most crimes committed between now and 2000 are going to become solved
I don’t believe this for sexual assault, domestic violence, or many forms of child abuse, elder abuse, and the like; because these crimes are quite often committed in private.
Sure, if you mean bank robberies, jewel heists, and other movie-plot crimes. But the clearance rate for those is already quite high.
Put another way: by having gravitational advantage, it’s able to exert power over people on earth while being highly resistant to people on earth exerting power over it.
It is a castle at the top of the largest hill, which means it can rain down whatever it likes on the people below while not worrying that they will climb up to put a stop to it.
You essentially never, as an agent, get both total knowledge over the exact scenario some number of people are in, except for some specific, bounded uncertainty over some parameters, with no way of communicating with anyone inside the settings, or any kind of longer-running tradition or doctrine or set of expectations from other agents as to what you “should” do.
Put another way: These sorts of thought-experiments ask me to imagine that I am in an epistemic (and practical) position that I cannot ever really occupy.
“Suppose that you have a button that, if you press it, kills a random person you’ve never met and donates a billion dollars to effective charities that can save a million lives with it. Do you press it?” How did I come to believe that the button works this way? How does it work? Why can’t I take it apart and remove the random-person-killer component while leaving the money-giver component intact? (Is it somehow sacrificing the one person to gain the billion dollars? Sacrificing to what? How does that work?) How the hell does it know which people I’ve met, to exclude them from being killed? (Is Palantir involved? I bet it’s Palantir. And maybe Flock.) Did the person who set up the experiment cheat and exclude themselves from being randomly killed? Which charities? How do I know they’re effective? When they find out where the money came from, won’t they refuse to accept it? Why should I have believed the answers to any of these questions?
Or, put another another way: Sufficiently-advanced trolley problems are indistinguishable from “Would you still love me if I was a worm?”
Or, put yet another way: My moral decision-making is entangled with the rest of my understanding of the world, in much the same way that fire is entangled with the rest of chemistry, including life. The thought-experiment asks me to imagine that I can breathe in a world where fire can’t burn.
Mythos-style models that create security vulnerabilities
Nitpick: Security vulnerabilities are typically created by mistake; by coders (human or AI) writing buggy code. Mythos is noted for discovering vulnerabilities, not creating them.
And that rhymes with professional deformation.
(Of course this conflict among housemates is a systems engineering problem; everything is a systems engineering problem. Now let’s consider the flow of Amazon boxes through the house, because there’s a bottleneck somewhere causing a space leak...)
Trojan Basilisk?
How does this model account for “bad” uses of prediction markets, e.g. cases where a market user artificially corrupts the outcome, illegally uses their access to national secrets to extract from the market, or otherwise uses the market extractively rather than predictively?
Bribery can happen when it’s already someone’s duty to do something, but they refuse to do it until they’re paid extra. The parent may think the child has a duty to accept the vaccination; either arising from the child’s self-interest, from the public interest (or categorical imperative), or from a duty to obey authority. By accepting the vaccination only when paid off with a snack, the child is acting not from duty but from desire for the snack. Thus, the child does not learn the habit of acting from duty.
There are other sorts of “bad faith” arguments besides misrepresenting one’s own beliefs.
For instance, an arguer may attempt to provoke you to outrage to make you look foolish in front of an audience, without ever misrepresenting their own beliefs.
Or an arguer may record the argument, and use phrasing intended to get you to speak specific words, then later edit the recording to make you sound wrong, evil, or stupid. Their goal in arguing was something other than to arrive at the truth: it was to produce a recording that could be manipulated to tell a lie. Therefore, the argument itself was in bad faith because of their intentions, even if they never in the argument misrepresented their own beliefs.
Aside from values drift, there’s also acquiring new information. I know more about my immediate needs than I know about what my needs will be next year. Even if my values don’t change, I’m not a perfect predictor.
Many popular languages today (notably the C family) ultimately descend from ALGOL, which is from 1958.
“Structured programming”, i.e. writing code as syntactically-delimited blocks, functions, and procedures rather than with numbered lines and GOTOs, was pioneered in ALGOL.
Popular languages today such as Python, Java, JavaScript, Go, and Rust may diverge pretty widely in features (and syntax), but all of these are ultimately ALGOL descendants; albeit with influences from other language families too.
(If your language has for loops, it’s an ALGOL descendant.)
Lisp and Fortran are also pre-1960.
Simula (and thus object-orientation) is from ’62, but influenced by ALGOL. Smalltalk is a Simula descendant. C++ is what you get if you try to build Simula ideas on top of a C compiler (and go a bit gaga for operator overloading).
There are some languages a little later than that, that look pretty different. For instance, APL is from ’68. Forth is from 1970. ML, which gave rise to Haskell, is from ’73.
True; although symbolic ambiguity (requiring a caption to explain) seems to be acceptable.
I thought at first that the current “bowels of Christ” tag meant “gut feeling”; that the “bowing out” top-hat meant “I tip my hat”; and I mistook Moloch as being Baphomet (!) and likely signifying “this is esoteric”.
LEDs strobe when they’re run at less than 100% duty cycle or on a half-wave rectifier. They should shine steadily when run at 100% on DC power.
Though I find the LED shimmer to be rather less unpleasant than fluorescent lamps (classic tubes or CF bulbs), which flicker in time with the AC power and have rather incomplete spectra that make pale-skinned people look greenish and often make an audible hum.
Elephants are also the subject of silly jokes that hinge on their bigness, and their possible presence or absence, e.g.:
Alice: “Why do elephants wear red nail polish?”
Bob: “I don’t know. Why?”
A: “To hide in strawberry patches. Did you ever see an elephant in a strawberry patch?”
B: “No...”
