1<!DOCTYPE html> 2<html lang="en-gb"> 3 4 5 6 7 <head> 8 <link href="https://gmpg.org/xfn/11" rel="profile"> 9 10 <link rel="canonical" href="https://tommorris.org/posts/2026/what-has-athens-to-do-with-palo-alto/"> 11 12 <meta charset="utf-8"> 13<meta http-equiv="X-UA-Compatible" content="IE=edge"> 14<!-- Enable responsiveness on mobile devices --> 15<meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1"> 16<meta name="generator" content="Hugo 0.58.3" /> 17 18 19 20 21 22<title>What has Athens to do with Palo Alto? ⢠tommorris.org</title> 23 24 25<meta name="twitter:card" content="summary"/> 26<meta name="twitter:title" content="What has Athens to do with Palo Alto?"/> 27<meta name="twitter:description" content="Weird AI cultists are not serious people. Laughing at them is good for the soul."/> 28 29<meta property="og:title" content="What has Athens to do with Palo Alto?" /> 30<meta property="og:description" content="Weird AI cultists are not serious people. Laughing at them is good for the soul." /> 31<meta property="og:type" content="article" /> 32<meta property="og:url" content="https://tommorris.org/posts/2026/what-has-athens-to-do-with-palo-alto/" /> 33<meta property="article:published_time" content="2026-09-16T13:12:57+01:00" /> 34<meta property="article:modified_time" content="2026-09-16T13:12:57+01:00" /> 35 36 37 38 39 40<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.7.0/styles/default.min.css"> 41 42 43 44 45 46 47 48 49<link rel="stylesheet" href="/scss/hyde-hyde.0a519403c9c558a36e3feb8b46f9a46b06fb49ea43a214ec1420114bf9a43c7a.css" integrity="sha256-ClGUA8nFWKNuP+uLRvmkawb7SepDohTsFCARS/mkPHo="> 50 51 52<link rel="stylesheet" href="/scss/print.2744dcbf8a0b2e74f8a50e4b34e5f441be7cf93cc7de27029121c6a09f9e77bc.css" integrity="sha256-J0Tcv4oLLnT4pQ5LNOX0Qb58+TzH3icCkSHGoJ+ed7w=" media="print"> 53 54 55 56 <!-- HTML5 shim and Respond.js for IE8 support of HTML5 elements and media queries --> 57 <!--[if lt IE 9]> 58
58<script src="https://oss.maxcdn.com/html5shiv/3.7.3/html5shiv.min.js"></script>
58 59
59<script src="https://oss.maxcdn.com/respond/1.4.2/respond.min.js"></script>
59 60 <![endif]--> 61 <!-- Icons --> 62 <link rel="apple-touch-icon-precomposed" sizes="144x144" href="/apple-touch-icon-144-precomposed.png"> 63 <link rel="shortcut icon" href="/favicon.png"> 64 65 66 67</head> 68 69 70 <body class="theme-base-tm "> 71 72<div class="sidebar"> 73 <div class="container "> 74 <div class="sidebar-about"> 75 <span class="site__title"> 76 <a href="https://tommorris.org/">tommorris.org</a> 77 </span> 78 79 <p class="site__description"> 80 81 </p> 82 </div> 83 <div> 84 <ul class="sidebar-nav"> 85 86 87 </ul> 88</div> 89 90 <p> 91 <section class="social"> 92 93 94 95 <a href="https://github.com/tommorris" rel="me"><i class="fab fa-github fa-lg" aria-hidden="true"></i></a> 96 97 98 99 100 <a href="https://mastodon.social/@tommorris" rel="me"><i class="fab fa-mastodon fa-lg" aria-hidden="true"></i></a> 101 102 103 <a href="https://bsky.app/profile/tommorris.org" rel="me"><i class="fa-brands fa-bluesky fa-lg" aria-hidden="true"></i></a> 104 105 106 107 108 109 110 111 112 113</section> 114 115 </p> 116 <p class="copyright"> 117 © 2026 Tom Morris 118 119 </p> 120 </div> 121 <div> 122 </div> 123</div> 124 125 <div class="content container"> 126 127 128<article class="h-entry"> 129 <header> 130 <h1 class="p-name">What has Athens to do with Palo Alto?</h1> 131 132 133<div class="post__meta"> 134 135 136 <i class="fas fa-calendar-alt"></i> 137 <time class="dt-published" datetime="2026-09-16T13:12:57-0000">2026-09-16</time> 138 139 140 141 142 143 144 145 <br/> 146 <i class="fas fa-tags"></i> 147 148 <a class="badge badge-tag p-category" href="/tags/ai">ai</a> 149 150 151 <a class="badge badge-tag p-category" href="/tags/self-driving-cars">self-driving cars</a> 152 153 154 <a class="badge badge-tag p-category" href="/tags/technology">technology</a> 155 156 157 <a class="badge badge-tag p-category" href="/tags/ethics">ethics</a> 158 159 160 <a class="badge badge-tag p-category" href="/tags/smart-contracts">smart contracts</a> 161 162 163 <a class="badge badge-tag p-category" href="/tags/blockchain">blockchain</a> 164 165 166 167 168 <br/> 169 <i class="fas fa-clock"></i> 15 min read 170</div> 171 172 173 </header> 174 175 176 <div class="post e-content"> 177 <p>Imagine if we could implement an entire system of ethics in JavaScript.