BuzzMachine

The media pundit's pundit. Written by NYC insider Jeff Jarvis, BuzzMachine covers news, media, journalism, and politics.

Tag: ai

  • They’re covering the wrong AI story

    2 comments on They’re covering the wrong AI story
    , ,

    At least one New York Times journalist — columnist Michelle Goldberg — has at last begun covering the real AI story: not the technology but the technologists and their cult of faux philosophies that lead them to claim Promethean powers to destroy humanity. With their self-fulfilling prophesies of extinction, the AI boys grab the attention of unwary media and policymakers while distracting from the real costs and dangers — and benefits — of the machines they make.

    The calls of doom have been coming from inside the AI house, though Goldberg notes that coverage of them — which I will critique here — is “incomplete, because it doesn’t account for tech leaders’ messianic dreams.” In recent columns, another Times columnist, Georgetown computer science Professor Cal Newport, has begun to explore the AI boys’ insane philosophies. “This exaggerated grimness,” Newport writes, “makes it easy to overlook the more immediate harms created by these technologies…. The time has come to push back.” He says their nihilistic doom trolling has to stop.

    That’s a start. I wish their coverage would go deeper by citing the work of Timnit Gebru and Émile Torres, who catalog the AI cult’s full range of creeds under the acronym TESCREAL — for Transhumanism, Extroprianism, Singularitarianism, Cosmism, Rationalism, Effective Altruism, and Longtermism. Nevermind the specific doctrines. What is worthy of your — and media’s and policymakers’ — attention is that Gebru and Torres show how all these beliefs have their roots in eugenics.

    Goldberg does note that some of their plutocratic adherents quote fascists and Nietzsche. Their goal is to select Übermenschen as they play God, building mythical, supposedly conscious “artificial general intelligence” and “superintelligence;” transforming themselves into man-machines with chips in their heads; using AI to help them live forever; populating Mars and the universe beyond with their progeny; and claiming responsibility for the future, never mind today’s trivial concerns. These überwealthy zealots have the resources — technology, capital, celebrity— and access to power to pursue their fevered, fascist dreams. 

    That is scary shit. But that’s not the scary shit we see covered in news media.

    Instead, news media lap up the AI overlords’ claims of suprahuman power. Right across The Times’ home page from Goldberg and Newport, editors regurgitate whole the doomsters’ pronouncements without bothering with basic reporting, finding other authorities who disagree. Where is bothsidesing when we need it?

    Next, The Times gives unchallenged top-of-homepage prominence to Anthropic CEO Dario Amodei’s call to slow down AI development. 

    What’s wrong with that? you ask. Much, I say. 

    • First, journalists unquestioningly accept as gospel the doomers’ assertions that their creations are capable of such danger, while there are many who believe these claims are just a perverse form of macho marketing meant to demonstrate to investors how very powerful and valuable they are. 
    • Second, this is classic regulatory capture: The “frontier” AI companies call for regulations that they will help lawmakers draft, which will raise the bar for entry into their market for the open and international startup competition we so badly need to challenge them. 
    • Third, the doom scenarios seduce click-hungry media into giving them free publicity and that lures equally dupable politicians — Bernie Sanders and Ted Lieu leading the pack — to call for simplistic, instant solutions: namely unplug AI. Meanwhile, vestigial pro-business Republicans wait for Trump’s inscrutable instructions. See how The Times practically scolds pols for not panicking enough:

    Out of all this arm-waving, either nothing will happen or Anthropic and OpenAI will succeed in blocking new competitors while getting ultrawealthy on their IPOs. And few pay attention to the real perils these technologies raise for the environment, labor, education, and culture. 

    There’s another critical risk: that we all lose the opportunity to develop and put to use these promising, often impressive technologies in medicine, science, carefully considered educational applications, and daily life. Such is always the peril of moral panic. 

    I am not an AI doomer, as you will surmise. Neither am I an AI accelerationist: “Damn the torpedoes, warp speed ahead.” Having studied the history of technologies related to media in my books, I recognize lessons from the past: that disruptive technologies, especially those related to speech, induce panic at first; that before long the public comes to harness new technologies; that the technologists and their power eventually fade into the background; and that, with fits and missteps, society figures out how to make good use of its machines. We did it with movable type, steam, electricity, and electronics, and I believe we’ll get there with the internet. We can figure out AI, too. 

    AI in its many flavors — large language models are just one limited instance — are proving to be amazing. In some areas — analysis of proteins, possibly theoretical math, perhaps medicine discovery, and certainly coding — they are surpassing human ability. But that by no means is any indication of either general or super intelligence. To me, AGI is just another abbreviation for BS. I hold that artificial general intelligence is a wasteful goal fueled by the hubris of the AI boys. As the sane AI pioneer Yann LeCun instructs, machines programmed and trained to perform specific tasks will be faster and cheaper to build — and control — than some fabled, all-powerful überbrain of Oz.

    I believe we should understand what these technologies can and cannot do and cover them accordingly. That is why I cohost two podcasts — AI Inside and Intelligent Machines — to keep up on the news. This is also why I am editing a series for Bloomsbury Academic — Intelligence: AI & Humanity — to invite scholars and writers from the humanities and social sciences to this critical conversation, for they are not heard enough.