A: “That’s how you know it works!”
The humor of elephant jokes largely centers around the “big if true” nature of an unlikely claim about elephants.
“How many elephants can you fit in a Volkswagen? Five: two in the front seat, two in the back, and one in the glove compartment.”
An elephant. Elephants are famously big, and if your neighbor says they have one in their living room, that’s probably not true.
Perhaps related: I suspect there’s also a skill transfer from music to computer programming. Specifically, I’ve observed a few musicians to have a much easier time learning programming than I expected for an adult with no prior programming experience.
This might have something to do with —
Comfort with patterns, iteration, and structure. If this is the thing, I would expect knitters and fiber artists to have some of the same advantage; I don’t know if they do.
Comfort with deliberate practice itself. Many new programmers get frustrated with the amount of practice and heedfulness it actually requires to get good. Trained musicians are accustomed to acquiring skills and methods through deliberate practice.
A nexus between music, programming, and math. (One of the musicians I’m thinking of also was a math major.)
How about businessmen? Some sources claim that Henry Ford repented his antisemitism in 1942 and took legal action to end the publication of his own most well-known antisemitic work, The International Jew. A few years later Ford died of a stroke (not his first) after being shown newsreel footage of the Nazi concentration camps that his (former?) ideology had enabled.
European colonialism often involved clearing a place of its previous inhabitants and resettling it
That’s how the English often proceeded in North America (and Australia, for that matter), but it’s not nearly so accurate a description of how the Spanish did things. Conquistadors aimed more to make themselves lords over an enslaved/enserfed native population, not to wipe them out and replace them with Spanish immigrants.
And there was a lively debate within Catholic Spain about whether it was theologically and philosophically correct to conquer native populations. (Do Christians have an obligation to forcibly stop pagans from committing atrocities like human sacrifice upon one another? Are the pagans “natural slaves” incapable of developing morality on their own? Or must we respect that they have rational souls and can be converted and assimilated through cultural manipulation?)
Okay, good. That’s what I thought, I just wanted to make sure I wasn’t making a not-knowing-what-the-conversation-was-really-about error. (“Never give anyone wise advice unless you know exactly what you’re both talking about. Got it.”)
The word “agent” sure means a lot of things.
In business and economics, “agent” is relative to “principal”. An agent acts on behalf of a principal; is intended to act for the principal’s interest; but can experience conflicts of interest (a “principal-agent problem”). An agent can be (at least in theory) held accountable for misrepresenting its principal. An agent is expected to actively pursue its principal’s interest (see e.g. Matthew 25:14-30).
In traditional AI, “agent” is relative to “environment”. Russell & Norvig define an agent as “anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators.” An agent receives percepts and takes actions. A “rational agent”, in turn, is an agent “that acts so as to achieve the best outcome or, when there is uncertainty, the best expected outcome.” (This in turn implies utility theory, probabilistic reasoning, and a bunch of other infrastructure.)
In philosophy, an “agent” is any entity that takes actions. In ethics specifically, a “moral agent” is one who takes moral actions; actions that can be judged right or wrong. “Agent” here is contrasted with “patient”; the doer of an action vs. the sufferer of its effect. (A “moral patient” is an entity that has moral interests; one that can suffer a wrong or benefit from a good deed. Not all moral patients are moral agents.)
In psychology, “agency” is the trait of choosing one’s own actions; of making decisions for oneself. We experience agency when we make our own choices for our own reasons and purposes; we recognize others’ agency when we see them make their own choices for their own reasons and purposes. (We may be in error about agency: we may mistake passive parts of our environment as being agents; we may underestimate our own freedom to act.)
However, Milgram used “agentic” to mean nearly the opposite: a person enters the agentic state when they “come to view themselves as the instrument for carrying out another person’s wishes, and they therefore no longer see themselves as responsible for their actions.” An obedient agent (in this sense) experiences less agency (in the sense of the previous paragraph); indeed, Milgram’s agentic state is an escape from that kind of agency.
A very different “agency” is found in the 2021 CFAR handbook, which defines the word to mean “The property of both having and exercising a capacity for relevant action; one’s ability to meaningfully affect the world around oneself and effectively move toward achieving one’s goals. In particular, agency implies the ability to move beyond default patterns and cached answers, and to think and act strategically.”
“Agent AI” has previously been used in contrast to “tool AI”, for instance by Karnofsky (2012). Karnofsky discussed making AI safe by limiting it to “tool AI”, with Google Maps as an example of a tool: it tells you the optimal route to your destination, unlike a Waymo, it doesn’t take actions in the world to take you there. This distinction seems somewhat quaint today, because …
… today, “AI agent” is often used to mean something like “a process that repeatedly uses an LLM for planning and composition of instructions, which it then carries out.” Such an “agent” is an odd fit for many of the above senses of “agent”:
It need not act on behalf of a principal, so it’s not a business agent. (It also can’t be held accountable.)
It is not a rational agent; it does not maximize an expected utility. (And a sensors-and-actuators agent is an odd fit for the shape of LLM-based agents too; but R&N’s “can be viewed” is quite broad, so I’ll allow it.)
It does take actions, so can be said to have some sorts of philosophical agency; its moral agency is pretty doubtful though.
Opinions on its psychological agency vary widely … including in opinions generated by LLMs trained to play “agent” characters.
Similarly with regards to its degree of Milgram-agentic obedience to human authority.
Most do not appear to be CFAR-agenty in the sense of having their own goals and actively pursuing them.
And the tool/agent distinction seems obsolete when you can build an “agent” out of a thin wrapper around a query/response “tool”.
Or, to put it another way:
In the agent era, the agents may not be agents, but they’re agents, and they’re getting even agentier.