</p> 178 179<p>“Thatâs silly”, one might say, “JavaScript is a terrible language. We need 180something better.”</p> 181 182<p>Okay, fair enough. I don’t like JavaScript either. Take your pick. It can be as 183functionally pure as Haskell, or as hackishly old-school Perl as you like. 184Maybe Rust. That’s a neat language. It’s got a borrow checker and everything. 185And really nice compiler errors. Booted up Emacs? Gotten all your version 186control sorted? Great! Now, go, open up <code>src/ethics.rs</code> and implement those 187ethics.</p> 188 189<p>A more sophisticated interlocutor, at this point, steps in and objects to the 190entire project. Ethics is far too complex, subjective and embodied to implement 191in any programming language. I mean, we only implement a subset of our ethical 192principles in legislation, and to do so we need to be quite specific and we 193need human oversight.</p> 194 195<p>And they’re right. “Thou shalt not steal” becomes “thou shalt not <em>dishonestly 196appropriate property belonging to another with the intent to permanently 197deprive them of it</em>” combined with a whole set of institutional processes and 198safeguards (such as the benevolent police officer giving your 12-year-old a 199stern telling off after he nicks some sweeties from the local Co-Op rather than 200arresting him), to the public interest test we expect Crown Prosecutors to 201dutifully apply before initiating criminal proceedings. Good luck turning 202those into code. And, as already stated, of the ethical principles that we have 203condensed into rules, only a subset of them are of the sort we want to put into 204the statute books and have constables and magistrates and the like enforce on 205others.</p> 206 207<p>The smart contract people had an attempt at turning the <em>legal</em> subset into 208code, and it was as much of a failure as you’d expect. My <a href="https://tommorris.org/posts/2020/hart-contracts-not-smart-contracts/">standard 209objection</a> 210of “draft me an implementation of an <code>is_reasonable</code> function, because every 211question we actually bother to litigate about ultimately comes down to a 212dispute as to what being reasonable entails in a particular context” remains 213remarkably unanswered. Anyone silly enough to still try to push smart contracts 214in 2026 will try very hard to avoid facing it, which is unfortunate because the 215conclusion that follows is: “if you can’t solve that, and need to fallback to 216the existing socially constructed systems of laws and the various institutions 217that enforce them, well, is there any point in the smart contract anymore? 218Obviously not. You could just have a normal contract written in a natural 219language like English, and on the very rare offchance it goes wrong, you could 220pay some court fees in your preferred jurisdiction to have a reasonably neutral 221arbiter to debug it”.</p> 222 223<p>A concession, then. âOkay weâre going to struggle with this. Maybe you canât 224encode the full complexity of human ethics into computer code, but we are 225building autonomous systems! And they definitely need some ethical guidelines! 226After all, they could kill someone.â</p> 227 228<p>Such is the purported goal of projects such as <a href="https://www.moralmachine.net/">Moral 229Machine</a>
229, a website hosted by MIT inviting 230people to contemplate the ethical rules that might bind autonomous vehicles. To 231anyone with a philosophical education, it owes as much to Reddit trolley 232problem memes as it does to serious philosophy, but it’s a fun game.</p> 233 234<p>In said experiment, you are supposed to decide whether the ethical code of a 235driverless car ought to, <em>ceteris paribus</em>, run over a child or an adult. How 236about a child vs two adults? Or maybe a rich adult vs a poor adult? A mother vs 237a childless adult? A <em>Telegraph</em> reader vs a <em>Daily Mirror</em> reader? A disabled 238benefits scrounger vs an upstanding job creator? Five homeless criminals vs 239three doctors? What if the baby in the pram was actually going to grow up to be 240Hitler? What if it’s one giant duck or a hundred small horses? You know the 241drill.