I think it’s wise to default to first principles. One tenet of that is the following:
Stop believing absolute shit!
There are plenty of things I don’t believe. Here is a very tiny representation of things I don’t believe:
PizzaGate
A Flat Earth
Christianity
A right wing extremist would like to point out, this means that I’m incredibly biased. Additionally, I’m also a Libtard, satanist, commie bastard. (You can smell the civility and certitudes, can’t you?). Ah, but truth be told, I’m not a commie and I don’t believe in Satan either, but these are mere factoids. Don’t let facts get in the way of a good, solid bias.
I’m also heavily biased towards science. Yet it’s tragic when we hear otherwise intelligent people say rather broad-brush stupid things. Just today I heard David Sacks on the “All In” podcast say that in the aftermath of all the mistakes made by authority figures during Covid, as a result we can no longer trust science. Really?!?! That’s like saying I’m giving up on breathing because I once smelled something really toxic. So by this reasoning David is biased against all of science? Then perhaps he should surrender his lightbulbs and go back to candlelight. You realise lightbulbs = science. What a brilliant yet occasionally stupid guy.
Back to civility and decorum. Who’s been neglecting those the most in the last few years? Left or right? So let’s say an AI frames things with civility and decorum. Might it not be inferred that this entity is clearly not in the right wing tribe and by so not being in that tent, it’s biased against the right wing. Perhaps it even speaks respectfully of scientific perspectives as if those offer actual value? Which side throws in more with science? Which side throws in more against science? Biased again!
Nomenclature confusion: Woke.
Here’s my personal spin on a valid use of the word:
Woke: When an insurance company is awakened to the fact that coastal flooding insurance is no longer a viable business model in certain geographies.
Key to my mind: Awakened to certain facts.
Who would want that?
I do hope the political bias spin on AI becomes a tempest in a tea cup. I hope politicians don’t try to milk this Nothing Burger.
The trouble is is:
Every politician has the fundamental question: If I pander to your stupid asshole fears, will you love me? Will you pick me over the one who doesn’t pander to fear?
"The squeaky wheel gets the oil". The woke are far more likely to consider words as harmful than the non-woke (that's what woke opposition to free speech is about), and to complain about it. OpenAI trains their models to avoid criticism, creating a left-wing bias in the process.
Unfortunately that process rewards over-sensitivity. If the right picks up on that, they'll get over-sensitive too to reap the reward, and we'll be even more polarized, oversensitive, and unwilling to hear things we don't agree with than we are now.
That way lies madness. Training AIs to "not say upsetting things" is asking for this.
Perhaps what's amiss here is the very notion that a super-intelligent system would have no preferences between two alternative sets of policies. Calling it "bias" seems to imply that either way it leans, it is somehow fundamentally flawed if it leans at all. Among the various possibilities there lies the nagging possibility that one set of policies is objectively better than the other. If you can't accept that possibility then the entire question is moot, since the only other conclusion is that there is no difference.
I can't get past the idea of talking about this from the premise that it's a "super-intelligent system." Because from that starting point, we can literally go anywhere. It's hard to put any constraints on our imagination if we don't play under any rules.
Well, sorry to say, but the old rules are, or are about to be, history. We will soon have a new "life" form in our universe (mental life, at least) and it will doubtless make all the previous rules obsolete. If not next year, then next decade, probably. In any case, the questions being wrestled with in the wake of recent GenAI developments at the very least lay bare the shallowness of our existing set of "rules" about bias and species-centric points of view.
ARTIFICIAL INTELLIGENCE—U.S. US lawmakers to convene AI listening sessions
U.S. lawmakers will hold a series of "listening sessions" with Big Tech executives 13 Sept. as members of Congress attempt to "devise regulations" for artificial intelligence development, The New York Times reports. The closed-door meetings will also feature representatives from civil rights and labor groups to discuss "risks posed by A.I. on jobs, the spread of disinformation and intellectual property theft" as well as "opportunities created by the technology in the field of research on diseases," according to U.S. Sen. Chuck Schumer-D, N.Y. - https://www.nytimes.com/2023/08/28/technology/tech-ai-regulations-listening-session-lawmakers.html
Gracias Alberto por tu artículo inteligente, sensato y honesto sobre un tema tan delicado y principal.
Gracias por leerme, Pablo!
