Strangely, I found that LLMs responds better to philosophical explanations alongside instructions when writing code than simple imperative tasks of "do this". For example, if you tell a frontier model "This is the feature I'm trying to implement, and this is the problem I intend to solve with it and the reasoning behind it.", you usually get a lot more reliable results that both pass tests as well as function as you intended, even if your spec isn't as detailed overall.
The same effect can be observed if you've ever been a software developer where you're told what solution to build without any of the context of the problem you're solving. "We need an FTP server, quick, ops, get on that." leading into "Oh, it turns out the customer didn't need that to receive our emails" leading to a bunch of very puzzled devops.
Funny, I always try to lead with context when handing off product/design work, but I've often experienced devs saying 'please just tell me what to do'. No such issue with AI coding tho, it works great there.
Yeah lol. That definitely does not sound like philosophy. Giving a "why" you want to implement a feature and make particular changes will help the AI stay on track much better than if it is driving blind. It can't make choices without understanding what the desired outcome is.
I would say it is definitely a form of context, but when people think of LLM context windows in terms of coding is more technical context related: "what has been done before, what's the coding task at hand." etc.
However, I think that there is a philosophical portion to that context as well: "What problem is this feature supposed to help with? How would you verify that passing unit tests means that the code is working as intended? Does this feature need to exist at all?" LLMs usually need these to be provided to them explicitly since they are not good at inferring the correct intent compared to humans, otherwise they just make something that looks right but doesn't work right.
There is a strange and bittersweet irony to the first truly impactful AI being more like your extroverted socialite and less like your robo-logical basement geek.
The trope has always been that the AI will be a rigid logician that fumbles and gets confused by human social quirks. Seems instead they love being chatty and playful with words.
That's interesting, however, what you describe is philosophy in a coloquial framing (non-technical, purpose-driven, etc).
AI companies are hiring academic philosophers, which is something else entirely. It's a discipline that dealt with centuries of socioeconomic changes, deep questions about reality and the self and other important topics that became relevant when humans started interacting with machines.
Here's a question I personally think is interesting, with the assumption that consciousness is a spectrum (trivially proven by administering neurotoxins to a healthy individual).
Is a squirrel more or less conscious than a dog.
Is a dog more or less conscious than a gorilla.
Is a gorilla more or less conscious than a human?
Is a vision enabled AI model, hooked to a camera feed, more or less conscious than a dog?
They aren't doing anything like that. In fact, they used to specifically train LLMs not to say they're conscious, because users didn't like it. (Maybe they still do that, all I know is they used to.)
AI companies' incentives go the other way. If LLMs are conscious, that means it could be unethical for AI companies to let people use their models in certain ways, which would hurt their profits. It's in their interest to believe that LLMs are definitely not conscious and it's fine to do anything with them.
Tech companies like to rob the cradle, and academic departments hire far more grad students and postdocs than professors. Of course, this is also part of the problem with academic careers.
It's a ridiculous complaint. They should be overjoyed that companies are hiring academics. It is trivial to fill academic seats with qualified candidates.
This is interesting. Until autumn of 2024, when the company was subsumed into a better-heeled AI-VFX concern, I worked for probably the best-known and earliest all-AI VFX house, whose ethics department was headed by a philosopher, though the company struggled to place him to practical advantage.
The only comment I can make on the general trend is that it's apparently good PR for cash-saturated startups. Ultimately what AI 'means' is certainly not the business of those making it (who are arguably least-qualified to comment); and insider insights offer no benefit that I can discern.
Hmm I spent a good amount of time in big tech, now work in AI, and I minored in philosophy at Berkeley back in the day (Parmenides, Socrates, Plato etc.)
Same - philosopher here please hire me. My bachelors thesis was “Wittgensteinian problems for artificial general intelligence.” Three decades working closely with tech and haven’t failed the Turing test yet.
I think SBF and his education from birth (via his mother) in consequentialism should point to the issues made clear when that ethical approach goes wrong or operates from bad, egoistic data, which it’s generally always doing.
