mars and Earth are far away from each other. Even if the communications transmissions are sent at light speed between them, it takes much longer than the 1 second of response time required to feel “instant.” How much longer? It depends. The planets orbit the sun at different speeds, so aren’t a constant distance apart. At their closest, it takes light 3 minutes to travel between Mars and Earth, and at their farthest—while not being blocked by the sun—it takes about 21 minutes. A round-trip is double that. So nothing akin to real-time chat is going to happen.

But I’m a designer, a sci-fi apologist, and a fairly talented backworlder. I want to make it work. And perhaps because of my recent dive into narrow AI, I began to realize that, well, in a way, maybe it could. It just requires rethinking what’s happening in the chat.

Let’s first acknowledge that we’ve solved long distance communications a long time ago. Gardner and Tulsa could just, you know, swap letters or, like the characters in 2001: A Space Odyssey, recorded video messages. There. Problem solved. It’s not real-time interaction, but it gets the job done. But kids aren’t so much into pen pals anymore, and we have to acknowledge that Gardner doesn’t want to tip his hand that he’s on Mars (it’s a grave NASA secret, for plot reasons). So the question is how could we make it work so it feels like a real time chat to her. Let’s first solve it for the case where he’s trying to disguise his location, and then how it might work when both participants are in the know.

Fooling Tulsa

Since 1984 (ping me, as always, if you can think of an earlier reference) sci-fi has had the notion of a digitally-replicated personality. Here I’m thinking of Gibson’s Neuromancer and the RAM boards on which Dixie Flatline “lives.” These RAM boards house an interactive digital personality of a person, built out of a lifetime of digital traces left behind: social media, emails, photos, video clips, connections, expressed interests, etc. Anyone in that story could hook the RAM board up to a computer, and have conversations with the personality housed there that would closely approximate how that person would (or would have) respond in real life.

Is this likely to actually happen? Well it kind of already is. Here in the real world, we’re seeing early, crude “me bots” populate the net which are taking baby steps toward the same thing. (See MessinaBot, https://bottr.me/, https://sensay.it/, the forthcoming http://bot.me/) By the time we actually get a colony to Mars (plus the 16 years for Gardner to mature), mebot technology should should be able to stand in for him convincingly enough in basic online conversations.

Training the bot

So in the story, he would look through cached social media feeds to find a young lady he wanted to strike up a conversation with, and then ask his bot-maker engine to look at her public social media to build a herBot with whom he could chat, to train it for conversations. During this training, the TulsaBot would chat about topics of interest gathered from her social media. He could pause the conversation to look up references or prepare convincing answers to the trickier questions TulsaBot asks. He could also add some topics to the conversation they might have in common, and questions he might want to ask her. By doing this, his GardnerBot isn’t just some generic thing he sends out to troll any young woman with. It’s a more genuine, interactive first “letter” sent directly to her. He sends this GardnerBot to servers on Earth.

Launching the bot

GardnerBot would wait until it saw Tulsa online and strike up the conversation with her. It would send a signal back to Gardner that the chat has begun so he can sit on his end and read a space-delayed transcript of the chat. GardnerBot would try its best to manage the chat based on what it knows about awkward teen conversation, Turing test best practices, what it knows about Gardner, and how it has been trained specifically for Tulsa. Gardner would assuage some of his guilt by having it dodge and carefully frame the truth, but not outright lie.

Buying time

If during the conversation she raised a topic or asked a question for which GardnerBot was not trained, it could promise an answer later, and then deflect, knowing that it should pad the conversation in the meantime:

• Ask her to answer the same question first, probing into details to understand rationale and buy more time
• Dive down into a related subtopic in which the bot has confidence, and which promises to answer the initial question
• Deflect conversation to another topic in which it has a high degree of confidence and lots of detail to share
• Text a story that Gardner likes to tell that is known to take about as long as the current round-trip signal

Example

• TULSA
• OK, here’s one: If you had to live anywhere on Earth where they don’t speak English, where would you live?

GardnerBot has a low confidence that it knows Gardner’s answer. It could respond…

1. (you first) “Oh wow. That is a tough one. Can I have a couple of minutes to think about it? I promise I’ll answer, but you tell me yours first.”
2. (related subtopic) “I’m thinking about this foreign movie that I saw one time. There were a lot of animals in it and a waterfall. Does that sound familiar?”
3. (new topic) “What? How am I supposed to answer that one? 🙂 Umm…While I think about it, tell me…what kind of animal would you want to be reincarnated as. And you have to say why.”
4. (story delay) “Ha. Sure, but can I tell a story first? When I was a little kid, I used to be obsessed with this music that I would hear drifting into my room from somewhere around my house…”

Lagged-realtime training

Each of those responses is a delay tactic that allows the chat transcript to travel to Mars for Gardner to do some bot training on the topic. He would be watching the time-delayed transcript of the chat, keeping an eye on an adjacent track of data containing the meta information about what the bot is doing, conversationally speaking. When he saw it hit low-confidence or high-stakes topic and deflect, it would provide a chat window for him to tell the GardnerBot what it should do or say.