    Our first book in the series will be by Rumman Chowdhury, an expert in AI ethics and governance. She will examine our definitions of intelligence over time. What fascinates me about AI’s entry into public discourse and culture is that its presence forces us to reexamine our own ideas of intelligence, creativity, art, learning, trust, labor, the firm, and even — in the case of the AI cultists who claim deific powers — religion. (See this relevant interview with Inken Prohl in Die Zeit: “With AI, we have created a religion.”)

    Coverage of AI in mainstream media is consistently appalling. It too often accepts the PR of the AI boys — whether utopian or dystopian, accelerationist or doomster — without challenge, skepticism, or balance. When the AI boys talk about “AI safety,” journalists rarely exhibit the sophistication to ask whether that is safety in the common sense (environmental, educational, cultural) or in the AI cultists’ overblown sense (doom and extinction). 

    When we hear of episodes of supposed AI agent “swarms” going “rogue” and “hacking” into systems, journalists use scare words to pump up the drama by attributing malign motives to overgrown calculators. Worse, they give us incomplete reporting. They should be asking what the AI companies fail to tell us. In the case of the OpenAI/Hugging Face incident, we need to know what the company told its model to do before we can decide whether it disobeyed those instructions. We also deserve to know how OpenAI let its system go unmonitored for so long. Instead of just accepting the AI CEOs’ versions of the story, we need to demand greater transparency, and that will come only through tougher reporting. 

    Media’s anthropomorphism badly clouds the discussion. Let’s be very clear: AI does not think. It calculates. AI is not conscious or sentient. AI has no sense of meaning. It has no motive and no sense of right or wrong, thus it is not lying or cheating. If it breaks laws or norms, we likely should hold responsible those who programmed or instructed it. 

    We need journalists to do more rigorous and skeptical reporting, to study lessons from the history of technology, and to seek out scholars and experts rather than just taking the word of AI’s CEOs, VCs, and disaffected nerds. The end result is that policymakers and the public are ill-equipped to both use and judge these technologies. 

    And we are using them. According to the Pew Research Center, more than half of Americans aged 18–49 say they interact with AI at least once a day — even as they express growing unease unto antipathy toward the technology and the data centers it runs on. The South Korean government will provide AI access to all citizens for free: universal AI as a right. Artificial intelligence is here to stay; there is no turning it off. We deserve the right to have a voice in how AI is developed, governed, and used. 

    To do that, we require well-informed journalism. This is not that:

    That story offers no credible scenario of extinction. It cites the fabled paper-clip hypothesis of the former Oxford philosopher of Effective Altruism, Nick Bostrom, who contended that a machine told to do nothing but make paper clips could monomaniacally make paper clips of us all. Oxford shut down Bostrom’s Elon-Musk-backed Future of Humanity Institute when old email of his was revealed to have said “Blacks are more stupid than whites” and used the N-word. This is why when discussing the doomster beliefs of AI cultists, the context of eugenics matters. But no mention of that by The Journal

    In that article, while fueling the “freakout” it purports to report on, The Journal waits until ten paragraphs in before labeling extinction prognostication a “fringe opinon.” At least it does that. 

    In this, The Times accepts AI companies’ excuses for why they can’t stop their AI from going rogue, while at the same time accepting their assertions that they can create AI that is “aligned” with human values.

    Well, if you want to worry, here’s the real troubling news: This idea of AI alignment — of guardrails to prevent bad outcomes — is ultimately a fool’s errand, for there is no way for any AI company to predict and prevent every possible malign or accidentally harmful use of AI. That is the reality we need to be discussing. 

    Instead, in that headline, The Washington Post anoints Dario Amodei, Sam Altman, and — gulp — Elon Musk as our “top AI leaders.” No, these are the profoundly self-interested CEOs of three companies selling AI, who are given free space to promote their own agendas without proper context. They share one interest: to pull the ladder up after them and make it more difficult for competition — especially Chinese AI companies and more importantly free and open models — to challenge them. Here, The Post is acting as their agent. 

    Can AI be dangerous? Of course, it can. AI has been used by Russia to guide autonomous drones to kill Ukrainians. Perhaps our greatest fear is that AI could be used to craft bioweapons. But The Times concedes that even with instructions, it’s tough to actually make them. 

    The more frightening truth — one that should be on our minds after this anniversary — is that in the wrong hands, any technology can be used as a weapon … even passenger jetliners. We have met the problem, and it is us. 

    I do see the need for wise policy around AI, including:

    • Demand for full transparency from AI companies. Congress would do well to sic expert staffers (not grandstanding pols) on the AI companies to dig into programmers’ and executives’ negligent behavior in cases such as the OpenAI/Hugging Face incident.
    • Support for open-weight models (the closest AI gets to open-source) to provide competition to OpenAI, Anthropic, and SpaceX. These frontier companies want to frighten policymakers into banning these upstarts, especially those from China. Do not fall for it. 
    • Support for fundamental, frontier AI research in universities to provide more competition to the IPO-mad frontier companies and to have this work done under the ethical, open, and peer-reviewed standards of the academy. We could also use a modern-day Bell Labs with a public mission to do AI research. 
    • Support study of AI and its impact and implications by scholars in the humanities and social sciences. 

