AI Can Ride My Bike with No Handlebars

I don’t remember the first time I heard Flobot’s Handlebars, but I have been hearing it a lot again recently in an attempt to connect with my past.

The song’s lyrics start with someone expressing happiness at the fact that they can ride a bicycle with no handlebars, and by the end, they threaten to end the world in a holocaust. There can be several interpretations of what any lyrics mean, but in this case, we can find a quote from the artist that clarifies the intended meaning:

The song is about the idea that we have so much incredible potential as human beings to be destructive or to be creative. And it’s tragic to me that the appetite for military innovation is endless, but when it comes to taking on a project like ending world hunger, it’s seen as outlandish. It’s not treated with the same seriousness… We have these little moments of creativity, these bursts of innovation, and every time that happens, that innovation is used to oppress and destroy people.

MC Jonny 5 / Jamie Laurie (main singer and founder of Flobots) to MTV News

The song speaks about how scale almost always corrupts human innovation; that the same core innovation spirit in us that lets us find meaning and joy in simple acts like riding a bicycle with no handlebars can bring us meaning (and in some cases, joy) of guiding a missile with no satellite; that the human ego at scale is a dangerous thing.

Which brings us to the world of AI. Scale is everything in the world of artificial intelligence today, and it has been a big data center contest lately. The companies behind the frontier models have also scaled to extreme massive levels, with valuations hovering around hundreds of billions of dollars.

Perhaps the song is quite fitting for this narrative as well.

AGI, Consciousness and the AI Arms Race

AI CEOs love talking about the end of the world for some reason. Another unqualified thing they often talk about is consciousness, with many frontier labs hinting that consciousness is just another milestone on the AI development roadmap, i.e., not only could their models represent and mimic people but actually be people. Almost all of them do not ponder much the ethical and moral implications of a truly conscious being in this context.

I think AI will probably, like, most likely sort of lead to the end of the world, but in the meantime, there will be great companies created with serious machine learning.

Sam Altman, OpenAI CEO

The idea that intelligent machines will destroy the world is not new. Many sci-fi works explore the situation; most involve a hyper-intelligent system that is either fully conscious as a being or can act as if it is. Most of them will not explain how it works, in the sense of how these computers or software turned into this thing.

The thing just shows up buck naked in a bar with a German accent and demands your clothes. It doesn’t tell you, “Oh yes, I understand what you’re saying and why you shall be curious. Yes, please sit down while I give you a presentation going through everything, all the scientific progress we made, how the machines actually work, what is consciousness and how we manipulate it, and et cetera.” That would make for a very boring movie.

Once we build a generally intelligent system, that basically we will ask it to figure out a way to make an investment return for you.

Sam Altman, OpenAI CEO

Consciousness is a subject that always seems to be at the center of the AI discussion but is never discussed seriously. For example, a tech CEO might boldly proclaim that their company will reach AGI (in fact, a specific form of conscious AGI) in the next three years. However, instead of following through on the legal, ethical, and philosophical framework of the company owning conscious beings, they would move on to how their ad business did 5% more year-over-year.

But is consciousness required or a part of the definition of AGI? The consensus seems to say no.

In recent times, we’ve also seen some pullback in the rhetoric, with leaders trying to define AGI as a capacity threshold rather than a claim of consciousness. For example, last year Sam Altman said: “I think it’s not a super useful term” when talking about AGI. It would be fair to assume that, instead of creating a living, breathing digital dog (which presumably is not just a matter of compute), the focus is now on creating a system that can replace human labor effectively, at least.

While a non-conscious AI system that can perform human tasks at a superlative capacity will mean you don’t need to worry about ethical and moral issues of deploying the said AI, it would still be a dangerous thing for society at large, just not in a Skynet type of way.

In his essay “The Adolescence of Technology”, Anthropic CEO Dario Amodei proclaims seriously, “humanity is about to be handed almost unimaginable power, and it is deeply unclear whether we possess the maturity to wield it.” Most of the essay seems to be an argument for more regulations on AI; you can read it at your own risk and make up your mind on it.

Either way, skynet or no skynet, we can assume that if the AI arms race goes well, then in the worst case (from the point of view of AI companies) is that LLMs do not gain consciousness with more compute, but with better harnesses and integration, AI can become genuinely useful in every field to the point where the “controllers” of AI, i.e. the companies that control them, will be given an almost unimaginable power and it is deeply unclear if they have the maturity to wield it.

At that point, the worries expressed by Mr. Amodei in his essay become realistic concerns for everyone. However, if his answer can be found in the blog post that starts with “There is a scene in the movie version of Carl Sagan’s book Contact where the main character, an astronomer…”, I deem it not serious enough. By framing his essay with a romanticized sci-fi reference to the movie Contact, Amodei treats an immediate geopolitical and economic crisis as a philosophical thought experiment, skipping over the difficult questions of accountability. (I know that sounds rich coming from the person whose entire hook of the article is an late 2000s rap song, but still.)

Even on a rudimentary reading, he skips most of the difficult questions and focuses on things like billionaire philanthropy and preventing hardware access to oh-so-evil China. It is a great answer if you’re a billionaire and control the said AI, but it would be possible to argue that it is not the best answer, or the most ethical answer, or the fairest answer for all.

This brings me back to the song, and these particular lines:

I can hand out a million vaccinations / Or let ‘em all die in exasperation / Have ‘em all healed from their lacerations / Or have ‘em all killed by assassination / I can make anybody go to prison / Just because I don’t like ‘em / And I can do anything with no permission

A world where that stands true for private US companies like Google, OpenAI, Amazon, Anthropic, or any private company does not sound like a good world. And it might be the world we are heading towards, AGI or no AGI.

The Human Ego Remains the Problem

It is a bit funny how tech CEOs will often claim that artificial intelligence, “the AI”, will take away x percent of jobs in y years. But they never say that their company will do it; they never say they’re investing billions hoping this happens, but rather “the AI” will take away the jobs and replace human beings, and we must be forewarned about such an evil winter that is to fall upon us.

Yet, from what we can see, there is no difference between OpenAI and the models it runs behind its closed doors, and the same for every private frontier AI lab. Yet there seems to be a sense that it’s acceptable to attribute full agency to the AI when it’s convenient while you own the shares and all financial gains.

You blame the same demon you command and profit from.

The moral problem in AI in this case isn’t really whether AI can be conscious or how to live in a world where that is true. But rather, should we trust in AI labs to be worthy of wielding such power? History tells us that no private entity is worthy of that power.

Note: You can find the public version history of my articles on my github: link.