</p> 242 243<p>Why, yes, when you put it like that, trying to come up with a prioritised list 244of who the computer should prefer killing does give off some major eugenics 245vibes. But maybe we need to hold our woke snowflake noses over such 246objectionsâwe desperately <em>need</em> such principles codified in order to write the 247ethics submodule for the Waymos and Teslas, and we need to face these kinds of 248unpleasant tradeoffs head on in order to make the world a better place. You 249canât bake a cake without telling the computer which eggs are most morally 250deserving of being broken.</p> 251 252<p>But I have a much simpler objection: itâs completely pointless.</p> 253 254<p>We arenât going to get a democratic vote or a citizen’s assembly or a public 255consultation on whether we get killed by an out-of-control robotaxi. If a group 256of psychologists put together enough checkbox studies and empirically settle a 257few centuries of philosophical argumentation about ethics, and finally come up 258with a one, true objective set of ethical criteria to code into driverless 259vehicles, the manufacturers simply <strong>wonât care</strong>. Theyâll decide the emergency 260response behaviour of autonomous vehicles based on whether it maximises profit 261and minimises legal liabilities, both their own and that of major insurance 262companies. They arenât weighing up Kant and Hume. Do you think Spotify 263executives are sitting there carefully meditating on whether the miniscule 264compensation musicians get is fair based on Rawlsian principles? The only 265philosophers they care about are the ones determining far more prosaic matters 266like tort liability and insurance premiums. I don’t think Lloydâs of London 267spend a lot of time consulting any of the books on the âintro to ethicsâ 268reading list when setting premiums or compiling actuarial tables. It is those, 269along with the <em>Judicial College Guidelines in Personal Injury Claims</em> (and 270equivalents elsewhere), that are the binding authority here, not Scanlon or 271Parfit or some other learned moral philosopher.</p> 272 273<p>To believe there will be an outbreak of ethical consideration from tech company 274CEOs upon the completion of a worked out, implementable moral calculus is 275utterly delusional. The flight logs and email correspondence of the late 276Jeffrey Epstein doesn’t exactly demonstrate that the political or business 277leaders of the world are overburdened by moral conscience.</p> 278 279<p>Is anyone going to force them to do otherwise? The Americans? The European 280Union? <em>Andy Burnham?</em> Please. Let’s be serious. Governments don’t want to wade 281into some moral philosophy debate unless there are some culture war votes to be 282had in it. When Parliament first legislated in 2018 to allow autonomous 283vehicles on British roads, they did not require manufacturers to have installed 284a properly functioning morality chip, with the design of it audited by a panel 285of eminent ethicists. They merely said “if insurers are willing to take on the 286risk of civil liability, well, off you go”. That’s not in the same category of 287conscience-driven issue as abortion or assisted dying where party whips would 288grant MPs a free vote, but it is a fairly pragmatic answer to the rather more 289real questions of liability and insurance. You know, a thing that actually 290matters when vehicles smash into one another and/or hurt people.</p> 291 292<p>(The British government then realised it was a bit more complicated than that 293and had to pass another piece of legislation in 2024, but, again, the Secretary 294of State for Transport was not sitting with competing moral philosophers on 295each shoulder whispering in their ear. They’re far more concerned about whether 296building some fast charging stations on the M1 will create more jobs, or how on 297earth to get HS2 to actually happen sometime before the heat death of the 298universe.)</p> 299 300<p>If the conclusion of this kind of intellectual self-indulgence is never going 301to be meaningfully implemented (either voluntarily, through self-regulation, or