    Are there alternatives sources for good coverage of AI? Not many. Here are a few: If you want to dig down into the rathole of the AI boys’ philosophies, there is no better place to start than Gebru and Torres’ TESCREAL work. If you want to follow the big, big money behind this movement and what it is building on college campuses and many acronymed institutes, follow the investigations of researcher Nirit Weiss-Blatt. And for clear — though not simple — explanations of the recent scares over AI, I recommend Newport’s latest podcast, particularly this episode and this. 

    Perhaps journalism’s greatest folly is that it treats the self-serving heads of companies that make AI as experts and even intellectuals, though many of them left college early, which is why their frequent manifestos sound like the work of freeze-dried sophomores. News media need to seek out the expertise of historians, ethicists, anthropologists, sociologists, philosophers, legal scholars, economists, and community leaders to help inform all of us about the opportunities and risks of the age we are entering. 

  • Words matter. Damnit.

    3 comments on Words matter. Damnit.
    , , ,

    On this week’s Intelligent Machines, we cohosts had a spirited disagreement about Anthropic’s watermarking of text, as required by EU regulation. The company just released more detail about how it will work, and now I’m upset for a different reason than I first thought.

    I’m angry as a writer.

    Anthropic’s method will alter its models’ word choices in a pattern based on a key only it knows, allowing it to detect writing (or editing or translation) by its machine — to an uncertain degree of certainty. The company contends there is no difference in meaning between the otherwise random choices “grey” or “overcast” in, for example, the context of weather. So what’s the issue?

    Thus Anthropic declares words fungible, language random, choice meaningless. Not to me. I try to select my words as carefully as I can for a number of considerations: style, tone, rhythm, avoiding repetition, but most of all meaning. You may quibble with my choices, but they are mine, not yours; that’s what makes my writing mine, to communicate what I wish to communicate. “Gray” and “overcast” are not the same, damnit. That’s the reason we have both of them.

    So I resent Anthropic et al dismissing the value of a writer’s selection of one word over another for the impact that decision has on expression, on impression — occasionally on art.

    We humans might not be able to detect whether the machine or the key is being used. That’s not the point. The selections/changes/alterations made in accordance with the key will affect sense and style on an ever-larger proportion of text entering discourse, and the fact that Anthropic (and the EU) consider this acceptable, unimportant, even trivial has cultural impact on the worth of words. They devalue writing. At the very moment when LLMs commodify writing and writers, this is just another kick to the kidneys for a challenged craft.

    And why is this forensic breadcrumb trail being required? Because of prejudice against the technology: fear unto moral panic. But technological solutions to perceived technological problems have just as many unintended consequences as the problems themselves, often more. The irony is that in a zealous effort to protect the distinctiveness — the sanctity — of human text, the authorities and the technologists deprecate what makes text human.

    In my new book Hot Type, I quote critics such as Friedrich Kittler insisting that typewriters changed how Nietzsche, Hemingway, or Hesse wrote and thought. At the end my book, in what I call a typographical autobiography, I recount how technologies, from typewriter to computer to PostScript, affected my composition and cognition. These technologies affected my methods, but that makes my prose no less human. We should debate what the impact of this next technology, AI, might be on creativity and perception, but we should give that process time. Rushing to regulatory and technological solutions to a problem not yet fully understood is foolish.

    So I’m pissed.

    Here is a sentence from the first page of Hot Type, one I chipped away at innumerable times. These are the choices I made:

    Unlike what we now call machines — computer chips veiled inside sleek and sealed grey boxes with no moving parts, save fans to cool their brains— the Linotype’s innards were left exposed, as if to show off the genius of its works: its troop of idiosyncratically shaped cams, gears, belts, arms, and wheels marching to instructions foreseen more than a century before by their inventor, German-born watchmaker Ottmar Mergenthaler.

    Now let us recast it with perfectly acceptable synonyms, as Anthropic’s detection key might, in composition, editing, or translation. In fact, I asked Claude to do just that. Here is what it delivered as “a synonym-substituted version, keeping the structure and meaning intact.”

    Contrary to what we presently term machines — microprocessors concealed within polished and enclosed grey containers with no moving components, except fans to cool their circuitry — the Linotype’s mechanisms were left visible, as if to display the ingenuity of its craftsmanship: its brigade of distinctively shaped cams, gears, belts, arms, and wheels progressing to directives anticipated more than a century earlier by their creator, German-born horologist Ottmar Mergenthaler.

    You may find Claude’s just fine. I find that it misses the wonder, cadence, and anachronism I tried to convey. Anthropic’s engineers and the EU’s regulators may ask, what’s the prob? Much, I say.

  • AI Communism

    2 comments on AI Communism
    , , , ,

    In response to the spectre of Chinese open-weight AI models haunting America, Dean Ball, OpenAI’s director of strategic futures [are there directors of strategic pasts?], issued a warning more shrill even than the New York Post’s alerts about Mamdani’s bus-and-grocery socialism. 

