Human → Superhuman → Ultrahuman

We will always have the kisses

The Kiss by Gustav Klimt, 1907–1908

I. Remember when computers used to lose?

The superhuman phase didn’t last much.

The first sign computers were no longer just “human level” at chess was Garry Kasparov’s famous loss against Deep Blue in 1997. The shock prompted him to invent Advanced Chess, where humans and computers teamed up as equals, the origin of the superhuman phase.¹ But it didn’t officially begin until 2005—the last time a grandmaster dared challenge an AI and was obliterated. That’s what superhuman means: above human.

Not one of us has won a match against an AI chess program ever since.

Eventually, Stockfish and AlphaZero became too good for humans to contribute to tactics or strategy. Wins were AI’s wins. Losses were often downstream of human mistakes. Praised grandmasters—even Magnus Carlsen—became a hindrance for computers.² Even players who, despite not being the best at chess, excelled at partnering with AI and understanding its perks and flaws—the original cyborgs—could no longer do anything.³

By 2020, “draw-death” was the likeliest outcome. Computers drew against human-computer cyborgs (even when books and internet access are allowed) because the human contributions were zero.

The superhuman phase didn’t last much—it began in 1997 and ended in 2020.⁴

In the span of little more than 20 years, Computers went from “brute-force methods may solve chess” to “[it’s] impossible for humans to compete,” to “impossible for humans to help.”

To help.

Twenty years isn’t enough time to absorb such a humiliation. In the superhuman phase, humans were sense-makers albeit not the main architects. We weren’t the best players but still understood the best plays.

After that, we barely stood in awe at the unfathomable unfolding of forces beyond our grasp. We became living handicaps, paralyzed by the tangible fear of our impotence. And like superstitious heathens dreaming up of angry gods in the deafening eruption of a volcano or a bright thunderstorm, we couldn’t help but bow down.

But wait, if the superhuman phase has ended, in which phase are we now?

II. An obituary to our stubborn dignity

Gwern wrote an obituary to Advanced Chess, describing the categories of skill that an AI technology goes through as it gets better: subhuman, human, superhuman, and a fourth category where chess belongs:

As automation and AI advance in any field, it will first find a task impossible, then gradually become capable of doing it at all, then eventually capable of better than many or most humans who try to do something, and then better than the best human. But improvement does not stop there, as ‘better than the best human’ may still be worse than ‘the best human using the best tool’; so this implies a further level of skill, where no human is able to improve the AI’s results at all rather than get in the way or harm it. We might call these different phases ‘subhuman’, ‘human’, ‘superhuman’, and ‘ultrahuman’. . . .

Advanced chess has been employed . . . as an exemplar of what increasing technological development may imply: not technological unemployment, but increasing partnership. . . . However, if advanced chess is going to be used this way, we should remember that after the superhuman phase, comes the ultrahuman phase. . . . Advanced chess players generally admit that at some point humans will cease to be net contributors; when was that?

It happened. AI is ultrahuman at chess.

Better than the best humans and better than the best humans using the best tools.

Don’t freak out. Machines have been ultrahuman in many activities for centuries: moving fast, lifting heavy weights, or making arithmetic operations. The pattern is clear. Technology slowly crawls up the chain: subhuman → human → superhuman → ultrahuman. Kasparov’s loss felt weird because chess borders what we consider to be an exclusive human ability.

It gets weirder the more we enter human territory—cognition and creativity. You don’t invite an F1 car to the 100m Olympic race. It feels intuitively fair. We grudgingly accepted our inferiority. But once we shrunk into our minds in shame, to hide from powers way beyond ours in the physical realm, we never expected to be besieged here as well. We have nowhere to go.

Kasparov fought to “defend our dignity.” Former Go world champion Lee Sedol did as much against AlphaGo in 2016—he later apologized “for being so powerless.”

Both lost. Humanity lost with them.

As technology conquers terrain, the end picture unfolds. Ultrahumanity.

III. Not all AI systems are near ultrahuman

Let’s not go that far. Let’s leave “ultrahumanity” for another essay.