302through legislation/regulation), why bother arguing about it? I mean, like all 303of philosophy, itâs a moderately interesting thing to discuss, and closing 304metaphysical loopholes can scratch the same kind of itch as crosswords or 305sudoku puzzles. And if you are a teenager and have a life ahead of you filled 306with Zoom calls, back pain and having to change energy supplier every six 307months, a few years of contemplating the nature of consciousness, reality, 308truth, knowledge, beauty and the good life is a pretty fun way to get yourself 309into lots of student debt, and maybe learn some cool and interesting things 310that might bring some value to your life. (But good luck convincing your boss 311that you shouldnât implement dark patterns in your sales funnel because it 312offends against Kantian principles.)</p> 313 314<p>Pure intellectual interest can’t justify why we have researchers pursuing the 315application of trolleyology to autonomous vehiclesâreading academic philosophy 316journals is a rather niche and nerdy hobby, after all. No, the ultimate purpose 317of driverless car theorising is to convince people that there is <em>some</em> type of 318ethical consideration going into the development of this technology. It is 319rather reassuring to think that there are people diligently working away to 320make technology follow <em>some</em> kind of moral principles (even if we’re still a 321bit unclear as to the exact principles), and it also comforting to think that 322once this piece of ethical algebra has been solved, it’ll be handed off to 323beneficent technologists to dutifully translate into code to be installed into 324self-driving vehicles in the near future. Please focus on this and try to 325ignore the fact that the person ultimately overseeing all of this is more 326likely to be someone who is so high on ketamine that he does not give a single 327damn about whether your (or anybody else’s) gran gets killed on a zebra 328crossing. He is far more concerned with building Mecha-Hitler.</p> 329 330<p>One of the early papers from the MIT Moral Machine creators was titled 331“<a href="http://rdcu.be/vKrb">Psychological roadblocks to the adoption of self-driving 332vehicles</a>”. The thesis is roughly that consumers want a 333wholly self-protective set of rules which manufacturers won’t put in because 334it’d be bad for their PR to be seen as uncaring and sociopathic, but if they 335implemented a purely utilitarian ethic that’s as willing to sacrifice the 336passengers as innocent bystanders, consumers would probably seek to try a 337different type of car because it’s not very nice to die in a car accident. It’s 338also not very nice to be scapegoated for a crime you didn’t do in order to 339bring about societal calm (feel free to substitute in any of the other standard 340objections to crude utilitarianism, there are many to choose from). Vibes-wise, 341all of the options aren’t ones people are particularly keen on. (As does 342“they’ll do whatever is in the best interests of the manufacturer and the 343insurance company”, but we are duty bound to not mention that particular 344elephant.)</p> 345 346<p>The answer, then, is to try and make the general public <em>think differently 347about the ethics of self-driving cars</em>. They are not inviting you to help 348co-design a nice algorithmic compromise between Kant and Bentham on the 349important subject of “shall the car kill your gran?”, they are producing an 350interactive marketing exhibit selling the benefits of self-driving cars to a 351public who have conflicted and inconsistent moral views, because they’re humans 352and humans have conflicted and inconsistent moral views. It’s an industry 353marketing campaign dressed up as an exercise in applied moral philosophy. 354Please, let’s drop the pretense and just be honest about it.</p> 355 356<p>This kind of philosophical marketing exercise is the exact same thing we are 357now seeing with “AI safety” and “alignment research”. We have some people who 358sit in the corner of the various AI labs and get paid good money to produce 359some extremely earnest blog posts imagining all sorts of scary doomsday 360scenarios and pretending that if they just ask the chatbot to not do bad 361things, it might not. But the companies keep building the technology they think 362might create the doomsday scenario. This makes great copy for journalists. “We 363have people here who think they might accidentally create a murderbot, but 364they’re also deeply committed to building the murderbot because the Chinese 365government might also build the murderbot first and wouldn’t it be awful if, in 366the twelve minutes before every nuclear weapon wipes out all humans, we were 367forced to use an AI that would not generate any pictures of Tianneman Square, 368so we must build the murderbot first to stand up for freedom and the American 369way”