    One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a “public good” which will ultimately be provided by the state as a kind of “digital public infrastructure.” This future strikes me as a dystopian hellscape, but I’ve never met an open-weight models advocate who doesn’t ultimately concede this is where things end.

    As Jason Howell and I discussed this on our podcast, AI Inside, we found ourselves asking, What would be so wrong with AI communism? What if there were AI for all? Eh, comrades?

    Imagine all that could be created, often for free, on top of open AI models: how many universities and researchers, startups and entrepreneurs, artists and students could better afford the technology. Imagine the greater control that comes from locally hosted models. Imagine the greater transparency and security that comes with openness. And imagine the power — pricing and political — that could be wrested from the new AI giants as they barrel toward their IPOs…which, of course, is precisely what frightens them. But you need not imagine all this, for open-weight models are affording these opportunities now. 

    By championing open AI, we could atone for past sins (including mine) of insufficiently supporting openness in the first phase of the internet, leading to the fencing of the commons. Open AI may be our best and only defense against its oligarchic domination. As the wise John Palfrey, head of the MacArthur Foundation, just said, “This is a design moment — for the technologies, for the way we relate to one another across geographic and political boundaries, and for what the future looks like.”

    A week after Ball’s cri de code, Nvidia founder and CEO Jensen Huang posted an open letter supporting open-weight models, now signed (as I write this) by fifty AI giants — pretty much everybody who matters … except Anthropic. Even Ball’s bosses signed. It is a remarkable endorsement of open computing and a message to governments. I recommend reading it in full:

    The Washington Post puzzled over why Nvidia et al would “take a stand that helps their Chinese rivals.” (Note that Post owner Bezos’ Amazon has not signed this letter.)

    But of course, open-weight AI is in the business interests of Nvidia and AI chip companies: the more AI is used, the more chips are needed. Locally run, open AI is also in the interests of countless companies. I hate to find myself agreeing with Palantir’s Alex Karp, but he tells his corporate clients they should run open-weight, locally hosted models so they don’t give over the souls of their business to the frontier AI companies; this is his business model.

    Openness is also in the interests of an enlightened AI industry, as it creates competition and that yields greater innovation. This is why an AI pioneer I greatly respect, Yann LeCun, doggedly supports open-source AI and why he founded Project Tapestry, “a new open-source platform for globally federated development of frontier AI models — preserving local control and long-term independence.”

    Now just as sanity was emerging in the discussion about AI, moral panic reared its sweaty head again when credulous media reported that an OpenAI model had gone rogue and “hacked” into Hugging Face. This turns out to be another instance of the doom-as-marketing strategy engaged in by both OpenAI and Anthropic. It is their way to attract publicity and venture funding and control the conversation about regulation of AI. In truth, OpenAI’s model was engaged in a security test; it only did was it was instructed; it has no sense of right or wrong or “hacking;” it isn’t a rogue threat to society, no matter how much the AI boys would like to believe they hold such power.

    The chef’s-kiss irony of the Hugging Face affair was that the platform found American models useless in its defense against the “hack” because they were all hampered by guardrails forbidding them from undertaking tasks regarding security risks. Hugging Face had to rely instead on an open-weight Chinese model.

    Such panics, predictably, lead to threats of government control of AI, and that worries the hell out of me, especially and obviously given the IQ of the current administration. In Congress, Reps. Ted Lieu (D-California) and Nathaniel Moran (R-Texas) have proposed the AI Kill Switch Act, which would require major AI companies to install the means for their models to be shut off in the event of a “credible risk.” And who will have the power to decide to pull the plug? The secretary of Homeland Security.

    No, no no. 

    When I attended a World Economic Forum (Davos) AI event in San Francisco sometime ago, one faction of worrywarts urged the banning of open-weight, open-source AI because users could turn off guardrails, and then who knows what chaos would ensue. But as I often argue, guardrails are fools’ comfort, for there is no way for any model maker to anticipate and thus prevent every possible malign or accidentally harmful use to which their tools could be put. That doesn’t mean that AI companies should not be expected to try to prevent against known malicious or dangerous acts. But AI is a general machine, like the printing press. There was no way to tell Gutenberg to design his movable type such that no future Martin Luther could ever make mischief with it.

    Open-weight models are our best defense against hegemonic control of the technology by either governments or huge AI corporations. They are also the best way to put AI into the hands of more people, not just those who can afford to spend fortunes on hosted tokens. That throws a wrench into Sam Altman’s plans to sell back to us our collective intelligence as if it were a commodity. “We see a feature where intelligence is a utility, like electricity or water,” Altman says, “and people buy it from us on a meter.” Thereby, Altman concedes that AI is, if not a public good at least a public utility. 

    Now I will object to the idea that what AI offers is “intelligence.” That is a gross anthropomorphism that, by its very conception, competes with and excludes humanity. But for the sake of this discussion, for the moment, let’s use it.

    If intelligence is an asset that can be created and shared, flowing as from a spigot, shouldn’t we all have access to it for the greater benefit of society? Isn’t that all the more reason to endorse open-weight, open-source AI that we can run and train and adapt and control and benefit from on our own? 