Chess is a board game—simple compared to the boundlessness of the real world. It has well-defined rules, a constrained action space, and perfect information. But AI has since conquered more complex games like Poker and Diplomacy or video games like Dota 2 and Starcraft. As long as the final goal can be formalized in computer-intelligible terms, AI wins. But that’s just games or simple tasks, right?

It’s not. The interstitial phase during which humans and AIs collaborate as equals can’t be eternal. AI, computing, and robotics keep challenging the corners of our supremacy, bringing down the pillars of our identity as top-of-the-chain.

But not all pillars are equally brittle, so a finer analysis is in order. Moravec’s paradox suggests those that seem more robust—because they take more effort on our part—are the first to fall. Chess feels hard but it’s easy. Walking feels easy but it’s hard.

So let me put Gwern’s four categories in concrete terms to avoid falling for such paradoxes:

• Robots like Optimus and Figure 01 are painfully subhuman in dexterity and movement (although improving). Agency and common sense reasoning are subhuman as well (although AI is rapidly entering the human phase in math).

• AIs like ChatGPT, Copilot, and Midjourney are at the human level in writing, coding, and painting—around average but nowhere near the best.

• Predictive AI is superhuman at recognizing and detecting stuff (we’re undeniably worse but may help against adversarial attacks and out-of-distribution cases).

• Finally, in the bottomless pit of the ultrahuman, we have games, like chess, Go, and poker.

AI’s prowess is inversely proportional to ours (and, in a way, jagged). But remember that technology slowly crawls up the chain (or slips down the drain, if you feel despair instead of excitement toward AI). In time, it won’t matter what was harder or easier.

I don’t know how, I don’t know when, but I can see those other things join the ultrahuman den.

IV. Finding humanity’s sacred places

Is there anything that won’t? Is there a sacred place for humanity?

Yes! Of course, there is! There are many.⁵ But to see them you need to stop thinking in terms of optimization, improvement, and being the best. “Subhuman”, “superhuman”, and “ultrahuman”—they only make sense in terms that capture that one dimension.

What’s ultrahuman love? Ultrahuman sympathy. Ultrahuman humanity…

Let me share a quote by philosopher Shannon Vallor from her essay, The Danger Of Superhuman AI Is Not What You Think:

. . . doesn’t granting the label “superhuman” to machines that lack the most vital dimensions of humanity end up obscuring from our view the very things about being human that we care about?

Yes, it does.

Ultrahuman technology is the best thing that can happen to us as long as we never forget that it can’t, by definition, enter these other dimensions, those we care about. Sometimes—most times—what we want is an equally imperfect human. I find it worth repeating here something I wrote long ago:

Humans like humans.

I don’t want to kiss a robot, however perfect the robot—or the kiss—may be.

We will always have the kisses.

Perhaps it’s time to stop thinking of ourselves as kings losing our kingdoms and begin to accept, embrace, and relish the imperfections that made us along the way.

1. Kasparov conceived Advanced Chess’ unique format for the superhuman phase specifically, to improve our understanding of the game, help us play better games, and measure the strength of human-AI teams. All of that happened until teaming up was no longer in the best interest of the AI.

2. Even if the AI could still be a victim of adversarial attacks, it was just too strong.

3. Interestingly, this group of amateur chess players compensated for their lack of chess skills—which the computer provided—with a substantial surplus of “AI skills”. This point is important so I’ll come back to it later.

4. Gwern made a prediction range of 2011-2020 and provided a vast amount of evidence I won’t be sharing here, concluding that by the end of 2015, when he published his write-up, the best AI + human teams were already mostly drawing against the best AI engines. He confirmed this prediction with AlphaZero in 2018.

5. Even Mark Zuckerberg says that a person isn’t just a mind. As a person, you’re active and have energy, he says. Interestingly, his remarks clash with those of many people in the tech industry who believe you can upload a mind in the form of bytes to the cloud and obtain a person. “That’s ridiculous.”