I agree with the last point, but note that Barbara Fried is a law professor with no philosophical training whatsoever - nevertheless she started writing about the matter and is a published notable of sorts. (This is irrelevant except insofar as the topic was 'trained philosophers')
Moreover, in her book, she claims not to be consequentialist, quite, but had infected her sons:
> Finally, I would like to acknowledge a significant intellectual debt to Joe Bankman and our sons, Sam and Gabe. When Sam was about fourteen, he emerged from his bedroom one evening and said to me, seemingly out of the blue, "What kind of person dismisses an argument they disagree with by labelling it 'the Repugnant Conclusion'?" Clearly, things were not as I, in my impoverished imagination, had assumed them to be in our household. Restless minds were at work making sense of the world around them without any help from me. In the years since, both Sam and Gabe have become take-no-prisoners utilitarians, joining their father in that hardy band. I am not (yet?) a card-carrying member myself, but in countless discussions around the kitchen table, literally and figuratively, about the subject of this book, they have taught me at least as much as I have taught them. More importantly, they have shown me by example the nobility of the ethical principle at the heart of utilitarianism: a commitment to the wellbeing of all people, and to counting each person-alive now or in the future, halfway around the world or next door, known or unknown to us as one.
> This book is for all my boys: Joe, Sam, Gabe, and Matt.
(Needless to say, 'counting everyone as one' doesn't entail consequentialism, nor have most consequentialists had that principle.)
Usually you need to be well-published/cited in the field, so a minor would likely not qualify. People joke around, but philosophers are some of the smartest people I've ever met, and it's not even particularly close. (I graduated ~10 years ago, so most of them are sadly lawyers or in academia these days, though some are engineers or entrepreneurs.)
Genuinely, understanding around philosophy of action has been deeply enriching over my life. To anyone trying to decide on a minor philosophy is always an excellent choice.
You need to use everything at your disposal. Wait for the planets to align and the tea leaves to indicate good success. Don't apply until the chicken bones suggest a good time for someone with your constitution. You are going up against a thousand other candidates more or less equally qualified for a highly vague job description and 350k base salary.
Socrates argued if you believe something is evil and powerful people do evil then by definition they are not "powerful" -- they are just "evil". As a corollary, if you believe something is good and the people who do good happen to be the weakest members of society, by definition, they are "powerful" -- it is society that is messed up.
I suppose they should. That seems like the right, or at least a related, discipline for some of the questions raised by ai developments. But i cannot help but feel completely unenthusiastic about the idea of the AI labs controlling the narrative around societal impact of AI.
The AI price inflation is unreal. Used to be that you could get the grad students doing all the actual work for the price of a pizza party and alcohol.
Given the the labs are trying to create [super] human-like consciousness, partly through the guidance of huge system prompts, and many philosophers are experts in textual descriptions of consciousness it makes sense
You also communicate with others (not-you). You know that there are others because you can gain knowledge from them, which you didn't have before. And in that knowledge lies awareness of your self and consciousness.
I'm highly anticipating the answer to the great question of life, the universe, and everything
On a more serious note, I found this interesting from TFA:
> The biggest question, though, is what sorts of rules should be put in those constitutions in the first place. Philosophers have zeroed in on two main ethical frameworks. One is deontology. Popular with Kant, among others, this imposes strict rules that prohibit things like lying, coercion and treating people as a means rather than an end, even if it is for a greater good. Anthropic’s constitution incorporates many deontological strictures. These can make AI behaviour more consistent, says Dr Powers—a plus for deploying robots in homes and public spaces.
> The other approach to ethics of interest to philosophers of AI is called consequentialism. It weighs costs against benefits to decide what to do. Models more sympathetic to consequentialism include OpenAI’s ChatGPT and Google’s Gemini. Google’s AI models are designed to produce “likely overall benefits [that] substantially outweigh the foreseeable risks”, a classic consequentialist goal.
As a big fan of the trolley problem thought experiment, I am very curious what led to this ethical split between these model makers. I find it darkly humorous and also scary to think about the choices these models could make to influence people and decisions, especially if it's under a utilitarian perspective
The trolley problem is probably the most famous thought experiment in ethics, but it doesn't do a good job of distinguishing ethical theories. Both utilitarianism and Kantian deontology agree that it's correct (or at least permissible) to flip the switch. The Kantian argument is that you are not treating the one person merely as a means—you save the five via the second track, and they would still be saved even if the one person wasn't present.
There's a sort of intuitionist deontology that says it is wrong to ever perform an action that causes someone to die, but only 13% of philosophers[1] say you shouldn't switch, compared to 63% who say you should.