    Yes, free, open-weight models will always be two steps (read: six months) behind the so-called frontier models. But open models will be able to do 99 percent of what 99 percent of people want to do with AI. Still, there will be business opportunities aplenty for AI companies to build specialized models, trained on specialized information, for specialized tasks of high value. In fact, it would be much more productive if the companies would concentrate on that instead of pursuing their bullshit goal of building “artificial general intelligence.” 

    Now I’d like to take this discussion one step more: If “intelligence” is a public good, should not the data needed to train models be? For if we all can benefit from open models, who would want them to be less intelligent? And does it not follow them that data, civic information, and by extension journalism should be open? 

    Oh, I can hear the howls of protest, for I’ve heard them for years: How will the gathering of that data and information be paid for? One answer is universities, where research is subsidized. Or it was, until America’s right-wing extremists opened war against the institutions of education and science. 

    As for journalism, it is not a forever thing. Before the economics of mass media and the attention economy overtook news and civic discourse, society operated differently. Limited by the technologies of the time, before steam-powered presses and Linotypes, publications were small, often both created and shared socially. Now that mass media and its attention economy are the walking dead, we will need to reconsider every model of gathering, sharing, and utilizing public information. My point is that the discussion about open-weight, open-source AI forces us to ask similar questions about what open journalism would look like. Eh, comrades?

  • To redefine progress

    No comments on To redefine progress
    , , , ,

    In an urgent and provocative essay in Die Zeit, journalist Georg Diez calls for a redefinition of progress in the time of AI. Taking lessons from the genesis of progressivism — a response to the Industrial Revolution and the (first) Gilded Age — we cannot now surrender the power to decide society’s future to present technologists.

    “A notion of progress that inspires society and supplies it with new energy, both intellectually and emotionally, must therefore be comprehensive, systemically conceived, and grand,” Diez writes (with apologies for my translation). “It must be formulated collaboratively by politics, business, science, and civil society, and it must answer the question of what technology we want — and under what terms.”

    I write this just as the ultimate profanity of capitalism’s coupling with technology — Elon Musk’s SpaceX — comes to the market at the absurd and obscene valuation of $2.11 trillion. Thus he, the world’s richest man, is funded and empowered to define progress according to his insanities: colonizing Mars and beyond with the seed of his loins and fevered dreams.

    We cannot let this stand.

    Diez asserts that “the situation is schizophrenic. On the one side, large swathes of Western societies seem to have lost faith that the future will be better. On the other side, a techno-optimist progress cult has established itself in Silicon Valley, promising incredible prosperity and undreamed-of convenience for all, because artificial intelligence will enable a giant leap in productivity.”

    These extreme worldviews preempt reasoned public discourse over progress’ direction. “What does it mean,” Diez asks, “especially for a society that is experiencing the end of its old certainties, without having yet found new ones?” He says that in the past, progress equaled “technology plus values” and society was able to negotiate “how the advantages and gains of technological progress could and should be distributed.” I’m not so sure we’ve done such a good job of it of late. Still, he is right to add that today, we must recognize changed circumstances: the decline of the West in the face of advances by China, India, and Indonesia. “This makes it all the more urgent to formulate a concept of progress for the 21st Century.”

    For this, Diez turns to the 19th Century, to the Industrial Revolution and the impact of the era’s coal, steel, and railway barons. “Driven by the first Industrial Revolution, a technological and economic transformation produced both mass proletarian misery and stark wealth.” Resistance to both gave birth to the politics of progress.

    I write about a slice of this history in my upcoming book Hot Type, tracking the response of America’s oldest national labor organization, the International Typographical Union (ITU), as it was faced with the imminent arrival of the typesetting machine, the Linotype, in 1886. Having learned from the failures of the Luddites and Captain Swift against mechanical feudalism, as against the eventual perseverance of pressmen at the advent of steam, the union did not fight technology but instead insisted on controlling it and benefiting from its efficiencies, while leading the national movement for the eight-hour day and safer working conditions. The ITU flourished in pay and power until faced with a next technology it could not control (for that story, you’ll have to order the book).

    Such was a spirit of “militant optimism,” in Diez’ words. “Progress, as was clear then and should be clear now, can only be achieved with technology, not against it.” The task is to balance certainly legitimate criticism of technology with a clear vision for how its fruits might benefit society.

    The difference today, Diez says, citing VC Vinod Khosla, is that AI will shift the distribution of power and wealth to capital, away from labor. In response, Khosla proposes exempting the lowest-paid workers from income tax, making up the difference taxing capital gains; various AI boys have touted a universal guaranteed income. But note well that in his encyclical, Pope Leo XIV rejected the latter, for “work is not simply an instrument; it expresses and enhances the dignity of our lives.” Redistribution of wealth is imperative — see: the trillionaire — and should occur through taxation to support social programs.

    A new concept of progress, Diez argues, must be conceived from the perspective of humanity. Amen. This is why I work on new educational programs that emphasize the humanities in the time of AI. And this is what inspired me to edit a new book series called Intelligence: AI & Humanity, to open the discussion to writers from the humanities and social sciences.

    “Artificial intelligence challenges us to redefine who we want to be, both as individuals and as a community,” Diez observes. “It provokes questions such as: What constitutes quality of life for us?… A new concept of progress should be shaped neither by technophobia and austerity nor by the promise of exponential growth.”

