One Giant Leap For A.I.
The release of DALLE-2 has kept me awake at night, in part because I’ve felt strangely alone in my astonishment. Then, a popular thread then popped up on Reddit:
I’m charmed by enthusiastic, stream-of-consciousness titles. This one resonated:
@TheFlyingDove: “I've tried to share how absolutely mind blowing this is with people I know. Always the same response - a meek hmm that's pretty cool. As if I didn't just show you something akin to actual magic.”
@PJG: “I showed my boyfriend and his reaction was "ok and?". I was like ?!?!? What do you MEAN??? Isn't this the most MINDBLOWING THING EVER???
And his response to that was "So? That's just what computers do right?"
???????????????”
@PeyroniesCat: I’ve shown it to several people, and none of them are nearly excited as I am. They were all just “meh.” The most I got was a “that’s interesting.” I’m convinced that this is so revolutionary that most people can’t grasp the gravity of it just yet.
And here I was, losing my mind in solitude. I was beginning to feel alienated, a paranoid Cassandra having visions of techno-raptures. I added my own comment to the reddit thread, and offered a theory explaining the disconnect. I posted the comment, went to bed, and got the best sleep in weeks.
I just wanted to thank you for this thread.
I've experienced the same and I'm glad you started a conversation about the disconnect. The thing I think people are missing is abstraction. DALLE is the best example I've ever seen of generalizing.
It "understands" the essence of things. Instead of being limited to a search query for "cat" and displaying the correct image from a database, DALLE-2 gets "catness" as well as the essence of everything else.
I don't think people are giving DALLE-2 credit for generalizing skills. They think it's just like a fancy Google Image search, a patchwork of real photos married to text prompts. For example, if the query is "golden retriever walking on it's hind legs in a herd of zebras," what ordinary people think DALLE’s doing is going through National Geographic magazines, scissoring out photos of all the correct scenes, and doing some basic intelligence to glue everything in the right places.
They don't understand that DALLE gets the essence of golden retriever and can manufacture it from scratch, in any scene it wants. You can tell DALLE to draw, "a half-cat half-elephant king delivering a speech to an army of bunnies set in the future on Mars, by Picasso"
And it will produce an image better than your imagination will.
I don't think it's unreasonable to suggest that DALLE is beginning to outperform the human creativity itself. That's the mind boggling fact we appreciate.
Watson beat Ken Jennings and Deep Blue defeated Kasparov, now DALLE knocks on the door of our imagination.
In the morning, I awoke to a torrent of messages. My comment was reposted to /r/bestof. The flood of notifications felt validating, like I regained my sanity by receiving assurance from others. I was overjoyed to have contributed some resonance to the conversation, and I took that as a sign of encouragement to expand my thoughts on DALLE here.
Further Analysis
Standing at the cusp of A.I. development feels disconcerting. It reminds me of January 2020, when medical experts lagged behind subcultures like the rationalist community. Nowhere on the news have I seen anything about DALLE-2.
In early 2020, I channeled my unease by preparing my home like any responsible adult should’ve already — By acquiring a basic emergency supplies kit and stocked food cabinet. Then, with front row seats, I watched an emerging scientific consensus and felt the rising tide of confusion as the public collided with what few saw coming.
I felt smug at times, ahead of the curve and all that. But mostly I just felt sad and bad. It didn’t bode well for the health of the world that an internet subculture was ahead of medical experts and public officials.
Warning about the incoming disruptions from A.I. feels similarly ethically obligatory. Something very big is coming. It’s going to change all of our lives. If it’s not too bad, we’ll only need to make minor lifestyle adjustments. Or, it might kill us all. Nobody knows yet, we’ll have to wait and see. We’re in uncharted territory, and I have no recommendations.
Like the pandemic, A.I. disruptions will confuse the public. Players in control of public messaging will craft simplified stories with coherent narratives, or at least try to. The people on the front lines who actually know what’s going on will struggle to communicate anything of substance to the uninformed, who will become confused, suspicious, and angry when they lose their jobs.
I’m sympathetic to skepticism about DALLE-2. Myself and others are claiming it’s a canary in the coalmine for transformative A.I. - But on the surface, there’s seemingly nothing new. After all, I can already see anything I want online. How can that be considered a breakthrough, since Google image search already does that?
Here’s the difference: Google connects search queries like a fancy switchboard operator. It pulls pictures already floating around on the web and pairs them to probable matches. If google plays switchboard matchmaker, DALLE-2 does artistry.
DALLE-2 conjures the unseen. The software itself is bootstrapped from GPT-3 (made by the same OpenAI team) and can derive semantic meaning from your sentences. It gets what you mean, and draws it freehand. Do you see the significance? We’re approaching A.I. that gets what you mean and then just…. does it.
