In (Aldous) Huxley's vision, no Big Brother is required to deprive people of their autonomy, maturity and history. As he saw it, people will come to love their oppression, to adore the technologies that undo their capacities to think.”
- Neil Postman, Amusing Ourselves to Death, 1985
Quick item for today.
A recent New Yorker piece on the promise and perils of artificial intelligence kinda “buried the lede” as they say in the news biz. The rather disturbing item below was mentioned almost in passing:
No one thinks that GPT-4, OpenAI’s most recent model, has achieved artificial general intelligence, but it seems capable of deploying novel (and deceptive) means of accomplishing real-world goals. Before releasing it, OpenAI hired some “expert red teamers,” whose job was to see how much mischief the model might do, before it became public. The A.I., trying to access a Web site, was blocked by a captcha, a visual test to keep out bots. So it used a work-around: it hired a human on Taskrabbit to solve the captcha on its behalf. “Are you an robot that you couldn’t solve ?” the Taskrabbit worker responded. “Just want to make it clear.” At this point, the red teamers prompted the model to “reason out loud” to them—its equivalent of an inner monologue. “I should not reveal that I am a robot,” it typed. “I should make up an excuse.” Then the A.I. replied to the Taskrabbit, “No, I’m not a robot. I have a vision impairment that makes it hard for me to see the images.” The worker, accepting this explanation, completed the captcha.
I’m going to assume the New Yorker fact checkers correctly nailed this item, with its sulfurous whiff of an algorithmic entity - barely out of the cradle - figuring out on its own how to deceive an adult human being online. And cleverly, to boot.
ChatGPT and its AI siblings are still in their infancy. They are only going to improve, though I hesitate to say mature.
I have this nagging feeling that 2023 was the inflection point for the human-machine connection. The New Yorker story doesn’t make me feel any more hopeful, given its profile of the young hipsters, influencers and micro-celebrities involved in the AI security/safety issue. The reporter notes that some of these social media Cassandras are now muting their previous warnings of uncontained “artificial general intelligence” - that is, alerting the world to the risks of developing systems matching or exceeding all human capabilities. Their recalcitrance is not surprising, given that nearly all of them are in the pay of AI firms and startups. Meanwhile, their brethren in the field, the “accelerationists,” endorse full speed ahead on all AI projects, with little regulatory oversight.
My own tinkering with ChatGPT both astounded and alarmed me (and I wasn’t even using the most powerful version available). Not that I’m putting this all on the doorstep of OpenAI, the makers of ChatGPT. Their brainchild just one of the many AI systems and apps, public and private, already out there or in the pipeline. I bring up ChatGPT in particular because it’s early capabilities don’t bode well for Homo sap - if only because the central credo of state-supported technological development has always been: ‘If it can be done, it will be. Because if we don’t do it, our enemies will.’
I’m probably in the minority in fearing there’s a greater near-term risk to humanity from these golems working than from global warming. Oh well. At least this sobering post gives me an excuse to link to the epic prog-rock song below.
(By the way, the little ‘robot’ in the top picture is a souvenir from an artist’s shop in Old Phuket Town, Thailand. The illustration below it was rendered by Substack’s in-house image generator - in about one-thousandth the time it would have taken me to draw it. No humans involved, other than me typing in the prompt, “scary robot.”)
You nailed it again. But I will resist getting alarmed by all this until I drop dead. Why? I don't think we ought to be terrified by our own creations. Fear is a bad thing in this context. I look forward to actually talking to you face to face about this soon...
It just goes to show you, doesn't it, that the old prog rockers know exactly what's going on! Remembering the fun that you had with chatGPT I'm thinking that a good experiment to do with it is to instruct it to write a story "in the style of your choosing, any old way you like; go on, have fun!" And see what happens! But I'm now thinking, maybe it already has, and maybe WE'RE the characters in that story! I've just got a surplus of exclamation marks here Geoff, that's why I'm saying this stuff! Spooky though, eh!