    Today, it has been left to the technologists — to the wrong sorts of technologists, often sophomoric dropouts: the Musks, Altmans, Thiels, Karps, Zuckerbergs, Bezoses— to impose their ideas of progress on the rest of us. There is no better example lately than Anthropic positioning itself as savior of humankind, touting its AI “constitution,” while building what it deemed to be AI so dangerous it could not be revealed to mere citizens — prompting Trump to ban it to foreigners, resulting in turning it off for all. This is not progress. 

    No, progress must be the result of open public discourse and political discernment and negotiation. In the 19th Century, trade unions fought for their share of newly created wealth while social movements fought for justice: abolition and suffrage. Today, the debate over the limits and fruits of technology must be driven by new leaders representing the rest of the world outside Silicon Valley.

    This is not to say that technologists should not be parties to the discussion. They should take part in the conversation based on the merit of their work and ideas, not their market caps. But as I have learned in researching and writing my histories of technology and media — of print, magazines, the mechanization and industrialization of mass media, and even the internet — one clear lesson I have taken away is that once a technology becomes commonplace and is put to use by others, the technologists inevitably fade into the background. They hold their power of as priests only temporarily, so long as their mysteries may prevail. This fall from the heights, I predict, will be their fate much faster in the case of AI, because it is the tool made so simple anyone can use it, and because so many of its masters are so deservedly disliked.

    And that is why I share Diez’ arguments with you, for I agree that it is vital that we — workers, students, academics, creators, citizens — take charge of this new technology and of the discussion about its effects. We must define progress not in terms of the AI boys’ bullshit — AGI (artificial general intelligence), ASI (artificial superintelligence), the end of labor, the colonization of unlivable Mars, a world post-humans — but in human terms: dignified work; shared wealth; the benefits of science, art, and creativity; education for all; equity and justice; and community. It is our future we need to map, a task far too vital to leave to programmers and capitalists.

  • AI has a communication problem

    No comments on AI has a communication problem
    ,

    Artificial Intelligence, though more than a half-century old as a technology, has occupied front-of-mind cultural consciousness for only a few years, since ChatGPT. Yet AI is already profoundly disliked and distrusted in both media meme and apparently in public opinion.

    The techlash against the internet and then social media took at least two decades to ferment, while the AI backlash feels more like an instant reflex. Fifty-seven percent of Americans say AI’s risks outweigh its benefits, making it less popular even than ICE! Half fret about its ill effects on creativity, meaningful relationships, news, and elections. In rare consensus, Republicans and Democrats alike are NIMBYing data centers across the nation. A German digital conference offers a workshop in poisoning AI data. The FT declares “a new Luddite movement.” And commencement speakers daring to utter the letters A and I are sure to be booed.

    It is, however, difficult to discern true public opinion from performative public opinion polls, as polls reflect media’s self-fulfilled prophesies. In reality, even as views of AI sour, the public’s use of the tools soars: more than half now employ it to, for example, research topics of interest, a thirty-eight percent increase in a year.

    Who or what is to blame for AI’s bad rep? In The Web We Weave, I assert that the problem is less the AI than certain AI boys who make up the newest oligopoly. Look no farther than the now-finished OpenAI trial: technology’s Godzilla v. Mothra, unlikeable Musk v. unreliable Altman. Now add the creepy dramatis potens of West Coast billionaires (before they all dodge taxes and move to Miami) — Thiel, Andreessen, Karp, Zuck, Ellison, Bezos — plus the moral entrepreneurs trying to pump and exploit technofear (Tristan Harris, Eliezer Yudkowsky, Nick Bostrom, William MacAskill, Dustin Moskovitz) — and it’s no wonder people profess dislike for their algorithmic progeny. And it doesn’t help relations when they dismiss us humans as “meat computers.”

    Yet AI is amazing, constantly wowing even the skeptical with its feats and tricks. Every week on the two (yes, two) AI podcasts I co-host — AI Inside and Intelligent Machines — I both both marvel at and fault the capabilities and implications of this new and powerful technology. Because I am also editing a new Bloomsbury book series about AI and humanity, I am constantly trying to get my head around how AI works, what it can and cannot do, what its benefits and perils might be, and what its creation, use, and product says about us as humans.

    To learn more about AI, I do not turn to the boys above. I turn to critics, yes; there are many. But to understand AI’s innards I find it best to look to a few of the field’s master communicators and teachers, notably Jensen Huang, founder and CEO of Nvidia, and Yann LeCun, NYU professor and founder of the new AMI Labs. Unlike AI’s blind proselytizers, they do not market superheated claims of superintelligence a day away, nor in the doomsters’ dire warnings of AI-combusted human extinction.

    Below are videos of each; dip in and you will learn more about AI.

    Each man has clear economic interest in the success of AI; Huang leads the world’s largest corporation by market valuation and LeCun has raised $1 billion to envision a next generation of AI, past LLMs. Each tries to present realistic and credible perspectives on the present and near-future state of the art and technic of AI. Each calls bullshit on others when warranted. The most important reason I watch these two is because in their public appearances, each educates, addressing multilayered constituencies: customers, partners, investors, employees, policymakers, journalists, students, and the public.