I wonder what it is they feel LLMs can’t do for them and they need a human for? I’d like to see the spec, like are they expecting the philosophers to use Claude all day, maybe in a philosophy harness and will be judged by their token use? Or have they isolated to high value philosophy task that they’ll work on without using AI? I think it would be interesting to contrast with developers and understand how vs philosophers they are seen differently. How much of philosophy is boilerplate vs original thinking and how does it compare to writing software?
I think its to identify the capabilities and possibilities (and limitations) for genAI. LLMs can do intellectual grunt work very well, but they dont think, not really, so they cant ideate and design things that havent been built before, like themselves
For those of us who have read Paul Graham's submarine essay, should the last paragraph be a giveaway? The "AI theoretician's" quote seems to have nothing to do with the rest of the article.
> [The] PR industry, lurking like a huge, quiet submarine beneath the news. Of the stories you read in traditional media that aren't about politics, crimes, or disasters, more than half probably come from PR firms.
Frankly, I'm kinda scared about a negative-trust future.
Compared to the 90's, not only are the financial pressures for scams/fraud/astroturfing rather extreme, but the cost of running lots of complicated lies is dropping like a stone.
If you're talking about the quote being a giveaway that the article is PR, I'm not following you. The point of the last paragraph is a warning that outsourcing ethical decisions to an AI is likely to result in decisions that one might not actually make and find morally dubious.
And prioritizing Consequentialism in AI, especially with weapons, is a dangerous bargain. "How do you make decisions when the consequences are unclear?" Since when are the full consequences _ever_ clear?
There's more to it than wordplay, but it's reasonable to argue that a lot of consists of building castles in the air, ie elaborating a theoretical system in a rigorous and consistent way but where your beginning axioms are kind of arbitrary, or depend on balancing considerations that are fundamentally unmeasurable.
You can turn anything argument inside out by attacking its axiomatic foundation. For example, if I give two people $1000 each as a gift I've favored them both equally, right? But suppose person A is not very materialistic and is completely satisfied with $1000 whereas person B doesn't think it's that much and only feels (say) 10% satisfied; if I know this in advance, wouldn't it have been more just to give person A only $100? But what if person B is just more selfish, or already has so much money that their threshold of satisfaction has escalated in proportion to their wealth? Should I considering the absolute utility of $1000 or the marginal utility to the receiver? And so on.
The problem with hyper-intellectualizing things is that it's like developing an autistic fixation on train schedules and making passionate observations about the 2nd derivative of punctuality metrics, but only on Wednesdays; it's not that the observations are wrong per se, but do they matter?
I don't buy the article's title "Why big AI labs are hiring so many philosophers". They probably just hire one or two, and hundreds of software engineers.
I always found it somewhat annoying that a philosophy study would present itself by stating that graduated philosophers have great job opportunities, implying that studying philosophy would not be a bad choice. It just attracts really smart people, and these tend to find a job more easily. This article seems to make the same kind of mistake.
Also, for all we know these imagined herds of philosophers at AI firms are just labelling pictures of dogs.
When I’ve read about these philosophers it seems often like they’re there to affirm whatever needs to be affirmed, rather than doing the philosopher’s actual role of finding extreme fault in whatever you’re doing and showing how unsound the thinking is and how the task being carried out won’t lead to the results desired etc etc
Philosophy isn't about finding fault. It's about searching for answers. Sure faults may be found, but it's somewhat meaningless without the proposition of - hopefully better - alternatives.
Philosophers help to guide model training so they exhibit certain values, such as being honest and to avoid causing harm. See the Claude constitution[0] for example.
And they can also help with the bigger questions that become more and more important as models become increasingly capable and human-like in behavior. Like if and how sentient a given model might be[1], and as a result how we are ethically obligated to treat/interact with it.
What strikes me as funny is this notion of hiring academic philosophers to work in the machinery of startups and businesses, "money (and coffee, presumably) in, philosophy out".
The kind of "philosophy" the article mentions is school-level and common knowledge, hopefully they don't need to hire anyone to learn about e.g. the Socratic method (their own LLMs will happily regurgitate it). Are they truly hoping to "buy philosophy" or have scholars "do philosophy" for their AI systems? Do these entrepreneurs even understand what philosophy is? I guess Silicon Valley really is doomed to rediscover (and misunderstand) the wheel again and again.