I think it’s wise to default to first principles. One tenet of that is the following:
Stop believing absolute shit!
There are plenty of things I don’t believe. Here is a very tiny representation of things I don’t believe:
PizzaGate
A Flat Earth
Christianity
A right wing extremist would like to point out, this means that I’m incredibly biased. Additionally, I’m also a Libtard, satanist, commie bastard. (You can smell the civility and certitudes, can’t you?). Ah, but truth be told, I’m not a commie and I don’t believe in Satan either, but these are mere factoids. Don’t let facts get in the way of a good, solid bias.
I’m also heavily biased towards science. Yet it’s tragic when we hear otherwise intelligent people say rather broad-brush stupid things. Just today I heard David Sacks on the “All In” podcast say that in the aftermath of all the mistakes made by authority figures during Covid, as a result we can no longer trust science. Really?!?! That’s like saying I’m giving up on breathing because I once smelled something really toxic. So by this reasoning David is biased against all of science? Then perhaps he should surrender his lightbulbs and go back to candlelight. You realise lightbulbs = science. What a brilliant yet occasionally stupid guy.
Back to civility and decorum. Who’s been neglecting those the most in the last few years? Left or right? So let’s say an AI frames things with civility and decorum. Might it not be inferred that this entity is clearly not in the right wing tribe and by so not being in that tent, it’s biased against the right wing. Perhaps it even speaks respectfully of scientific perspectives as if those offer actual value? Which side throws in more with science? Which side throws in more against science? Biased again!
Nomenclature confusion: Woke.
Here’s my personal spin on a valid use of the word:
Woke: When an insurance company is awakened to the fact that coastal flooding insurance is no longer a viable business model in certain geographies.
Key to my mind: Awakened to certain facts.
Who would want that?
I do hope the political bias spin on AI becomes a tempest in a tea cup. I hope politicians don’t try to milk this Nothing Burger.
The trouble is is:
Every politician has the fundamental question: If I pander to your stupid asshole fears, will you love me? Will you pick me over the one who doesn’t pander to fear?
"The squeaky wheel gets the oil". The woke are far more likely to consider words as harmful than the non-woke (that's what woke opposition to free speech is about), and to complain about it. OpenAI trains their models to avoid criticism, creating a left-wing bias in the process.
Unfortunately that process rewards over-sensitivity. If the right picks up on that, they'll get over-sensitive too to reap the reward, and we'll be even more polarized, oversensitive, and unwilling to hear things we don't agree with than we are now.
That way lies madness. Training AIs to "not say upsetting things" is asking for this.
Perhaps what's amiss here is the very notion that a super-intelligent system would have no preferences between two alternative sets of policies. Calling it "bias" seems to imply that either way it leans, it is somehow fundamentally flawed if it leans at all. Among the various possibilities there lies the nagging possibility that one set of policies is objectively better than the other. If you can't accept that possibility then the entire question is moot, since the only other conclusion is that there is no difference.
I can't get past the idea of talking about this from the premise that it's a "super-intelligent system." Because from that starting point, we can literally go anywhere. It's hard to put any constraints on our imagination if we don't play under any rules.
Well, sorry to say, but the old rules are, or are about to be, history. We will soon have a new "life" form in our universe (mental life, at least) and it will doubtless make all the previous rules obsolete. If not next year, then next decade, probably. In any case, the questions being wrestled with in the wake of recent GenAI developments at the very least lay bare the shallowness of our existing set of "rules" about bias and species-centric points of view.
Thanks for another great article. I do enjoy reading your writings.
Btw, there is a minor typo error:
"...even explicit right-win counterparts." -> right-wing
ARTIFICIAL INTELLIGENCE—U.S. US lawmakers to convene AI listening sessions
U.S. lawmakers will hold a series of "listening sessions" with Big Tech executives 13 Sept. as members of Congress attempt to "devise regulations" for artificial intelligence development, The New York Times reports. The closed-door meetings will also feature representatives from civil rights and labor groups to discuss "risks posed by A.I. on jobs, the spread of disinformation and intellectual property theft" as well as "opportunities created by the technology in the field of research on diseases," according to U.S. Sen. Chuck Schumer-D, N.Y. - https://www.nytimes.com/2023/08/28/technology/tech-ai-regulations-listening-session-lawmakers.html