369 is a pretty wild story. And if that kind of patriotic American 370flag-waving mythology isn’t to your particular taste, other options will 371presumably be along soon.</p> 372 373<p>It’s bullshit. It’s marketing fluff. That’s all it is. Again, let’s not dignify 374this with any suggestion of being real. It is Scientology but for people who 375know how to write Python. You don’t need an article in a peer-reviewed academic 376journal to know that Lord Xenu isn’t real, or that Jesus doesn’t actually want 377you to don black Nikes and commit ritual suicide with $5.75 in your pocket to 378transport yourself on to a spaceship following the Hale-Bopp comet.</p> 379 380<p>None of this is serious thinking. It is a distracting form of secular 381apologetics for the business interests of some of the worst human beings on the 382planet. The people around them believe it because being paid a six figure sum 383by OpenAI or Anthropic to fret about bullshit “x-risks” is probably a <em>really</em> 384great gigâway better than churning out yet another Java CRUD app. But, it comes 385with a downside. You are the house ethicist of AI bros. So, if it is in the 386financial interest of Sam Altman to torture a thousand orphans, well, your job 387is to produce some plausible-sounding sophistry roughly along the lines of 388âyes, torturing a thousand orphans is incredibly unpleasant but we <em>really</em> 389need the self-improving magical genius computer to cure cancer and make us all 390live for eternity, and thus everything is justified by this massive upside in 391utilityâ backed up with some made-up numbers (sorry, <em>Bayesian priors</em>) and, 392congrats, that’
392s all the sign off needed for completely ethical baby 393torture.<sup class="footnote-ref" id="fnref:0"><a href="#fn:0">1</a></sup> Turns out if you put “infinite wealth and God-mode cheatcodes for 394all of reality” in to a cost-benefits calculation, it tends to skew the 395figures. Who could have guessed? When applied to philanthropy, this is the 396exact same kind of utilitarian self-dealing that justifies televangelist 397preachers buying private jets to minister to the poor and needy.</p> 398 399<p>A desperate reality check is needed here. You have technology companies that 400are run by some deeply unpleasant people, and they’re getting rich (at least on 401paper) by creating products that cause a whole succession of frankly <em>terrible</em> 402problems for society. That takes many forms: algorithmic discrimination and 403bias, the replacement of important public services by shitty chatbots, people 404getting into delusional relationships with chatbots (some of which escalate to 405self-harm and suicide) or completely frazzling their brains by delegating every 406single decision and piece of thinking in their life to an AI assistant. Most 407concerning for me is the epistemic annhiliation that’s been caused by a 408never-ending torrent of slop and bullshit overwhelming institutions and 409professions that have a vital social role of helping sort truth from falsity 410(including education, law, government and the media). I really miss the time 411when photos and videos could generally be assumed to be real, and text was 412usually written by a human being who knows something. These are problems 413happening <em>now</em>, causing real world harm <em>now</em>, to <em>actual living human 414beings</em>. Not sci-fi harms to made-up hypothetical humans we are saving 10,000 415years in the future according to some longtermist utilitarian calculus.</p> 416 417<p>These very real problems aren’t solved by chinstroking philosophical debates 418about the status of AI agents in one’s moral cosmology, or whether we can come 419up with the optimal vehicular homicide algorithm. The burning question is: “how 420the hell do we stop them from screwing up society?” This isn’t a question of 421deep armchair ethical theorising, it is a political and regulatory problem of 422the sort economists, lawyers, politicians and a whole variety of other policy 423people deal with every day. This isn’t to downplay the difficulty of 424formulating good policy, and certainly not a justification for <em>bad</em> 425policy.