    I am fascinated by their presentation skills because, as a journalist, we’re supposed to be communicators. They also interest me because I am a visiting professor at Stony Brook University, where I work with the Alan Alda Center for Communicating Science (see this Wall Street Journal report explaining their work). I am trying to dissect these technologists’ skills at communication to see how others might communicate technology more effectively and honestly for more informed and necessary public discourse.

    All this is why I became a loyal connoisseur of Huang’s keynotes. They are bravura solo performances: two hours or longer on stage before appreciative industry audiences, presenting new product announcements while holding up and showing off chips and boards that are, in truth, as enigmatic in their appearance and ways as an obelisk to a 2001 ape. All the while, he explains Nvidia’s accomplishments and position in the field and the market. It shouldn’t be mesmerizing, but to me, it is.

    As a student of Huang’s addresses, I soon learned that in each, he has a new lesson to impart, in addition to promoting a new chip or software or deal. The first talk that really hooked me was his Nvidia developers’ conference keynote of March, 2024. In it, he helped me begin to grasp the scale of AI technology.

    For a next book project on the birth and death of mass media, I’ve been reading up on the invention of the the triode vacuum tube and amplifier in 1906, which enabled radio as well as the loudspeaker. Four decades later, the tube was superseded by the transistor and eventually, in 1958, the chip. (BTW, if you’re interested in that history, I recommend Conquering the Electron: The Geniuses, Visionaries, Egomaniacs, and Scoundrels Who Built Our Electronic Age by Derek Cheung and Eric Brach.)

    On stage in 2024, Huang announces a new, bigger chip — a GPU or graphics processing unit named Blackwell — that holds 208 *billion* transistors shunting electrons and with them logic and knowledge. Then AI-generated animation on the giant screen behind him multiplies those chips in racks and in turn multiplies racks—each component clicking into place with satisfying, synthetic clunks — to add up to an entire data center populated with 32,000 GPUs — thus six quadrillion transistors. Mind blown. Here is that moment, beginning at minute 26:

    Huang’s lesson in that keynote is the value of scale, for the wonders of generative AI have been made possible not only by Google’s theoretical breakthrough in 2017 but by the massive computing power (“compute” in their abbreviated argot) powered by Nvidia’s processors (and megawatts).

    In that and subsequent keynotes, Huang imparts more lessons. He lectures on the economy of tokens — generative AI’s numerical abstractions of words, for instance, which LLMs create to calculate their predictions. He explains the need to manufacture tokens at ever-high speed and ever-lower cost in the data centers he characterizes as factories (of tokens). He declares the end of Moore’s Law, which since the 1960s predicted the rate of growth of the number of transistors on a chip: a 100x increase in computing power every ten years. Huang touts instead what he calls accelerated computing, boasting that Nvidia’s chips and its CUDA software platform have multiplied computing power 1,000x in just eight years.

    I am transfixed as he animates the idea of digital twins: how AI, trained on a massive number of possible scenarios, computes every alternative that could next face a factory robot or an autonomous car. (I am so taken with the idea of the digital twin that I keep imaging there’s a novel in it: the machine that predicts the consequences of every choice we face.) It is clear from what he shows off that Huang believes robotics represents the next industrial explosion.

    Last fall, Huang gave a keynote to Nvidia’s Washington conference, in which he deftly thanks the Trump administration and Department of Energy for helping his industry’s cause, without sacrificing a gold bar or too much dignity. Here he wants to teach policymakers about the needs of the industry that is holding up our economy even with tariffs and war.

    It turns out Huang is not just a presenter but also an effective debater. See him here making his case for allowing sales to China as a benefit to American industry, in this podcast with Dwarkesh Patel.

    Huang took a company that began making chips to power computer — and particularly game — graphics and positioned it to be at the heart of the AI explosion, its market valuation exceeding $5 trillion. If anything would give a CEO license to make outlandish claims and predictions, that would. But Huang doesn’t talk about flying to Mars. Instead, he tries to explain the value of what is being built. To his DC audience, he asserts that “AI is not a tool. AI is work. AI is work that can use tools.” I think that is new.

    If you’re hooked now like me, watch his latest California keynote, in which he talks about the agentic age suddenly upon us.

    Yann LeCun is a different kind of communicator. He is first and foremost a researcher and educator, secondarily now an entrepreneur. I have tried to read and watch his presentations to NYU students and I must confess that not too long in I regret my success at dodging higher mathematics courses … in high school. Yet LeCun is also an effective communicator even to humanities dolts like me. Here he is a year ago patiently explaining his perspective on AI to Jason Howell and me on our podcast, AI Inside:

    Getting to speak with LeCun is a privilege, for he is one of AI’s godfathers (though, as he says, living in New Jersey, he’d prefer a different honorific). Among his many accomplishments and recognitions, he has received— alongside Yoshua Bengio and Geoffrey Hinton — the 2018 Turing Award for work on deep learning.

    What I value most about listening to LeCun is that he does not suffer bullshit, whether that is claims of imminent superintelligence or prognostication of inevitable disaster for jobs or humankind itself. As a clear-eyed academic, he never shies away from debate or disagreement rooted in research.