• To the stalling GARDNERBOT…
• GARDNER
• For now, I’m going to pick India, because it’s warm and I bet I would really like the spicy food and the rain. Whatever that colored powder festival is called. I’m also interested in their culture, Bollywood, and Hinduism.
• As he types, the message travels back to Earth where GardnerBot begins to incorporate his answers to the chat…

• At a natural break in the conversation…
• GARDNERBOT
• OK. I think I finally have an answer to your earlier question. How about…India?
• TULSA
• India?
• GARDNERBOT
• Think about it! Running around in warm rain. Or trying some of the street food under an umbrella. Have you seen youTube videos from that festival with the colored powder everywhere? It looks so cool. Do you know what it’s called?

Note that the bot could easily look it up and replace “that festival with the colored powder everywhere” with “Holi Festival of Color” but it shouldn’t. Gardner doesn’t know that fact, so the bot shouldn’t pretend it knows it. A Cyrano-de-Bergerac software—where it makes him sound more eloquent, intelligent, or charming than he really is to woo her—would be a worse kind of deception. Gardner wants to hide where he is, not who he is.

That said, Gardner should be able to direct the bot, to change its tactics. “OMG. GardnerBot! You’re getting too personal! Back off!” It might not be enough to cover a flub made 42 minutes ago, but of course the bot should know how to apologize on Gardner’s behalf and ask conversational forgiveness.

Gotta go

If the signal to Mars got interrupted or the bot got into too much trouble with pressure to talk about low confidence or high stakes topics, it could use a believable, pre-rolled excuse to end the conversation.

• GARDNERBOT
• Oh crap. Will you be online later? I’ve got chores I have to do.

Then, Gardner could chat with TulsaBot on his end without time pressure to refine GardnerBot per their most recent topics, which would be sent back to Earth servers to be ready for the next chat.

In this way he could have “chats” with Tulsa that are run by a bot but quite custom to the two of them. It’s really Gardner’s questions, topics, jokes, and interest, but a bot-managed delivery of these things.

So it could work, does it fit the movie? I think so. It would be believable because he’s a nerd raised by scientists. He made his own robot, why not his own bot?

From the audience’s perspective, it might look like they’re chatting in real time, but subtle cues on Gardner’s interface reward the diligent with hints that he’s watching a time delay. Maybe the chat we see in the film is even just cleverly edited to remove the bots.

How he manages to hide this data stream from NASA to avoid detection is another question better handled by someone else.

An honest version: bot envoy

So that solves the logic from the movie’s perspective but of course it’s still squickish. He is ultimately deceiving her. Once he returns to Mars and she is back on Earth, could they still use the same system, but with full knowledge of its botness? Would real world astronauts use it?

Would it be too fake?

I don’t think it would be too fake. Sure, the bot is not the real person, but neither are the pictures, videos, and letters we fondly keep with us as we travel far from home. We know they’re just simulacra, souvenir likenesses of someone we love. We don’t throw these away in disgust for being fakes. They are precious because they are reminders of the real thing. So would the themBot.

• GARDNER
• Hey, TulsaBot. Remember when we were knee deep in the Pacific Ocean? I was thinking about that today.
• TULSABOT
• I do. It’s weird how it messes with your sense of balance, right? Did you end up dreaming about it later? I sometimes do after being in waves a long time.
• GARDNER
• I can’t remember, but someday I hope to come back to Earth and feel it again. OK. I have to go, but let me know how training is going. Have you been on the G machine yet?

Nicely, you wouldn’t need stall tactics in the honest version. Or maybe it uses them, but can be called out.

• TULSA
• GardnerBot, you don’t have to stall. Just tell Gardner to watch Mission to Mars and update you. Because it’s hilarious and we have to go check out the face when I’m there.

Sending your loved one the transcript will turn it into a kind of love letter. The transcript could even be appended with a letter that jokes about the bot. The example above was too short for any semi-realtime insertions in the text, but maybe that would encourage longer chats. Then the bot serves as charming filler, covering the delays between real contact.

Ultimately, yes, I think we can backworld what looks physics-breaking into something that makes sense, and might even be a new kind of interactive memento between interplanetary sweethearts, family, and friends.