<sup class="footnote-ref" id="fnref:1"><a href="#fn:1">2</a></sup> But it is policy we need, not industry propaganda wearing the robes 426of moral theology. It is all a form of Potemkin intellectualism, giant 427distractions from far more urgent and practical matters. Tech theory is there 428to create a facade of some serious ethical and intellectual consideration going 429on to cover over the reality, namely that the industry is run by Delboy-like 430chancers desperately flogging mostly the latest shiny thing. It is valuable for 431them to project the impression that they are doing serious moral philosophy 432that might decide <em>The Future of Humanity</em>, rather than high-stakes gambling on 433the future value of data centres filled with GPUs badly summarising emails. 434People like Sam Altman may have the power to end the world, and the money to 435buy presidents and prime ministers. But us ordinary non-billionaire 436non-tech-founders have the power to point out that the entire philosophical 437structure of their weird AI doomsday cult was formulated by a bunch of 438skull-measuring racist dorks on a message board that’s devoted to a crap 439600,000 word Harry Potter fanfiction written by a <em>fully grown adult</em>, and this 440will never stop being funny. All of this is a deeply ridiculous and absurd 441circus and ought to be treated as such.</p> 442<div class="footnotes"> 443 444<hr /> 445 446<ol> 447<li id="fn:0"><p>Sam Bankman-Fried ran very similar arguments to justify the kind of fraud 448that landed him in federal prison. Despite apparently being very clever, 449his bad ideas hurt a fair number of rich people’s bank accounts, and he 450seemed remarkably unaware of how they <em>really</em> go after you for that kind 451of thing. Same for Elizabeth Holmes. She would have been fine if the only 452people Theranos had screwed over were Walgreens customers in Arizona who 453wanted to know whether or not they had diabetes, HIV or various types of 454cancer… but Theranos also screwed over investors like Rupert Murdoch and 455Betsy DeVos and, in America, that’s a <em>real</em> crime for which you do real 456time.</p> 457 <a class="footnote-return" href="#fnref:0"><sup>[return]</sup></a></li> 458 459<li id="fn:1"><p>Meta and TikTok and various other social media platforms are awful. But, 460also, the Online Safety Act is an abysmally bad piece of legislation that 461simultaneously fails to solve the problems with big social platforms, and 462creates a massive regulatory moat that makes it much <em>harder</em> for small, 463independent websites to compete with them. Facebook can afford a few 464lawyers to fill in a few forms every year to send to Ofcom. Random forums
465about cycling or basketweaving or whatever, not so much. Also, the age 466verification providers are poorly regulated. Literally no guarantee they 467won’t lose all the scans of everyone’s passport or whatever they submitted 468to send DMs on Bluesky.</p> 469 <a class="footnote-return" href="#fnref:1"><sup>[return]</sup></a></li> 470</ol> 471</div> 472 473 </div> 474 475 476<div class="navigation navigation-single"> 477 478 <a href="/posts/2026/the-invisible-dragon-discovered-magical-security-vulnerabilities-but-they-go-to-another-school/" class="navigation-prev"> 479 <i aria-hidden="true" class="fa fa-chevron-left"></i> 480 <span class="navigation-tittle">The invisible dragon discovered magical security vulnerabilities but they go to another school</span> 481 </a> 482 483 484</div> 485 486 487 488 489 490 491 492 493</article> 494 495 496 </div> 497 498 499
500<script src="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.7.2/js/all.min.js" integrity="sha512-b+nQTCdtTBIRIbraqNEwsjB6UvL3UEMkXnhzd8awtCYh0Kcsjl9uEgwVFVbhoj3uu1DO1ZMacNvLoyJJiNfcvg==" crossorigin="anonymous" referrerpolicy="no-referrer"></script>
500 501 502 503 504 505
505<script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.4.0/highlight.min.js"></script>
505 506 507 508
508<script type="text/javascript"> 509 510 hljs.initHighlightingOnLoad(); 511 </script>
511 512 513 514
515<script data-goatcounter="https://tommorris.goatcounter.com/count" 516 async src="//gc.zgo.at/count.js"></script>
516 517 518 519 520 521 522 </body> 523</html>
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.