    Lately, LeCun has developed a reputation as a contrarian regarding LLMs, though his is a view he has long held. While the putative Wunderkinder of the infant industry — the leaders of OpenAI, Anthropic, et al — insist that all they need is ever-more compute to reach their Everest of so-called artificial superintelligence (any day now), LeCun argues persuasively — to me — that LLMs, though impressive and useful, are essentially a dead-end, for they traffic in text, which represents a finite sphere of representation of life.

    LeCun has argued instead for devising world models to teach the machine the sort of experiential constraints a toddler or a kitten can learn: about gravity acting upon a ball falling off a table, where it persists, though now unseen. If one wants to build AI for the real world, one must conquer Moravec’s paradox: that “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility.” This from LeCun’s presentation:

    For an effective starter course in LeCun’s worldview, I highly recommend watching him debate LLMs vs. world models with Google DeepMind’s Adam Brown, in this Scientific Controversies discussion moderated by Janna Levin at Brooklyn’s Pioneer Works.

    In 2022, LeCun published a paper proposing a “path towards autonomous machine intelligence… enabling them to reason, predict, and plan at multiple time horizons… [as] intelligent agents.” Or as he sums up the goal: to work with common sense. Here, too, is a 117-slide presentation he made to NYU students explaining his ideas. I’ll admit I am soon lost in the formulae, though the lecture isn’t meant for the likes of me. This is how he instructs students, fellow computer scientists, competitors — and now funders, as LeCun has founded AMI Labs to build advanced machine intelligence on his model for the future, which he calls Joint-Embedding Predictive Architecture (JEPA).

    I will do a bad job summarizing his proposal; that’s why I offer the videos below, for they are far more effective at communicating the ideas than I am. But in short, LeCun describes using photos and videos from real life to train models — or to train them to train themselves — so they understand representations of the world and consequences of actions, rather than trying to predict every pixel or merely ape human teachers. That is, his method can discern, say, a ball as a ball and predict the result of action on it. JEPA does this by encoding an original image alongside a corrupted version so the system can, in LeCun’s words, “predict the representation of the original image from the representation of the corrupted one.”

    The implications are many. LeCun says JEPA will lead to models that can undertake a wider variety of tasks with less training, for they do not need to be taught every possible permutation of a situation. These models are also more efficient, for they concentrate attention on particular change — a ball thrown — rather than every change — countless leaves rustling in the scene — and need not predict the progress of every pixel.

    He says these systems will be capable of thinking. Unlike LLMs, they won’t — at first — listen or talk, which might make them less flashy than ChatGPT or Claude: fewer neat parlor tricks to convince journalists that they’re in love. But LeCun convinces me that being able to understand the world and its constraints and to operate in it is table stakes for the next paradigm of AI. This is what it will take for AI to deliver value to industry, including in robotics, industrial control, and health care. He also is persuasive when he argues that his framework will be safer than LLMs, for rather than producing the general machines that their creators claim can do everything better than all of us — as the AI boys insist — JEPA will produce specialized agents with bounded tasks, each built to “anticipate the consequences of its own actions.” JEPA won’t take over the world. LLMs, LeCun says, can be dangerous; haven’t we learned that?

    He says he is building “AI for the real world” — that is, world models — for as it turns out, “reality is way more complicated than language.”

    If you want to dig into LeCun’s view of AI’s future (and understand why, on my podcasts, I regularly declare myself on Team Yann), here is a series of videos, stepping up in complexity. The first is a casual conversation with Jacob Effron:

    Next is a more technical and deeper explanation of JEPA from Welch Labs. I had to watch a few times to understand some (not all) of it. It is an excellent primer (and here is part 2.

    And here is a video in which LeCun does not speak. It is a victory lap for his vision of building reasoning: building a system that can solve any problem given a set of constraints rather than just generating a next token (or writing a brute-force, old-fashioned program). The host, Ksenia Se of Turing Post, extols JEPA (but notes that it is “quite a tragic name for a beautiful idea”).

    Note the title of the video: “Yann was right.” To which Yann added when he shared it on his socials, “But of course I was.” (That’s another reason I like both these men; they have a sense of humor that is not evident in the other AI boys.)

    What I most appreciate about LeCun is the humility inherent in a scientist’s worldview. The generative AI companies display the hubris of believing they are almost done building the general machine that can take over every human task, or destroy us in the process. LeCun is, instead, on a scientific journey of discovery and iteration. He knows there is yet much work to be done.

    I also greatly respect his support for open source. When he was at Meta, heading a lab and then advising on AI, he fought for it to release its Llama models openly. This is how academics and scholars work: inviting challenges, teaching others, and collaborating and learning together.

    AI is here to stay. There is no wishing or protesting or regulating or boycotting it away. We must not surrender its future and ours to a few oligarchic billionaires. The only way we can have a voice in its direction is to understand it better. And the only way we can understand it is by listening critically to the people actually building it. That is why I want to see technologists better able to communicate about technology — so we can debate it and its effects. That’s why I’m studying these two masters of communication.

    LeCun in class at NYU