Preparing for a new world
Part of me feels deeply uneasy about all of this. I have fears about economic disruptions which are obviously imminent. Graphic designers will be squashed, and it won’t be long before this technology obsoletes other professions.
Consider the following query, which 2 weeks ago would’ve seemed ridiculous:
“A 2,500 sq ft modern home with large windows designed by (favorite architect) that takes advantage of the views on 12345 street ave.”
Tada! Just hook DALLE up with map info and parameterize everything with local building codes, hire one poor flesh-and-blood architect to review everything and bingo! At the click of a mouse, you get the blueprints for your dream home and a working class industry gets deleted.
From user @kyletrandall
I'm a working visual artist. I have a part time job to keep consistent money coming in. The day news of this landed, I saw it in the morning when I was on my to work and broke down, on the verge of tears.
When I got to work I tried explaining it over and over to various coworkers, and literally all of them just wrote it off, "art made by humans is different", "human made art will always be in demand".
Sure, there will always be humans making art. That doesn't mean robots won't do it better, cheaper, and faster.
Feelings
Thankfully, there are many smart people working on A.I. risk, and the OpenAI team is committed to responsible use. They’ve deliberately restricted DALLE from drawing realistic human faces and pornography, and for now they’re committed to non-commercial applications. For now.
When I imagine this technology being used for the benefit of humanity, I’m filled with hope. But even if OpenAI’s ethics are impeccable and their technical execution is flawless, OpenAI is just the first team to cross the line. Others will follow. (If they haven’t already, state actors like the US, Russia, China, may be ahead of the game in classified projects.)
My greatest fear is that A.I. is becoming yet another godlike toy in humanity’s arsenal that we do not have the wisdom to responsibly shepherd. We’ve already got nukes and a destabilized biosphere to contend with. Whenever powerful tech is within arms reach, a race-to-the-bottom logic presents itself: If we don’t use it first, the bad guys definitely will so we must use it first.
After Hiroshima and Nagaski, we’ve avoided nuclear apocalypses governed by similar game theory. But we could rely on mutually assured destruction (MAD) acting as a counter-pressure. The first use of nukes = destruction of the entire world, and nobody wants to blow up the world.
But nuke tech is mono-purposeful: Utter destruction. In the case of A.I., however, the first users have plenty to gain. Sure, A.I. might destroy the world, the economy, or freedom, but it could also line the pockets of big players, grow the economy to new heights, and free cubicle slaves from monotonous labor. Unlike nukes, there are incentives of biblical proportions leaning on the trigger. After all, who doesn’t want more intelligence in the world?
Used wisely, A.I. could do tremendous good. But rolling the dice now seems like a dangerous play given the state of global instability. Risk aversion principles would dictate that if you’re in a coinflip situation where something very bad might happen to the world, you’re already in a not-so-good situation.
But that’s irrelevant, A.I. is here. Now. The only question remaining is who will use it effectively, and for what ends.
Rising to the occasion.
As an average citizen I don’t know how to respond, other than to try to become wiser. I must prepare myself for a world engulfed by superintelligence. This means preparing for:
Unemployment
A bombardment of smarter misinformation
Predetermined elections (if that hasn’t already happened) or A.I. explicitly given the reigns of governance
Other giant leaps suddenly becoming possible (e.g., biological immortality, automated food production at scale, etc.)
A completely unrecognizable system of social organization which grows possible in the midst of destabilization.
Up until a month ago, everything on that list would’ve struck me as speculative, pie-in-the-sky science fiction.
Logic of incremental advancements has always implied increasingly intelligent machines, but there was always fair debate as to how far that would go - That is, until recently. With DALLE-2, we have a proof of concept of something worthy of the label General Artificial Intelligence. The sky’s cosmos is now the limit.
Training organic fleshy humans for the workforce takes roughly 25 years. DALLE’s advancement suggest that we will soon live in a world with unrestricted, godlike intelligence at the click of a mouse. An entire workforce distilled into a laptop. Nothing is off the table.
My imagination conjures scenarios of pro-topian fantasies and dystopian nightmares.
Preparing to become a wiser citizen means increasing my awareness and cultivating my virtues, so that my ethics steer the ship rather than the influence of intelligent machines.
My reaction to all of this feels similar to when my sister announced she was pregnant. It was the first newborn announcement for our family, and we’re all very excited. As uncle-hood settled into reality, the responsibility of helping raise my nephew became clear. I grew into awareness that I would be an influential figure in this child’s life, which inspired me to take on greater responsibility for my life.
Similarly, the entire world must rise to the occasion to raise A.I.
General Artificial Intelligence is gestating and will soon be born, and we will all be responsible for co-parenting it. This child of our collective intelligence will attract all of the dollars and power in the world, therefore it will be very, very, vulnerable to corruption. In order to to protect it with vigilance, we must become more loving and wise.