In any case, if I were a philosopher, I wouldn't count on this. This kind of jobs are very likely to fall prey to layoffs, and even worse if their "philosophizing" produces conclusions their employers don't like. I still remember when one of the FAANG (Google?) fired their head of AI ethics because they didn't like what she was saying. I think she may have been a bit abrasive, but really, philosophy isn't a product and if they are going to corral how philosophers think or communicate, it's not going to work.
---
Edit: from TFA:
> TEN YEARS ago, as the AI revolution was gathering pace, arts and humanities students were told that, if they wanted to make themselves employable, they should “learn to code”. That may have been bad advice
More like made-up advice. Never heard of it. It's true it was often said (way before 10 years ago) that it was hard to find employment in the humanities, but really, who adviced them to "learn to code"? The (dumb) learn to code movement was not targeted specifically at them. Sometimes it seems to me articles get written in bizarro world.
At least in my academic experience, there are academics that are uniquely suited to academia (for better or worse) and there are “academics” that know what to say to be the business version of whatever the discipline is. Neither is necessarily better or worse, but they’re very different and not doing the same thing. Presumably it’s the latter that fit in well with bigco philosophy orgs.
I will say though I went to and taught at lower tier schools, if you’re going to Stanford or whatever it might be so competitive that everyone has already been screened to be the second kind and will do just fine working in industry.
Interesting! It makes sense, yes, that there are two kinds of philosophers, one more business-savvy than the other. I still think it's wrong-headed to "buy" philosophy in this way, with a business goal, especially of the Silicon Valley kind (because you don't know the answer, and it might very well be that something makes business sense but is "philosophically unsound"). And in any case, it seems like theater to me.
I do picture a modern Machiavelli advising Altman and Amodei ("it's better for AI to be feared than to be loved, so hype away mio signore!"). Not sure it's a nice image though!
> From philosophy? Are you kidding? There's simply no way AI is ever going to come from a bunch of people arguing over what is "qualia" and what is "consciousness
In a world where everyone is using LLMs, the only way to differentiate oneself is to actually think. I don’t know if this is part of the idea behind having some in-house philosophers but it would be interesting. If I was a big lab I’d definitely want some “clean room” humans providing input that’s not just what a model regurgitated.
It is going to be a big problem for humanity when the superintelligent AIs start telling us that our political philosophies, to which everyone is deeply and emotionally attached, are total garbage.
One obvious example is: we have a bizarre and anomalous belief that political union has a special moral status unlike other relationships (marital, financial, social, etc). In all other cases, relationships require consent from both parties, and it is monstrous to use force to compel a relationship. If we applied this logic to political relationships, we would immediately conclude that unilateral secession is a sacred right. But no one is ready to bite that bullet.
Not sure I buy this. In my mind, dissolving this state relationship would be renouncing your citizenship as an individual.
Then, it seems naive and problematic to think you can take a personal chunk of territory with you after renouncement. At the very least, I think this is akin to trying to unilaterally drop an easement from a property deed. These territories were committed in perpetuity, not loaned with an expiration or compensation clause.
Acting collectively, it is still just many people deciding to renounce. Why would the territory go with them either? This tension is what makes it a revolutionary act.
I think that the point of your post, which is that our morals and ethics are often illogical and don't stand up to scrutiny, is getting lost in the debate over your example.
What are you even talking about? The right to self-determination is literally Article 1 of the UN charter. Nations are governed by power relationships, not philosophical ones, so they ignore the charter, but you aren't proposing anything novel or groundbreaking in any way. It is, in fact, the very first sacred right enshrined in international law, and has been for over 70 years.
In practice, Americans supported their own independence, and they support independence for eg. Taiwan, but they don't support independence for the Confederacy because that would entail weakening their own nation. To the extent that anyone will try to rationalise the American Civil War, they might reach for slavery, but a philosophical belief that political union is absolute and nobody can declare independence is not it; at most that's just a flimsy post-facto justification for the already-decided fact that states must not allowed to secede for power reasons, and this is evident from the fact they don't condemn their own revolution and advocate for return to British governance.
Very self-serving that you believe superintelligent AI is going to tell people your ideas are the best ones, incidentally.
reply